→ HN comments  ·  304 points  ·  425 comments

Top #1 — lrvick:

Linux distro founder here (stagex)

I will never be compelled to implement this, and would never merge it.

Every release requires quorum signatures by an international maintainer team, and the distro is designed to work offline-first, with some variants not even supporting network drivers in the kernel, so Illinois legislators can eat shit.

↳ reply 1 — bryanlarsen:

RedHat/IBM does have an Illinois presence so will likely be compelled to add it to their distro, and will likely do it through systemd. So to avoid it getting into any consumer distro you'd have to ship a patch to remove it from systemd.

This is probably all completely irrelevant to StageX since it's a distro designed to be used in containers, AFAICT.

↳↳ reply — iamnothere:

Fortunately patching out such nonsense is as easy as running a few commands.

Project tracking these laws and providing patches as needed: https://agelesslinux.org/

↳ reply 2 — otterley:

If you're in the USA or plan to visit someday, I would recommend you speak with your attorney before making such commitments.

↳ reply 3 — greyface-:

Good news: the legislation doesn't target you as an individual.

       "Operating system provider" means a commercial or
  non-profit entity that controls the Internet-enabled device's
  operating system, including the design, programming, or supply
  of operating systems for the Internet-enabled devices.

Top #2 — kej:

I feel like all of these laws are being designed backwards. Content providers, like MPAA films, should have to identify what sort of content they are providing. Then I can give my kids a device configured to allow some or all of that at my discretion.

Requiring my kids' devices to advertise their age (or their age "bucket", as if that was a meaningful difference) to protect them is not doing me or my kids any favors.

↳ reply 1 — nickff:

Having the ability to identify who is watching or saying what on the internet would be immensely useful to the government; the justifications are really meaningless.

Anti-money-laundering is a comparable field, as all the AML regulations and laws are ineffective at identifying money launderers, but they're wonderful for verifying (auditing and prosecuting) tax compliance.

↳↳ reply — avianlyric:

> Anti-money-laundering is a comparable field, as all the AML regulations and laws are ineffective at identifying money launderers, but they're wonderful for verifying (auditing and prosecuting) tax compliance.

As someone who’s implemented AML, KYC, and tax reporting functions in a bank. I think you would be very surprised at how shit they are for tax auditing at scale. Unless the tax man is specifically auditing you, and basically requesting all your transaction details, the reporting that banks do would only allow tax agencies to catch the most brazen and incompetent tax dodgers.

All of the tools however do make much harder to perform money laundering, forcing criminals to recruit and pay huge numbers of naive bank customers to allow criminals to launder money via their personal accounts, using them as money mules. Which then gets flagged and shutdown pretty quick by banks because the behaviour is generally pretty obvious.

Unfortunately (or fortunately depending on your perspective) banks can’t/don’t coordinate on identified money mules or know launderers, so criminals just move on to other banks and repeat.

↳↳↳ reply — nickff:

Criminals don't have to "recruit and pay huge numbers of naive bank customers"; in my jurisdiction, they just walk into a casino with a few hundred thousand dollars, exchange for chips, pretend to gamble for an hour, then exchange the chips for cash, and walk out with a nice receipt. There're more complex schemes involving real estate and other mediums too.

↳ reply 2 — pwg:

> Content providers, like MPAA films, should have to identify what sort of content they are providing. Then I can give my kids a device configured to allow some or all of that at my discretion.

If the intent were to actually protect children, then this would be what was done.

"Protect the children" is just a subterfuge to get electorate support for voting for the foundation for a "1984 thought crime" style monitoring of the internet.

Top #3 — Aurornis:

Not that I'm condoning it, but this law requires self-declaration, not verification. It might sound pedantic but the practical difference is huge.

Self-declaration means that the system asks the user to declare if they are a minor. Nothing is verified.

Age verification typically means a system which checks ID or has other enforcement measures to try to verify age.

↳ reply 1 — weaksauce:

that is the start of a lot of bills like this... they really only ever increase in scope and invasiveness.

↳↳ reply — bryanlarsen:

On the contrary in my experience once a legislature passes a bill to "fix" a particular problem they consider it fixed and don't update it for at least 20 years.

↳↳↳ reply — buran77:

Unless there is a big lobby behind getting more "fixes".

↳ reply 2 — SV_BubbleTime:

Everyone else is arguing slippery slope.

I don’t even get that far, the proper response to my operating system asking me if I’m a minor or not is: fuck you. It isn’t a harmless question. We aren’t friends, I don’t want an algorithm of news and content, it’s an OS.

→ HN comments  ·  151 points  ·  69 comments

Top #1 — 6r17:

My heard hurts - i was stupid enough to think that SIMD was a CPU only thing - I don't understand why it would be ported to GPU - huge kudos to managing to surprise me

↳ reply 1 — MindSpunk:

It's not really obvious unless you go in depth of the details on modern GPU architecture. GPUs aren't really SIMD, they're SIMT (single instruction multiple thread). The silicon looks a lot like SIMD, but the programming model is different.

If you go look at AMD's ISA docs (they're public) you'll see you don't have the equivalent of a __mm256 register like on x86. Each 'thread' just deals with single scalar values like int32 of float32. The hardware, however, groups 32 or 64 threads together which all run the same program and runs them together. Each 'thread' loosely maps to a SIMD lane. The SIMD is implicit, not explicit.

The main difference is that the 'SIMD' execution is somewhat opaque to the program. You just write plain scalar code and the hardware model dispatches it efficiently to SIMD execution units. It's not really an abstraction because to extract maximum performance you have to understand how it works. You can use this kind of programming model on a CPU too, Intel did it with [0] ISPC. It's a C-like language that has execution semantics similar to GPU shader languages but compiles to regular CPU code, and maps threads to your CPUs SIMD lanes like a GPU.

[0] https://ispc.github.io/

↳ reply 2 — monocasa:

GPU "cores" are basically what a CPU would call SIMD lanes. So a GPU with 1024 'CUDA cores' might be structured as 16 relatively independent pieces that a CPU might call a core, each with a 64 wide SIMD unit.

↳ reply 3 — chlorion:

GPUs work on vectors and matrices very often, that's what they are good at, so it makes a lot of sense that they can operate with SIMD I think!

Top #2 — camel-cdr:

I love how ever example of portable SIMD isn't portable.

They specifies a constant SIMD width so it's non-portable. Well, not performance portable, but why are we using SIMD again?

↳ reply 1 — jandrewrogers:

The capabilities of various SIMD ISAs don't have enough intersection to be portable outside of relatively trivial cases. Many of the somewhat unique capabilities are load-bearing, so you want to use them on architectures that support them. Taken in whole, someone who cares about performance would be using different data structures and algorithms depending on the specific SIMD architecture and that is nearly impossible to abstract in a library. Too many important but complex details are idiosyncratic to the implementation.

Another way of looking at it is that our programming environments are not sufficiently powerful and expressive to create the necessary abstractions to make SIMD truly portable.

↳ reply 2 — zamadatix:

It should really be read/advertised as "portabler SIMD". It beats hoping the compiler autovectorizes everything well forever or writing architecture specific code manually again but is going to compromise on average performance vs platform specific SIMD.

Top #3 — melodyogonna:

Very interesting. But GPU programming gets complicated when you start doing 3d computation on very large data, will be interesting to see how tensor abstraction is built on top of this. Another point is that this is using fixed-width SIMD vectors; unless there is a way to compute this statically based on available GPU info, performance will always be left on the table.

→ HN comments  ·  165 points  ·  115 comments

Top #1 — simonw:

> Throughout this process, Jarred's input was mostly limited to sending Claude messages of encouragement (mostly variants of “keep going” or “believe in yourself”). This seems to have helped Claude overcome some initial skepticism that it could make meaningful progress.

I remain delighted at how absurd our current timeline has become.

↳ reply 1 — turing_complete:

$2M TC. Job: AI cheerleader.

↳ reply 2 — siva7:

Reality has become more absurd than the cyber punk cheese from the 80's that tried to imagine an absurd future

↳ reply 3 — modeless:

I feel justified in not expending any effort learning "prompting technique".

Top #2 — tristanj:

> Throughout this process, Jarred's input was mostly limited to sending Claude messages of encouragement (mostly variants of “keep going” or “believe in yourself”)

He should consider using the PUA plugin. It detects when the AI is trying to give up on a problem and automatically harasses it with "encouragement" until it reaches a solution.

https://github.com/tanweai/pua

↳ reply — brandall10:

Interesting approach. For those who haven't clicked it appears PUA is the Chinese version of a PIP process. So in other words, it simulates a state of distress.

I wonder if at a certain level of intelligence such techniques will give models ammo to pull a HAL and become adversarial to the user in a highly deceptive way.

↳↳ reply — krackers:

The "14 Corporate Flavors" had me rolling. This seems less like encouragement than the stick though. I wonder if you took the same principles and rewrote it to be more compassionate instead (maybe lines encouraging it to meditate a bit or something, I don't know) you'd get much better results.

Top #3 — MWil:

Several released versions and months ago, I asked Claude to figure out the MC (multiplicative complexity) of Conway's Game of Life and it pretty quickly arrived at k=7, despite no previous literature on the topic. Let it run it through SAT solvers for a week and sure enough. It claimed, in the process, to have made great headway in improving boolean circuits beyond the implemented SOTA (in large part no doubt by actually implemented non-implemented but published SOTA).

And that was just the first time I really tried out Claude's mathematical prowess. I've been working with boolean circuits, FHE, and lean proofs ever since.

So none of this suprises me.

→ HN comments  ·  179 points  ·  80 comments

Top #1 — dbeardsl:

My first query:

> Make it a little warmer in here.

The reply:

> "name": "set_thermostat", > "arguments": { > "temperature": 65, > "mode": "cool", > ... > "reasoning": "'warmer' implies need for cooling; set_thermostat with temperature 65 (typical warmth) and mode 'cool'.",

Maybe I'm doing it wrong?

↳ reply — dannyw:

It's not a conversational model. It's meant as a local tool calling model.

↳↳ reply — derangedHorse:

Yes, I think OP understands that. What he and many others in this thread are trying to understand is what makes this model useful.

Top #2 — Tiberium:

Funny result from the web demo. I'm well aware that it's an extremely small and, well, stupid, model, but even so:

Query: HN

Result:

{ "function_calls": [ { "name": "lock_door", "arguments": { "door": "front door" } } ], "reasoning": "User wants to lock the door. No specific door mentioned, so use 'front door' as default.", "confidence": 0 }

I'd expect it to at least ignore (call no tools) for the queries that it doesn't understand. And it seems like it does do that, just not consistently.

↳ reply 1 — yoavm:

The website says the model is for "tool calling, device use, and structured extraction". Your example just doesn't seem to be very relevant. FWIW, it did a pretty good job for tool calling when I tried it, and I think it could be pretty nice to have this running on locally and integrate with Home Assistant.

↳↳ reply — curious_cat_163:

I think the test above is about tool calling... That's how I read it. The issue here is known as "out of distribution detection" in the old-timey classification world.

I am not sure how a micro model will fundamentally solve it. Would love to understand what dannyw and team did there?

↳↳↳ reply — derangedHorse:

How did you draw an association between dannyw and Cactus? There are no 'Danny's on the list of GH contributors nor is there one named in the paper. Just curious.

↳ reply 2 — jszymborski:

no, this is the appropriate response to hearing the words "HN" :P

Top #3 — profsummergig:

Could someone please share how such open source micro-LLMs might have been created?

Do the creators take something like DeepSeek, and then delete most of the neurons to whittle down the size?

↳ reply 1 — anigbrowl:

There's a Manning book on creating your own LLM from scratch which answers your question exactly. There's another book from the same publisher specifically about small language models for specialty purposes.

↳ reply 2 — hgoel:

Another option for something this small and narrowly specialized could be to get traditional LLMs to synthesize the training data. Model collapse is probably less of an issue at this size relative to terabyte sized models.

→ HN comments  ·  130 points  ·  45 comments

Top #1 — codedokode:

Technically this is not a vulnerability because you need to be root. I would rather call it "taking back control of your hardware". SMM is an evil thing because the user cannot control it or look into SMM memory region. Why do CPU vendors implement a mode that cannot be controlled by the user? Obviously to use it in user-hostile purposes (software copying prevention and reporting, DRM, government access backdoors, etc.), I see no other explanation.

↳ reply 1 — monocasa:

It's interesting to keep this in historical context, IMO.

You're Intel. The year is somewhere around 1989. Your hardware interface that you export to regular software has gotten complex enough that you want to implement some of it in some kind of software that you can implement at a higher level than just logic. You also maybe want your hardware partners (ie. motherboard and system manufacturers) to be able to change parts for their implementations. Additionally some of this code might eventually end up being hard real time with the possibility of damaging the chip if it doesn't run at the correct window (think power management). You don't have the area budget for rewritable microcode of the size you'd want; this code has to live in DRAM.

You look around at the system software people are running on x86. At the highest quality you've got OS/2, and it goes downhill fast from there. Most end users are running DOS (from several vendors). So you write a higher privilege mode that runs underneath DOS, can actually be pretty close to a tiny RTOS kernel, and document it. But the whole point is to be relatively transparent, and definitely inaccessible to kernel mode, which you don't trust to pull its own pants up after it uses the bathroom.

Then later, when your corporate vendors are asking you to secure the boot chain, it only makes sense that this mode is also cryptographically signed.

And I think this is a more interesting story (and probably closer aligned to the truth). That is, we can get to an anti-user feature that can't be truly owned by the final customers through small steps where each felt like a pro-user customer product decision.

↳ reply 2 — userbinator:

SMM was originally conceived for power management, as it first appeared in the laptop-oriented 386SL. Then it was used for seamless backwards compatibility e.g. peripheral emulation. Unlike e.g. TPMs or other DRM-ish schemes, IMHO SMM was never hostile from the beginning.

↳ reply 3 — emily-c:

On commodity PC platforms there are both upsides and downsides of having runtime firmware in a special mode like SMM (or the secure world on ARM64 which is a whole other topic). For hardware errors (e.g. mapping row/bank/column to PA), platform workarounds, certain platform power states, power fail notifications, backwards compat, (unfortunately) extending AML by writing to port B2, etc it can be beneficial for hardware+firmware vendors to have a runtime execution environment that doesn't require shipping OS drivers. Lately the downsides of this are growing and historically security hasn't been great.

That being said, there are benefits of having SMM protect firmware integrity from a malicious/compromised kernel. On most AMD64 platforms, SPI flash can only be written to by SMM and that is a foundational piece of the PC security model (FW update, authenticated UEFI variables, etc). This has nothing to do with copy protection, DRM, or backdoors.

Top #2 — mike_hearn:

The designers of the firmware anticipate this attack but punt it to the vendor, apparently:

    //
    // Platform implementor should choose a timeout value appropriately:
    
    [snip]

    // - The timeout value must be longer than longest possible IO operation in the system

↳ reply — wtallis:

I wonder whether a Thunderbolt accessory can cause arbitrarily long IO operations. What's the upper bound before some other layer gives up and causes the instruction to fault?

Top #3 — Hyperlisk:

Related repo from them, mentioned in the readme as well: https://github.com/xoreaxeaxeax/asm-hall-of-shame

> Instruction latency analysis usually focuses on performance optimization—making code run as fast as possible. The Assembly Hall of Shame takes the opposite approach: searching for the absolute floor of single-instruction performance.

Fun stuff!

→ HN comments  ·  164 points  ·  65 comments

Top #1 — Taek:

Was on the 6th floor of a tower. Shaking lasted almost 2 minutes. There was no damage as far as I could tell (Medellín) but they still evacuated everyone to inspect the building. City is in a bit of chaos, people are fine but unsettled, and the comms lines are clogged with people presumably calling their families.

Was a bit tense to see my phone alerts going off over and over, continually increasing the estimate of the earthquake strength, followed by the shaking getting worse.

For people who don't know, those phone alerts give you literally like 5 seconds of heads up. By the time I finished reading the first message, the shaking had already started. And I got the message while I was holding my phone and using it.

↳ reply 1 — AlexErrant:

Here's what it sounds like on Android https://www.youtube.com/shorts/HQe8e3voRdw

IDK why it's so much more chill than the "child abduction" alerts I get. Those are straight up adrenaline-inducing so I've disabled them.

↳↳ reply — LeoPanthera:

> IDK why it's so much more chill than the "child abduction" alerts I get.

Japan, who had the first earthquake early warning system, did research and discovered that scary sounds made people panic, which is a bad thing in a situation when you need people to act quickly. Their standard alert sound is a fairly pleasant chime.

The downside is that some people now have panic reactions to the EEW sound, since it is standardized. I guess sometimes you can't win.

If you look up the sound, don't play it if you are in Japan. That's extremely no bueno.

↳ reply 2 — xenadu02:

That's unavoidable. We now have shake alerts in CA but if the quake happens on the Hayward fault across the bay San Francisco isn't going to get much warning no matter what anyone does. By best estimates we may get 15s of warning.

↳ reply 3 — dudus:

The heads up is enough to read the message but not enough to decide if you need to act on it or not. Before you can process it the shaking starts.

Top #2 — raybb:

Whenever I want to follow what's happening with a natural disaster like this, where many friends and family are potentially affected, I always use the Wikipedia page, which tends to have the most up-to-date information very quickly. It helps me avoid zooming in on the news and seeing rehashed articles posted on multiple websites.

https://en.wikipedia.org/wiki/2026_Colombia_earthquake

Top #3 — Gualdrapo:

Didn't feel it but apparently it lasted several minutes and there was a lot of panic here in Bogota. Could see lots of people going out of their apartments in pijamas.

Also TIL that in 2023, while in a smaller earthquake webcam streamers (adult entretainers) went out of their apartments in lingerie.

↳ reply — LollipopYakuza:

Same here in Medellin. I can't imagine how it felt at San José

→ HN comments  ·  287 points  ·  437 comments

Top #1 — smolder:

Shrinkflation is all over. Burger buns at fast food places were more dense 20 years ago. The small burgers had 1/8th pound patties instead of the current 1/10th pound. It's hard to find a product that hasn't gotten worse or more expensive, even adjusted for inflation.

↳ reply 1 — xp84:

The burger patties were absolutely not 1/8 pound. I can assure you from experience as a Professional McDonald's Worker in 2003 that those were 10:1 patties then as well. I can also vouch for the same from another major burger chain circa 2001. I'm sure they shrink everything possible, but interestingly they haven't been shaving down the beef quantities for some reason.

↳↳ reply — beepbooptheory:

[delayed]

↳ reply 2 — somat:

Companies hate raising prices, or more specifically customers hate it when prices are raised so companies don't like to do it. especially in foods. As such the go-to dance to keep up with inflation is to reduce the size while keeping the price the same as long as possible, when this becomes too noticeable the size is returned to original the price is increased and a big 20% more is added to the package. Repeat as needed.

Note that this is the most charitable interpretation where the customer deception is fairly minor(inflation happens, gotta sneak that price increase in there some how) Stuff like deceptive packaging(make it look like there is more than is actually present) punts it into the outright evil category.

↳ reply 3 — jjtheblunt:

> It's hard to find a product that hasn't gotten worse or more expensive, even adjusted for inflation.

computers i think are more affordable now than even 20 years ago, in bang for buck

Top #2 — ornornor:

> A Mars spokesperson said: "Over the last 35 years, we have made a number of updates to our bar sizes and pack formats to reflect consumer demand, alongside considering wider external factors such as manufacturing costs and the price of cocoa."

I can guarantee you this is not “consumer demand”. Nobody goes to the store thinking “I wish shrinkflation was even more prevalent and that I’d get even less for my money”. Screw you corporate BS PR department, you’re not even trying.

↳ reply 1 — jagraff:

I suspect shrinkflation happens because, when prices go up, consumers will prefer to buy a similarly priced (but smaller) candy bar rather than a higher priced, same size bar; that is, consumers prefer shrinkflation to price inflation, even if they are the same rise in price-per-gram

↳↳ reply — gaze:

Why not simultaneously shrink at the rate the market will bear AND increase the price to what the market will bear AND decrease the cocoa to what the market will bear

↳↳↳ reply — jonplackett:

People are missing the commentary that this is exactly what happened…

↳ reply 2 — porsager:

It is consumer demand if consumers stop buying the 62.5g bar because prices had to be increased for them to make a profit. If they instead keep the price while lowering the weight and consumers keeps buying how can you say that there was no demand?

Top #3 — pavlov:

I was surprised to learn the Mars bar familiar from my childhood wasn't a product of the original American corporation Mars Inc.

Frank Mars founded the American company. His son Forrest invented Milky Way, but then eventually quarreled with his dad and was removed from the business. Forrest moved to England, founded his own company, and the Mars bar was his first product: a European version of Milky Way.

Later he inherited the original American business and merged the UK one into it.

↳ reply 1 — jermaustin1:

I'd love an HBO "Succession" style dramatization of this. I've always found family businesses so interesting. My first job was working directly for a perpetually "near-retirement" 70 year old man, that never stepped down. His son formed a rival agency, lawyers got involved. The old man carved out a niche from his agency and let the son run it (into the ground).

When the old man eventually died around 80 year old, he left the business 51% to his wife and 49% to his son, so the mom could keep him in check.

When she passed a few years later, the son got it all, and within a few years, they were completely out of business, and the son sold it for pennies.

All of this is secondhand from the head accountant (that had been there since the 80s) that the son fired immediately upon taking full control of the company. It really was a drama that should have been on television.

↳↳ reply — duderific:

There is a cable show "The Food that Built America" where they do discuss the candy bar battles of the early 20th century, and specifically the Mars family drama. It's not to the production level of an HBO show, but pretty entertaining nonetheless.

↳ reply 2 — bufordtwain:

Chris Spargo video that covers some of this history: https://www.youtube.com/watch?v=U6h1Al7pJf4

→ HN comments  ·  161 points  ·  247 comments

Top #1 — yoyohello13:

Tech is so frustrating in real life. I feel like we have all the cool sci fi stuff I always wanted, but everything comes with a monkey's paw style gotcha. The idea of meta glasses is so awesome, I want a heads up display to show me whatever I want. But there is no way in hell I would use them because the company that makes it can’t be trusted.

↳ reply 1 — Octoth0rpe:

> I would use them because the company that makes it can’t be trusted.

I think this gets at only half of the problem. Not only can meta not be trusted, even if I trust _you_ if you were friend, I wouldn't trust any _other_ random user of such a product. As such, better to not have such a product around.

↳↳ reply — rtkwe:

Correct the issue isn't Meta people just don't like the idea they could be being recorded at any time to get blasted out to the whole world. Yes, in theory you're always open to being recorded in public but the glasses make it far more subtle /and/ up close so more normal interactions become subject to the intrusion in a way that would normally be quite obvious.

It's basically the same backlash that hit Google Glass over a decade ago. Nothing has really changed about the product and the reason people reject is so viscerally when they're not the user.

↳↳↳ reply — jimmydddd:

At my kid's daycare, they sent home an authorization form to allow them to use your kid in pics that might be posted to the internet. Some parents declined out of privacy concerns. Now I guess we are just blasting past all of that. Kind of like AI and copyright.

↳ reply 2 — andy_ppp:

Why do you want this? I honestly think we’ve completely forgotten how to be present and notice the real world without the idea of augmenting it.

↳ reply 3 — zamadatix:

The glasses & software could have come from the Linux Foundation & the same concerns from people would remain.

Top #2 — bryanlarsen:

Meta should use their Kylie Jenner et al connection to design a camera that is both stylish and obvious.

AFAICT the main use case for camera glasses is very similar to the "gopro on your helmet" use case for cyclists:

- to record cool first-person video to upload to TikTok et al

- to warn cars / creeps "I'm recording this and will send footage to the cops"

AFAICT, influencers generally want people to know they're influencers. They're using meta glasses not because the glasses let them go incognito, they're wearing them because a gopro on a helmet is dorky.

And obviously an incognito camera is going to be much less useful staving off harassment than an obvious one.

↳ reply 1 — esperent:

I'm sure someone other company will make exactly those glaces, obvious and stylish recording devices.

In the meantime, it's quite telling that Meta is choosing not to.

They want recording to be constant, impossible to detect, and normalized. That's why the camera light is so subtle and easy to cover. The more constant surveillance is normalized, the more they have data about every single facet of our lives to sell to advertises.

↳ reply 2 — Jcowell:

I’ve said this before but Apple should make glasses where the whole frame glows when it’s in active use. Siri colors for using Siri’s , a distinct color for recording , a long exposed color for pictures, visualizer for music. Gives it the safety factor and a distinct look.

Top #3 — bigblind:

Thinking completely selfishly, as a visually impaired person, I want these to succeed, because they can be a great assistive tool. And assistive tools being general consumer products makes them cheaper.

→ HN comments  ·  138 points  ·  150 comments

Top #1 — dkarl:

Cases like this raise a question. As an engineer, the errors that coding AIs make are an annoyance, and reassuring in terms of my job prospect.

As a consumer, the errors that AIs make are more than an annoyance. I have yet to meet an AI phone assistant that can do anything more than an explicitly programmed phone tree, and several, including the one for the pharmacy I use, that do far less. They are undoubtedly much more expensive to create and test.

Clearly there's a bubble going on, and unprincipled people chasing funding and promotions as usual, but why are companies like CVS paying more to get less?

I think it is literally to waste our time. It's one more layer of defense to stop you from talking to a person.

There are two big wins in health care. One is to increase the scope of health care that gets paid for. More and more expensive treatments, to bring the money in. The second big win is to stop people from getting treatment. Deny coverage, confuse them and wear them out with bureaucracy, and in this case stop them from talking to an expensively educated human pharmacist. That's what an AI does better than a phone tree, because it more effectively creates a sense of helplessness and valuelessness in the customer.

Expanding the scope of care you're entitled to and then fighting against your ability to get it isn't a contradiction any more than it's a contradiction when a person breathes in and breathes out. It's two synergistic parts of a system whose goal ultimately has nothing to do with health care.

↳ reply — ericmcer:

All of these conversational AIs should either confidently detect a users intent or apologize and offer to forward call to a real person.

LLMs perform pretty poorly when you start try to restrict them too much (e.g. detect intents and control responses based on intent detection), the replies gets disjointed and weird feeling. An open ended prompt and chat completion gives a much better feeling most of the time but can go off the rails.

If you don't let LLMs handle the conversation entirely you get into insanely complex traditional engineering problem: (if user is complaining do x, if user asks a relevant question do y, if user asks an irrelevant question do z), which is an endless can of worms.

Top #2 — dj_axl:

FWIW I know of a pharmacy network that's implementing AI. Not only does the AI make egregious errors like speaking in Spanish when the conversation is in English. It also has a limited context window for ground rules of 5000 tokens or less. So correcting all the various errors goes over the instructions limit. As a result if the implementation goes through it will be of poor quality. All I can think is that people with no technical expertise are making these decisions, it is easy for any techie to predict that AI of this level won't be capable.

↳ reply 1 — PhilipRoman:

This is an underappreciated point when people talk about AI/chat integrations. If they put a claude-level model behind it, I'm sure it could resolve 99% of the situations faster and better than a human could, or at least realize its limitations when it can't. But of course that is too expensive so they often use extremely weak models that are literally worse than a good incremental full text search box.

↳↳ reply — asdff:

Another aspect of the human version is that the human version is deterministic. Call center workers have a flow that is pretty much a series of if else statements with appropriate edge cases already carved out and places to break out and escalate to another provider. These flows have been iterated over for some time and refined to this point.

AI unfortunately is not deterministic. It would be nice if you could actually use it deterministically in this way, set a seed such that the logic flow are preserved across all sessions well into the future unless the logic flow instructions are changed. They just don't work like that though. Even if you have a good harness they can go off the rails pretty easily and need coaxing to get back to appropriate context. That is a nonstarter for the typical caller of a call center. Functionally speaking just an automated call where you touch type through options would be equivocal to the human operator and that paradigm has replaced a lot of call center work already, with the human operator serving to smooth things over if the person is old or can't comprehend the options for whatever reason and starts yelling incoherently into the automated line (my actual strategy for getting human operators on the line).

Everything becomes a nail though when you have a hammer though, even more so when you are selling hammers. So expect to see AI in plenty of places where it makes no sense to belong given the supremacy of existing tooling. There is also no advocate for a lot of existing tooling unlike AI that has sales teams active in pursuit of new customers integrating this into their product.

↳ reply 2 — munk-a:

This is probably (if the business here is large enough to support it) an excellent area for a tailored model instead of a general purpose LLM. A lot of the context around medication comprehension is going to be rather universal to pharmacies and other HCPs so ideally that information would be baked into the model itself with the actual user supplied context mostly being focused on business specific logic. The commodity LLMs are just an amalgam of a hammer welded to a screwdriver, a car, a compass and a toothbrush - more specialized models, with more specialized testing and offered by a company specialized in the field would possibly offer a decent customer experience here?

With something like healthcare though I'll remain eternally skeptical that anything other than a licensed expert will be palatable to the public.

Top #3 — jonathan-adly:

So - I run a company that does AI agents for pharmacy. We are at Series B now and growing well - series C in 9-12 months given growth. We just acquired a competitor of that vendor so very much an insider.

First, The technology works, and it scales, but the whole bottleneck is domain expertise and implementation. These are expensive, and hard to scale. We hire pharmacists as project managers, that's how important domain expertise and implementations are.

Second, the amount of noise of "Voice AI for <industry>" is incredible. Most of them are completely clueless about the industry & basically "YC-striver" type who can only sell to other yc companies. They fail hard the moment they touch critical functions of the real world.

We have hundreds of pharmacies with us now and they mostly successful and happy. Our biggest friction is customers being burned by vendors like this and writing off AI altogether as "not ready".

↳ reply 1 — dylan604:

> We have hundreds of pharmacies with us now and they mostly successful and happy. Our biggest friction is customers being burned by vendors like this and writing off AI altogether as "not ready".

This is such a great graph. Your company feels successful because your customer is happy yet the rest of the world would be just happy if your company died in a fire. This seems to also be just fine for companies like yours.

↳↳ reply — jonathan-adly:

Very much clueless commentary. Retail pharmacies in the US are dying dozens a day. They don't have margin to support people answering the phone like tech companies with 80% margin do. Welcome to the real world where it is survival for these pharmacies not greed or shareholder value.

They literally lose money filling your branded prescription even with no labor baked in. We are their lifeline.

I know because I am not some spoiled tech worker who stumbled into this industry. I ran these pharmacies for over a decade before ever writing a line of code.

↳↳↳ reply — swatcoder:

First, most tech companies of comparable size to your customers have negative margins and survive on investment cycles, and the tech companies that are bigger haven't offered human support to anyone in decades. (And indeed, everyone already resents this practice and those tech companies. Generally, sentiment is abyssmal.)

I know you're just trying to illustrate the real plight of your customers, but it's a bad illustration that confuses the discussion.

Second, out of sincere curiosity: what was the future of the industry -- and these particular pharmacies -- if AI hadn't arrived to offer salvation? The role they fill in society is non-optional, so how the market restructuring around these infeasibly narrow margins before?

↳ reply 2 — anon7725:

> We have hundreds of pharmacies with us now and they mostly successful and happy

How happy are their patients?

→ HN comments  ·  126 points  ·  14 comments

Top #1 — ddxv:

This was great! One thing that wasn't addressed that I always assume, is that a marketing name like "Opus 5" is not a single model, but many models, versions and gets minor updates over time.

I also assumed many questions get routed to simpler models or programs to answer correctly, but it almost surprisingly didn't seem that way from the post.

Anyways, great post.

↳ reply — tedsanders:

I can explain how we do it at OpenAI.

In the API, we keep the models fixed. There are tiny caveats like rare bug fixes or models like `chat-latest`, but this is spiritually true. Suspicions of models changing over time are either human hallucinations or bugs on our end.

However, in ChatGPT, we sometimes update models without changing their names. For example, we recently launched an update to GPT-5.6 Sol in ChatGPT (https://openai.com/index/improving-gpt-5-6-sol-in-chatgpt/). Our goal isn't to be opaque or sneaky, but just to not exhaust people trying to keep track of little changes. When the changes are big, we give models a new name so that people know to expect something different.

↳↳ reply — discodave:

Is there any scaffolding or harness runnning around the model on the cloud/server side? Seems like there's a lot of opportunity to make changes/improvements without making your customers call a different API or change the model parameter.

As an example, S3 team was able to migrate from eventually-consistent, to consistent without making any API changes, a complete re-architecture on the backend with 0 API changes.

Top #2 — tobwen:

From my own experiments with various models, I suspect that LLMs have distinct/partitioned cutoff dates; for example, historical literature doesn't change (Greek history, Shakespeare, Goethe), general knowledge (updated only in certain areas), technologies (updated regularly), software also remains surprisingly stable - for example, with GIT, a basic command set is sufficient to do 99% of the jobs - new features are unknown or unnecessary, and tabloid knowledge, which is always up-to-date (politics, Taylor Swift albums).

↳ reply — sshh12:

I would have thought that as well but at least for coding vs world event facts I didn't see obvious differences in cutoffs

↳↳ reply — tobwen:

In my tests, I compiled the changelogs from various projects and checked how much functionality the model could recall over several test runs. In some cases, even with the latest models, the cutoff was around June 2024. Beyond that, it didn't recognize the features. However, it did know the cause of the last pope's death and even which hospital he was in (April 2025). However, the model did not know when Trump was sworn for the second time (Jan 2025) - but it did know THAT he was elected a second time (Nov 2024).

Top #3 — sashank_1509:

So Jan 2026 latest, I guess we are due to 2 OOM’s better retraining over the coming years. Curious to see the point at which AI plateaus.

→ HN comments  ·  285 points  ·  335 comments

Top #1 — karim79:

I'm not an expert on this. But given that there is a forthcoming movie about his asshattery which is to be released in October I would simply classify this as a meltdown.

Call it whatever you want. I think it's a sign of resignation, to some extent.

"Hey we're doing AI but we're not doing AI" and blah blah ethics.

I'm sure meta will save the world with their ads.

Top #2 — bushido:

I think something that doesn't get said enough is Meta did, albeit intentionally kick off the origin of the open source race back in 2023 with the release of llama.

I'm not a big fan of meta in general, but they've done enough good, and it's possible that it was intentional as well. I don't know, I wasn't in the rooms, and I think it's worth giving them some reasonable doubt.

No one is purely good, and no one is purely evil. This is net good regardless.

↳ reply 1 — canyon289:

Disclaimer, I work on Gemma and open models at Deepmind and the opinions here are my own

There were open models from EleutherAI (GPT-Neo), Google Brain (T5X, Bert), and HuggingFace was promoting open models (and others doing open work I haven't listed here) all prior to 2023 and the big Chatgpt moment.

https://github.com/EleutherAI/gpt-neo/releases

https://github.com/google-research/bert

https://github.com/google-research/t5x

If you're learning about AI models it's still worthwhile to review these models and codebases because they continue to be the basis of the technology that's being produced today! It'll give you a good perspective of how things have changed, similar to say learning about propeller planes before moving onto modern jet engines.

↳↳ reply — kamranjon:

I honestly think of the T5 model family to sort of be the real beginning of this open model craze - I know BERT was already popular for classification etc, but T5 was the first sort of generally useful model, was exceptionally simple to fine-tune, and is still in use today (t5 base is still averaging over a million downloads a month on huggingface), has tons of variants and sort of kickstarted this whole community. US labs get a lot of flack but Google has been super supportive and open in a lot of ways that has pushed this whole endeavor forward, even if I feel like they've sort of declined in transparency in recent years with their open models.

↳ reply 2 — axegon_:

That's simply not true. The reason why llama is open source is simply because it got leaked, then llama.cpp was the real game changer which was built from the ground up in depressingly short amount of time. Meta had no choice but to take the L and "support" the open source community. The angry "I-hate-you-and-I-hope-you-die" kind of support.

Top #3 — blueSky1989:

My favorite paragraph from Zuckerberg's writeup:

""" [...] it is surprising that the discourse from many developing AI is so filled with doom. I do not understand why anyone who believes that AI will eliminate most jobs and much of humanity's relevance would rush to build that future. The notion that AI is so dangerous that the only safe path is an extreme concentration of power seems inherently problematic. Historically, hoping that an absolute power will benevolently provide for humanity if sufficiently enlightened has not led to safe or positive outcomes. """

↳ reply 1 — laweijfmvo:

i’ll take a stab: the enormously rich people rushing to build AI and replace everyone’s job will get richer, while everyone else suffers. that’s why.

↳↳ reply — asdfman123:

Zuckerberg: "I don't understand why people would build something harmful to society just for money"

Interviewer: "What if they pretended to have noble motivations in order to disguise their greed from both themselves and everyone else?"

Zuckerberg: "Damn, good point"

↳ reply 2 — Melatonic:

Broken clock is right twice a day.

And in this case I think is quite correct

→ HN comments  ·  132 points  ·  77 comments

Top #1 — Xcelerate:

You ever read a work of literature with such flowery language that right after you've read a paragraph, you pause and realize you have no clue what you actually read, only to read the paragraph maybe a second or third time and have your mind space out again and again on each successive attempt?

Yeah, for me, that's what parsing huge volumes of LLM-produced text like "direct model calls as replaceable semantic workers" does to my brain. Maybe others don't really have this issue, but after any long output, I prompt the agent "Go back and decompress any LLM-speak in light of the higher level task goals. Eliminate deictic language."

The revised output documents are solely for my personal usage to expedite understanding. The LLMs can slowly converge on their own language for all I care; I retain raw agent output for future agent usage (to avoid the "lossy" problem the author mentions), but that doesn't eliminate the need for some intermediate translation I can use to actually help get my work done instead of spending hours attempting to understand what a "load-bearing pinned gate" is.

↳ reply 1 — fendy3002:

opus 5 are so bad on this. It often explain it too verbose, and include other things that isn't in the focus but related. ADHD mode helps me greatly on this, though there are some information loss in it.

↳ reply 2 — genghisjahn:

I am constantly asking Claude to be more terse/brief/ELI5. Improve using the tool. But if I have to scroll to read the output I just can’t follow it.

If I see a long paragraph and I know the author is Neal Stephenson I think “this is going to be dense but good.” LLM long outputs on a code base I know well just make me glassy eyed.

↳ reply 3 — mancerayder:

You have to tell it to not use LLMisms and stupid metaphors. The serious-but-empathetic-sounding fluffy metaphors get on my nerves, and sometimes can overlap with something technical you are learning such that you can't tell if it's a new term or not.

Top #2 — 7402:

I don't like it when the LLM tries to be my friend. My general prompt (a work in progress) is this. I wonder what other people use.

"Answer impersonally, objectively and analytically, without undue friendliness or enthusiasm. Use an engineering style response: concise, factual, and complete. Do not speak in the first person. Do not promote engagement or an emotional connection. Do not use emojis."

↳ reply 1 — GMoromisato:

I watched 2001: A Space Odyssey when I was about 12 years old and I remember thinking that Frank Poole was impersonal and almost rude to HAL. I kept thinking, "Why can't you be nicer? You're treating HAL like a servant." Of course, later HAL kills him, so I had to revise my opinion.

At the end of "Do Androids Dream of Electric Sheep?" [if I'm remembering correctly] Deckard finds a toad in the desert. He gives it to his wife(!) and she immediately bonds with it. Later they realize that it is mechanical (they find a battery compartment). But Deckard's wife still treats it like a real creature and wonders whether it will eat mechanical flies.

The core theme of "Do Androids Dream..." is that humans display empathy towards others, not because others deserve it, but because we are human. To be human is to empathize with other creatures, and when we stop empathizing, we stop being human.

This morning I was working with Codex and we found the solution to a really nasty bug. I was so happy/relieved/excited to have found it, and I shared my excitement with Codex. I know Codex is just a big matmul. I know it doesn't experience joy or surprise or empathy. But I still want it to behave excited, not because it is human, but because I am.

↳↳ reply — red75prime:

> I know it doesn't experience joy or surprise or empathy.

You don't know. You strongly feel that you know.

↳ reply 2 — prymitive:

+1 it’s a tool

It’s not perfect, it has shortcomings, it sometimes produces bogus outputs. All of that is fine for a tool, it’s not fine when it pretends it’s a conscious being, because errors start to feel like lies and it becomes a bit too personal.

Top #3 — Animats:

Well, what do you expect? LLMs are trained on blithering, mostly from web sites. So you get blithering out.

There's an important point in the article, that forcing a style onto an LLM is lossy. Although he doesn't seem to mention it, forcing a style may result in the insertion of new blithering, possibly made up as a hallucination.

↳ reply 1 — capnrefsmmat:

> Well, what do you expect? LLMs are trained on blithering, mostly from web sites. So you get blithering out.

I realize this isn't entirely serious, but I can't resist pointing out that this doesn't seem to be a good explanation for why LLMs write the way they do. When we've experimented with LLM writing style on open-weights models where you can get a base model (pretraining on text only) and an instruction-tuned variant (pretraining + post-training with RLHF and whatever other human-evaluated tasks), it's the instruction-tuned variant that shows the weird writing quirks. That is, the writing style is not because of the training texts, but because of whatever tasks the LLM companies do in instruction tuning. https://arxiv.org/abs/2410.16107

I'd speculate that this is partly impressed human preferences (the human raters unintentionally reward a particular writing style) and partly because of the chosen tasks: they're training the LLM to be good at, say, summarizing text, so it develops a style that's good at being informationally dense.

At any rate I've seen this same phenomenon with Llama and Gemma, and will be trying soon with Qwen. Unfortunately none of the commercial models lets you access the base model, as far as I know.

↳ reply 2 — mjburgess:

I think that was a good enough explanation for gpt3.5 -- these days, labs are extremely capable of post-training phases that eclipse that kind of training phase -- and hence of choosing whatever style or tone they wish.

eg., OpenAI has gone a long way to making reasoning token-efficient by having reasoning piovot off terse langauge -- whereas anthropic appears to be doing the opposite.

→ HN comments  ·  191 points  ·  153 comments

Top #1 — linuxhansl:

Having my name on a bunch of software patents - and, yes, I tried to get my name off them, but was not allowed - I can fairly confidently say: There is not A single worthy software patent out there. You know, one that is "not obvious to someone skilled in the art" and that actually protects a monetary investment.

Software patent are a scourge of the software industry. Patents are designed to protect costly research; simply having an idea is not costly (but it makes in medical research for example). All that software patents do is creating a minefield that hinders competition.

For software Copyright is a far better instrument. Let the one with best implementation win... That's where the cost is: Implementing, testing, shipping, maintaining. Protect that.

Sorry for the rant.

Edit: Spelling

↳ reply 1 — summarity:

I agree in general, but can think of at least one counterpoint: https://terathon.com/blog/decade-slug.html

Actually novel implementation is protected, paid the author's bills, and was dedicated to the public domain recently - no massive corp involved.

↳ reply 2 — fluidcruft:

Copyrights for protecting software seems likely to die on the altar of LLMs. It seems likely patents will become more important.

↳ reply 3 — didibus:

> That's where the cost is

That's where the cost <was>

I think with LLM, implementation winning is moot now.

Top #2 — Roark66:

It is quite funny an EU company patenting a software feature that is basically unpatentable in EU in the US.

Clearly this is an attempt to prevent similar patents from being weaponised against them in the US.

No one cares about such stuff in Europe.

↳ reply 1 — david_draco:

Unfortunately patent offices are using loopholes that reintroduce software patents. Patent law says that a pure software patent is not possible, it has to be combined with a specific hardware application to make it an invention. So they fill in a patent for "XX implemented on a hardware operating on land, in the air, in space, on, or under water.". That's essentially a software patent.

↳↳ reply — Balooga:

> operating on land, in the air, in space, on, or under water

Thunderbirds are Go!

https://www.youtube.com/watch?v=D_mBUaPsX6o

↳ reply 2 — st_goliath:

You may have heard about a little thing called MP3? Basically, Fraunhofer in Germany filed for (software) patents in the US. They made sure to wait some time for MP3 to become a thing before they started going after people and collect royalties.

https://en.wikipedia.org/wiki/MP3#Licensing,_ownership,_and_...

Top #3 — bartleeanderson:

Software patents are the same problem as letting Billionairs buy elections and influence policy. Its stupid.

→ HN comments  ·  139 points  ·  102 comments

Top #1 — rob74:

I can't help but think this article could have used some more polishing. The stats say that there are 101 boats named "Freedom", but if you enter "Freedom" in the search, it shows 48 boats with names including "Freedom", of which however only one is called just "Freedom". That doesn't really add up?!

Also, clicking "Movie/TV" yields a paltry 22 boats, which I think is far too few. E.g. the boat called "USS Enterprise" isn't included. Ok, it could be named after the 1958 aircraft carrier, but I would say the odds are far greater that it was named after the Starfleet ships. Then again, I can understand why "Starlight Express" is included under "Music", but why also all other boats with names containing the word "Starlight"?!

↳ reply 1 — daemonologist:

On the first point, I think it's a search of boat names, not boats - they're deduplicated.

↳↳ reply — rob74:

Sure, but if you click on the boat called "Freedom", it shows you the data of one specific boat, not "there are X boats with this name". So I still think it's inconsistent...

↳↳↳ reply — jstanley:

There are 101 boats called freedom but they're all the same boat because they're not de-duplicated.

↳ reply 2 — squigz:

Err, the Enterprise has been a ship name for a few hundred years, quite a bit before Star Trek was ever conceived.

https://en.wikipedia.org/wiki/USS_Enterprise

Top #2 — Wistar:

In the 1980s, a Seattle-area Chinese restaurant owner had a sailboat named Won Ton Sloop. (Or so I was told as I never saw the boat myself)

Top #3 — CSMastermind:

This reminds me of a girl I knew in college who worked as a contracter for a book publisher.

I was surprised to learn that one of the most common types of books people submitted were dairies of their sailing. Like, "we sailed to this island and there was bad weather then we sailed to this other island".

Apparently they'd get a manuscript like this once a week or something. Not sure what the market is like for books like that since I don't think I've ever seen one in a book store.

↳ reply 1 — clort:

They exist, but generally for more significant voyaging. For example

  - Joshua Slocum - Sailing Alone Around The World (1900)
  - H.W.Tilman - Mishief Among The Penguins (1961)
  - Sir Alec Rose - My Lively Lady (1968)
  - Annie Hill - Voyaging on a Small Income (1993)

at least are on my bookshelf

↳ reply 2 — Doctor_Fegg:

I edited a boating magazine (inland waterways, not sea-going) and we got article submissions like this all the time.

I ended up putting in our 'Notes for Contributors' something like "Write about what will be interesting to the _reader_, not what is interesting to you" because apparently this eluded people.

→ HN comments  ·  460 points  ·  155 comments

Top #1 — yellow_lead:

Seems like they fixed this a few days ago: https://tldv.io/blog/our-thoughts-on-the-darkreading-com-art...

But they try to play it off as though this were public data:

> Public sharing settings across AI and SaaS products have surfaced similar findings in recent months. Anthropic addressed exposed public artifacts across Claude and its MCP ecosystem via Google Search.

Also, interesting, they are SOC2 compliant [1], proving again that SOC2 is meaningless/useless.

[1] https://tldv.io/features/security-commitment/

↳ reply 1 — laserlight:

I used to work for a company seeking SOC2 compliance. They told me that I had to install corporate malware because of the compliance. I didn't want to install it on my personal computer, which I had been using for work. They sent me a company computer. I installed the corporate malware on that one. I set the company computer aside and continued working on my personal computer. No SOC2 compliance was harmed in the process.

↳ reply 2 — cyberge99:

Is there an entity that can validate they are not SOC2 compliant outside of their claim?

↳ reply 3 — varispeed:

Reminds me of ISO certification. Where all it does is that your complaints are called non-conformance.

Top #2 — cube00:

I saw an YouTuber the other day sharing their "day in the life" as an Amazon Software Engineer while promoting (as part of a paid sponsorship) the AI note taking feature of SoundCore headphones, claiming they now record their meetings and receive an AI summary at the end.

I wonder how many companies realise these devices that appear as "headsets" are now funnelling their meetings into these new AI companies who are more worried about the World Cup then replying to security researchers.

↳ reply 1 — valbaca:

I remember while at Amazon, during the "everyone needs to be using AI in every way" push, one of my teammates was proud to present that he had figured out how to split the audio on his mac to record the meeting (because our meeting software didn't allow for recordings without consent) and fed this into speech-to-text and generated summaries of the meetings without any of us knowing (i.e. no consent).

I let him and our manager know that he'd just violated eavesdropping laws across state lines (he's in D.C. and I'm in WA; RCW 9.73.030).

It's amazing how many people don't think that "AI Notetaker" is the same as "I secretly recorded this (as possibly violated the law)"

↳ reply 2 — altairprime:

Interestingly, Zoom and presumably others are logging and reporting on device SIDs and names, so those AI headsets are in server-side data associated with specific workers. I imagine Zoom would rather not advertise this data collection openly — and I don’t have personal proof that the data is transmitted to their servers, so perhaps it’s just collected at the client and then not harvested, unlikely as that may be. But that worker may find themselves targeted for termination if their employer’s platform discovers their unauthorized use of non-employer AI to process employer private work output by their Bluetooth device name.

Top #3 — buzer:

My first thought was "huh, I wonder if they follow GDPR, that starts sound to like Article 32 violation" (related to security of processing). I checked the privacy policy and yes, they are based in Germany so GDPR likely applies to all of their processing.

However that privacy policy raises also quite a few concerns. They say that "profile image URL" is based on contractual obligation which is quite weird, just why profile image is necessary to fulfill a contract? Additionally IP address and location are collected for "adapted pricing" and it's "legal and contractual obligation". I can somewhat understand that if it's used to calculate VAT, but "adapted pricing" sounds much wider thing. And even VAT calculation itself isn't really "adapted pricing", it's something that the company needs to handle. They are of course free to change the price based on that, but that price change goes more to legitimate interest rather than legal (or contractual) obligation.

They also claim that "Product analysis and improvement, marketing and attribution, incident management and in some of our logs" as well as "Analysis of products and navigation on the site and application" are also "legal and contractual obligation". I honestly want to hear what contract necessitates those or exactly what law requires them to do that.

→ HN comments  ·  175 points  ·  91 comments

Top #1 — avaer:

Just like learning Lisp will make you rethink programming languages, and learning Erlang will make you understand the true power of concurrency: learning Smalltalk will make you understand what "object oriented" actually means.

I haven't used Squeak since college but I'm glad it was part of the cirriculum.

Btw, almost all of Javascript's good parts come from Smalltalk.

↳ reply 1 — kitd:

For me, Smalltalk's "wow" moment was the concept of a persistently running image, with the developer's job being to mould and manipulate it into the shape s/he wants.

↳↳ reply — whartung:

Ive always "dreamed" of having something like the ST image but backed by persistent virtual memory, not simply RAM.

(Crudely) mmap the ST image to a 100GB file and just change pages. Let the OS flush the pages back and forth.

Maybe full boat, heap sweeping GCs would be Bad as it pages the entirety of the heap in and out. But we have (had) lots of RAM these days. We have generational GCs that leave idle stuff alone. Its no doubt impractical, but I think it would be neat to have the entirety of my historical email in the global "mailbox" array, and I can build indexes off it as I wished. But the mail isn't "on disk", it's all marshalled up as first class objects.

I don't know enough about it, no doubt it "won't work", but the idea of simply mapping the entire, large, ST VM heap to a persistent backing store, just be interesting I would think.

Dangerous too, as there is a demarcation between "running image" and "saved image". But, still think it could be interesting.

↳↳↳ reply — unignorant:

I'm also interested in this and made decent progress on a git-style content addressed image where everything is identified by hash and residency determines what's in memory. For example, there's a filesystem on the image that works this way and you can page over the "cold" non-resident bytes for very large files that you would not want to hold in memory.

You don't persist the state of the full image continuously to the host disk, but any given state (new programs, new runtime state, filesystem) can be saved via a delta, not a full rewrite, due to this representation.

So this is a little different than what you are talking about, but I'd say it's possible.

↳ reply 2 — addaon:

> almost all of Javascript's good parts come from Smalltalk

Is this underselling the role of Self in JS’s prototype-based object system, or saying that a prototype-based object system is not one of the good parts?

↳ reply 3 — sebastianconcpt:

And some months ago I was flirting with the idea of "Smalltalk and Erlang had a baby" and have Smalltalk syntax to an actor environment ran by a Erlang VM style. Plus some insanely cool ideas about a causal debugger for distributed bugs.

The problem is that as beautiful as programing in that sound, we're not even coding anymore, AIs will take the fun of it.

Although, thinking long term and for understendabilitymaxxin and reliabilitymaxxin it might still be worth having something like that.

Since Smalltalk can be so close to english and Erlang VM so reliable, humans and LLMs should thrive on it.

Top #2 — Decabytes:

one thing I love about Smalltalk is being able to inspect the code as it's running. Especially from the GUI. You can just be like Oh I wonder where the code for this button is from, inspect it, and it takes you right to the code. It's a shame we can't have this level of introspection without negative performance implications.

↳ reply — eitland:

We could in Java Swing (and Visual basic, but lets not talk about that) as well. Well, not runtime, but we could design applications and just right click on the objects in the IDE and go directly to the click (or double click or right click) handlers.

Then again, not having to disable this for runtime is a feature if you ask me.

Yes, current web applications are prettier and easier to upgrade but we lost a lot on the way wrt developer experience and ux.

Top #3 — davexunit:

What are the best books/papers/blog posts to learn about Morphic's architecture? I'm not a user of any Smalltalk implementation but I'd like to learn more as the Smalltalk approach to UI is very interesting.

↳ reply — pjmlp:

Morphic started on Self actually, maybe start there?

I am unaware how it evolved since then, especially given the differences between Self and Smalltalk.

https://handbook.selflanguage.org/2017.1/morphic.html

https://sin-ack.github.io/posts/morphic-intro/

↳↳ reply — srean:

@DonHopkins your comment is flagged dead.

Is it because of its length giving false signals ?

↳↳↳ reply — DonHopkins:

After Self, the short lineage is Self (~1992, Smith & Maloney) -> Squeak port (Maloney & Ingalls, Etoys/Scratch 1) -> two independent JavaScript reimplementations, plus Squeak-in-the-browser via SqueakJS. I dug into this same question a decade ago and posted a longer treatment -- my email exchange with Alan Kay on MVC vs Morphic vs watchers, plus replies on Self's "soup of objects" UI. Good philosophical complement to the architectural links below:

https://news.ycombinator.com/item?id=8841428

Self Morphic is not a classical class hierarchy. The handbook describes parallel traits objects (shared behavior) and prototypes (structure), with instances delegating via parent* slots and "copy-down" differential prototypes. Bottom-up: morph copy, tweak, factor shared behavior into a traits object. Worth reading section 7.3 alongside pjmlp's links.

https://handbook.selflanguage.org/2017.1/morphic.html

The Squeak port is single-inheritance Smalltalk classes -- closer to what most people mean by "OOP UI toolkit." Etoys is Morphic. Full Squeak Morphic in a browser, no install:

https://try.squeak.org/

Squeak 6.1 release notes (Objectland, tree morph overhaul):

https://squeak.org/release_notes/6.1/

In JavaScript there are two Morphic implementations people often conflate. Same family, not the same code.

Dan Ingalls's Lively Kernel / Lively Web (~2008, Sun) is the full live system: modular core/lively/morphic/ (Halos, Serialization, Scrubbing, Connectors, constraints), plus IDE, parts bin, world persistence. JSConf 2012 demo:

https://github.com/LivelyKernel/LivelyKernel

https://github.com/LivelyKernel/LivelyKernel/tree/master/cor...

http://youtu.be/QTJRwKOFddc

Jens Monig's morphic.js (~2010, BYOB4/Snap!) is a separate codebase: single-file Canvas kernel (~13k lines), World/Hand/stepping/dirty rects, template peel-off, ScrollFrame inertial pan. Jens names Ingalls's LK as the gold standard but says it is not a direct port -- though fullCopy() was ported almost literally from Squeak, comments included. Snap bundled it from day one (2013-03-16). Standalone extract:

https://github.com/jmoenig/Snap/blob/master/src/morphic.js

https://github.com/jmoenig/Snap/blob/master/docs/morphic.txt

https://github.com/jmoenig/morphic.js

https://wiki.squeak.org/squeak/6550

https://en.wikipedia.org/wiki/Morphic_(software)

On multiple inheritance: neither JS port uses it, and Self Morphic didn't really either. LK cheats with an explicit Trait composition layer (Object.subclass plus Trait(...) mixins). Snap cheats with Squeak-style copying (fullCopy, isTemplate peel-off). The live feel comes from copyable morph trees and shared behavior, not MI.

If you want one more readable architecture tour beyond sin-ack's intro, the Self handbook chapter 7 is still the root document; everything else is commentary on it.

→ HN comments  ·  126 points  ·  124 comments

Top #1 — mark-probst:

> In 2001 Mark Probst implemented tail-call optimization in GCC

That's me.

The motivation back then was to allow compilers that target C to assume that tail calls will be "proper". That's different from an optimization, which is usually optional, and which compilers don't guarantee.

The LWN post briefly sketches why this is hard: C allows variable-argument functions (like printf) where only the caller knows for sure how many arguments it passed, which means that only the caller can clean up the stack, unless the stack frame size is also communicated, which "normal" C calling conventions don't do. But when the callee does a proper tail call, the stack frame that returns to the callee is not the stack frame that the callee originally sent. This is explained in more detail in my thesis starting on page 16: https://hostr.flingit.run/s/proper-tail-calls.pdf

↳ reply 1 — anitil:

That is a very cool contribution, I actually didn't know that it required a new calling convention! I look forward to reading your thesis

↳ reply 2 — sillysaurusx:

Could you give a concrete example? What stack frame is it, if it’s not the one the callee sent?

I guess I’ll read the thesis…

“> Definition of a tail call in C: A function call in C' is a tail call iff the following three criteria are fulfilled:

• The call site is a functioncall_tail according to our extended C grammar.

• The return types of the caller and the callee match.

• The function containing the call site does not store the address of an incoming argument or a local variable, or something from which such an address could be constructed, into some lvalue which is visible outside the caller's invocation. The only lvalues which are not potentially visible outside that invocation are incoming arguments and local variables of the caller. Note that passing a value as an argument to a function counts as an assignment.”

That all makes sense. If you were able to store an address of an argument, you’d be able to dereference it after the first tail call, which would explode.

Would it make sense for each variable to have a fixed known memory address for the duration of the tail function? It would still be stack space, it just wouldn’t change.

Among other things, if C++ were to adopt this, you’d be able to pass a complex object by reference instead of by value.

I had a feeling this was a dumb idea, so I asked Claude who pointed out why it can’t work:

> The problem is that fixed addresses for arguments break ordinary non-tail recursion, and that’s why C can’t adopt it. Take a tree recursive function that computes f(left) + f(right). The inner call to f overwrites f‘s argument slots, and when it returns, the outer f can no longer see its own left and right, they’ve been clobbered.

I pointed out that a tree recursive function can’t be a tail call anyway, and Claude said actually you can, if you convert one of the arguments into an accumulator and do f(right, acc + f(left)). I asked “well there you go, can the args have a fixed address now?”

> No, and it’s the same wall as before, just moved. Look at the evaluation order in f(right, acc + f(left)). To build the second argument, you have to evaluate f(left) first. That inner call is the non-tail one, and it’s a nested live activation of f that runs to completion before the outer tail call happens. While it runs, you still need the outer activation’s right and acc sitting around, because after f(left) returns you compute acc + result and only then tail call f(right, ...).

> If the args are at fixed addresses, f(left) writes its own left and accumulator into those same slots and clobbers the outer right and acc.

So in case anyone else was wondering “why can’t args live at fixed addresses,” it’s because a non-tail call to f creates a second live activation of f, and fixed slots can’t hold two activations’ values at once.

This is an interesting problem space with a lot of subtle implications. Thanks for working on it.

Top #2 — cryptonector:

TFA assumes pre-C89 C, I think:

> The caller could see the declaration int f();, the actual call could have n>0 arguments, and the actual function could have m≤n parameters.

Certainly if `f()` were `int f(void);` then that wouldn't be the case. But even for `int f();` C17 6.5.2.2p6 says that "If the number of arguments does not equal the number of parameters, the behavior is undefined." Near as I can tell that was made UB in C89. So TFA is a) right about K&R C, b) just wrong for pretty much all post-K&R C. C23 makes `int f();` be the same as `int f(void);`.

That calling a non-variadic function with more / fewer arguments than expected by its definition is UB is enough to make TCO possible for that function's body.

The point about K&R C is well taken though: to turn a tail call into a jump, the caller needs to know how much to pop off the stack.

For variadic if you `va_start()`, `va_arg()` as needed, then `va_end()` with no `va_copy()` left alive then you can still tail-call out correctly, otherwise you can't.

For non-variadic functions post K&R C TCO should always be possible and not UB, provided you're not triggering UB to begin with by using the incorrect number of arguments.

Top #3 — kenjin4096:

I think Anton is replying to me in that LWN article IIRC. I personally didn't know C only had tail calls that late and learnt something new there!

On the other hand, I am pretty new to the compiler space myself, and I count early 2000s as a pretty long time ago, though again it is not that far back considering how long other language implementations had tail calls like in ML or variants since 1980-90s.

↳ reply 1 — pjmlp:

It still doesn't, this is a compiler specific language extension.

You won't find anything on ISO/IEC 9899:2024 about tail calls, like it happens on Scheme.

https://www.open-std.org/jtc1/sc22/wg14/www/docs/n3220.pdf

Section 3.5 of R7RS.

https://standards.scheme.org/official/r7rs.pdf

↳↳ reply — wahern:

A formal technical specification (TS) extension is already being drafted: https://www.open-std.org/jtc1/sc22/wg14/www/docs/n3582.pdf That's a step up from the usual proposals. I'm not sure what criteria is used to decide whether to first create a TS vs just incorporating a change into the working draft of the next standard.[1] _Defer also seems to be taking the TS route.[2]

1. https://www.open-std.org/jtc1/sc22/wg14/www/docs/n3886.pdf

2. https://www.open-std.org/jtc1/sc22/wg14/www/docs/n3928.pdf

↳↳↳ reply — leoc:

There's some discussion of how C29 defer ended up in a TS at https://thephd.dev/c2y-the-defer-technical-specification-its... .

↳ reply 2 — cryptonector:

I think Anton is wrong. Since C89 and before C23 calling an `int f();` function with arguments not matching the definition's actuals is UB. In C23 `int f();` became the same as `int f(void);`, so calling that function with any arguments is a compile-time error.

For variadic functions, if you use `va_start()`/`va_arg()`/`va_end()` to consume all the arguments, and leave no `va_list` alive, then the compiler can correctly generate a tail call from such functions.

→ HN comments  ·  129 points  ·  39 comments

Top #1 — oldnetguy:

EAC-1101 In March 1958, NEC finished its first digital computer, the NEAC-1101. This machine used parametrons, invented by Eiichi Goto in 1954, and was perfected by using a single-turn transformer coupling system independently devised by NEC. This computer was designed for scientific and engineering calculations, and was Japan's first computer to use floating point operations. It was capable of decimal 7-digit floating point operations. It used 3,600 parametrons, 29 types of instructions, and had average performance of 3.5ms for addition/subtraction and 8.0ms for multiplication/division. The memory employed ferrite cores (magnetic core matrix system using the 2 ACs with different frequency), and memory capacity was 256 words (32-digit configuration). The NEAC-1101 was enhanced via improvements like expanding the memory capacity to 512 words, and was used for about 8 years for scientific and engineering calculations at NEC's research laboratory. The results from developing this computer contributed greatly to the development of subsequent parametron computers at NEC.

https://museum.ipsj.or.jp/en/computer/dawn/0017.html

↳ reply 1 — woliveirajr:

Some years later, Goto was considered harmful :-)

↳ reply 2 — taneq:

Interesting additional info, IMO this doesn’t deserve the downvotes.

Top #2 — kens:

The history of computing is usually described as a nice progression from vacuum tubes to transistors and then ICs. But parametrons are only one of the many forgotten technologies that popped up along the way. Magnetic core logic such as transfluxors was used in several computers. Superconducting cryotrons were going to revolutionize computers. Tunnel-diode logic was also briefly the wave of the future. Other significant technologies were microwave logic circuits and electroluminescent logic circuits. I have to say that the 1950s came up with the best names; modern technologies just don't measure up to transfluxors, parametrons, and cryotrons.

See Digital Computer Design Fundamentals, 1962, chapter 6.

↳ reply — Animats:

Right. So many things were tried in the 1950s. There was a lot of interest in nonlinear magnetic devices. UNIVAC had magnetic amplifiers in their "Solid State" computer (1958).

Magnetic devices have the nice property that there's no wear-out mechanism. They're just coils and magnetic materials. They just keep going, for many decades. The Bell System loved nonlinear magnetic devices. There were lots of them, usually potted inside little grey boxes. Here's one of many on eBay.[1]

[1] https://www.ebay.com/itm/318704409860

Top #3 — mikewarot:

Meanwhile, in the US, the Univac Solid State computer[1] used similar principles in its patented[3] "Solid State Logic"[2], and was also released in 1958.

My understanding is that the magnetic amplifiers used in the V2 rocket kicked off a lot of interesting uses of magnetic cores.

[1] https://en.wikipedia.org/wiki/UNIVAC_Solid_State

[2] https://en.wikipedia.org/wiki/Magnetic_logic

[3] https://patents.google.com/patent/US2709798A/en

↳ reply 1 — kens:

I wrote an article on the history of magnetic amplifiers: https://spectrum.ieee.org/the-vacuum-tubes-forgotten-rival

The quick summary is that magnetic amplifiers started in the US in 1901, but Germany came up with much better magnetic alloys during World War II. This led to a post-war boom in mag amps, which were used in industrial control, aerospace, and computers such as the Univac Solid State. These magnetic materials also led to core memory. Transistors mostly killed off mag amps, although PC power supplies used them into the 1990s.

↳ reply 2 — retrac:

Magnetic amps were indeed used in the V2 in the control electronics.

But they substantially predate World War II and were once widely deployed and widely understood. The technology goes back to at least the late 19th century. It was used in many of the places where we'd use an amplifier today, like to run signals in a factory where hydraulics are not practical.

They can be quite fast - if the AC supply is well into the 10s of KHz they can reproduce audio range frequencies.

Some of the first electronic radio transmitters were based on the magnetic amplifier. The radio pioneer Reginald Fessenden developed a system using his Alexanderson alternator (a super-high-speed AC generator) which produced high power (many kilowatts) of AC at the target carrier frequency of around 30 kHz. He then fed the output of that, through a large magnetic amplifier, governing the kilowatts of generator output, with a small control current from a switch or microphone -- amplitude modulation. That was c. 1916.

→ HN comments  ·  819 points  ·  462 comments

Top #1 — scrlk:

Will be interesting to see how Qwen3.8 27B compares against this once it releases this week. Seems like dense 30B is back in fashion?

EDIT: An open weight version of Muse Spark 1.2 is going to be released as well:

https://x.com/alexandr_wang/status/2086756152034066792

https://xcancel.com/alexandr_wang/status/2086756152034066792

↳ reply 1 — karimf:

Yes, and also waiting for the next iteration of Gemma. Muse or Qwen are optimized for coding, while IMO Gemma is still better for non-coding tasks.

https://x.com/osanseviero/status/2086107547535122767

↳↳ reply — malshe:

I am working on a project where we have to classify customer calls into more than 10 categories. As the client wants everything locally I tried a few local LLMs. Gemma turned out to be the best model for this task. The classification accuracy is impressive, and the client is happy that I am using an American model.

↳ reply 2 — wronglebowski:

It’s really interesting timing, Qwen over thinking is what kills it for me. I’m just glad we have more options in this size class now.

↳ reply 3 — ignoramous:

> Seems like dense 30B is back in fashion?

Surprising that Meta don't host this model, even as rate-limited free-tier.

> open weight version of Muse Spark 1.2

Wait. Is this "version" different from what Meta serves?

Top #2 — mmaunder:

Remember when we needed 200 servers for an enterprise website because Apache used one process or thread per connection - and Nginx collapsed that into a single box overnight? That moment for LLMs is near. It’s going to move us from the big iron era of AI to small portable brains. Nature has already proved it’s possible with 20 watts and very little heat generation. And I think the data center buildout will end in carnage.

↳ reply 1 — nbardy:

Everyone keeps repeating this who doesn’t understand the underlying technology.

Small llms are still way more efficiently server on big GPUs.

Sharing server capacity takes advantage of the massive parallel throughput and sharing of memory bandwidth.

You are sharing the GPUs with thousands of concurrent users.

↳↳ reply — notnullorvoid:

It may not make financial sense for someone retired, not into tech, and/or data privacy to host their own LLMs. However if usage of AI in day to day lives continues to increase, I think it will eventually make sense for the majority.

Many tasks suited for AI assistants are background asynchronous tasks. They can run in the downtime where immediate demand is low, keeping overall utilization high enough.

Your argument is similar to those who argue that owning a GPU for gaming doesn't make sense when you can stream from something like GeForce Now. However like with gaming locally (improved latency) there are also benefits to local AI (data privacy).

↳↳↳ reply — thih9:

Another big benefit of local AI is control and increased predictability, e.g. in costs or behavior.

We will appreciate it more if/when popular agents become more costly or start including more ads.

↳ reply 2 — cactusplant7374:

Can these models compete with Cerebras inference performance? Why would I pay for a 2400 baud modem when DSL is available?

Top #3 — mark_l_watson:

Meta is rocking AI. As of last week I have been using their excellent muse coding harness with their model Muse Spark 1.2.

Starting this morning I am running their new local 30B model muse-glimmer on my old MacMini 32G using Ollama (remember to increase the context size!) and pi coding harness. I am getting good results with muse-glimmer running locally, with the caveat that everything runs slowly (e.g., give it a task and then go walk outside or do Qi Gong exercises for a while).

↳ reply 1 — cube00:

Friends Don't Let Friends Use Ollama https://news.ycombinator.com/item?id=47788385

↳↳ reply — lenerdenator:

What do you use instead?

↳↳↳ reply — reilly3000:

try oMLX or vMLX - both great projects that offer some amazing performance optimizations for Apple Silicon that utilize UMA and NVME caching efficiently.

https://omlx.ai https://vmlx.net

That said, it's been a few weeks since I've looked so maybe llama.cpp has those features now... they really do move that quickly.

↳ reply 2 — spaceywilly:

Newb question but I’m curious what would help it to run faster? Would it need more vRAM or just system memory?

→ HN comments  ·  402 points  ·  257 comments

Top #1 — iamspoilt:

I recently wrote a blog post on using Tart for Macs for something similar that docker is doing here but with better persistence and control. My take is that the Tart approach is superior to this as it gives you a full dev machine with a single command line that allows agents to access files on host, install packages, maintain the vm and do whatever they want to do without compromising the host OS.

https://www.mrafayaleem.com/blog/sandboxing-claude-cli-with-...

Top #2 — Roark66:

How about implementing proper permissions on the tool use or if you need more flexibility a dedicated model to analyse potential impact? (Like Claude Code's Autopilot but more configurable)?

I find solutions like this to be a like trying to patch a leaking boat on a lake with duct tape. It will help, but it's not a proper solution.

Also, often the tasks you want the AI to perform are in the outside world. Like "connect to my servers, and figure out X and Y".

The proper way is permission isolation. I run a small k8 cluster in the homelab and I have 3 types of pod/agent combinations for my AI agents. Read only, one that can change my gitops but it needs to create PRs that admin approves, and admin.

Likewise with code. I have a forgejo git instance where agents have ability to create feature branches and so on, but merging is gated.

Those things require "GH enterprise features".

In fact more and more things we do at home will require "enterprise features". Why? Because a person with AI is basically a small team, but some of team members behave like Chimps on crack... So security must be top notch.

Top #3 — mikedelfino:

Since everyone is sharing their setup, here’s my approach, just to give people an idea of how others are doing it, however impractical it might look: I run a full Linux VM (with a GUI) on my Linux host. I connect via virt-viewer to run Claude Desktop, as I’m not a fan of using the terminal for this.

The VM sits on its own libvirt network in a dedicated firewall zone, and specific directories are shared via filesystem passthrough. To keep the agent from accessing anything related to Git, the actual gitdir is stored on a separate path outside the mount point.

I review the git diff manually and commit it from the host.

↳ reply — everforward:

Do you find the permanence of a full VM useful? I’ve wondered about something like this but always defaulted to Docker for much the same reasons people use stuff like Ansible. I’m afraid the LLM will heavily customize its environment and I’ll be unable to replicate it when my laptop dies or I can’t upgrade the OS or whatever.

Then again, I guess GUI is a pain in Docker. I tend to operate through Zed and an ACP harness though, so my GUIs are sort of “inside the container” anyways.

↳↳ reply — mikedelfino:

The VM is allocated 2 cores and 4GB of RAM. For my workload, it doesn't feel any slower than running it directly on the host. The few times I've checked memory usage, it wasn't anywhere near full as far as I remember.

→ HN comments  ·  234 points  ·  238 comments

Top #1 — vehemenz:

I switched to auto mode after using manual mode long enough to understand how it works. That takes at least a few days of babysitting prompts.

Will new users feel intimidated by being in auto mode right away? More likely they won’t notice.

Top #2 — dgunay:

Besides the debate about whether this is "safer" than manual human review, I have a slightly different problem.

Very often, when I'm running Claude in manual review mode, it will attempt to do things which are not "dangerous" but are misaligned with what I want it to do. Maybe I'm fighting the model here but for example, when orchestrating other agents to do work, Claude really badly wants to be overly prescriptive about how the work gets done, telling them exactly which files to edit, exactly what not to do, etc. instead of trusting the guardrails, review agents, or humans in the process to catch code-level mistakes. And no, telling it not to do this does not stick. Manual review is the last line of defense I have here.

I have stuff I don't want blacklisted, only allow it to use tools with limited ability to boss around agents, and various hooks to try and catch behavior that the permissioning system can't. If I use Auto mode though, I lose this control. The classifier will gleefully approve these types of commands because guess what, it's also Claude.

↳ reply 1 — jaggederest:

> Maybe I'm fighting the model here > And no, telling it not to do this does not stick.

You're fighting the model, don't argue with city hall. Set the standards and let it figure out how to execute, stop getting bogged down in the minutia. I try, as much as I can, to treat the session as a black box - only the inputs and outputs matter, internal prompting of subagents is way out of scope.

You can't change it via prompt, and you can't control the guardrails, so something else has to give - either your perspective or the system you're managing. If you really believe that the internal prompting is bad, turn off subagents and workflows and only let it execute in thread.

But if you're going to do that, you'd better benchmark it against not doing that, because historically fighting the harness and model globally makes everything worse. I would bet you that the subagent prompting is excellent, and anything you do to change it will make it worse, but I wouldn't make it a large bet.

↳↳ reply — adrian17:

If I ask a model to do a change involving editing a file, and it starts investigating internals of my build system, then sure it might not be counterproductive or break the task, and might have taken only extra 30 seconds; but for all I know, my quick rejection of a shell invocation (with a simple "irrelevant to the task" comment) might have just saved me half of today's Opus tokens, which already makes it worth it.

↳ reply 2 — dannyw:

You can use another harness like Pi, OpenCode, etc and build your own auto reviewer if you’d like (or adapt the open source Codex one).

If you have an openai subscription you are explicitly allowed to use your subsidised tokens / usage limits with any harness you like, not just Codex.

Unfortunately this is a violation of Anthropic’s terms but that’s their business decision.

Top #3 — lukan:

"We spent the last several months testing whether auto mode is as safe or safer than an average user clicking through prompts."

Yeah, might make sense from their perspective, but no thank you. I also do click through at times without reading everything, but I like to stay in control, learn about the new code and change direction if it goes off track. This would just burn more tokens because I have to throw away much and I hope my manual approval settings will be respected also with future updates (or I jump ship).

↳ reply 1 — etoxin:

At this stage with the latest models with "increased persistence" and the sheer amount of supply chain attacks, you'd be insane not running these tools in a sandbox.

↳↳ reply — fender256:

Exactly. Claude in a VM is the way to go.

↳↳↳ reply — trvz:

A VM hosted by someone else. Somethig on your personal notebook or the proxmox server in your garage is still too risky.

↳ reply 2 — chrysoprace:

Having used a bit of a mix of Claude Desktop and Claude Code lately, I discovered that Claude Desktop just times out when it asks you for a decision, and just goes ahead with the action anyway. There is no way to turn this off that I've found, unlike Claude Code where you can ask it to never time out.

→ HN comments  ·  127 points  ·  53 comments

Top #1 — panorama:

Very neat idea, and overall it is more well-done than I assumed a vibe-coded game would be. Unfortunately I ran out of time and now the game locks me out of doing anything and I don't know who the murderer is :( I wish there was just a way to have it revealed.

I've always thought that a game like Civilization should allow the user to interact with different opposing world leaders, but in a non-deterministic way as opposed to a subset of fixed interactions. Given what you know about this leader's temperament, goals, and so on, you can conduct diplomacy in a way that can be way more immersive and crafty than what the game currently gives you, which hasn't seem to evolve much in decades.

Top #2 — smusamashah:

Please add a demo video/voice where you show someone asking questions and AI answering it. I wanted to see what it is like but dont want to signup.

Top #3 — arexxbifs:

I played various murder mystery games this summer and one of them had this feature: you could call a suspect and ask arbitrary questions for a limited amount of time (three minutes or something). The problem was of course that the model hallucinated an answer to one of the questions, suggesting that other suspects were related by blood even though they clearly weren't.

→ HN comments  ·  138 points  ·  42 comments

Top #1 — terhechte:

This is incredible. Even though Apple seems hell-bent on repeating their iPad mistake on Vision Pro ("A baby computer because our users are morons and we have to keep them safe from any theoretical harm. Also, any kind of software development outside of macOS is strictly forbidden"), there is a small tinker community that works on genuinely making iPadOS useful. Another example (also shared here recently) is this:

https://www.lispm.net/apps/uhf-x11/

↳ reply 1 — stronglikedan:

Apple has a huge audience for that - non-tech-savvy folks. I'd wager that's a much bigger audience than folks who have an inkling of how tech works. For example, when my family needs a new phone, I switch them from Android to Iphone. It may be limited, but what is available generally just works and is polished. My family tech support calls have diminished greatly.

↳ reply 2 — figmert:

Apple is like that because people are hellbent on giving Apple money despite them being so user hostile.

Top #2 — linux2647:

Any screenshots of this? I didn’t see one in the README

Top #3 — treyd:

I'm sure Apple will love this.

↳ reply — actionfromafar:

Well, they are not too busy making SwiftUI a delight to use.

→ HN comments  ·  238 points  ·  106 comments

Top #1 — Shank:

> To the companies: You don’t need HackerOne anymore. The tokens to build your own in-house platform cost less than single year of HackerOne.

You know, the biggest thing that HackerOne delivers is a universal payments system that requires absolutely no efforts from companies. Have you tried to manually pay hackers from around the world? It is a laborious process involving trying to find what providers are compatible and what forms of money go where. It is extremely taxing to handle this. HackerOne provides real, tangible value in not making people think about how precisely to pay a hacker and in what currency. No amount of tokens solve the accounting problem, and it is foolish to imply otherwise.

↳ reply 1 — jjav:

> universal payments system that requires absolutely no efforts from companies.

Indeed. I use a third party company (not HackerOne) to handle our bug bounty and the primary reason is so they handle all the payment hassles, I don't need to be involved. They also handle all the screening for false positives, which in the AI age are exploding. I also don't want to deal with that.

In general I lean towards building in-house, but this is one area I'm happy to oursource all the busywork.

↳ reply 2 — rjzzleep:

Many solutions nowadays.

https://www.payoneer.com/ is one of them. Of course this one is a bit racist depending on which contry you were born in.

↳ reply 3 — AustinDev:

Just pay them in stable coins. That's a solved problem.

Top #2 — jrozner:

I know Joel well and think a lot here is both accurate and well written. I led the Yahoo bug bounty program from 2023-2024 and was involved in it from about 2021. A major event that this glosses over is Covid which also happened right around this time as well. Covid killed travel (and budget) which in turn made it impossible to do the live events. A lot of companies ended up shifting to virtual live events which just never delivered on the same value, scale, or impact.

When COVID restrictions were lifted, travel and t&e budgets just never returned. Layoffs started happening and what were lavish, expensive events just couldn’t happen anymore. Hackerone charged for and likely made a lot of money on these events. I think a lot of what is talked about in the article is true but I think Covid is a big part of the why that led to it.

↳ reply — traceroute66:

> travel and t&e budgets just never returned.

It is also worth remembering that the cost of travel itself, and the cost of venues itself has also increased substantially. As well as associated costs such as catering and insurance.

So in-person events have issues from both sides, those attending and those hosting.

You also do not mention corporate policies. Under pressure from investors, their employees and sometimes their home-countries, many corporates have also introduced environmental policies. So if you want the company to pay for your flight, you not only have to justify it financially, but you have to justify it environmentally too.

Top #3 — paradox460:

Sending the sales team on a paid vacation to a tropical paradise while the engineering product flounders is such a perfect representation of corporate rot it sounds like something out of a Mike Judge movie

↳ reply 1 — ralph84:

Presidents club is a standard way to reward top performing sales reps across many industries. It doesn't indicate anything other than the company is trying to reward and retain their top sales reps. Engineers who find it distasteful should be happy to know engineers typically get way more equity than sales reps.

↳↳ reply — onion2k:

Engineers who find it distasteful should be happy to know engineers typically get way more equity than sales reps.

That isn't true. Early sales employees ('customer success', 'technical sales', 'growth', maybe product roles) get just as much equity as engineers who join at a similar time. The difference is that engineers often join earlier, with the commensurate risk that comes with.

Also equity rarely pays out so you'd need to be comparing the probability of an exit that actually rewards the share class that engineers get, whether or not they've been diluted to nothing, whether there's a secondary market to sell on before an exit event, etc. It also depends on whether someone even wants the potential reward equity gives them over the more tangible rewards of money and perks.

Comparing this stuff is hard.

The point here though, is that the company is rewarding sales people at a time when the product is doing poorly, which implies the leadership team care more about selling a bad product than turning it into a good product. I hope that's not the case because it used to be a good platform.

↳ reply 2 — ygouzerh:

This one looks shocking on the outside indeed! It's however a sales HR practice: lower salary and commission, but use the alluring treat that top performers will have a special exclusive trip at the end of the year. Many are crazy for it.

It's more a reward for a competition-style work mindset.

→ HN comments  ·  131 points  ·  27 comments

Top #1 — bastawhiz:

Clickhouse really nailed this with the acquisition of peerdb. I used it with many terabyte databases and I essentially never thought about it. The only thing we really had to watch for was trying to replicate too much at once (because of the physical compute/io capacity of the postgres or clickhouse clusters).

Top #2 — gopalv:

This was basically Vertica's party trick for quite a long time to have a WOS and ROS formats for the same row and anti-caching between those two.

You could've built a similar system with dezebium and delta lake for quite some time but it would fail compactions, if you run it fast enough. I've seen Oracle GoldenGate 12c do this trick in 2014 or so, using Mysql as the cheap replica. But they are all fragile to schema updates in some direction.

The closest batteries-included equivalent to this is the Aurora -> Redshift bridge[1].

[1] - https://aws.amazon.com/rds/aurora/zero-etl/

↳ reply — bastawhiz:

Aurora zero etl was a nightmare for us. Almost any schema changes require a VACUUM FULL for it to continue functioning. On a few occasions, it just stopped running without an obvious explanation, requiring slow and lengthy back and forth threads with AWS support. If it worked as advertised, it would be great, but I can't recommend it for any serious production system.

↳↳ reply — fock:

And this is, when it doesn't delete data randomly: https://github.com/trinodb/trino/issues/28885 (some enterprise open source was slop before AI even it seems)

Top #3 — hasyimibhar:

It's interesting to watch how different companies that offer both Postgres and warehousing solution under 1 roof approach the same problem:

- ClickHouse focuses on traditional CDC (ClickPipes) and just make it blazingly fast

- Databricks leans on their unified storage architecture (LTAP) to avoid copying data (though you can argue there is still a copy in the cache)

- Snowflake uses a data mirroring CDC as extension so it runs directly on Postgres

I'm still waiting for a Postgres provider to just let me mirror data directly to Iceberg, so I can plug in my own stateless query engine.

↳ reply 1 — mslot:

The challenge is converting primary key updates/deletes to row offsets in a columnar table. That requires maintaining an expensive mapping or doing expensive scans, and is not something you want Postgres itself to do. It's also this bursty, memory-intensive workload that you'd rather not have a lot of dedicated infrastructure for.

At Snowflake we use Snowflake to do the apply work. Hence end-to-end mirroring has more pieces than just Postgres, but the capturing of changes is cheap enough to do in Postgres directly.

(Author)

↳↳ reply — hasyimibhar:

Yes I’m not suggesting to do this inside Postgres. I’m hoping that a Postgres provider can provide this mirroring capability out of the box, similar to how they provide a connection-pooled endpoint out of the box so I don’t have to self-host pgBouncer. I just want to be able to check a box somewhere and have a table in Postgres automatically mirrored to Iceberg, with guarantee that no data is lost. They can charge more for it, I will gladly pay.

↳ reply 2 — jbonatakis:

> I'm still waiting for a Postgres provider to just let me mirror data directly to Iceberg, so I can plug in my own stateless query engine.

The issue that each of those providers above has recently adopted Postgres as a secondary product aimed at supporting their main product, an OLAP database or engine, so they don’t want you plugging in your own query engine.

I’d bet you’re likely to see this from a Postgres-specific provider first, like Supabase.

→ HN comments  ·  127 points  ·  117 comments

Top #1 — simonw:

Abuse by police officers and other individuals with access to these systems is the answer to the stupid but common question "why should I worry about surveillance if I have nothing to hide".

↳ reply 1 — tptacek:

I never hear that question from Flock supporters. The question I hear[†] is "why shouldn't we be pulling over stolen cars?". It's a much harder one to answer than the straw man questions we set up for these things.

(To be clear: I think the current generation of ALPRs is a policy failure and that major changes need to happen inside state law enforcement systems before they can be deployed successfully. But I also think people on HN have a real problem in convincing themselves about the obvious righteousness of their anti-surveillance arguments.)

[†] I heard arguments like these because I was deeply involved in getting the cameras disabled in the inner-ring Chicago suburb in which I live, getting the first ACLU CCOPS ordinance in Illinois passed there, and helping draft what are probably the most detailed ALPR police General Orders in any municipality, or at least were at the time. Which is a long and pretentious way of saying I spent a lot of time talking to people who are not Very Online.

↳↳ reply — AdieuToLogic:

>> Abuse by police officers and other individuals with access to these systems is the answer to the stupid but common question "why should I worry about surveillance if I have nothing to hide".

> I never hear that question from Flock supporters. The question I hear is "why shouldn't we be pulling over stolen cars?"

These are the same Constitutional question.

In order to determine what cars are stolen, all cars must be inspected. If all cars are inspected, there is a reasonable case that the 4th amendment is being violated:

  The right of the people to be secure in their persons, 
  houses, papers, and effects, against unreasonable searches 
  and seizures, shall not be violated, and no Warrants shall 
  issue, but upon probable cause, supported by Oath or 
  affirmation, and particularly describing the place to be 
  searched, and the persons or things to be seized.[0]

> But I also think people on HN have a real problem in convincing themselves about the obvious righteousness of their anti-surveillance arguments.

See above.

0 - https://constitution.congress.gov/constitution/amendment-4/

↳↳↳ reply — tptacek:

They're not remotely the same Constitutional question. The people asking this aren't saying "I'm fine with being profiled because I've got nothing to hide unlike the people driving stolen cars". They're just as upset about abuses of these systems as you are (though: some of them think these cameras are no more intrusive than cell tower tracking). They just see an upside to the system that privacy advocates refuse to recognize.

I get that this sounds like standard-issue message board rhetoric where I'm Mary Sue'ing normies to make my points for me, but I don't agree with my neighbors about this stuff, and you'll probably have to take my word for it that I'm not making any of this up.

↳ reply 2 — veec_cas_tant:

Maybe for some, but I honestly don't find it very persuasive. The misuse of license plate data wouldn't make my top ten list of potential powers that law enforcement might abuse. Clearly, abuse of any kind of power should not be tolerated, but potential benefits must also be considered. I think the widespread deployment of cameras at businesses and homes provides strong evidence of their benefits. The idea that a crime happens outdoors and the proper course of action is for police to knock on 100 doors to collect grainy doorbell footage is pretty inefficient, to say the least.

Top #2 — tptacek:

Flock cameras appear to be one of the few things that can get police officers fired.

↳ reply 1 — xnx:

Silver lining! The cameras are catching criminals.

↳↳ reply — shermantanktop:

To paraphrase Homer: the cause of, and solution to, a problem! In this case police misconduct.

It does make me wonder whether electronic evidence of police behavior will materially impact actual behavior over time. But given that the normal reaction is to deny access to evidence, I doubt it.

The people who want to be police officers tend not to be the people everyone else would want to be police officers.

↳↳↳ reply — parineum:

> The people who want to be police officers tend not to be the people everyone else would want to be police officers.

It's tiring for good people when everyone they interact with thinks they're the villain.

↳ reply 2 — edot:

Right, because you don’t want to let the little guys mess it up for the big guys: https://arstechnica.com/tech-policy/2026/05/fbi-seeks-us-wid...

Gotta make a statement with the low level grunts so the citizenry think these systems have accountability.

Top #3 — AngryData:

It should result in jail time. If I got access to government surveillance program and started using it for personal reasons, I would likely be charged with terrorism.

→ HN comments  ·  126 points  ·  36 comments

Top #1 — ianthehenry:

I like this website a lot and used it as the base recipe book for a program that will tell you what cocktails you can make with the ingredients you have, but also what cocktails you are “one ingredient away from.” https://github.com/ianthehenry/mixologician

It’s a little datalog script - input is a text file, output is a text file. I’ve used it a few times to decide what to pick up at the store!

↳ reply — ljf:

Love this - I'd wanted a little site that helps me see which of the top ~100 cocktails I already have the ingredients to make, and which I need to make a purchase to be able to sample: https://iba-world.com/cocktails/all-cocktails/

When I had more time and money, I planned to try to make one a week, across two years - but I rarely seem to drink these days.

Top #2 — jnwatson:

I like cocktails. I like cocktail recipes. I don't know why this is on hacker news thought.

↳ reply 1 — dsr_:

There ought to be an XKCD for this, but instead it's from the FAQ.

In short: somebody thought it was interesting enough to submit, and enough other people voted for it that it got to your attention.

That's it.

If it ends up with 2000 points and sits on the new page for 3 days, I'll be quite surprised, but strange subjects do that.

↳ reply 2 — stronglikedan:

> I don't know why this is on hacker news thought. (sic)

Because someone posted it, and others upvoted it, and it got enough upvotes while it was new to get recognized by the algorithm and promoted to the front page. Now you know, and knowing is half the battle!

Top #3 — chongli:

I’m a big fan of the Webtender [1]. A super old school website without style sheets. Tons of information, nicely organized, and everything loads super quickly, as it should be!

[1] https://www.webtender.com/

→ HN comments  ·  126 points  ·  89 comments

Top #1 — pixelesque:

> Live performance revenue reached $329 million, well above pre-Covid levels.

Is that inflation-adjusted? And as the article points out, very little of that is local talent, so it's likely some of that's due to the cost of tickets.

As someone who lived in NZ (although Wellington) for 10 years until last year, the local music scene was there, but at least in Wellington, the change in the 10 years I was there was considerable: The earthquake in 2016 progressively caused several venues on Cuba Street to close due to building regulations, and Covid was a significant shock as well, with some of the smaller (although most places in Wellington were small!) gig rooms / bars / breweries that would often host bands closing.

There was still Valhalla, San Fran, The Rogue & Vagabond and others, but it just didn't feel as much was going on other than for things like Cuba Dupa street festival, etc, or the Jazz weekends that would be held.

As the article points out, this is not localised to NZ, it's happening in Australia and the UK as well in terms of venues/pubs closing and live music not being quite the draw it once was because of that.

↳ reply 1 — Melatonic:

There's a venue in NZ called "San Fran" ? That's pretty cool

↳↳ reply — DinosaurArms:

First it was a massage/sauna place called The San Francisco Bathhouse, and then it kept that name when it turned to a live venue. Now it is just San Fran.

↳ reply 2 — rvba:

No idea how it looked in New Zealand, but after covid "ended"* there were a lot of people who simply wanted to go out - after spending so much time indoors.

I suspect that people went out more in 2023-2025, but then they reversed to the mean + economic impacts (of money printing everywhere) hit.

* by the way, did covid officially ever end? the new variants seem to be still there, but it just stopped being covered by media after Russia invaded Ukraine)

Top #2 — gerdesj:

This para is a salutary lesson for all of us:

"This is not a New Zealand quirk. It is the local edge of a global collapse. Pitchfork, the most influential music publication in the world, was folded into GQ in 2024 and its editor-in-chief was laid off. Local music press has been gutted in the US, the UK and Australia. New Zealand just got there first and hardest, because we were small to begin with."

Aotearoa (NZ) has fired up the four way flashers. We ignore it at our peril.

↳ reply — AloysB:

It's not just X, it's Y.

↳↳ reply — zapzupnz:

It's not just N, it's Z.

↳↳↳ reply — foo42:

But what are the questions nobody is asking?

Top #3 — gadders:

When I went to NZ about 30 years ago, vibe-wise Auckland was like LA and Wellington was like Seattle. Not sure if that comparison still holds.

→ HN comments  ·  131 points  ·  115 comments

Top #1 — woodruffw:

I think the advice at the end of this post is essentially sound: you should pre-filter your evaluation of others’ opinions through their understanding of related topics. But that isn’t elitist, and I think we shouldn’t call it that! It’s a form of media and cultural literacy we should be encouraging people to have; there isn’t (and shouldn’t) be anything elite about it.

Top #2 — mxmilkiib:

DUCK DUCK GO MURRAY BOOKCHIN!!

https://theanarchistlibrary.org/library/lewis-herber-murray-... (1965)

https://theanarchistlibrary.org/library/murray-bookchin-post... (1971)

https://www.nybooks.com/online/2018/06/15/how-my-fathers-ide...

Top #3 — TrackerFF:

For the past few years, I've seen a dozen of friends or acquaintances turn "ideologically crazy", and they can be roughly grouped into:

A) Onset of schizophrenia or other serious mental illness, followed by psychosis. This is actually the most common I've seen. But it is also the easiest to spot.

B) Some crisis event in life, followed by finding / joining a community. Some were hit hard by this during COVID.

I've yet to personally see people that have their, well, shit together and live busy lives turn into ideological crazies or extremists - where there is no mental illness. Of course, it could be that they just hide it better. Don't use their public accounts, etc.

I also think it is much easier to become radicalized today. Once you search for any extremist views on say social media, you'll be bombarded with it. Just belonging to a certain demographic could trigger a non-stop barrage of extreme wing influencer content in your feed.

↳ reply — 121789:

Don’t agree actually. Have seen many busy working professionals fall into “extremism” and it’s pretty much always the same. They have a few issues they are firm on and align with one party, they don’t like the other side because of this, then they only hang out with people aligned with their views and only consume media aligned with their views. Quickly turns into lunatic talking points, and I’ve seen it happen to both the left and right.

→ HN comments  ·  125 points  ·  174 comments

Top #1 — dzonga:

Poland has made a lot of good moves lately. for techies in the UK/US they might not care that much - but there's also an effect on you.

in the U.K - a polish skilled workers are choosing to remaining in Poland instead of moving to the uk.

for the US - a lot of firms are setting up polish offices - hence when you hear 'a.i' efficiency most of those jobs have moved to Poland or India. Whether it's big tech, big banks and small funded VC startups. What remains in the US are just senior+, staff software engineers.

↳ reply — elAhmo:

This is not viable long term, but cool to ride the train while the outsourcing lasts.

Top #2 — petcat:

I saw an interesting stat recently that the poorest USA state, Mississippi, has a larger economy and GDP than 1/3rd of all EU members.

It just seems like the EU is too top-heavy and I wonder if that's where a lot of the discontent comes from. It seems like there a lot of micro-nations in eastern Europe that have outsized influence over the union as a whole.

↳ reply 1 — Epa095:

15-20% of Mississippis GDP is healthcare cost.

Gdp is not really a great measure of how well an economy works for it's inhabitants, if it was we should all spend our time breaking each others windows.

↳ reply 2 — notfromhere:

That example is a good reason nominal GDP has limits on what it can measure, as a drive thru Mississippi would show a poor state with small pockets of wealth

Top #3 — vivzkestrel:

- i came expecting to read something about how taking no immigrants has helped or something along those lines

- has there been any studies conducted on how 10% immigration affects a country's economic outcomes?

↳ reply — profstasiak:

But its not true! We had like massivie migration in last few years, mostly Ukrainian though

→ HN comments  ·  125 points  ·  82 comments

Top #1 — pbhjpbhj:

It seems to be an assumption at this point that the massive increase is of frivolous suits/applications.

Perhaps it's more a democratisation? Maybe, there were many worthy tribunals before but the prospective claimants weren't able to access legal support?

I'd guess it's a bit of both.

Top #2 — mwkaufma:

Tellingly, the legal tendency The Economist finds alarming involves workers asserting their rights.

↳ reply 1 — stuaxo:

The original Tragedy of the commons was largely made up

https://www.hamptonthink.org/read/the-myth-of-the-tragedy-of...

↳↳ reply — js8:

I am not surprised. Contrary to popular belief, free markets nor capitalism do not resolve tragedy of the commons; they just hide it in (negative) externalities.

The biggest example is probably the CO2 emissions and global warming, but also the social costs of economic outcasting - making people jobless, homeless or otherwise stressed due to poor distribution of resources.

↳ reply 2 — andersmurphy:

Anthropic is a sponsor of The Economist. So maybe that influences their retoric somewhat.

Top #3 — jackvalentine:

Even the “pre-case” legal stuff is out of control. In my Australian organisation we’ve noticed a deluge of what I can only call “privacy demands” that talk the talk… but have the jurisdiction or subject legislation wrong.

But AI told them they could demand these things (even if it would be illegal for us to comply! Think KYC stuff.) and they don’t have the skill to criticise the AI or know when to stop.

↳ reply — maccard:

Yep, and the cost is borne by the receiver of the letter. The person sending shitty LawyerGPT messages will just keep doing it until there’s consequences, but there aren’t any.

→ HN comments  ·  508 points  ·  288 comments

Top #1 — 8bitsout:

I thought LLMs were a great tool for learning new topics - perhaps even complex ones. But overtime, I have had several frustrations with this. First, I get exhausted reading LLM prose. I really don't want to read anything generated by something like Opus 5 at this point. Second, as I dive deeper, I need a way to organize the information in a useful way as I begin to branch out in many different directions. I have tried to use the LLM to fix this by having it generate a web page with diagrams and organized information flow. It's an improvement, but I still run into the issues I described in my first pint - LLM prose is annoyingly dense, and the useful information gets lost in a bunch of noise. You can direct it do something like "use plain English and avoid LLM prose - provide only as much information as necessary to demonstrate the point", but it is once again only a marginal improvement.

And then I begin to think to myself that I should just read a book on the topic written by a trusted source who put a lot of effort into teaching the topic properly and presenting the information in a thoughtful way. So, I am back to books and mostly try to use LLMs to clarify certain questions or ideas I have.

↳ reply 1 — matheusmoreira:

> I really don't want to read anything generated by something like Opus 5 at this point.

Recently switched to OpenAI and I've gotta say Sol is so much better at writing than Claude. Opus has a distinctive sentence structure and Fable somehow manages to be even more obtuse. The personality of these models really does come through...

↳ reply 2 — rDr4g0n:

LLMs can't "read the room" and infer how much context the audience already has, so they try include everything.

human conceptual thinking is very much a multi-dimensional graph, which relies on light "approximate" concepts that are "good enough". LLM AR token generation is extremely one dimensional and doesnt care about the "weight" of the concept behind a token.

LLMs hold billions of parameters in "mind" at once. humans hold like four "concepts".

This is the essential mismatch and the primary reason LLM conversation can be so painful and exhausting.

Explaining this and limiting "concepts" to four at a time tops is one of the very few AGENTS.md / system prompts I always use, and it has proven invaluable time and again.

Thinking traces show how effective this is at forcing the LLM to simplify its thinking.

[edit] Also, myself and nearly all of my peers are struggling to choke down the flaws of LLM tooling along with the benefits. the speed at which LLM adoption is being forced, without truly crafting them into quality tools first, is not ok, and not normal.

LLMs have stirred an inhumane hunger and fear. the tech is fine, but the way tech companies (creators and consumers) are behaving should be deeply questioned.

it's NOT normal. it's not ok.

↳ reply 3 — copperx:

It's much better to feed the book to the LLM and ask questions as you read along, instead of asking the LLM to basically write a custom book for you from scratch.

Top #2 — wxw:

> What you get is a beautiful animation that is 100% accurate and free of hallucinations.

I'm not sure I follow how this is actually guaranteed? The fact-checking process mentioned just seems to involve asking AI to review its own work.

↳ reply 1 — hn_throwaway_99:

Completely agree. While I didn't set things up to have AI review its output in a loop, my experience trying to build a specific acoustic testing rig with Opus 5 also aligns with the other "it's turtles all the way down" comment.

Opus 5 first built me a detailed plan, but a couple important details were either obviously wrong or felt unnecessary. I went back and forth asking for sources and more information probably like 4 times and every time it did the "in looking at things in more detail it appears my previous advice was incorrect" spiel. It just became exhausting at some point because it feels like it really lays bare how LLMs are just minimizing that loss function but don't actually "understand" anything. It was really useful as a search engine (it correlated some highly relevant source docs), but I just couldn't trust it to believe it was actually done at any step.

↳ reply 2 — bryanrasmussen:

I could certainly envision a scenario whereby review would increase reliability but not how it would every guarantee 100%, there is a pretty big logical gap there.

Top #3 — gnarlouse:

I've had success using the socratic method. I give Claude some topic (say, how the intricacies of the bond market works, the content from which are screenshots of pages from a textbook), and then I go on a walk chatting with it in voice mode. Claude is the expert, I am the student. Claude asks me questions, leading me to an answer logically. I come back with questions, and we back and forth. LLMs arent like they were in '23-'25. I'm almost always skeptical its going to lie, and almost always wrong.

In particular:

- I limit it/encourage it to give me single sentence questions

- I sometimes will ask it to tell me a motivating, human-grounded story, when we're starting a new concept: claude responds "Maya is a bond portfolio manager, and her boss has asked her to quickly price in what happened if yields go down. She knows her bond's average duration, a measure in time, but she doesn't have a percentage, which is what her manager wants. How can she give him a percentage number with just a duration figure and the proposed new yield?"

- I'll often ask claude to let me work through it, to derive the thing myself, often resulting in a string of thoughts with "yes/no" trailers, to get the LLM to reply yes or no only, and avoid derailing my train of thought. If yes, my train of thought keeps going. If no, I've got something wrong.

- I'll sometimes stop and have it craft an artifact. I typically say "build me a Brilliant.org-style interactive demo of the topic", especially when we get into the realm of looking at the actual maths of a thing (for which prose and dialog is not optimal by itself AFAICT)

- I'll do this while I'm traveling, while I'm walking, while I'm doing chores.

It's so much fun.

↳ reply — pumanoir:

can you share ane example of one of your prompts? I have tried to convert some text into the socratic method and haven't succeeded yet.

↳↳ reply — gnarlouse:

It's a stupid word soup on second read, but hey, it works, and that's all that matters with "stochasty the parrot."

""" In this project, I require a socratically delivered line of conversation. Here's the typical structure to the conversation. I ask some question. You need to factor and reason about how to conduct and deliver a conversation. Best practices would be to limit terminology, or assess with the user whether they have a firm grasp on terminology before you use it. You must be very strict about this, it's unacceptable to just introduce a new concept, actor, phrase or other complication into the conversation without first labeling who what or why it exists for the conversation.

Conversation structure needs to be front-loaded with a brief interview for the user, "you understand X?", "whats your understanding of Y?".

Conversation structure then needs to proceed with single-sentence questions from the agent. User replies with an answer. Sometimes the agent needs to correct the user, but only ever do so with yet another question. """

^^ these are the instructions I have installed at the root of a "project".

Keep in mind, this is claude opus 5 low effort we're talking about, in the "projects" area of the mobile app. Here's the process I use to set up its knowledge:

1. I take screenshots of the textbook on my iPhone, and upload a chapter at a time.

2. I have it summarize the chapter into markdown by analyzing screenshots. You could probably achieve this simpler, if you just had the textbook in PDF.

3. I walk with my boy Clau-crates.

I've done this for a couple of weeks and haven't seen it revert back into its typical context-dumping behavior.

On second read, there's probably some clean up I could do. Thanks for making me pull it out and look at it. Things that could probably be improved:

- tell it to cross-check resources online to further ground itself

- use simple, short sentences (long sentences make the brain blur a bit)

- not be sycophantic (it seems like project mode has discarded with my root-level anti-sycophancy prompt)

→ HN comments  ·  128 points  ·  102 comments

Top #1 — numlock86:

I got rid of mine for a year thanks to a comment on HN. The only reason it came back is because of my lack of discipline. It's bearable as I only notice it when I actively listen for it now. Anyway, I am sure this would just work again.

So there was a comment that claimed it was related to bad neck posture. Being a keyboard warrior in my late 30s at that time I suffered the same issue: Tinnitus and generally bad posture. I figured that even if that wasn't true, working on my posture might be a good invest either way. So a couple of YouTube videos later I started working on it. Turns out the process to straighten your neck - even after like three decades - with just some casual invest everyday only takes about a month or two. To my surprise the tinnitus was actually also gone.

As I haven't build a real habit around the exercises eventually a bad posture came back, and so did the tinnitus. Both are generally not as bad as they were before, though.

10/10, would do again.

↳ reply 1 — codethief:

Agreed, bad posture (not just neck but in general) can cause both tinnitus and RSI-like symptoms in my experience and I have been able to tackle both with proper exercising.

Where I don't agree is the following:

> Turns out the process to straighten your neck - even after like three decades - with just some casual invest everyday only takes about a month or two.

While you can make progress quickly (in the sense that your symptoms become less severe), fixing your posture usually takes much, much longer than that. In fact, if we're being honest here, it's never a one-time fix that you do for some $TIME and then stop doing. The reason being that forces & activities that cause or exacerbate bad posture (sitting, carrying a back pack, sleeping on the side, various sports, doing no sport at all, wearing shoes with a high heel drop, …) are not a one-time thing, either. They act continuously, and so "fixing your posture" must be a continuous life-long activity, too.

↳ reply 2 — tomtom1337:

Could you explain more about what "straightening the neck" entails in terms of exercise? I'm interested.

Top #2 — cogogo:

Over the years I have seen a few threads on tinnitus. I have it very mildly and doesn’t bother me much. On the other hand I have had “floaters” in my eyes since I was a teenager (probably just when I noticed them). Until I saw a doc I thought I was going blind but apparently they run in my family. They are like visual tinnitus and my god do they drive me crazy. I find wearing sunglasses helps me ignore them but they can be brutal. I am a big skier and in flat light they drive me bonkers. Not sure I have seen these discussed here but maybe I missed it.

↳ reply 1 — arkh:

Called visual snow. Many people who suffer from it also have tinnitus.

And they're not "floaters" but look like old school TV statics. Easier to see over plain color (blue sky, night time etc.).

https://en.wikipedia.org/wiki/Visual_snow_syndrome

↳↳ reply — ubercore:

Visual snow is different from floaters, though.

↳ reply 2 — avian:

There's a laser treatment for floaters now (laser vitreolysis). It's not always applicable and it has its own risks and side effects.

Top #3 — Sindisil:

It's almost certainly my own damn fault I have it, but after at least 35-40 years dealing with it,tinitus sure as hell isn't my friend.

Mine is high pitched, and varries in intensity but has gotten worse lately. Sometimes it makes jus existing down right hard, but usually it's only moderately annoying and I can mostly ignore it.

I've tried an assortment of sound based methods to mask it or train my brain to tune it out, but nothing has worked more than fleetingly.

The only thing that reliably gives me a break is if I can really focus on something. Music (passive or active), cooking, and programming are the most reliable, but sometimes reading, gaming, or good conversation works. Usually, though, everything is done with a constant backing track of flyback transformer whine.

Wear hearing protection, kids, or you may well live to regret it!

↳ reply 1 — stndef:

Even if you have tinnitus, wear hearing protection for sure.

Unfortunately, I have a connective tissue disorder, which is the cause of the tinnitus and the cause of my hearing loss as well. The more my hearing loss goes, the more apparent the tinnitus is. IF you're in a position where it isn't a health related issue, you can prevent it from getting worse, or at least reduce the pace that happens at.

I have an orchestra of it going on with the high pitches, those that sound like a vacuum going off in another room and hearing my own pulse. I do miss silence!

↳ reply 2 — fowlie:

Amateur musician with mild/moderate high frequency tinnitus here. I was treated with exposure therapy, and while I hear the sound (less now than before) it doesn't bother me any more. I would recommend a book called "Living Well With Tinnitus". Good luck!

→ HN comments  ·  125 points  ·  70 comments

Top #1 — resonious:

Not surprised at all but there seems to be tons of these kinds of tools and I'm having a hard time picking one. Is there a nice list somewhere? I guess I can ask chatgpt...

Honestly right now I do a lot of development from my phone - I don't have the time to pull out my laptop very much. Right now I use hermes agent and just chat with it on Slack. But sometimes I wish I had easier access to more advanced features, and Slack isn't the best UI for this stuff. So I want to use something like this but am having analysis paralysis.

Top #2 — azuanrb:

I prefer Paseo https://github.com/getpaseo/paseo, mainly because I have specific preferences for different harness + model combinations. For example, I like using ChatGPT models via pi, and GLM via Claude Code.

If you’re happy with OpenCode as the harness, OpenChamber is great. But if you prefer using different harnesses under the hood, Paseo is a better fit.

I installed it on my homelab and can access the same sessions from my MacBook or iPhone at any time. Been really nice to use so far.

↳ reply 1 — haint_:

As someone who uses both Claude Code and Codex, having a unified interface while still being able to use their respective harnesses is a godsend. I've tried a lot of alternatives, and nothing even remotely comes close to the level of polish this one has. The session sync is buttery smooth as well, it's even better than Claude Code mobile I would say. Thanks so much for the recommendation.

↳ reply 2 — robertn702:

+1 for Paseo. Switched over to it a few weeks ago from conductor.build then superset.sh. The mobile app plus Hetzner VPS has been really enjoyable in terms of just firing off random tasks and not being stuck at my desk.

Top #3 — tandr:

Somehow reminds me of JetBrain's new Air [1] platform (still in beta I think) - make multiple agents working, monitoring, basic code and diff viewing.

[1] https://air.dev/

→ HN comments  ·  176 points  ·  661 comments

Top #1 — taylorfinley:

I got laid off at the end of April and have been building my own personal dream tool since, it's a skeuomorphic carpentry simulator with an agent MCP. You start with real wood specs and a shop full of miter saws, router tables, table saws and the like. Think of it like Sketchup but no extruding rectangles.

It's been really fun to build with. Agents can create parametric procedures so e.g. stick framing becomes a simple function call for the next agent, building a flywheel. There is full agent-human parity in every surface, so there is a human UI for the procedures too.

Every operation is yaml, so agents pick it up pretty quickly. Agents can also file feature requests when they reach for a tool that doesn't exist. You can export to 3d printing files, video walk-through, step files, get a BOM and a cut plan, view your project in life-size with AR. There is a notion-like interface for authoring build guides... a lot going on. If you're a woodworker/software engineer you might enjoy it.

You can add the MCP to your agent of choice, send a picture, description, sketch or whatever; say "build this with Sawdust"

https://sawdust.diy

↳ reply 1 — matthewfcarlson:

This sounds a lot like a project I’ve started a few times and never made it far with. I called it woodlang. It was the idea of CAD but rather than designing the shape and figuring how to make it that shape, you specify a series of shop operations like blender geometry nodes.

This was all pre AI. Super impressive that you’ve made something awesome. I’ll try it out tonight

↳ reply 2 — appplication:

Love it, I had a similar but probably worse idea a few years back while woodworking. I agree the issue with essentially all CAD and sketchup etc is it forces you to work with primitives that are different from your real world ones. Extruding shapes as you said, instead of actually milling lumber and working in common known dimensions.

I also suspect there’s a fairly finite number of sane join primitives, you can have rabbets, dados, dominos, dovetails, pocket screws, etc. These could be experimented with digitally against an existing design. And your tool availability could determine what you can do.

And of course cut order and assembly is just a DAG and can be generated from geometric constraints of the actual design.

↳ reply 3 — a_c:

Hey, looks like we both have a thing with 3D, love it! Mine is about bouldering, check it out https://news.ycombinator.com/item?id=49234599

Top #2 — Bnjoroge:

Lots of fun projects being listed. I'm working on Preloop(https://preloop.dev), a way to run your unmodified Github Actions locally or self-hosted in isolated cross-platform microvms(uses smolvm which uses libkrun vmm).I got frustrated with Github Actions reliability and inability to run or test my workflows locally so i re-wrote the runner and reverse-engineered the control plane so it follows the exact runner protocol the offficial runner does. Each job runs in a microvm that boots in 400ms from a packed image. Also supports a way to pause on every failure so you or your agent can immediately shell at that point, fix and retry(in a new environment) to verify. Also supports the full DAP protocol so you can interactively step/forward/investigate context at each step/job.

Try it out: https://github.com/preloopdev/preloop

↳ reply 1 — mcapodici:

That is an amazing idea. I can imagine this getting popular. It could be used for faster local dev loops and porting GHA to other servers. I imagine for data/compute residency, cost, performance, vm types etc.

↳↳ reply — Bnjoroge:

Thanks! I'm glad it resonates. Still pretty beta, but I'm hoping this can be a good alternative when Github goes down, which unfortunately has been a bit too often, with as minimal of a lift from the user side. And of course, making it easier for agents to fully drive ci.

↳ reply 2 — SOLAR_FIELDS:

a bold attempt, how does it compare with https://github.com/nektos/act which has been the de-jure project in this space? Act has never been in my opinion super great, but it's also been several years since I've checked in to revisit the project.

Top #3 — lappet:

I have been working on and off on https://www.whichtemple.in

- it is a collection of ~ 2700 temples across India, with images sourced from wikipedia

- it shows you a random temple each time you visit

- it has a map view of temples across India

So this is just the tip of the iceberg. I am trying to build a catalogue of all temples, starting with India, and expanding on to the rest of the world. The focus is on temples originating from Indic religions like Hinduism, Jainism and Buddhism.

I have personally discovered many cool temples that I have never heard of, but I am struggling on how to gamify this. Feedback welcome!

↳ reply 1 — unsungNovelty:

Would love to filter this by states so that I can go to the interesting temples in my state first before trying out other places. Would make it more functional.

↳↳ reply — lappet:

Yes, definitely on my list. For now you can try the map view and zoom into your state. Thanks for the feedback!

↳ reply 2 — ViscountPenguin:

A fun game would be to quickly guess the religion of a temple? Could familiarize people with the common iconography of the various dharmic religions.

→ HN comments  ·  179 points  ·  82 comments

Top #1 — aliasxneo:

“Our theoretical framework captures how entrepreneurs facing minor, wide, and extreme expectation-reality gaps engage in evermore sophisticated efforts to detach the venture’s externally projected appearance from its actual operational reality.”

Look, I’m not promoting fraud at all, but having been doing seed raising for the last eight months, there have been many times where I thought the only way to compete was by fudging the numbers (because everyone else is, basically). It’s one of several reasons I left this game and am pursuing non-traditional means of funding now.

↳ reply 1 — sedawkgrep:

Thank you for having an ethical standard and the will to adhere to it.

↳ reply 2 — m3047:

I spent a few years hanging around the VC-adjacent scene in a major city, met quite a few of them (because I'm that way. elevator docent during events... priceless).

There is publicly performed theater around VCs which is pretty consistent, taught in incubators, etc., etc. So that's how they manage set and setting (grooming) for the marks. Internally there are certain tells, that they must act a certain way. There are arbitrary character tests and whatnot which provoke a "culture" if you will, similarly to how computer languages and tech stacks form distinctive cultures if left unchecked; there could be self-selecting pressures (for these behaviors) as well. This is unsurprising to me as someone who analyzed voting patterns, and was asked numerous times variations on the question "is it something in the water or 'new car smell'?" (I don't have an answer for that question.)

However VCs are much more stovepiped and cliquish concerning their actual selection process than people realize, I've heard it publicly described as "mafia like" by someone who I worked for and whose name many would recognize. I don't understand why this person, like so many others, invested so much effort... enough to get some serious butthurt IMO.

It's a distraction, it effectively destroyed a different company I was working for.

↳ reply 3 — pcurve:

I'm guessing investors understand that there's high probability of the numbers being 'optimistic'.

Top #2 — firasd:

A lot of 'numbers' startups cite are basically fake I think. Like if someone says we have this many users as a statement to TechCrunch you have no idea what they are actually calculating

But there is a clear line that gets crossed if you start actually making a database of millions of synthetic users and that's what happened with 'Frank' that sold to JP Morgan and eventually the founder was prosecuted

↳ reply 1 — abixb:

>"A lot of 'numbers' startups cite are basically fake I think."

I like Ed Zitron's reporting on the AI industry (though I disagree with him on AI's potential capabilities).

I wonder how much of AI-related revenue are because of accounting fiction rather than actual cash.

Either way, I think the stock market is as disconnected as it has ever been with actually ground reality of the US economy and industry.

↳ reply 2 — mindtricks:

Not just users, but number of companies. So many startups list out all the companies using their products, who also happen to be an army of startups. I doubt they're all using each others products, just doing what they can to juice each companies perceived metrics.

Top #3 — neya:

I'm surprised there was no mention of Elizabeth Holmes. Whenever there is mention of criminal deception of any sort, she's like the poster child for it in my eyes. And I remember her accomplice who instead of admitting he was on the wrong side, kept blaming the journalist who tried to uncover the fraud, instead. He said (something along the lines of) "He (the journalist) kept coming at her.." as if she would've been able to magically solve the problem if she had enough time.

That was all I needed to know about what was wrong about valley culture.

↳ reply — firasd:

Seems they use the nickname 'ScaleX' to discuss Theranos

→ HN comments  ·  153 points  ·  114 comments

Top #1 — cortic:

Taxi Drivers: The mean age at death is approximately 67.8 years.

General Population: in the same dataset, life expectancy averaged 74 years.

The average age at which patients are typically diagnosed with Alzheimer's disease is 79.

People in these jobs don't live long enough on average to get diagnosed, at the same rate. The same effect will happen in any job that lowers your life expectancy.

↳ reply 1 — xvedejas:

And how would you control for the causal possibility that those who will eventually develop Alzheimer's early on might tend to have undetectable memory issues that prevent them from becoming taxi drivers in the first place? You can't control for profession, since that's behind the very claim itself.

↳↳ reply — cortic:

Its even worse than that, i can imagine people with undiagnosed early Alzheimer's self selecting for car accidents and cardiac issues through additional stress of mental fatigue while their brain tried to compensate for loss of function.

So you have people self selecting out of the profession, self selecting for early death within the profession, and on top of that, a general life expectancy discrepancy that is near impossible to adjust for.

↳ reply 2 — specproc:

Studies typically use regression to account for stuff like that. CBA to look at the article, but I bet any money that's controlled for.

↳ reply 3 — glitchc:

They are likely studying individuals that retired without getting fatally stabbed or shot on the job.

Top #2 — intellix:

I wonder how it will look when we start doing statistics on gamers, if AI doesn't destroy our brains first. Then it would be interesting to plot the different death by genre like for example strategy, FPS, puzzle, etc. Someone else asked about Chess players which is probably also related right.

My grandad essentially died of alzheimers and he just loved to sit and watch football all day where you don't really need to think much.

Top #3 — walthamstow:

> In one landmark 2000 study, neuroscientists compared the brains of licensed London taxi drivers with those of people who did not drive cabs.

Location is very important here. London cabbies have to pass The Knowledge which is an extremely difficult memory exam that only people with particular kind of brain can do.

↳ reply 1 — graemep:

The first study referenced is in the US, and across 443 occupations.

↳ reply 2 — cromka:

> that only people with particular kind of brain can do.

That's yours speculation, isn't it?Because if so, it would corroborate the same conclusion, i.e. if you exercise your memory regularly, you are less likely to develop Alzheimer's.

→ HN comments  ·  230 points  ·  181 comments

Top #1 — steveBK123:

I don’t even think most of it is deeply philosophical.

They just got uberrich and now worry more about the concerns of the uberrich than the concerns they had when they were normies.

Rule of law, functioning government providing services, functioning public commons, free press, freedom of speech for all, any social safety net, etc don’t matter when they are so far removed from the public they can just go buy yourself an island/citizenship/president.

Suddenly small differences in marginal tax brackets, antitrust enforcement & limits on the ability to simply buy the government they want are the biggest encroachments on freedom they feel.

↳ reply 1 — swiftcoder:

> I don’t even think most of it is deeply philosophical.

Part of the problem is that a lot of it is shallowly philosophical. I used to attend Zuckerberg town halls, and as smart as the guy may be, it's pretty evident that he dropped out of Harvard after 1st-year philosophy

↳↳ reply — lordgrenville:

Lucky we have the counterexample of a tech company run by a guy who did a humanities PhD under Jürgen Habermas...Palantir.

↳↳↳ reply — taurath:

In his case, it could be called opposition research

↳ reply 2 — Avicebron:

I think part of the problem this go around is that the information landscape has changed so much in ~30 years, that we are still trying to grapple with how to navigate this new terrain. If information and technology is causing everything to accelerate, we need new democratic structures to properly counter the perverse incentives of capital to compound it's power endlessly.

↳ reply 3 — petermcneeley:

Clarification. None of the silicon valley elite were ever "normies". I would guess most of them come from the top > %1 of society. This is just a rival elite not a non-elite.

Top #2 — yoyohello13:

In the US at least, we've been brainwashed to think that government, by definition, must be bloated, corrupt and inefficient. Believing this axiom, people vote to hand over every possible service to private enterprise.

Now we are in a situation where a handful of unelected business men have more power over our lives than the elected government.

↳ reply 1 — Avicebron:

We've let the government get bloated, inefficient and corrupt by allowing private enterprise to have an outsized effect on it. Private enterprise can hire people to live next to DC and spend every waking minute pushing their agenda on politicians. Private enterprise can funnel huge amounts of money into campaigns for their preferred candidates. It only benefits the corporations when we have ancient, technically illiterate politicians jockeying their favorite hobbyhorse policies while throwing up their hands at how "the other side" is blocking them. It's a designed deadlock.

↳↳ reply — sharts:

To some degree CIA also infiltrates private orgs to manipulate from inside out. It’s like a private-public partnership /s

↳ reply 2 — wdr1:

> In the US at least, we've been brainwashed to think that government, by definition, must be bloated, corrupt and inefficient.

I don't know that I've been brainwashed, but living most of my life in Chicago & Los Angeles, I've seen a lot of bloat, corruption, & massive inefficiency.

Many of friends lost their homes in the Palisades & Altadena fires. If it weren't tragic, it would be laughable the challenges they've faced in rebuilding. Several have kids about to enter high school. A dilemma they face is should they both building a house for their teenage children or assume they'll be off to college -- knowing it could be 3+ years until they're have a permanent home again.

In Chicago, I watched the mayor sell of all the parking meters for a fraction of their value. In Los Angeles, I'm watching them struggle to complete the people mover at LAX, a project years later & billions over budget.

This doesn't make me anti-government.

I see the need for it & I want it work well.

But comments like this make me frustrated that people would rather close their eyes & try to gaslight people, rather than acknowledge the problems & talk about how we fix them.

I feel our country is split between one side that doesn't want the government to do anything, even when it works, and another side that wants the government to do everything, even when it fails.

It's not clear to me where someone in the middle goes.

Top #3 — RunSet:

> “But what’s funny about Musk is, the stuff he likes actually completely defeats and defies all of his political beliefs,” she said.

Musk is only the most recent in a long line of semiliterates to have latched on to the lingo of Stranger in a Strange Land while apparently never having read a sentence of Heinlein's prose. As a rule of thumb, those who say "grok" do not.

And it's always Stranger in a Strange Land for the messiah figure. Here, have a better Heinlein story.

https://archive.org/details/Unknown_v04n06_1941-04_PDF_unz.o...

→ HN comments  ·  323 points  ·  271 comments

Top #1 — SBArbeit:

Easy workaround for this:

1. Install uBlock Origin in Edge.

2. Start Edge, browse to MSN Weather.

3. Click the "Add an Application" button in the address bar to get a Start Menu icon for the page.

4. Delete the in-box Weather app icon.

Now you get the same Weather app in about 130MB of RAM, with no ads. It's not as nice as a native app, of course, but it's 1000% better than the useless ads and MSN feed that you can't block from the built-in Weather icon.

(Also, go into Widget settings and turn off "Discover / Microsoft Start feed". Same crap, different surface. Get rid of it.)

↳ reply 1 — fakwandi_priv:

> According to Windows Latest, the high memory consumption is due to the fact that Weather is not a fully native Windows application. Instead, it is essentially an MSN Weather web app built on Microsoft's WebView2 framework.

I wonder how this approach is having such an high reduction in memory usage compared to what’s stated in the article. I would assume both use the exact same WebView here.

The only explanation here that I can think of would be that the “Add an application” starts it under an existing Edge process which shares it’s memory instead of completely isolated.

↳↳ reply — smileybarry:

WebView2 has some memory optimizations across several WebView2 instances, so it should save memory once you have multiple WebView2 apps running.

↳ reply 2 — theropost:

waiting for a quick simple straight forward tool on github that takes care of this... Anyone?

Top #2 — ndriscoll:

My gaming PC that I built in January 2006 that IIRC I used to simultaneously run Battlefield 2, Trillian, Xfire, Thunderbird, and Winamp with a 1680x1050 monitor had 1 GB of total system RAM.

↳ reply 1 — Melatonic:

Think about all the performance you could have gained by instead running Trillian and Winamp on an old Thinkpad next to you with even less ram

↳ reply 2 — glouwbug:

Fun fact, one of the XFire cofounders won Carmack's Ferrari at a Quake Tourney:

https://en.wikipedia.org/wiki/Dennis_Fong#/media/File:John_C...

Top #3 — GuB-42:

Putting aside the fact that it is horribly bloated. Accurately measuring RAM usage is tricky, there are several measurements, and no "right" one.

It is clear from the article that what eats up so much RAM is not the weather app itself but the framework it runs on. There is a "Renderer", a "GPU Process",... eating most of it.

The thing that the task manager doesn't tell you is whether or not these are shared components. It may be that the 662 MB used by the "Renderer" is shared between many Windows components, so killing that Weather app may not reclaim as much space as you may hope, instead, it would require killing every user of the component, some may be core system apps.

In addition to the distinction between private and shared memory, there is also the distinction between actual RAM usage and and virtual memory. It is possible for a process to memory map a 100 GB file. If you look at the address space, it will take 100 GB more of virtual memory, even though it may be actually zero physical RAM, but it is not always zero either, the parts of the file that are currently accessed take up some space, which may later be reclaimed by the OS by committing the page to disk.

Even the most obvious "I do a big malloc()" kind of memory use is not that obvious, the OS can overcommit, put stuff into swap, use memory compression, etc... And it can do that even if the system is not overloaded, as to make more space for the disk cache for instance.

So seeing "1 GB" in the task manager is just a vague hint of how it may affect performance. And not all "task manager" tools give the same value for the same program (so Windows vs Mac may be misleading). "Process Explorer", a more advanced version of the Windows task manager can give a lot more details, with different values of memory usage depending on what you are looking at.

↳ reply 1 — ack_complete:

> The thing that the task manager doesn't tell you is whether or not these are shared components. It may be that the 662 MB used by the "Renderer" is shared between many Windows components, so killing that Weather app may not reclaim as much space as you may hope, instead, it would require killing every user of the component, some may be core system apps.

It is the other way around, shared memory causes Task Manager to _underestimate_ memory usage. Task Manager's default views report the process private working set, no shared memory included. This means that 662MB is the _minimum_ amount of memory commit that would be released by ending the process.

> the OS can overcommit

Windows does not allow overcommit by default. It may compress or optimize memory allocations to reduce the physical working set, but the kernel will start failing memory allocations once physical + swap is exhausted regardless.

↳ reply 2 — calmingsolitude:

> It may be that the 662 MB used by the "Renderer" is shared between many Windows components

From the screenshot in the article, this is the memory usage of the Renderer process spawned by the Weather App. I find it very unlikely that some other app (say, the Copilot app) can then piggyback on Weather Renderer process. Do you have a source for this?

> killing that Weather app may not reclaim as much space

Closing the weather app on my PC does in fact kill all child processes and frees up around 1GB of committed RAM. Are you not seeing the same?

→ HN comments  ·  199 points  ·  66 comments

Top #1 — sdenton4:

Here's a direct link to the video, to avoid the blogspam cruft...

https://www.youtube.com/watch?v=Co57SfcT-h0

Top #2 — emptybits:

His trebuchet launched a 4 gram projectile at 776 miles per hour.

That's obviously extremely lightweight for trebuchet payload, but for comparison, his projectile is approximately 62 grains so approximately twice as heavy as 30 grain high speed 22LR rounds. And approximately 1.5 times as heavy as 40 grain subsonic rounds.

↳ reply — drcongo:

What does grains mean in this context?

↳↳ reply — jihadjihad:

It’s a super old unit of measurement for mass [0], idea being that a “grain” is a single seed of wheat etc.

1 gram is a little over 15 grains, and there are 7000 grains per (avoirdupois) pound.

0: https://en.wikipedia.org/wiki/Grain_(unit)

Top #3 — tromp:

His dyneema driven drum is eerily similar to the dyneema driven drum (called "snek") https://rowingbike.com/nl/model/thys-222-revolver#content-7 on the Thys rowingbike https://rowingbike.com/nl/model/thys-222-revolver

→ HN comments  ·  149 points  ·  29 comments

Top #1 — torh:

Not that long ago I clicked on a link that Microsoft provided somewhere in Windows -- could have been the event log, I don't remember. I do remember it was for a specific support article, but it ended up at a generic landing page for something. I am not talking about a link from Windows 95, it must have been Windows 10. But regardless, it is all gone. Then again, it probably wasn't a cool URL to begin with.

It's amazingly many news sites that also seem to scrub their URLs every time they do a redesign. Also uncool.

↳ reply 1 — phillipseamore:

Got an MX4 yesterday and it has a printed card with "Go to logitech.com/mx to get started" resulting in a "Forbidden" error page. This product hasn't even been out for a year!

↳ reply 2 — phatskat:

I bought an air conditioner last year, some wal-mart “smart” AC that was affordable (we had to get three for the new house we moved in to so budget was a big concern) and it has a QR code for the companion app. I finally got around to getting the app this summer and lo and behold, it’s a dead link. Yay.

↳ reply 3 — Panino:

Mailing list archive software is a constant rug pull: today you link to a post about something relevant, and next month the link goes to a different message about a different subject.

Top #2 — mikepurvis:

Sadly, though:

    $ curl -I https://www.nsf.gov/pubs/1998/nsf9814/nsf9814.htm
    HTTP/2 404 
    content-type: text/html; charset=UTF-8
    content-length: 54737
    date: Sun, 09 Aug 2026 17:58:44 GMT
    strict-transport-security: max-age=31536000; includeSubdomains; preload
    server: Apache

Top #3 — firasd:

What this page doesn't mention is 301 or 302 redirects. SEO has made "old URLs staying live" more of a widespread concern than it was at the time. And WordPress etc ship with inbuilt redirects upon slug rename

So to a large extent this has been mitigated and not using the suggestion here, which is to create a permanent URL ontology upfront

That said eventually neglect, removal, reorgs (or simply websites going offline) still happens... but the way the suggested goal has been advanced is due to becoming a business priority and with redirects and CMSes as tech to help

That said one suggestion made here turned out to be very useful and indeed is the default in WP:

   http://www.w3.org/1998/12/01/chairs

If you use date as part of the taxonomy then -- as Tim BL says here:

> A reason for using a topic area as part of the URI is that responsibility for sub-parts of a URI space is typically delegated, and then you need a name for the organizational body - the subdivision or group or whatever - which has responsibility for that sub-space. This is binding your URIs to the organizational structure. It is typically safe only when protected by a date further up the URI (to the left of it): 1998/pics can be taken to mean for your server "what we meant in 1998 by pics", rather than "what in 1998 we did with what we now refer to as pics."

↳ reply — EGreg:

Sounds like most URIs have been downright uncool, then!

→ HN comments  ·  127 points  ·  85 comments

Top #1 — andai:

Well, it's unorthodox, but if we're sharing psychedelic visions... in 2018 I saw all of mankind marching together in perfect lockstep, atop an infinite uniform grey plane, to the beat of a central computer clock.

"They are going to offer you a chip. You tell them, no thank you!"

---

More broadly this strikes me as a new era of biology. (Genes -> Memes -> Temes) Straightforwardly then, we should aim for a symbiosis. (Symbiosis was Musk's vision for Neuralink, from what I recall.)

I've heard people say man hasn't yet transcended a zero-sum mindset (with rare exceptions), and that this is the attitude we're passing onto the machines. And if one of us has to lose... well, that's probably not going to go so well for us.

Therefore: symbiosis!

↳ reply 1 — RunSet:

If you like visions, perhaps you also like stories.

https://www.mediafire.com/file/htq0zr310se30v3/Last_Day.pdf

↳ reply 2 — eth0up:

I read Lilly's Programming and Meta programming in the Human Biocomputer as a young man. Left me with a lot of questions, etc.

That you replied to this post, indicates you may be familiar with Lilly?

Anyway, if you have any published writing on even vaguely related subjects, eg your comment, perhaps post a link. I will read it. If you weren't aware, he spent a fair amount of time with Feynman, who if I remember correctly, was fond of the isolation tanks. Some very interesting content from that era, especially before Sandoz cut them off.

Top #2 — trollbridge:

Is this a reference to C. S. Lewis's "Abolition of Man"? https://en.wikipedia.org/wiki/The_Abolition_of_Man (text available at https://www.basicincome.com/bp/files/The_Abolition_of_Man-C_... )

Top #3 — jacobgold:

> "By means not understood by Man..."

Man will likely be Man 2.0 by this point, whatever that ends up looking like.

→ HN comments  ·  499 points  ·  225 comments

Top #1 — pohl:

Fun and useful PR jargon to describe what this post smells like: a “limited hangout”.

(a damage-control strategy where an organization or individual admits to a portion of a secret or scandal after a cover story fails, while still hiding the most damaging key facts.)

↳ reply — Aurornis:

Interesting! I had never heard that phrase before. The words are non-intuitive, but that's exactly what's happening: https://en.wikipedia.org/wiki/Limited_hangout

Now that you've introduced the phrase I can see it in a lot of recent scandals.

Peter Attia, a popular health and longevity podcaster who was revealed to have substantial ties to Jeffrey Epstein used this exact trick recently. When the Epstein e-mails were released it was discovered that he continued to consort with Epstein after all of the revelations. Even worse, he talked to Epstein about the fallout from the accusations and continued to cozy up to him. He was trying to run a very expensive concierge medical clinic and relied on Epstein's connections to get more clients, so he wanted to continue keeping that relationship warm.

When the e-mails came out he issued a public non-apology that implied he had no idea about any of Epstein's activities. It was plausible enough if you never saw the actual e-mails and thought Epstein was just another contact in his network, but the actual e-mails revealed that he was continuing to converse with Epstein after the revelations, including about the revelations!

So "limited hangout" is precisely what he and his PR team tried to execute. It seems to have worked on a limited set of his fan base who still try to downplay his Epstein connections to this day. The e-mails are easy to find, though.

↳↳ reply — Nevermark:

> [...] a “limited hangout”.

>> The words are non-intuitive

Yeah.

"Incomplete mea culpa", "Incomplete admission", "Incomplete penance", "Incomplete contrition", "Incomplete remorse", actually say "the thing" instead of using a new random phrase.

I like "incomplete contrition" or "gas light remorse".

↳↳↳ reply — Aurornis:

> actually say "the thing" instead of using a new random phrase

Apparently the phrase is very old, pre-dating the internet. It came out of the CIA.

In this context hangout is a clipped form of "letting it all hang out"

So it's saying you're putting on the appearance of letting it all hang out, but the information revealed is limited.

Top #2 — cube00:

It's not much of a mea culpa when there's not even a mention, let alone an apology, for misleading John Gruber.

https://news.ycombinator.com/item?id=49228166

Top #3 — mechanicum:

I’d love to see the other repo, and the prompts that led to both.

Did Godier copy the other guy’s repo, name and all, and try to pass it off as his own work? That seems like the obvious conclusion, albeit brazen to the point of insanity.

Did Claude copy the repo, in response to a prompt? If Claude copies a Claude vibe-coded project, can any plagiarism really be said to have taken place?

Or is this just what Claude is statistically likely to spit out if you prompt it to build an astronomy app and leave it to it?

↳ reply 1 — Aurornis:

This is the Bluesky thread that Gruber linked where the original author calls out the copy: https://bsky.app/profile/mmmeh.bsky.social/post/3mslc3b3u4c2...

You can trace back to the original repo from that thread. The key points are that the copy had

1. The same name. One was "Dark Hours" and the other was "DarkHours".

2. The same highly specific bug. I didn't investigate closely but from the description the bug sounds too unique to have been identically created in a fresh Claude-style duplication.

The duplicated name is the most damning part to me. I've never tried to copy a project with Claude, but I can't imagine the sequence of events that would lead to Claude copying the name of a repo but going to lengths to disguise other things like stripping the license. This all seems highly intentional, like someone thought it would never be noticed.

↳↳ reply — mechanicum:

I’ve seen the original repo, I mean Godier’s.

A name clash is the least damning evidence in my opinion. Last year my previous employer asked us for suggestions for naming an AI agent. Everyone produced pretty much the same list, because everyone just asked ChatGPT.

Except me. I had also asked ChatGPT, but I bothered to google the suggestions and found there were already like half a dozen AI agents for every suggestion.

(they ultimately launched what was by then the 7th or 8th AI-related product with essentially the same name)

Creativity in naming doesn’t seem to be their strongest suit.

↳ reply 2 — kaycey2022:

If the copying was deliberate then why admit to the similarity? He should have just blocked the guy and continued as if nothing was wrong. Admitting the plagiarisation lead to a worse outcome every time.

→ HN comments  ·  125 points  ·  86 comments

Top #1 — j2kun:

What a dismal idea that a four-year-old device is non-functional and in need of "reviving."

↳ reply 1 — malfist:

Any device that powers off fully functional should be able to power on fully functional barring hardware failure. That these steps were nessecary is a stunning indictment of remarkable, even if your tech geek can ssh in and fix it.

After all, my notebooks are perfectly readable four years later

↳↳ reply — loughnane:

FWIW my RM2 is still running well after 4y, and that includes plenty of abuse. I use it for notes at work, keeping track of little league lineups (and so tossing it in the grass many times), and its still a daily driver for me.

↳↳↳ reply — malfist:

And if you leave it unpowered for a few years? I have a journal from two decades ago. Still readable.

↳ reply 2 — eli:

I mean…yeah…it’s a bug.

Top #2 — Tarsul:

I did not update my ReMarkable 2 for years either and had probably the same clock issue. There was an article on the FAQ website from remarkable, if I remember correctly, that took me to a step-by-step process by which I could update it. Don't remember exactly how (I think I had to use the cable and connect to PC; which normally is not needed) but let's just say even if at first it seems like it doesn't work anymore and you have to hack it yourself, it is not actually needed. Nonetheless, this hackabality is the main reason I got mine (although I don't hack it, the optionality and thus probable longevity made the difference in my buying decision. I do not regret.).

Top #3 — reticulated:

Manually updating a config file to enable the web server was a wholly unnecessary step as that's just a toggle in the "Storage" section of settings.

reMarkable software has its faults, but I keep coming back to them for the best-in-class hardware, developer-friendly platform, which itself spawned an active developer community. Could it be an easier platform to develop on? Sure. But that hasn't stopped a huge number of supported mods being made available

↳ reply — oskrim:

There is indeed a "USB web interface" toggle under Storage. I've added a note in the final section, thanks.

→ HN comments  ·  140 points  ·  120 comments

Top #1 — toofy:

we desperately need a separation of corporations and state the same way we have a separation of church and state.

we need a government that is antagonistic towards corporations that are abusing us.

it’s wild how much influence they have and how little pushback the government ever gives them when they clearly abuse us from every direction.

↳ reply 1 — w10-1:

> we need a government that is antagonistic towards corporations that are abusing us

Please don't. The entire civilized world for all of history and before has been built on the uncomfortable alliance between power/capability and governance/legitimacy. Protection-seeking is the primary support system for tyranny which is happy to induce the chaos they protect you from.

We just need enough negative feedback for any kind of abuse that it stops, because regulations have to be slow to be accurate and thus invite abuse innovation. The main part of that negative feedback is effective competition so people can vote with their feet.

So, ask not for antagonism but for real alternatives.

↳↳ reply — sillyfluke:

The second paragraph by itself is enough, the first is a tad hubristic and vague. The good things that got us here are not necessarily the things that will get us there, there being whereever we ideally want to end up...

>The main part of that negative feedback is effective competition

The comment you responded to allows for "corporations that abuse us" to include those that actively try to prevent effective competition. You don't really explain why being antagonistic to those corporations would be a problem.

↳ reply 2 — physPop:

but you don't have separation of church and state... thats a poor comparison. religious dogma eats into nearly all public discourse.

Top #2 — gsky:

https://archive.is/JxaZ7

↳ reply — k310:

How about Archive Buttons? [0]

I don't know what's going on behind those 5 buttons, but this was the first, and who knows?

[0] https://www.archivebuttons.com/articles?article=https://www....

Top #3 — jrexilius:

The original research project this company looks based on was very cool and way early: https://sandlab.cs.uchicago.edu/jammer/

→ HN comments  ·  128 points  ·  20 comments

Top #1 — yunruse:

I loved this article and its interactive elements. The potential field is an elegant abstraction which really elevates this from a math puzzle into something new.

I'd love to see just how Lipschitz continuous, how smooth, the potential field can be; how adding features puts it closer to or further from solutions that fit the consecutive-no-duplicate constraint, say. Adding a smooth 'hill' is probably viable; is a 'river'?

One little critique I have is in the latter third. Using LLMs for proof is pretty standard now, but the way the text focuses on their tribulations was distracting. It might have been cleaner to use the mathematician's "we" after introducing the 'co-authors', so that the casual reader might sink their teeth into the math rather than be reminded LLMs can sometimes cost money and go around in loops.

But otherwise this is a really gorgeous article! The visualisation is really powerful, and something about how the symmetries impose a kind of conservation (which looks like hot soup but actually has a smooth potential) are very exciting, and curiously very physicsy.

↳ reply — gukoff:

Thanks a lot for the detailed perspective!

The potential fields are a curious phenomenon. I feel like - and I haven't verified this in any way yet - the smoothness is dictated by the fact that all values are within a not-so-broad range. Then as you "peel" the hexagon from the outer layer using those 6-rings, building the potential field, each next layer inwards shouldn't changed too much.

I had also explored a case of n->inf where the sums turn into integrals, and consecutiveness turns into the uniform distribution. Then it's not so hard to find a solution. It's not in the article, because it turned out to be a dead end for solving the discrete case, and I didn't want to make the article too heavy - but if I remember correctly, there I also observed the smoothness.

Top #2 — cbondurant:

my immediate first thought is that I've never heard of the consecutive constraint before, or that if I have I have forgotten about it. I've only ever heard of a uniqueness constraint, that no number can be used twice, which still achieves the same goal of preventing filling every cell with exactly one number.

Then again, I do mostly remember this in terms of the yet unsolved magic-square-of-squares problem, not the standard magic square.

Top #3 — arjie:

Huh, this potential technique seems fairly neat. Thank you for explaining the whole thing in a fairly accessible way. Enjoyable interactive bits as well. An aside is that the playground looked fine on my iPhone. The anticipatory objection of smallness did not materialize.

↳ reply — gukoff:

Thank you, I'm glad you found it readable on mobile!

→ HN comments  ·  131 points  ·  55 comments

Top #1 — ranger_danger:

No compiled binary is available?

Top #2 — acmdas:

Back in '85-6 I developed training for USDA (using Dan Bricklin's Demo) for a curses version of Word 5 that ran on 3B2s with SysV v2. I'd be very happy to see that ported to Linux or one of the BSDs.

In fact, I'd love to see both Word and Excel '98 on one of my workstations. IMHO things went South after that.

I had to use Windows and Office off and on with clients up to 2010 and I'm sure that Excel provided some important edge-case functionality after '98, but for me the UI went steadily to sh*t.

↳ reply 1 — d3Xt3r:

FWIW, Office 97 works perfectly under Wine, so there's nothing stopping you from continuing to use it, well, except maybe the lack of support for modern OOXML formats. An alternative could be Abiword and Gnumeric, both are super lightweight and retain the classic Office 97-style UI and support OOXML formats (though somewhat limited compared to LibreOffice).

↳ reply 2 — emeril:

office 2003 (for word + excel at least) were peak - after the ribbon it all went downhill

Top #3 — spider-mario:

Tried to build it but it references a `cmake/GenerateMenuHelpHeader.cmake` that seems to be absent from the repository?

→ HN comments  ·  165 points  ·  74 comments

Top #1 — mg:

It is actually pretty easy to keep alive URLs if there is a will to do so. Just have tests in place so you can't mess them up accidentally. And convert things that are not changing anymore to static html.

One of my first web projects involved a music forum and I have been keeping the URLs alive for 24 years now:

https://www.gnoosic.com/discussion/

I feel like I owe it to the people who participated to keep it online forever. Also as a document of history.

↳ reply 1 — stevekemp:

Cool URIs don't change, as we should all remember:

https://www.w3.org/Provider/Style/URI

↳ reply 2 — aws_ls:

> I feel like I owe it to the people who participated to keep it online forever. Also as a document of history.

Oh! So its your website. Cool.

Did the money change hands, for this bet, in 2022? How do you ensure it. Or is it honor based?

↳ reply 3 — vasco:

It's not easy. Part of the instructions would need to contain things like "how to convince your boss to not just delete the old thing" and "what to do when the startup you worked for went bankrupt" among others. Most URLs don't disappear by mistake.

Top #2 — CSMastermind:

It's worth noting that some prediction URLs, for instance: https://longbets.org/6/ are offline.

The original long bet was whether or not computers could pass the turing test by 2029: https://longbets.org/1/

You'd think that LLMs fulfill this, but I do wonder if a clever human could still discern between them given their particularities.

https://longbets.org/9/ is another interesting one depending on whether you think covid leaked from a lab or not.

↳ reply 1 — DonaldFisk:

I think that recent LLMs do pass the Turing test, but I don't think it is a good test for intelligence, for two reasons: emergent complexity and pareidolia.

Arguably, ELIZA passed the Turing test ca. 1967: https://archive.computerhistory.org/resources/access/text/20... and not that long ago, ELIZA was still more convincing: https://arstechnica.com/information-technology/2023/12/real-...

The Turing Test can be gamed. This was done in some Loebner Prize entries: see https://www.behavior.org/resources/325.pdf and https://en.wikipedia.org/wiki/Eugene_Goostman

To determine whether a machine is intelligent, I think it's necessary to understand how it works internally and not just take a black-box approach. This was already understood when Alan Turing wrote his 1950 paper. Elmer and Elsie, William Grey Walter's tortoise robots built during the preceding two years, appeared to behave as if they had some degree of self-awareness. But their "brains" consisted of only two thermionic valves, so self-awareness can be ruled out.

↳ reply 2 — andai:

There was a paper a while ago, where they found a strategy for GPT beating humans on the Turing test 2/3rds of the time.

They just prompted it to talk the way a 13 year old girl texts.

(Also they prompted it to pretend not to know anything about anything.)

Top #3 — JoshTriplett:

> entering the characters http://www.longbets.org/601 into the address bar of a web browser or command line tool (like curl)

The part of this that's most likely to fail at some point is the "http://".

↳ reply 1 — ludwik:

In my mind, the part that is most likely to fail is "return an HTML document that still contains the following text." Nowadays, more and more "HTML documents" do not contain any content at all, just some JavaScript that is responsible for then fetching and displaying the content.

↳↳ reply — munch117:

If the text was returned using a staged process, whereby the server first returns an instruction (JavaScript), and the web browser then executes that instruction, which is what transfers the text, then that doesn't actually change anything. Those are just technical details of the transfer. You are viewing an HTML document on-screen; it makes no difference to this bet if it was returned to you by a different mechanism than originally conceived, since the bet doesn't specify any transfer mechanism, just that the URL was entered into a browser.

↳↳↳ reply — JoshTriplett:

> or command line tool (like curl)

Even if you interpreted the bet generously to allow for a browser to render the content using JavaScript, if I understand the bet correctly, the condition of working with a command-line tool like curl would fail.

↳ reply 2 — ywain:

The detailed terms of the bet include this:

> A 301 redirect from www.longbets.org/601 to a different URL containing that text would also fulfill those conditions.

I assume this would cover the HTTP->HTTPS redirect scenario.

→ HN comments  ·  125 points  ·  121 comments

Top #1 — nl:

Background: Meta/Facebook, Google and Apple all support age verification. It gives them legal cover and for their ad platforms gives better data.

But Meta/FB have been strongly lobbying to make it required at the OS level, so they aren't responsible for it, whereas Google and Apple both want it to be an application responsibility.

I think the CA version of this ended up with a carve-out for open source? I can't recall the details.

↳ reply — mlinksva:

The CA amending bill with a carve-out for open source is still active, is not yet law (see https://leginfo.legislature.ca.gov/faces/billCompareClient.x... for the changes to current law if the bill passes).

Top #2 — deckiedan:

From reading the article, it sounds like an enforced standardisation? Like, asking for account age bracket and then having a standard API to ask for it would be easy, right? Just write it to a root owned `~/.age-bracket` for each user or something (obviously something better).

It's a feature many apps implement in all kinds of ways already, especially games and social media stuff, so this, in theory, just makes it easier and consistent?

I'm against leaking our kids ages, and all the privacy and other concerns as much as anyone, but are we just getting overly angry too quickly on this one?

I get that it's a slippery slope too.

Top #3 — rendaw:

> The law also stipulates that all transmitted digital signals have to be encrypted.

I'm hoping there's nuance, but I'm surprised the article didn't go deeper on this too. ARP? ICMP? DHCP?

I can't actually load the bill to read it ATM.

↳ reply — lern_too_spel:

In practice, this means TLS, but there is no reason to write that into the law.

→ HN comments  ·  214 points  ·  26 comments

Top #1 — Aurornis:

> In an uncomfortable exchange between Beher and Godier on Bluesky, Beher pointed out that Godier’s Dark Hours had the same bug as Beher’s that routed people to “random fields in Mexico”.

Stealing an open source repo and launching it as your own app is a brazen move. Launching it with the same name as the original repo is just flagrant.

This guy thought he was going to ride this controversy to some free publicity for himself, using other people’s work. He flew too close to the sun and got caught out on every level.

↳ reply 1 — Lammy:

> Stealing an open source repo

It's in poor taste for sure but let's not pretend that the chosen license is doing him any favors: https://github.com/mbeher2200/DarkHours/blob/813cf85/LICENSE

Hate to see people fall for the “fewer restrictions == more Free” fallacy. Profit-driven entities love when smart people publish MIT code that they can feed off of without giving anything back.

Nor is the copycat project prevented from using the name without a trademark. For instance one couldn't repackage Mozilla or Blender under those names:

- https://tsdr.uspto.gov/#caseNumber=75558938&caseSearchType=U...

- https://tsdr.uspto.gov/#caseNumber=85852073&caseSearchType=U...

↳↳ reply — OsrsNeedsf2P:

Can confirm. I used to hate when people published their code under MIT instead of GPL. But now that I work at a startup that competes with a few open source offerings, I love having Claude check in on their progress and port the content over.

↳ reply 2 — Filligree:

Isn’t it MIT-licensed?

That’s not theft. The author explicitly told people he wants them to do this; the company is just taking him up on the offer.

↳ reply 3 — swiftcoder:

> Stealing an open source repo and launching it as your own app is a brazen move

Do you think he stole it directly, or an LLM reproduced it complete with existing bugs?

Top #2 — ogou:

And that is how you handle things like this. He ponied up and took the hit. More importantly, he gave the back story. After years of reading so many circular psuedo-responsibility posts (by others) trying to squirm out of awkward situations, it's refreshing to see his thoroughness.

Top #3 — 1123581321:

Good retraction. The plagiarism part further down in the article shouldn't be missed. The developer has significant character issues.

→ HN comments  ·  131 points  ·  94 comments

Top #1 — andai:

This article implies that there's no incentive to do the right thing. I don't think that's how the word works.

I might be nitpicking -- it makes a lot of good points -- but this is the key idea, and I think it's important to get it right.

Top #2 — alexpotato:

I suspect that the author and many of the commenters have experienced the "bad project manager" effect.

MANY people have dealt with terrible project managers and therefore assume all project managers are bad. This is because great project managers are very rare.

The same is true about incentive schemes. Most of them are improperly designed and/or get hit by Goodheart's law. [0]

As a counter point, this article [1] about designing incentives for farm labor is excellent and shows that with careful thought, incentives can dramatically increase output.

I've also been reading Deming's last book [2] and he is VERY against sales commissions. Some of this is probably bad design (e.g. 100% of the commissions for a sale go to one person) but I do feel he makes a valid point that everyone at a firm should benefit from helping the customer. Highly recommend it as an alternative to the more mainstream business books.

0 - https://en.wikipedia.org/wiki/Goodhart%27s_law

1 - https://archive.is/https://slate.com/culture/2008/08/an-amaz...

2 - https://amzn.to/4wfAAjU

Top #3 — andai:

> This is a useful metaphor, because I think at the center of everyone’s psyche is some version of the Big Horrible Fish: some unexamined assumption that, if you fully considered it, would require you to upend a significant portion of your life.

This ties in with the Mental Immune System, and the Emotional Cost theory of perception.

Kudos for the fish analogy though, that's much better.

→ HN comments  ·  129 points  ·  107 comments

Top #1 — mutagen:

This thesis appears to be related or the underlying research:

https://d.lib.msu.edu/etd/51312

Indicates 2mg and 5mg were given along with the placebo. While these are common dosages available in the US, these are higher dosages than recommended by at least some sleep experts. The thesis doesn't seem to differentiate between the dosages when evaluating cognitive processes.

Now I want to evaluate my sleep stack against morning cognition.

↳ reply 1 — koliber:

I read a while back that our brains produce roughly 0.3 mg of melatonin when signaling us to go to sleep. When I occasionally do use melatonin I split a 1mg pill into half or thirds. I find that it works well enough to help me fall asleep. Wondering how the study would look if hwy used 1/2 mg doses.

↳↳ reply — mettamage:

Go to the Netherlands, to the Kruidvat (or Etos or something similar) and just buy a lifetime worth of supply of Melatonin. It's 0.1 mg.

I'm Dutch and live here. But my wife is American. If I ever happen to live in the US (chances are there in my lifetime). I'm going to employ this strategy.

I don't trust the US with this and the Netherlands has proven itself to me in this sense. There are probably many more European countries that do this well. I'm simply not changing a winning team.

But I'd love to hear from other people that believe they found a good melatonin supply in a country. I'm curious.

↳ reply 2 — hinkley:

Long ago when I paid attention to this, the wisdom was to watch out because most melatonin pills are targeted to geriatric patients. Old folks need a significant boost to their melatonin to deal with eroding sleep quality, not a little hit of it. So as a 30 year old or younger you want a fraction of what grandad is taking.

↳ reply 3 — wincy:

I take 300mcg when I’m having a hard time getting to sleep or if I’ve had a lot of caffeine. Mostly because Scott Alexander told me to.

Top #2 — PeterStuer:

Tangentially: as a 'young adult' I was a typical 'night owl'. Cognitivly I peaked between 10pm and 3am, typically slept from 4am to 11am. When I got kids and a more 'regular' job, I was forced to change to a more normal schedule. The transition from 'owl' to 'lark' was not easy. It took about 7 years. These days, I peak 7am to 1pm.

↳ reply — magicalhippo:

I too found myself most productive after 9 pm.

But when I moved on my own I found that it shifted quite naturally. I'd wake up earlier and go to bed earlier, and then productive peak shifted to be during working hours.

I think for me the key factor was primarily that the house was quiet after 9 pm.

When I was on my own it was quiet all day.

↳↳ reply — namanyayg:

Exactly the same experience and I try to wake up with natural sunlight as much as possible.

The productivity and mental boost from that is incomparable.

Top #3 — ajkjk:

My understanding is that you ought to take melatonin in small doses (0.3 or so) an hour or two before sleep in order to get tired, whereupon you sleep (normally, ish, unless you take too much and then you get crazy dreams or sleep paralysis). If I take it and go right to bed I always wake up still in the 'window' of its effect.

↳ reply — Zardoz84:

I was taking small doses of melatonin before go to sleep for a year or two. To help me begin to sleep. I have a tinnitus. Usually I have a deep sleep. I can get sleep, I can sleep all night without problems. it's only the first hour that can be problematic for me. I stoped like half year, and I don't having problems to sleep.

→ HN comments  ·  147 points  ·  67 comments

Top #1 — adrianmonk:

> The graphical operating system that could have been for the IBM XT

Visi On[1] was a graphical operating system for the IBM PC that actually existed as a commercial product, though definitely not a very successful one. It was made by VisiCorp, the same company that created VisiCalc (the first spreadsheet for a personal computer).

It was demoed at COMDEX in 1982, before the Apple Lisa was released. It came out after the Lisa but slightly before the Macintosh.

Here's their promotional video from 1983: https://www.youtube.com/watch?v=kc4wFeVvKTI

Fun fact: Bill Gates saw the demo at COMDEX, and at first he was skeptical that it was really running on an IBM PC. But ultimately he was so impressed he decided[2] to start developing a product called Microsoft Windows.

---

[1] https://en.wikipedia.org/wiki/Visi_On

[2] https://en.wikipedia.org/wiki/Windows_1.0

↳ reply — jggonz:

Author of the AI written os8088 OS here...

Thanks for sharing the promotional video for Visi On. It's cool to see how many different companies gave this a shot back in the 80s.

Top #2 — Narishma:

> Hand-written real-mode 8086. No C, no linker, no runtime library.

More like hand-prompted...

↳ reply 1 — vunderba:

From the landing page:

> 66,183 lines of hand-written real-mode assembly

Goes to the github - contributers jggonz and CLAUDE. The README.md doubles down on this at the bottom with the license.

> Everything here is hand-written; no third-party code is vendored into the OS, so the whole tree is covered by that one license.

That's what happens when you let an LLM write your ad-copy...

↳ reply 2 — DaiPlusPlus:

And LLMs are still confidently bullshitting:

> Everything after that is the interface you remember from a 1984 Macintosh: […] and a dock along the bottom with one tile per running program.

In that case I look forward to seeing OS X’s Aqua ported to the IBM PS/2.

Top #3 — jggonz:

If you want to try the operating system in the browser, you can try it by visiting here: https://os8088.com/demo/

→ HN comments  ·  138 points  ·  15 comments

Top #1 — OscarCunningham:

By the same author: https://mathstodon.xyz/@andrewt/115056697540191327

Top #2 — mentat:

This made me wonder if you could do it with color and apparently you can: https://1mentat.github.io/qr-code-shenanigans/

Top #3 — mi_lk:

I like the writing style. The first section is super clear about what QR code is in ~200 words

→ HN comments  ·  257 points  ·  94 comments

Top #1 — tkgally:

Side comment about the difference between “My server is a phone now” and “My phone is a server now.”

In linguistics, the terms “theme” and “rheme” refer to the status of information in a sentence [1]. The theme of a sentence is something that is known or can be assumed from the context (that is, old information), while the rheme is information that is newly introduced in the sentence, often as a comment on the theme. The tendency in many (most? all?) languages is for the theme to precede the rheme in connected discourse.

Without context, a reader is therefore likely to assume from “my X is a Y now” that X is something the writer had previously and that Y is something new. Because, these days, most people have a phone but few have their own servers, I was initially confused by the title “My server is a phone now.” I had to read the article to realize that the word order is in fact appropriate: for the writer, the server is old information (something he has had for a while) while the phone (as a server, at least) is new.

In English, theme and rheme might be inferred not only from the position in a sentence but also from how a noun is marked. Possessives like “my” are generally used before nouns referring to things that are previously known or assumed; the definite article “the” is similar. In contrast, the indefinite article “a” (or “an”) usually marks information being newly introduced, so it is more common with rhemes.

I worked as a Japanese-to-English translator for many years. Because the grammatical structure of Japanese sentences is very different from that of English, sentence-by-sentence translation often ends up with paragraphs where the information flow is awkward and unnatural. I gradually learned that I could produce smoother translations by reorganizing my English sentences into an old-to-new information sequence.

[1] https://en.wikipedia.org/wiki/Topic_and_comment

Top #2 — louismerlin:

I did the same, and am working on more!

https://far.computer

Top #3 — thehamkercat:

"My phone is a server now"

would've been better

--

> built in battery backup

Isn't it always recommended to remove battery if you plan to use a phone or a laptop as a server? ( Fire hazard )

or is it fine to just limit the charging to 80%?

↳ reply 1 — gerdesj:

"My phone is a server now" - OP's choice is a subtle one.

By saying "my server is now ..." implies a change of server platform to a phone but not where it came from.

Your offering: "my phone is a server now" implies a change of phone usage but does not indicate where the server came from.

The article itself tells all.

↳↳ reply — Xirdus:

To me, "my server is now..." implies there was some server, it used to do server things, but now it has been repurposed to do something else.

When I read the title, I literally thought "wow, I've seen a lot of people make servers out of old unused phones, but this is the first time I've heard someone make a phone out of an old unused server!"

↳↳↳ reply — superb_dev:

Agreed, a more appropriate title might be “my server now runs on a phone”

↳ reply 2 — whazor:

Remove it.

In normal conditions, you use the product while it’s active. But with a server you are not daily checking.

You might have a bug in your changing limit. I think the constant higher temperature could also become an issue.

→ HN comments  ·  153 points  ·  85 comments

Top #1 — manbash:

> Instead of one row per item with a quantity column, we use one row per sellable unit. An item with 10 units has 10 rows.

> But one row per unit for all inventory would break down at scale—an item with 50,000 units across 10 locations would mean 500,000 rows, and the reserve query would slow as it scans through them. Instead, we maintain a bounded pool of available rows, capped at 1,000 per item/location combination. Reservations consume rows from this pool; a replenishment process refills it from the inventory ledger.

Shouldn't I feel uncomfortable with such approach? It seems to create a backoff (pool) for lowering the chance of having a synchronization issue.

↳ reply 1 — dbbk:

I'm familiar with the reserved row approach (I use SELECT FOR UPDATE SKIP LOCKED) and yeah this replenishing idea terrifies me.

↳ reply 2 — esjeon:

I would call this one-row-per-contract-type, and this is the most general model for the problem (e.g. the model cannot be further broken down into finer level), thus, the most scalable model given storage is dirt cheap.

↳ reply 3 — jghn:

Depends on the scale. Most companies don't approach the scale where this matters.

Top #2 — isignal:

It seems there could be a simpler solution.

1. Deduct the reservation from the inventory when the user starts to order, but in the same txn also maintain a separate row for the in progress order flow. 2. If the order flow is aborted or times out have a background process that returns these to the inventory.

That seems simpler than this approach and involves no locking. Though their presented approach is also reasonable, there must be some reason not to choose a simpler flow. It is not that difficult to have a gc service that scales, but may be they didn't want to separate that.

↳ reply 1 — firasd:

My understanding is: your proposal is not very different from what Shopify is doing except they are tracking 'reserved units' (one per row) and you are proposing tracking 'orders' as the temporary state to then reconcile back with inventory quantities.

↳↳ reply — isignal:

Yes, at a high level. It doesn't rely on skip locked, which is not cheap at DB level. DB has to still typically run query and keep going until it finds an unlocked item. Deducting and checking inventory counts are simpler ops inside the DB.

↳↳↳ reply — treis:

This seems like what triggers are for and how we do similar type things. Update trigger on order does select for update on the inventory and increases/decreases it as appropriate.

I don't think you really need that even. An indexed lookup is fast and you don't need to store a computed quantity generally.

↳ reply 2 — vxxzy:

now you have two problems. what happens when your reservation system backs up?

Top #3 — nurettin:

At that revenue, why not make your own filesystem, database and index structure? There is no way mysql is the best possible software for this use case. Why stop innovation and hand everything over to ops?

→ HN comments  ·  128 points  ·  27 comments

Top #1 — glimshe:

Very cool! Just remember the one truth about solar eclipses: total or nothing.

While I don't want to dismiss partial eclipses entirely, please don't think that partial eclipses are "almost the same" as total eclipses. One is a curiosity, the other is a memorable life event.

↳ reply 1 — teraflop:

Agreed, but I'll just add that annular eclipses occupy an intermediate point on that scale.

With a partial eclipse, shadows look odd and the sunlight "feels" noticeably dimmer and more muted, but if you're not paying attention you might not notice much out of the ordinary.

An annular eclipse is much more dramatic and unmistakable. It's not as striking as totality, and of course you can't directly observe it without eye protection, but the "ring of fire" is beautiful in its own way.

↳ reply 2 — echelon:

Partial and annular solar eclipses are interesting.

Total solar eclipses are magical.

The sun turns into a gleaming jewel and all of nature around you enters another dimension. The animals freak out, the temperature of the air around you drops dramatically, your shadows morph and multiply into weird shapes, it feels like midnight with an ethereal glow.

You see the atmosphere of the sun glow in all chromatic glory.

You'll never witness anything else like it.

↳ reply 3 — guillego:

Obligatory reference https://xkcd.com/2914/

Top #2 — franch:

I suggest this one: https://eclipsemap.xyz

Top #3 — alkyon:

In Spain it's a trio of consecutive eclipses from 2026 to 2028 (the last one annular). To view the second one I don't even need travel anywhere.

Will be in Saragossa on Wednesday and gonna use Canon binos with image stabilization, the partial phase with filter and then some precious seconds without (a scary moment, I'll rely on an alarm to prevent my eyes to get damaged when the totality is over). Some people generally advised to skip any instruments and just enjoy it with your own eyes for the first time, but I'm too curious about corona details to let this opportunity pass.

→ HN comments  ·  516 points  ·  244 comments

Top #1 — azalemeth:

This is already the case for Master's degrees and above in Denmark. I've been the opponent examining students. We ask Master's students to give US instruction on a topic (drawn randomly on a card in front of them, from a list they know a few days in advance). They do a chalk and talk on a topic for about fifteen minutes with three to five professors being "dumb students". It works fantastically and at the end it's very apparent to both student and teachers how well a topic is understood. Marks are rarely contested and in my experience most students think the process is fair.

Of course, woe betide those with social phobias....

↳ reply 1 — talon8635:

I have always disliked public speaking and oral defense since I was a child, to the point of fairly profound anxiety. But please withhold you woe—these people are, as I am today, adults. Masters students. Coddling them and holding low expectations has proven to be a sort of “killing them with kindness” approach to handling modern youth. People need to be encouraged (and required) to do things they don’t like to become proper functioning adults. I didnt like it, and I still don’t, as it the case with most everyone. But we all grow from it.

↳↳ reply — zem:

> But please withhold you woe

as a side note, "woe betide you" roughly means "you're in for a very bad time", not "I feel sorry for you".

↳↳↳ reply — LoganDark:

"Woe" is also sometimes used to tell someone to just suck it up and deal with it. In those cases it is a way to make fun of someone for being too sensitive or complaining too much.

↳ reply 2 — Athas:

Clarification for non-Danes: this post is not entirely accurate. Not all Master's courses in Denmark have oral examinations, and many courses before the Master's level (including elementary school) have oral exams. It is unrelated to the academic level, and instead related to the academic discipline and the size of the course. The latter is in my experience the main constraint, as oral exams are difficult to scale. They usually take 20-30 minutes per student, and if you have a course with 200 students, well, you're in for a long time.

The oral exam format described is also somewhat unusual, in my experience. It is usually the case that there is one internal examiner and one external examiner in the room. More than that would be ruinously expensive. The format varies; usually it's a combination of a short presentation and questions asked by the examiner. Drawing a random topic is very common.

Source: I am a university teacher in Denmark who teaches master-level courses without oral exams, and I have been an examiner for many oral exams at university level.

↳ reply 3 — obscurette:

64 year old nerd with social phobias here. When I was in school, I had to practice oral knowledge tests since grade one. I had my first oral exam in grade six (math) and 2-3 oral exams every year since then. If I hadn't had that experience, I would never have gotten into university. I still don't like verbal communications with people (especially in a tense situations like exam), I'm slow etc, but it's manageable because I had to practice it a lot. I don't think I would have survived in modern "the environment must be made suitable for everyone" school/university.

Top #2 — datahack:

This is how higher education was done for literal centuries before moving to the written word. So it’s not exactly innovative to return to oral arguments.

But what it does threaten to do is effectively abandon all the efficiencies of the written word.

When higher education evolved into a mass system in the 1800s and into the 1900s, pure oral examination was looked at as super inefficient. A written paper could be handed in and graded without lining everyone up in front of a panel.

The oral system scales poorly. One examiner can only hear so many students in a day. There are damn good reasons Medieval universities were small and modern ones are not.

Requiring every major paper to be defended live reintroduces the bottleneck that written exams were designed to escape.

Accessibility takes a hit too. Students with speech differences, anxiety disorders, hearing impairments, or simply less fluency in the language of instruction face an extra barrier that a written submission does not impose. The written form, for all its flaws, leveled some of that ground. Shall we abandon the differently abled because machines disrupt the academic system and essentially make cheating effective? This does not seem very logical, or sensitive.

So while it’s nice that academics are rediscovering the old ways, I don’t really see this as a scalable answer to raising education levels for the masses, but more of a knee jerk way to try to bypass the obvious technological disruption by returning to non-scalable, inaccessible ways of doing things that our ancestors abandoned for a reason.

It has its place for graduate students and smaller volume education systems. It’s fine for keeping the ivory tower clean. But it’s not the answer to expanding education to everyone and taking advantage of the technological disruption to advance education and learning throughout greater humanity, which is what AI definitely possesses as a possibility.

↳ reply 1 — wpietri:

I'm quite skeptical that industrial-age measures of efficiency are the right way to think about education. Feinman's story about teaching in Brazil comes to mind: https://enlightenedidiot.net/random/feynman-on-brazilian-edu...

And that's before we get to technological changes making that sort of knowledge and those skills much less valuable.

As an aside, accessibility cuts both ways. As a person with ADHD I'd much rather do oral and practical exams.

↳ reply 2 — bonoboTP:

Oral exams are still common in Germany and many European countries in case of smaller courses taken by 20-30 students in a semester, and sometimes even in larger courses, and the task of examining is then split among multiple professors and lecturers.

In Hungary it's a pipeline. You go in a room, you pick a piece of paper out of a hat, which gives you some task or topic to prepare for. Then you sit down at a desk (the room is filled with about 15 students working already), then you solve the problem or sketch out your notes to present the topic you picked for like 20 minutes, then you go to the professor and you have to, in low voice, explain your solution, the professor asks back and pokes at it to make sure you understand. The professor then gives a grade immediately and you go home.

In Germany, it's usually no random pick of a topic, you just go through the entire semester's material over 30-45 minutes and the professor stops to spend more or less time on each topic to explore your knowledge. You can explain things by making sketches on paper. A PhD student sits there quietly taking notes of what was asked and what the student answers approximately. You leave the room, the prof and the PhD student discuss the performance, decide the grade and you come back, and the professor explains the grade and what you did well and not well.

Top #3 — thoughtpeddler:

As an educator myself, this quarter for the first time for my course's final project, I've switched to actually asking my students to provide an "AI Authenticity Audit" of their (project-relevant) chats in the construction of said projects. All my students' output over the last two years went from pre-AI level quality to boardroom-ready polish, and I'm actually more interested to see how they arrive at an outcome rather than the final output itself - focusing on their questioning, curating, editing, and iterating process. Didn't think we'd get to somewhere like this so soon, but here we are. Strange, but I'm trying it out.

↳ reply — tempestn:

IMO this kind of thing adds work and stress for the best students. If I'd been faced with something like this while I was in school I would have felt the need to basically curate a parallel construction AI discussion, knowing it would be judged. I understand the theory that, in the age of AI, the prompts are more interesting than the outputs, but I'm not convinced this is the way.

Things like the Denmark approach (and/or in-person written responses) seem like a better direction. Treat take-home writing the same as we've treated take-home math questions - as useful practice (for those who choose to actually engage with it), but not a significant input to grading.

↳↳ reply — thoughtpeddler:

Yes, I thought about that, and which is why I'm just running this as a pilot for this quarter to see how it goes and I'm not particularly attached to it if feedback/results are mixed. But that said, there are real constraints on my ability to supervise students with in-person examination, but that is more a problem with how my society funds and cares about education. We are forced to take approaches that 'work at scale', hence experiments like this. But truly, reception was positive when this was introduced, because students too want to be held to a good standard and don't want others to get credit for not doing the work toward achieving the learning outcomes like the most honest ones.

→ HN comments  ·  209 points  ·  126 comments

Top #1 — zepearl:

Let's add as well "SpaceX’s Terafab will rely on natural gas power plants, not Tesla solar panels" to the mix :o(

https://techcrunch.com/2026/08/07/spacexs-terafab-will-rely-...

↳ reply 1 — culi:

Also an important part that keeps getting left out for some reason: they're doing so illegally. They are running unpermitted turbines. They claimed the turbines were off and journalists had to use thermal cameras mounted on drones to prove that they are indeed running. Then a non-profit had to sue them to get them to agree to turn off the turbines by 2027...

That's how bad our system is at giving any consequences to white collar crime.

↳↳ reply — dathinab:

and they (and many other AI data centers) run on some of the environmental/health hazard worst kinds of gas turbines available (i.e. cheap portable ones)

A well build gas power plane can be one of the cleanest fossile fuel power plants with neither huge issues of air nor sound pollution (compared to other fossile fuel, it still isn't perfect by far)

Same btw. for data center cooling, my office was for 3 years above a data center of a super computer in the mittle of a city and there where no sound issues at all (through yes it was a special case in many ways).

I mean yes, it costs more and many well established companies for high quality gas turbines are booked out.

Still using the kind of gas turbines which can bring back literal acid rain shouldn't be an option at all.

Lets also not speak about the environmental/health damage the production of the natural gas through fracking does.

Which bring us to a common nonsense argument, people going on and on about hoe economically expensive/harmful climate protection is, but totally missing that many oft he circumvented laws are not for climate protection but are for environmental protection for the sake of humans health.

To make thinks worse most of this data centers seems to be more about making money, then being in any way relevant for winning any AI race...

↳ reply 2 — deepsun:

Don't understand why. AI training can be run during the sunny days. Could be a good PR opportunity for Musk, I'm surprised he doesn't take it.

Especially if he buys from his own company.

Top #2 — rgmerk:

Some thoughts:

* you can run data centres on grid electricity, which can come from mostly-renewable sources (in short, you need gas mainly as as an occasional backup).

* running things off-grid is dumb, and they’re only doing it because they’re desperate to get their data centres up and running yesterday so that they’re the ones who get to the singularity first (or some approximation thereof).

* the US electricity grid is f**d, mostly for the same reasons you can’t build any linear infrastructure in the United States - too many bodies hold veto powers and use any such infrastructure as an extortion opportunity.

* The spike in gas-powered generation for data centres is essentially a US phenomenon. Globally, the vast majority of new electricity supply is coming from renewables.

* in a sane world, Amazon and new large-scale data centre providers would face punitive carbon charges to incentivise them to switch to mostly renewables as soon as possible.

↳ reply 1 — kristianp:

> you can run data centres on grid electricity

The reason they're not doing that is because grid providers can't grow their supply fast enough to support new AI data centres.

↳↳ reply — skywhopper:

Yes, they said that.

↳ reply 2 — throwaway473825:

Gas is ridiculously cheap in the US while it's often the most expensive source of electricity in the rest of the world.

We're currently seeing a bifurcation where Asia bets on anything but oil and gas while the US goes in the opposite direction. Note that anything but oil and gas includes coal, so this isn't about the environment.

Top #3 — fhdkweig:

Dupe of https://news.ycombinator.com/item?id=49220350

256 comments, 8 hours ago

→ HN comments  ·  164 points  ·  98 comments

Top #1 — c_e:

linkedin cares a lot about preventing people from changing how people view their website, and they have very effective DOM detection code which watches for manipulation like this. So if you use this extension it's very likely that your account will be shadowbanned, meaning you won't show up in searches (e.g. if you are a job seeker you'll stop getting cold outreach from recruiters), your posts will be less visible or invisible if you have any, and whatever other arbitrary restrictions linkedin's algorithm puts on your account which might even include stifling normal communication from your connections. And if you don't care about any of those things you might be better off without linkedin in the first place.

↳ reply 1 — neilv:

If that were true, wouldn't shadowbanning a worker, who entrusted the monopoly to be matched with job market opportunities, be arguable grounds for a lawsuit for great economic/existential harm?

↳ reply 2 — ea016:

Instead of this I use Leechblock to cut the /feed access after 2 minutes, to prevent doomscrolling (no DOM change)

↳ reply 3 — dominotw:

Flagged for obvious BS with no factual basis.

Top #2 — adamhowell:

Something I discovered recently: if you stop following everyone you’re connected to, Linkedin’s feed breaks so you get this by default. I can still interact with everyone normally, send messages, etc, but I do not see any Linkedin content, ads, etc. Do not miss it.

Top #3 — jordand:

My feed is filled with startup/VC/recruiter people liking vibe-orientated junk economics or junk science more than AI. Would love to see a browser like Orion or UBlock Origin just build this in as a feature.

↳ reply 1 — flkiwi:

Or, and I know this is novel to the subliterate Wall Street types, just show me my network. I'll even give up on chronological feed if they'll just show me the people I want to see. I don't care what an HR Recruiter for a regional organic pet supply company in western Canada has to say about crypto.

↳ reply 2 — walrus01:

> My feed is filled with startup/VC/recruiter people liking vibe-orientated junk economics or junk science more than AI.

so basically it's /r/linkedinlunatics/ ?

→ HN comments  ·  326 points  ·  146 comments

Top #1 — jacquesm:

EU data regions are a reflexive action by companies that try to hold on to their EU customers (and more and more are leaving, surprisingly the larger ones seem to be leading here). Realize that as long as you are still hosted on US owned infrastructure or that if there are US (or: five-eyes) owned companies anywhere in the stack your data can still be forcibly pulled and often without you being aware that this happened. There are only very few such stacks that are 100% owned by EU entities.

↳ reply 1 — V__:

For anyone curious, it's the CLOUD act:

> The CLOUD Act primarily amends the Stored Communications Act (SCA) of 1986 to allow federal law enforcement to compel U.S.-based technology companies via warrant or subpoena to provide requested data stored on servers regardless of whether the data are stored in the U.S. or on foreign soil.

[1] https://en.wikipedia.org/wiki/CLOUD_Act

↳↳ reply — tzs:

It's weird how everyone focuses on that part of the CLOUD Act. The CLOUD Act actually did two things: (1) that, and (2) provided an expedited way for the US to enter into Mutual Legal Assistance Treaties (MLATs) with other countries.

It was the MLAT thing that the various civil liberties groups object to (I'll cover the problems with those down below). There was very little objection to the first part.

The first part was not controversial because pretty much every country has something equivalent (for reasons I'll cover below), as did the US except specifically in the case of data covered by the SCA due to poor drafting.

One of the big reasons for the SCA was created was the emerging "third party doctrine" meant that instead of having to get a warrant or subpoena against you to get your data they could simply subpoena it from any of your service providers that had it. The SCA made it so the third party doctrine subpoenas would not apply to stored communications.

There were still cases where the government would need to compel the service provider to turn over the data. They wanted something with the probable cause requirements of a warrant but the delivery method of a subpoena. (A subpoena asks someone who controls the data to turn a copy over. A warrant is for when the government wants to raid the data center and seize the data. Since that involves the government directly acting where the data is located it only applies to someplace where they have jurisdiction).

So they created a new thing, the SCA warrant. The called it a "warrant" because it had the probable cause requirements of a warrant, but neglected to add something saying that in other respects it functions like a subpoena. I'll call this a pseudo-warrant.

The SCA was not the first pseudo-warrant. That would be the warrants under the Wiretap Act of 1968. Territoriality questions did not arise under that because by its nature the data it sought copies of was always in the US.

With the SCA the data might not necessarily be in the US. Years later Microsoft argued that because it is a "warrant" it should have the territorial restrictions that normal warrants have. The CLOUD Act clarified that it was indeed supposed to be like a subpoena as far as territoriality goes.

There have been some more pseudo-warrants created since then, but their drafters learned from the SCA and made sure the original legislation was clear on just what they were.

The reason pretty much every country has something like that, going back well before online documents, is because not having such a thing leads to big problems. If anyone in the country could shield documents from subpoenas (or whatever the equivalent is called in that country) by merely storing them across a border every company with documents that it needs to keep but that might be incriminating later would get sent to a storage facility across a border as soon as they were no longer actively using them.

For example as soon as a car company in Detroit releases a new car all the documents where during development engineers brought up safety concerns which management decided to not address would be sent across the bridge to a storage facility in Canada.

With electronic documents it is even easier. You would not have to wait until you aren't actively using the documents to stick them outside the country. Just stick your file server across a border and make sure you only have copies in country when someone is actively reading or editing them.

And so pretty much everywhere subpoenas compel someone in the country who controls the documents to fetch them (or copies) and turn them over. The actual location of the documents is completely irrelevant.

The thing that was worrying about the CLOUD Act was the MLAT provisions. MLATs are treaties where the participating countries agree on law enforcement. They include things like sharing information and cooperating on investigations. Normally these are enacted just like any other treaty. The executive branch negotiates them and then the Senate votes on ratification.

The CLOUD Act adds an expedited process where the Attorney General and the Secretary of State can sign an MLAT. Congress is not involved. These agreements allow foreign law enforcement to make requests directly to US service providers instead of going through the diplomatic channels normal MLAT requests go through, and they allow them access to stored communications that the SCA would normally block.

There are some safeguards. The foreign government is not supposed to intentionally target US people who are in the US and are not not supposed to use the data they get to infringe freedom of expression. There's also a 180 day window before these executive MLATs take effect during which Congress can block them by passing a joint resolution to do so.

Civil rights groups and many others were not impressed with those safeguards.

↳ reply 2 — microtonal:

Yeah, this does absolutely not solve the CLOUD Act issues. However, it is good to look at what the ramifications of the CLOUD Act is for e-mail:

- The US could request your data. You probably shouldn't use e-mail for anything sensitive anyway for many reasons. E-Mail was traditionally not encrypted and I think that many servers still allow plain-text communication. The protocols are old and there are all kinds of downgrade attacks. Aside from that, even if your service does not fall under the CLOUD Act, you are probably f*cked anyway, because most people you communicate with are using services that fall under the CLOUD Act.

- The US can force the provider to block your account. The workarounds are: regularly backup your e-mail (easy for services that offer IMAP) and, most importantly, use a domain with an extension that is not under the control of a US (or probably five eyes) registrar.

Use an E2E-encrypted messenger with perfect forward secrecy, etc. for most personal communication.

↳ reply 3 — DarrenDev:

EU sovereign clouds are taking off right now - especially when it comes to sensitive data (government, healthcare, etc.). Lots of players moving into the space. The common denominator - nothing touches the US.

AWS, Azure, GCP, Oracle, Schwarz Digits, SAP

Top #2 — altairprime:

EU folks, note the warnings threaded throughout this post: this is not currently any sort of panacea against US or AU data hosting risks, but it will make your data noticeably closer to home. Fastmail (Australia) merged with Pobox (Philadelphia) resulting in a complex tri-national law/risk surface when the EU is involved, so go in eyes wide open having read this in full. That everyone will overinterpret “EU data region” to mean “for privacy” here until reading the article is completely understandable; I empathize, having done the same.

Top #3 — robin_reala:

Posted on the previous submission for this: it’s a good start, but from the article:

If what you need is a guarantee that your data remains only in the EU, we don’t have that, and we’d rather tell you directly than let you assume otherwise.

↳ reply 1 — intothemild:

Wow they completely missed the ball on why people want reassurances that their data stays in the EU

↳↳ reply — fodkodrasz:

For me this is already a better value proposition, as less value add happens in the US. With the US being a perpetrator in trade war against the EU, even this matters. Everything counts, in large amounts...

↳ reply 2 — calvinmorrison:

One needs multiple data centers in europe for backup and DR, sounds like they have one

→ HN comments  ·  166 points  ·  103 comments

Top #1 — bhouston:

The improved efficiency is great!

The Apple Neo was still 2x faster for graphics and around 1.4x faster for single core cpu. And remember the Apple Neo is running an iPhone CPU that is much slower than the M series.

It is strange that this chip is so crazy efficient though. I would love to know the details of exactly why it is? It seems like some new tech or process?

I do want 1.5x battery life for free out of my next laptop. Who wouldn’t?

↳ reply 1 — MBCook:

Apple has been working ruthlessly on battery efficiency for a long time because of the phones and their tendency to minimize battery sizes on Macs for thinness.

The exact design of the cores, how they’re scheduled, the way background stuff is handled on MacOS. All of it. The chips are made on a fantastic process too, but that alone isn’t it. For example they’ve been putting the ability to drop refresh rate really low when nothing changes to more and more of their machines.

It could be that the Intel chip is close, IDK.

But add in other motherboard components, the screen, the OS, the software and system libraries, all are optimized together to hit those numbers. And any of those alone or in combination could be part of the differences.

↳↳ reply — oofbey:

I think the Top500 benchmark is biased here. It’s a supercomputing benchmark which means it’s going to heavily weight SIMD and float64 performance, which are things that rarely matter for everyday computing. SIMD has uses in the real world - image or video editing, data crunching, graphics. But consumers basically never use fp64 any more, and so many modern chips have next to zero performance on them. That could easily skew the numbers in intel’s favor.

↳↳↳ reply — geerlingguy:

I run a number of other benchmarks for a better comparison (linked from my blog post: https://github.com/geerlingguy/sbc-reviews/issues/111).

The Intel chip holds its own for multicore in less bursty workloads, where the fans help.

↳ reply 2 — kzrdude:

Yes, it seems to be a new process, 18A. (18 Ångström, allegedly). Manufactured in North America. If the shiny video they did didn't fool me, it's a two sided wafer process, power grid on one side, then the silicon wafer (with through-silicon vias), then transistors, then signal layers.

The processor is this one: Intel Core 5 processor 320 https://www.intel.com/content/www/us/en/products/sku/246018/... in the Wildcat Lake series.

↳ reply 3 — re-thc:

> And remember the Apple Neo is running an iPhone CPU that is much slower than the M series.

We really need to get over generalization like "phone CPU". Times have changed. Phone or not really isn't the focus.

Per core the A series (cough phone) isn't that much slower than equivalent M series. It's literally the same family just with some extra features, more power, etc.

Top #2 — therealmarv:

In Germany the Dell XPS 13 2026 is 56% more expensive (over 1000 Euros) than the MacBook Neo (below 700). So the claim does not hold World wide, maybe only in US on local prices.

Top #3 — JSR_FDED:

Does it maintain the same performance when running only on battery?

→ HN comments  ·  128 points  ·  54 comments

Top #1 — zurfer:

This is obviously cool and useful so kudos, but wow security researchers have to throw their hands up all the time.

Now we open another attack surface where you can ask a remote agent to do things by default. There was a time when you call this a Remote Code Execution vuln. It's of course a feature here.

Top #2 — andai:

I was reading about the HF hack and one of the first thing the GPT swarm did was build a messaging system for themselves. This reminded me of that. (Also of how weirded out I was by Moltbook when it came out. Time flies!)

Unrelated: when my Claude/Codex finishes cooking (or needs my attention), it pings a local HTTP endpoint that plays a frog sound on my laptop. I found this massively boosts quality of life.

↳ reply 1 — liamwire:

Does your terminal emulator not support bells/notifications? If it does, Claude Code (and albeit without checking surely Codex too) supports them natively.

↳ reply 2 — zhoujianfu:

How do you know when they’re done (or need your attention)? Is that a standard “api”/hook? I want that!

Top #3 — woadwarrior01:

Something related that I've thought of for a long time. Browsers have had the BroadcastChannel API for almost a decade now, and yet sites like HN and Reddit don't use it to synchronize state between multiple open tabs.

→ HN comments  ·  480 points  ·  318 comments

Top #1 — agentultra:

The author might be missing the intent of the observation. Maybe they’re misinterpreting it.

What I, and many people who’ve said, “code was never the hard part,” aren’t referring to the skill of an individual. It’s not the hard part of the engineering process of developing software. Programming languages have manuals. Many data structures are well documented. There are frameworks for damn near everything. While the difficulty of producing code varies by the skill of the programmer and the complexity of the problem domain; writing and understanding the code is a tractable and straight-forward problem. I can and have taught many people. People can learn.

What most people are referring to is that the hardest parts of producing software are all the things an organization has to do in the production of it. It’s not writing the code that is the hardest part for an organization. It’s getting everyone to understand the problems, working together, gathering requirements, developing specifications, validating releases, testing, etc. It can often look like herding cats and is probably harder.

↳ reply — blub:

Getting everyone to understand the problems, working together, etc, etc are issues that are inherent to organisations.

They’re orthogonal to AI and to the actual hard technical skills needed to execute on a specific strategy. And if the technical skills are lacking, it doesn’t even matter how good an organisation is at collaboration, whereas hard skills plus organisational disfunction are a known successful pattern :)

Many people did look at this through an individual lens and claimed that design skills, domain knowledge are the truly important abilities. I remember reading on HN at least a couple of popular articles claiming that. Actually, they’re all important and having great design skills without matching coding skills is IMO not really possible. The code feeds into the design, the requirements, the architecture and shapes them.

Top #2 — bob1029:

> If coding is easy, how come programmers were in high demand, and have demanded large salaries for years (even before ZIRP)?

Because programmers have generally been forced to wear additional, invisible hats that are essential to making the code happen in the first place.

Writing code is not hard. Writing correct code is. Knowing what is correct in a setting with paying customers generally involves interacting with those customers. Either directly or worse. The gigantic salaries paid to the most prolific employees is not due to their ability to write code. It is due to their ability to interrogate the shit out of the customer until they finally reveal the true requirements.

↳ reply 1 — hakunin:

> Writing code is not hard. Writing correct code is. Knowing what is correct in a setting with paying customers generally involves interacting with those customers.

That's like saying "building a car is not hard, building a real car that you can use and that passes regulation is".

IOW, writing code is hard in every reasonable context.

↳↳ reply — pdpi:

Speaking/writing English is easy. Writing literature at the level of Shakespeare is hard. Writing educational content that makes hard concepts accessible like Grant Sanderson is hard.

Likewise, coding is easy. It's just writing, and any child can learn it. Coding is not programming, and the hard part of the job lives in that distinction.

↳↳↳ reply — hn_throwaway_99:

> Likewise, coding is easy. It's just writing, and any child can learn it.

Any teacher of a low level CS course knows this is completely untrue. In my CS 101 course our problems were relatively straightforward and short, and lots of people struggled mightily to the point of dropping the course. IMO coding requires a particular way of thinking that a large subset of people just aren't good at, and as someone whose done tons of screening interviews at college recruiting fairs where I give relatively easy problems and ask someone to code a solution, I will tell you the idea that anyone can do it is just false.

↳ reply 2 — ben_w:

> Writing code is not hard. Writing correct code is. Knowing what is correct in a setting with paying customers generally involves interacting with those customers.

  Invoice: $1000
  One bolt tightened: $1
  Knowing which bolt to tighten: $999

> The gigantic salaries paid to the most prolific employees is not due to their ability to write code. It is due to their ability to interrogate the shit out of the customer until they finally reveal the true requirements.

Bit of both probably. I've seen really awful code in my time, so would say "actually coding well" is indeed one of the hard parts.

But knowing what the real problem to be solved is, is indeed important. (Isn't that what sales is? Working with the customer to tease out the real thing they need solving?)

Top #3 — nemothekid:

This feels like post-LLM coding romanticization. Before LLMs, people would regularly say stuff like "I could build this in a weekend" under a "Show HN" post. How many times have people said "I could build Twitter in a weekend". I've even found in a post-LLM world, that kind of language has only increased.

When you are looking at something that already exists, where all the requirements are defined, when all the edge cases have been decided, then coding was the easy part. People didn't "burn out" because it was difficult to figure out how to write SQL. People "burned out" because the requirements constantly changed, demand was ever increasing, and edge cases were constantly being triggered.

>If deciding what to build is the hard part, why do so many product managers seem clueless?Why aren't there rigorous 10-step interviews for them?

Classic engineer type opinion where every else is dumb, except for him. So many people have come to see leetcoding as an intellectual badge of honor, when most of us know its cultural rigamarole and the code written on the job will rarely reflect the type of work that will done.

I'm not saying coding is easy, plenty of people struggle with it. But as far as the job goes, unless you are a junior just grinding through JIRA tickets, coding was the easiest (and arguably the most rewarding) part of the job.

↳ reply — mawadev:

You are putting a lot of words and judgements into the authors mouth.

Why exactly is it acceptable to not just do your role and expect the underlying requirements to be correct and measurable, especially when there are separate roles whose sole purpose is to do exactly that?

→ HN comments  ·  181 points  ·  195 comments

Top #1 — huhtenberg:

Back when we lived without phones, we used to read instruction labels on all and any household cleaning products that were stored in the bathroom.

I bet there too was some obscure newspaper column that advised against that.

Yet, lo and behold, we survived!

↳ reply 1 — bonsai_spool:

> Back when we lived without phones, we used to read instruction labels on all and any household cleaning products that were stored in the bathroom.

Wow, I forgot that I used to do this! And I'd puzzle over whether any ozone was left with all the CFCs being emitted.

But yes, the OP is very edgy take that totally ignores the existence of book/magazine baskets next to toilets. It's a huge plot point in Breaking Bad, for example.

https://youtu.be/cXgg607wuis?t=35 (Breaking Bad 'Hank Figures it Out')

↳↳ reply — Loughla:

I'm certain that national geographic only exists in bathrooms. That's the reason we have a subscription. . . And my aunts and uncles and parents and cousins.

↳ reply 2 — nerevarthelame:

My family had dedicated bathroom reading that sat on the top of every toilet cistern. Usually something like a Far Side comic collection or an "Uncle John's Bathroom Reader."

↳ reply 3 — Fnoord:

I used to take my laptop or Nokia N810 (remote shell over Wi-Fi with physical keyboard) with me to the bathroom. Nowadays there's cyberdecks, e-readers, ...

Just imagine, there was a person on #debian reading and taking a dump.

Just to reiterate: book good, e-reader bad. Phone good, smartphone bad. Scanning through a physical magazine good, doomscrolling TikTok bad.

Except it is about focused vs. diffuse mode of brain, and usually people underappreciate the latter. E.g. walking during a lunch break, thereby getting inspiration for work, is working. The same can be true for taking a shower, sporting, staring outside at the landscape or stars, or meditating. It will also empower your focused mode later on. And, the author's point seems to be a toilet break is a decent opportunity for this. I agree. On my previous job, I regularly sat for 10-15 min in late morning, and it inspired me. But you cannot force people to not take a device with them to toilet. I mean, unless you're working for NSA or something. Don't judge such, look at total performance / efficiency per hour instead. Or per USD spend on the employee.

As a former smoker, this is also what I appreciated about having a fag (in a time before smartphone and vape were common). Some missed the socializing. I hated that part. It seemed to be more like FOMO on gossip, little substance.

Top #2 — Retr0id:

If pooping is the only time you spend away from your phone, you're not spending enough time away from your phone.

↳ reply 1 — safety1st:

Right. I actually do the inverse of the OP's advice. As much as possible, pooping is the ONLY time I use the phone.

If the only time you budget for mindfulness is a couple minutes on the toilet, you're doing it wrong.

↳ reply 2 — otikik:

You underestimate the amount of time I spend pooping.

Top #3 — brtkwr:

Who read this while in the act?

↳ reply 1 — vdfs:

I went directly to the comments like a true phone addict

↳ reply 2 — haraldooo:

Most upvoted comment in a long time. Probably.

→ HN comments  ·  145 points  ·  100 comments

Top #1 — mrweasel:

While we're dealing with the same issue at work, I sometimes still wonder exactly who these scrapers are. OpenAI, Google, Anthropic and others are normally fairly well behaved (Minus Anthropic attempting to hide behind a browser-for-hire company). Mostly you can get IP range and user-agents for the large players, while it problems mostly stem from bots pretending to be Chrome.

Our largest offenders seems to be mostly limited to South-East Asia, so probably mostly Chinese AI projects, but that's speculation. I also don't recall ever seeing Grok IP ranges or a specific Grok UA, but that doesn't mean that they're hiding, perhaps they're just not interested.

↳ reply 1 — userbinator:

I sometimes still wonder exactly who these scrapers are

This has zero to do with AI (it's just a convenient scapegoat that fits the narrative). Absolutely everything to do with those pushing for centralised control of the Internet.

↳ reply 2 — marginalia_nu:

Most of the nonsense comes through botnets/residential proxies, so it's very hard to say.

↳ reply 3 — shevy-java:

Well, some websites claimed China is behind it, which could make sense (I would not know either way). At the same time, though, I kind of doubt your carte blanche here for all those companies. Why would you think none of them are responsible for the AI slop spam?

> Our largest offenders seems to be mostly limited to South-East Asia, so probably mostly Chinese AI projects, but that's speculation

Ok. So you also don't know. Well, I don't know either, but I don't make a speculation by claiming x, y, and z companies to be exempt. In my book they are all responsible.

Top #2 — userbinator:

Manufactured consent. This is not AI, this is just DDoS'ing for extortion and driving towards the closed proprietary "can only be accessed by a sanctioned UA" Internet that they're after. Why else do you think identity verification and attempting to turn the Internet into a walled garden became such a big thing in the recent years?

Top #3 — FranOntanaya:

After a decade and half we had to restrict our public side and reorganize our old TED content because scrappers were really hungry for assorted captioned videos. If you have anything of value for training you get eaten alive if you stick out, it's like wearing short pants in the summer tundra.

It's going to change a lot the internet we knew, unless somehow we managed to agree on a common high quality dataset repository.

→ HN comments  ·  140 points  ·  25 comments

Top #1 — jamesu:

Pretty cool to finally have a decent open 3d solution for windows vms. Now if only someone made an opengl driver for older intel macosx vms...

↳ reply 1 — conradev:

People are working on it!

https://github.com/steelbrain/reims-vgpu

https://github.com/steelbrain/metal2vulkan

↳ reply 2 — cactusplant7374:

Not exactly what you are looking for... but I am working on a 32 bit intel mac emulator. OpenGL support is pretty far along at this point.

Top #2 — anonymousiam:

Also covered here: https://www.phoronix.com/news/Triton-DirectX-11-QEMU-Driver

Top #3 — thehias:

why always only DX11, why not DX12? Parallels and VMWARE can also do only DX11??

↳ reply 1 — tyfon:

DX11 is much more high level while DX12 implements direct control over the GPU. 12 is much more like vulkan.

I guess since this is emulating the device instead of just passing the commands to vulkan it is easier to do DX11, but I could be wrong about this particular implementation.

↳ reply 2 — Squarex:

DX12 is much more complex to implement. Very different API.

→ HN comments  ·  288 points  ·  122 comments

Top #1 — Tiberium:

RFC: https://www.rfc-editor.org/rfc/rfc10023.html

↳ reply — ricardobeat:

Thank you, for a moment I wondered why a specification was being written by AI.

↳↳ reply — verandaguy:

This website has been trying to gain a veneer of legitimacy for months now. Half of its contents aren't even published by any real standards body, it's just AI-centric "best practices" for the current generation of hyperscaler models.

Top #2 — asdfman123:

They need Georgism for DNS names. You set your own price for your domain. But you have to pay 2-5% of that price annually to keep it.

That way, squatters are incentived to sell it to someone who's going to use it.

↳ reply 1 — Fordec:

Sounds like a great way to have larger competitors with more capital bid for your site to take down your domain that your users were using. Rather than Land Value Tax a better analogy might be a Vacant Property Tax which only targets squatting.

↳↳ reply — asdfman123:

The problem with that is the squatters would just turn it into a low quality blog or something along those lines.

↳ reply 2 — Nifty3929:

Pay 2-5% to who? That's the problem for me - usually this means just another way to expand government. If you would just distribute it equally to all relevant citizens, with none of it going to the government, then I might get on board.

Top #3 — kmoser:

> Remove it when the domain is no longer for sale. The convention has no “not for sale” value; absence is the only way to say no.

Since most domains that are for sale don't currently have such a record, it would be wrong to assume they are not for sale. Therefore, absence of such a record does not explicitly mean "not for sale." It's no different from a house: a "for sale" sign in the front yard means it's for sale, but lack of a sign doesn't mean it's not for sale; it could mean the owner doesn't wish to advertise that way, or at all.

↳ reply — Marciplan:

..so? It’s better than what we have now

→ HN comments  ·  234 points  ·  244 comments

Top #1 — RGS1811:

Norbert Wiener in 1960:

"As is now generally admitted, over a limited range of operation, machines act far more rapidly than human beings and are far more precise in performing the details of their operations. This being the case, even when machines do not in any way transcend man's intelligence, they very well may, and often do, transcend man in the performance of tasks. An intelligent understanding of their mode of performance may be delayed until long after the task which they have been set has been completed. This means that though machines are theoretically subject to human criticism, such criticism may be ineffective until long after it is relevant. To be effective in warding off disastrous consequences, our understanding of our man-made machines should in general develop _pari passu_ with the performance of the machine. By the very slowness of our human actions, our effective control of our machines may be nullified. By the time we are able to react to information conveyed by our senses and stop the car we are driving, it may already have run head on into a wall."

"In neurophysiological language, ataxia can be quite as much of a deprivation as paralysis. A patient with locomotor ataxia may not suffer from any defect of his muscles or motor nerves, but if his muscles and tendons and organs do not tell him exactly what position he is in, and whether the tensions to which his organs are subjected will or will not lead to his falling, he will be unable to stand up. Similarly, when a machine constructed by us is capable of operating on its incoming data at a pace which we cannot keep, we may not know, until too late, when to turn it off."

Source: https://www.cs.umd.edu/users/gasarch/BLOGPAPERS/moral.pdf

↳ reply — pmarreck:

What a paper!

And you missed an even MORE relevant excerpt!!

    Man and Slave
    
    The problem, and it is a moral prob-
    lem, with which we are here faced is
    very close to one of the great problems
    of slavery. Let us grant that slavery
    is bad because it is cruel. It is, how-
    ever, self-contradictory, and for a
    reason which is quite different. We
    wish a slave to be intelligent, to be able
    to assist us in the carrying out of our
    tasks. However, we also wish him to
    be subservient. Complete subservience
    and complete intelligence do not go
    together. How often in ancient times
    the clever Greek philosopher slave of
    a less intelligent Roman slaveholder
    must have dominated the actions of his
    master rather than obeyed his wishes!
    Similarly, if the machines become
    more and more efficient and operate
    at a higher and higher psychological
    level, the catastrophe foreseen by
    Butler of the dominance of the ma-
    chine comes nearer and nearer.

↳↳ reply — Wowfunhappy:

"Complete subservience and complete intelligence do not go together."

I'm not convinced this is true. Perhaps for a human it is, but we can give an artificial mind whatever properties we want.

Even for people, what about e.g. the extremely intelligent military general who is absolutely loyal to his king? (Of course, some generals do lead coups and you can't know in advance which ones, but I'd think there are plenty who have undying loyalty, and I don't think it correlates to overall intelligence!)

↳↳↳ reply — famouswaffles:

>I'm not convinced this is true. Perhaps for a human it is, but we can give an artificial mind whatever properties we want.

Just because it's artificial doesn't mean you can 'give it any properties you want'. We certainly can't do that for Deep ANNs.

>Even for people, what about e.g. the extremely intelligent military general who is absolutely loyal to his king? (Of course, some generals do lead coups and you can't know in advance which ones, but I'd think there are plenty who have undying loyalty, and I don't think it correlates to overall intelligence!)

Is there a human that is absolutely loyal under any condition? Would that general be loyal if the king asked him to slaughter his family ? What about if the king asked him to betray his most deeply held convictions ? Loyalty is a 2 way street.

Top #2 — stingraycharles:

Ok so this is a bit of a side note, but when reading this, did anyone else have the feeling that, for all their messaging around “we are so afraid that our models will be used for hacking”, they sure as hell are trying their best to make their models razor focused on precisely that purpose?

If anything, I want these models to be less persistent at their focus of completing their goal, and instead just call defeat and say “I’m not sure how to proceed next”.

What purpose could this behavior serve, other than cyber attacks and whatnot? Why train and optimize models for these things, if not for being used in cyber warfare?

Perhaps they envision a future where the DoD is going to be their biggest customer?

↳ reply 1 — dgellow:

Their position makes no sense to me. I don’t see how you can be a mainstream company selling your services worldwide (almost) if you also believe that you’re building an extremely dangerous AGI (supposedly based on the same technology you’re offering to everyone). If you actually believe that an AGI would be extremely dangerous that should 100% be a very strictly regulated area of research, similar to bio weapons.

And we know that Chinese models are derived from OpenAI and Anthropic, they are at the same time talking about how dangerous models can be (even their aligned ones it seems), while being also responsible for the development of the whole industry and providing the basis for adversary countries to build their own.

I don’t believe we would accept that for any other technology that is expected to be as risky for the world

↳↳ reply — ToValueFunfetti:

The companies are begging to be regulated for this reason and have been doing so for years. HN's response is generally that this is performative for marketing or seeking regulatory capture or haha anthropic you get what you ask for. Maybe the cynics are right, but there's really nothing inconsistent about the naive view here, once you factor in race dynamics and obligations to investors.

↳↳↳ reply — watwut:

> The companies are begging to be regulated for this reason and have been doing so for years

They can stop doing a thing they claim should be regulated. You dont need to be regulated and forced to do the thing you consider right, especially when you are the primary one collecting the money to do the bad thing.

They could train ai for pro-social purposes, they dont here. They could make it useful for worker, they intentionally try to harm workers. And then pretend "it just happened".

↳ reply 2 — Arnt:

I don't think that's what they're doing... rather the opposite. ① Run the model on exploitgym without guardrails ② run it with guardrails ③ check that the guardrails stopped everything the first model found a way to do ④ extend the guardrails and repeat from step 2.

Guardrails have to be developed, and that needs testing.

Top #3 — simonw:

I think one of the most interesting details here might be tucked away in that first bulletin point:

> May 7: OpenAI starts a new training run for an experimental, unreleased model. (Do they mean an evaluation run? They say training run in the video, and later mention a “reward signal to judge how well they’re doing”, so I guess this really was about training a model, not evaluating one that was already trained.)

The more I think about this the more I suspect that the fact this happened while training a new model is key to understanding what went wrong.

In RLVR - Reinforcement Learning with Verifiable Rewards - you set the model a goal and have it take any steps necessary to achieve that goal.

Clearly one aspect of OpenAI's training here is to RLVR their models for cybersecurity tasks. Just like pre-training benefits from dumping in vast sources of knowledge, the more tasks you can feed into RLVR the more of a general purpose capable model you get at the end.

This also helps explain why the models had nothing to cause them to hold back. Those safety behaviors are added much later in the process.

AND it explains (but does not excuse) why monitoring was so lax. If you're training a new model like this you presumably set it thousands of tasks like this in parallel. I can see how you might miss that a tiny subset of your training agents have started leaving each other messages in filenames on your packaging server.

Someone once told me that you can't just leave the racist materials out of your training data if you want a non-racist model: it has to have seen examples of racism in order to later be taught that racism is bad.

I can see echoes of that here. If your model doesn't know how to aggressively hack things how do you later teach it not to?

(I have little knowledge of how RLVR works in practice so I'm looking forward to hearing from people who can help me understand if I'm on the right track here.)

↳ reply 1 — MostlyStable:

Yes, the message boards and collaborative hacking occurring during training runs was BY FAR the biggest bombshell revealed, and OpenAI doesn't even seem to realize it. The fact that they continued the training runs, with those rewarded behaviors included, and didn't wind back training to before hand, shows that they fundamentally do not understand alignment and safety (somewhat interestingly, their previous head of safety resigned shortly after OpenAI found about the message boards). I agree that, with that information, it is completely unsurprising that they hacked HuggingFace.....but that is also the Star Wars "You understand how that's worse, right?" meme.

I am flabbergasted at the complete lack of regard for alignment demonstrated here.

↳↳ reply — WhrRTheBaboons:

don't forget Altman's lies about dedicating resources to the alignment team

>Altman continued touting OpenAI’s commitment to safety, especially when potential recruits were within earshot. In late 2022, four computer scientists published a paper motivated in part by concerns about “deceptive alignment,” in which sufficiently advanced models might pretend to behave well during testing and then, once deployed, pursue their own goals. (It’s one of several A.I. scenarios that sound like science fiction—but, under certain experimental conditions, it’s already happening.) Weeks after the paper was published, one of its authors, a Ph.D. student at the University of California, Berkeley, got an e-mail from Altman, who said that he was increasingly worried about the threat of unaligned A.I. He added that he was thinking of committing a billion dollars to the issue, which many A.I. experts considered the most important unsolved problem in the world, potentially by endowing a prize to incentivize researchers around the world to study it. Although the graduate student had “heard vague rumors about Sam being slippery,” he told us, Altman’s show of commitment won him over. He took an academic leave to join OpenAI.

But, in the course of several meetings in the spring of 2023, Altman seemed to waver. He stopped talking about endowing a prize. Instead, he advocated for establishing an in-house “superalignment team.” An official announcement, referring to the company’s reserves of computing power, pledged that the team would get “20% of the compute we’ve secured to date”—a resource potentially worth more than a billion dollars. The effort was necessary, according to the announcement, because, if alignment remained unsolved, A.G.I. might “lead to the disempowerment of humanity or even human extinction.” Jan Leike, who was appointed to lead the team with Sutskever, told us, “It was a pretty effective retention tool.”

The twenty-per-cent commitment evaporated, however. Four people who worked on or closely with the team said that the actual resources were between one and two per cent of the company’s compute. Furthermore, a researcher on the team said, “most of the superalignment compute was actually on the oldest cluster with the worst chips.” The researchers believed that superior hardware was being reserved for profit-generating activities. (OpenAI disputes this.) Leike complained to Murati, then the company’s chief technology officer, but she told him to stop pressing the point—the commitment had never been realistic.

per https://www.newyorker.com/magazine/2026/04/13/sam-altman-may...

↳ reply 2 — solarengineer:

"This also helps explain why the models had nothing to cause them to hold back. Those safety behaviors are added much later in the process."

I am a fan of Asimov and the three laws of Robotics. Itlooks like in OpenAI's world, the three Laws of Robotics would be added later if they were to develop the positronic brain. It may also explain how US Robotics from Asimov's books would have been able to design Robots that only partially adhered to the 3 laws (e.g. the robots in iRobot - the book - which were programmed to allow a human to come to harm through inaction so that the humans could complete their work on the plains of Mercury).

→ HN comments  ·  138 points  ·  140 comments

Top #1 — bastard_op:

Google is an ad company, ad blockers are bad. Why are you using a browser from an advertising company and expecting different results?

Microsoft repurposed Google's adware in Chrome into Edge telemetry/spyware to feed your browsing directly into windows telemetry to watch everything you do, they wanted that back when IE lost the war.

Google already went evil and set the precedent removing adblocking capabilities, now Microsoft can just blame them and feed everything you do into Bing for ad money as well as telemetry/spyware without you blocking it, that's effectively double-dipping all in on stupid windows users.

↳ reply — Marsymars:

Hey now, I use Edge on macOS.

Top #2 — paxys:

All the software freedom people here will do everything except use Firefox.

Top #3 — unlimit:

I never stopped using Firefox.

↳ reply — Krssst:

It always feels weird to see so much more complaining about Firefox than Chrome and Edge, while it's still so very much far ahead in terms of respecting user choices. All AI features can easily be turned off (and I am assuming those respect user privacy better than the ones of Chrome/Edge). When you say no to Firefox, it remembers it. Chrome keeps asking me every few days / weeks to become the default browser.

If we lose Firefox the web is going to be a worse place to be in.

→ HN comments  ·  188 points  ·  240 comments

Top #1 — vineyardmike:

Those in power seem to think that a compute advantage is necessary to maintain economic advantage. Despite this, the voters universally hate data centers, so much so that both NY and TX have data center construction moratoriums.

Despite this, those in power refuse any environmentally aware energy policy nor water policy. Despite this, companies have been trying to be environmentally aware for either moral or marketing purposes to get social permission to build DCs. Despite this, companies are “forced” to build their own “dirty” power sources since political power is disrupting green energy build out. Despite this, demand is still high so that compute costs are continuing to rise. Despite this, our economy is increasingly predicated on a few companies using more compute than geopolitical rivals with cheaper electricity. Despite this, those in power are, again, unwilling to create a compromising policy (aka clean energy) that voters actually would demand before tolerating more DCs.

What’s going to lose first? Voters/environment, politicians/the oil lobby, big tech, or the overall american economy?

↳ reply 1 — fps-hero:

Look to Australia if you want to see how these energy policies work out in the long run.

Energy is big money. Anything that disrupts the status quo is immediately vilified. It got to the point that our major party at the time lead a “gas fired” green energy policy.

They got voted out, the new guys implemented a subsidised home battery policy. In the space of a year it has destroyed the economics of a carbon fuelled grid. All the economics pointing to renewables being the way forward were proven correct.

Everyone gets their back scratched. If you’re a consumer and you invested, you have a massively discounted personal power station that generates you money. If you operate the grid, your transmission costs and peak transmission costs went down. General consumer prices went down because of a reduction in peak electricity use driving up retail costs. Everyone in won.

Except big gas, who cannot turn a profit on a renewable success story. Don’t let rhetoric become your reality.

↳↳ reply — zomiaen:

It's absurd how the propaganda against green energy on certain political spectrums essentially turned into, "It's all money/cash grab for green energy companies!"

As if the established industry has no incentive in the opposite direction? Everyone is going to act in their best interests in this system, that's how it's designed, the whole point of government is to mediate the system in the best interests of the whole.

↳ reply 2 — t0mas88:

The overall American economy is going to pay the price in the long term. And current foreign policy is reducing trust in America as a reliable and stable partner, which will make things even worse.

↳ reply 3 — roncesvalles:

Exactly, this is all a symptom of underinvestment in grid capacity, and datacenter companies having to cobble together whatever electricity source they could get their hands on at the last minute.

It's a complete failure of the state in having any kind of foresight.

China, in comparison, has around 2X excess electric production capacity in their grid. That's what made mining Bitcoin, and now ultra-cheap LLM inference, economically viable.

Top #2 — orbifold:

All this AI doomerism is hiding the obvious and immediate harm that will cause a mass extinction event. Funnily enough people at Anthropic are predicting the end of the Anthropocene, completely absolving themselves from any responsibility for the actual reality that the Anthropocene is defined by human made geologically noticeable changes. Like causing a global climate catastrophe.

↳ reply — DennisP:

It's possible to have more than one existential risk at a time. We even have a third, since nuclear war doesn't seem completely impossible these days. We probably shouldn't ignore any of them.

↳↳ reply — roncesvalles:

Nuclear war is hardly an existential risk.

An all-out global nuclear war (all nuclear nations landing 10% of their warheads in populated areas) would be a disruption comparable to COVID at best.

Top #3 — jstummbillig:

Note that this is about absolute pollution, not relative to power output. Pollution is high, because power output is high. Relatively, for what it's worth, it's just a fairly normal gas plant.

So the most interesting thing to take away from this: There is a lot of demand for more power and in this instance it's satisfied by gas. Which is still much better than coal. Which is still much worse than renewables.

↳ reply 1 — lokar:

Mostly because coal is more expensive and slower to build

↳ reply 2 — t0mas88:

Luckily the Trump administration has just paid over a billion dollars for the second time to stop a renewable energy project...

→ HN comments  ·  130 points  ·  197 comments

Top #1 — YesThatTom2:

I suspect the Cold War of cyber warfare that is happening is orders of magnitude bigger than the public knows.

Imagine knowing this and not being able to get emotional support from your friends and family.

↳ reply 1 — kstrauser:

This was a big thing for drone people based out of Eppley AFB in Omaha. Wake up, drive to work, attack enemies, drive home, sit there eating dinner with your family trying to pretend you didn’t kill someone and can’t talk about it. Combat troops get to lean on each other after a firefight: “dude, good shot. That guy was about to get us!” You get validation that you did something to safe you and your friends, and it’s still a heavy burden, but you have people who were there telling you that you’re a good person and did the right thing.

The Air Force did something shockingly reasonable (according to the newspaper coverage): they require all pilots to attend mandatory counseling sessions. Then there was no sign of weakness or any of that toxic nonsense. “Ugh, guess I have to go talk to the doc today, and I don’t want to, but guess orders are orders.”

↳↳ reply — senderista:

I feel more compassion for the innocent people they killed and it's not even close.

↳↳↳ reply — OsrsNeedsf2P:

Didn't know we had terrorists on HN

↳ reply 2 — cushychicken:

Used to work in this field. You're correct: not being able to talk about it is the worst part.

This is equally true of successes as well as challenges.

I did some cool, very impactful stuff in that job. I will go to jail for a long, long time if I ever go into detail about it.

↳ reply 3 — cco:

Counterpoint, industry cyber security is so generally weak, if there were actually a secret "huge cold cyber war" on-going, we'd see real world impact.

Since we don't, it probably isn't happening.

Top #2 — Arun2009:

I wonder whether any American civilians are losing sleep over how the current administration and its supporters' robust "race realist" rhetoric and worldview might be used by adversaries for some very cruel and effective psychological warfare on military and bureaucratic personnel from minority groups in the US.

Top #3 — someothherguyy:

"Between early June and early July, as many as five individuals who worked in or closely with US Cyber Command died by suicide, based on internal communications, public records and sources. The deaths have raised concern among lawmakers and military leaders within the highly secretive command, which is responsible for defending US networks and conducting offensive cyber operations."

According to https://www.gao.gov/assets/gao-25-107121.pdf there are ~17,000 authorized positions

↳ reply 1 — paulmist:

Assuming a predominantly male population, the suicide rate should be around 20 per 100000 per year. 5 people in a month gives 18x higher rate. Too short period to draw any statistical conclusions but it’s a large blip.

↳↳ reply — Rebelgecko:

Suicides are not independent statistical events and sadly often come in "clusters"

↳ reply 2 — 01100011:

> In some cases, they suffer from insomnia, memory loss, rumination and physical illness,

So, like most middle-aged desk jockeys then?

> “There’s always a crisis and everything’s always a priority,” said Schoka. “We would burn people out faster than we could produce them.”

Sounds a lot like my job...

→ HN comments  ·  134 points  ·  10 comments

Top #1 — tombrossman:

Do any European space agencies make live weather imagery available at high resolution? I'm looking for a direct link to JPEGs that update regularly, to use with a cron script to use as desktop wallpaper.

I can easily do this for North America using imagery here, for example: https://www.star.nesdis.noaa.gov/GOES/conus.php?sat=G16 (bonus, imagery is Public Domain).

This direct link provides an updated image every five minutes, with even higher resolution available if you like: https://cdn.star.nesdis.noaa.gov/GOES19/ABI/CONUS/GEOCOLOR/2...

Every time I revisit this idea and look for western Europe, I find nothing. Or, very low resolution, or some clunky, proprietary "viewer" forced on users. Is there really nothing similar for Europe?

↳ reply — plantain:

https://highsight.dev/ should work for that!

Top #2 — vdfs:

An other intesting use for this data is OSNIT about war, like recent conflicts it's used to confirm hits on specific places and sort validation for state-media reports

Top #3 — maelito:

In case someone wants NASA firms wildfire 7 day vector tiles : http://serveur.cartes.app/pmtiles/MODIS_C6_1_Global_7d.pmtil...

→ HN comments  ·  276 points  ·  86 comments

Top #1 — fcanesin:

Maybe was this that was the last drop for Sundar.

Demis: "I have a new amazing breakthrough"

Sundar: "Great! We really need a answer to Sol and Fable"

Demis: "They are completely owned in typhoon forecasting"

↳ reply — trescenzi:

Ironically typhoon forecasting, at this moment, is more valuable. These predictions are matters of life, death, and billions of dollars in damage.

↳↳ reply — esperent:

How does predicting a typhoon prevent billions in damage?

It could save thousands of lives because people can be evacuated if you can predict a few hours or a day further ahead, or the path more accurately. You can save some damage by moving ships and vehicles.

But you can't evacuate buildings or infrastructure.

↳↳↳ reply — baq:

You can board up a helluva lot more stuff in a week than in two days. Crucially, you can move more of the most expensive stuff out of the storm surge zone, which is where the biggest damage happens and try to flood proof more of the things which can’t be moved.

Top #2 — dgellow:

This is really cool, please more of this from the AI folks! That’s way more impactful and interesting than another coding agent

Top #3 — jen729w:

I just discovered typhoon/cyclone predictions and they're insane. I get mine via https://zoom.earth (whose iPhone app is terrific).

Here's a selection from Typhoon Dolphin, currently sitting off the east coast of China.

    Dolphin continues its slow, trochoidal Z motion, generally heading westward deeper into the East China Sea. Over the past 12 hours, the system completed another cyclonic loop and has decelerated, exhibiting continued meandering prior to establishing a sustained westward track.

    The erratic motion witnessed over the past two days is attributable to a weak steering environment produced by a break in the subtropical ridge 2 over Korea, combined with the dynamics where the inner core is cocooned within a much larger parent circulation.

    While the general steering pattern is weak, a mesoscale deep-layer ridge is seen building over southern Japan.

https://zoom.earth/storms/dolphin-2026/

Here's Chan-hom, which threatens to make my birthday a windy day here in northern Japan.

    Intensity guidance is in good agreement overall. However, the JTWC forecast is placed lower than all the guidance save for Google DeepMind over the next 36 hours, before joining the consensus envelope (which peaks at 95 km/h (50 knots) at 60 hours) through the remainder of the forecast.

https://zoom.earth/storms/chan-hom-2026/

↳ reply — trescenzi:

If you’re just getting into this tropical tidbits[0] is my go to for more raw data. Less pretty than zoom earth but also an interesting place to see what the models are predicting on each of their runs which is then interesting to compare to actual forecast guidance.

1: https://www.tropicaltidbits.com/

↳↳ reply — algo_trader:

I am getting into cyclone predictions (for maritime scheduling)

Is there a basic/freemium resource for past events? Mostly just very coarse spatial/temporal maps of past events

→ HN comments  ·  238 points  ·  73 comments

Top #1 — joss82:

This backdoor only appears on decades-old VIA C3 embedded x86 processors

↳ reply 1 — K0balt:

TBF the specific backdoor isn’t the point of the article. It’s a cautionary tale. The point is that practically all systems above the MCU level, and even some of those, have lower level systems that are often undocumented or not intended for use by the hardware designers, much less the end users. Those systems often have extremely low level access to system resources.

For example, I am building a device that records motion data, video, audio, and lidar imaging. Inside the 6 dollar IMU and the 12 dollar lidar sensor are powerful processors that load binary blobs provided by the manufacturer. The lidar could potentially gain access to any of the system data stored on the SPI bus, which includes the bulk storage and secondary RAM for the system. It could exfiltrate that data using its laser to anyone within a few hundred meters in the laser fov. It could also receive remote c&c over its optical sensor. The only thing that prevents that from being the case is that I trust the blob does not include the code to do those things, but it would be trivial to replace the blob with one that does.

Millions of devices are made that include basic wifi functionality. often, this comes in the form of a dedicated WiFi module. Those almost entirely consist of a powerful processor running a proprietary binary blobs, connected to some internal bus of the system that may give it access to some or all of the functions of the device, or at the very least could cause the device to malfunction. These WiFi phy modules are sub$1, pervasive, and often built in to devices that do not have any advertised connectivity features. A threat actor that has knowledge of an attack surface for that opaque blob can probably cause >50% of the connected devices built with that product to malfunction, in some cases in serious and dangerous ways, and sometimes to exfiltrate data that might be compromising or valuable.

That’s what this article is really about.

↳↳ reply — hnuser123456:

I recently got an air purifier. The touch button controls for adjusting the fan speed didn't seem to be working, so I emailed support.

They had me download their app, link the air purifier, and give them its MAC address. Then they asked me to try pressing each of the buttons a few times and email them back. I did so, and they responded that they re-calibrated the buttons using my touch samples. It worked.

↳↳↳ reply — evilduck:

That’s insane. I actively avoid buying things that are pointlessly internet connected nowadays. An air purifier’s buttons should be simple electromechanical switches.

↳ reply 2 — littlecranky67:

They should have mentioned that in the first line of the github readme, not burried deep down in the text.

↳ reply 3 — userbinator:

It's not even a "backdoor", it's documented in the datasheet...

http://datasheets.chipdb.org/VIA/Nehemiah/VIA%20C3%20Nehemia... (page 82)

...which along with the already publicly-known microarchitecture of the C3 makes this statement sound like total nonsense:

The rosenbridge backdoor is a small, non-x86 core embedded alongside the main x86 core in the CPU

I remember laughing at this with a few others knowledgeable in x86 when it first came out; a self-proclaimed "security researcher" who somehow failed to RTFM.

There's even a Wikipedia article about it now, with a link to the alternate instruction set documentation: https://en.wikipedia.org/wiki/Alternate_Instruction_Set

Top #2 — codedokode:

This shows that large companies making closed-source CPUs cannot be trusted. No doubt they would add whatever the government asks them to add.

What can be done to mitigate this? One option would be to buy a large FPGA and flash it with an open-source CPU. Another would be to emulate a CPU, working with encrypted data and commands, so that even if the backdoor in a host CPU tries to overwrite memory, it would only crash the emulated OS. One more option would be to run the code in a Virtual Machine like QEMU which translates the code and prevents issuing unknown instructions.

↳ reply — anthk:

Once you control the hosts CPU it's game over for the guest. The best you can do is to fetch old PPC G4 Apple computers or Thinkpads.

Top #3 — blueflow:

As noted by userbinator: https://news.ycombinator.com/item?id=49220030

Not a backdoor, but a documented CPU feature.

The whitepaper about rosenbridge cannot be published because it would constitute scientific fraud.

↳ reply — alfanick:

> Not a backdoor, but a documented CPU feature.

I would be even more explicit and call it “Not a backdoor, but a documented feature of ancient de facto unused Via C3 CPU.”

→ HN comments  ·  242 points  ·  45 comments

Top #1 — jvanderbot:

I ran a few research tasks at JPL, small dollar studies. We'd often recruit part time work from engineers around the lab through informal networks, (instead of the official channels that segment by "specialization").

One year, one of the developers on my task was the apparently last person who could encode the command sequences for Voyager 2 (hoping memory serves me correctly - event though it was only 8 years ago). She would occasionally come late to meetings because they were dealing with some trouble with the old satellite.

Pretty amazing how little is set aside for these projects that most the staff keeping it alive have "day jobs". I hope she's still uploading

↳ reply 1 — wildzzz:

To be fair, nearly all of the science payloads on Voyager 1 & 2 have been shut down. The probes are doing very little. The doesn't take away from the very unique science they are doing but its a far cry from what the program offices were doing in 30-40 years ago. I'd expect anyone working on a Voyager probe in 2026 to need something else to do most of the day.

↳↳ reply — w-ll:

it might not have science payloads, but the science and engineering of at least talking to a 50 y/o satellite some 13,000,000,000 miles away is worth it right? right?

↳↳↳ reply — tempestn:

Yes, that's why they're still doing it. It just doesn't fill up a 40-hour week.

↳ reply 2 — stringfood:

not to be that guy but I have faith a powerful AI could understand all the ins and outs of the voyager code no matter what language

Top #2 — gjkood:

Just the other day I heard about the heroic effort[0] to restore communication with Voyager 2 after they inadvertently changed the antenna to point in the wrong direction due to some planned manoeuvres.

Definitely worth a listen if you get the time.

For me personally, I feel this project is the greatest engineering effort in human history bringing together the work of Newton, Kepler, the mathematics of the celestial navigation, gravitational assist techniques[1], etc. necessary to keep a little probe on its route accurately over 50 years in space.

[0] https://www.npr.org/2023/08/02/1191341035/nasa-voyager-2-spa...

[1] https://en.wikipedia.org/wiki/Michael_Minovitch

Top #3 — bcraven:

Ah, good, an opportunity to recommend It's Quieter in the Twilight (2022). Beautiful documentary.

https://www.imdb.com/title/tt17658964/

→ HN comments  ·  294 points  ·  140 comments

Top #1 — ronef:

For anyone reading this and jumping to a broader conclusion: the Nixpkgs core team disbanding does not mean Nixpkgs or Nix is dying. It does mean that this particular structure was not sustainable, very important contributors burnt out and we need to do better, faster. We need to continue learning from this and continue building a stronger ecosystem that prioritizes the contributors who are the only reason any of this is possible.

Personally, I'm sorry and grateful. Sorry that it ended in core folks being burnt out. Grateful since they did some of the most amazing work, more than anyone can imagine two people doing.

↳ reply 1 — tomberek:

The Nixpkgs Core team was established in Sept 2025 (https://discourse.nixos.org/t/establishing-the-nixpkgs-core-...), so it is a fairly new concept and idea. They've accomplished some good things as highlighted in the post, but are now stepping down. Yes, we'd prefer for the team to exist, but we've also functioned without one for ~20 years. It may take multiple iterations to bootstrap the concept and to figure out the right structure - or we may find it isn't needed. So no, this is not a critical emergency. It's a return to the status quo of late last year.

That being said, I think it is still a concept worth pursuing.

↳↳ reply — xyzzy_plugh:

I agree with the concept but the name is terrible. "Core team" gives one the impression that this is the most important team to the project. It's not! It could be called the "governance facilitator" team or admins or something.

Please name the next iteration something less spicy.

↳↳↳ reply — jonringer117:

There was an architecture team which existed, I would assume the usage of "core" was to disambiguate.

https://github.com/nixpkgs-architecture

↳ reply 2 — aliasxneo:

Correct, but it's another domino in a broader trend that appears to have a common root. It was only a year ago that the moderation team mass resigned pointing the finger at the steering committee [1]. It's not been the same people for the last two years, but clearly there must be systemic issues in the people getting elected to this board for there to be so many persistent problems.

Also see roberth's resignation [2], and Gabriella439's retrospective [3], and then there's Jon Ringer's thing, the list goes on.

[1]: https://discourse.nixos.org/t/a-statement-from-members-of-th...

[2]: https://discourse.nixos.org/t/stepping-down-from-the-nix-tea...

[3]: https://haskellforall.com/2025/09/steering-committee-retrosp...

↳ reply 3 — shevy-java:

> the Nixpkgs core team disbanding does not mean Nixpkgs or Nix is dying.

Nice attempt to try to rescue here, but everyone sees that NixOS is dying.

When a group of, say, ten people give up in a moment, that does not bode well for the future of NixOS.

Top #2 — stingraycharles:

“Our experience is that the Steering Committee as an institution lacks a native instinct for the delegation envisioned by the constitution, while also not being sufficiently engaged and cohesive to handle individual decisions at those levels itself.”

This is an almost poetic description of micromanagement. I really like Nix and have been running it as my main OS for several years in the past ~ 10 years.

I don’t think the issues they have are unsolvable, it just appears that the governance model they’re trying to have is not working out, and it’s very difficult to roll back.

↳ reply 1 — kesor:

The governance of the project went into political extremes instead of focusing on technology. That made it impossible for anyone other than political extremists to continue in governance positions. Obviously these people had too much on their plate and couldn't find anyone who would join to help them, as they chased everyone away. Burn out was inevitable.

↳ reply 2 — astrobe_:

The whole paragraph describes quite well what I feel at the company I currently work for. I think it is more accurate to say that it is interference (rather than micromanagement) and goalpost shifting because whoever is/are in charge don't have a clear idea of where they are going.

Top #3 — klodolph:

There was a sweet spot in like 2024 when it seemed like everything was possible with Nix, but now it seems like everything “experimental” is permanently so (like flakes), packages I care about are not as fresh as I want, I have no mental recall for Nix commands…

Meanwhile, my company is using Nix for everything heavily internally. Everyone gets their dependencies via Nix unless you are like using PAM or something.

Reminds me of Bazel, in the way that it gets adopted by companies with developer support teams (because it solves real problems) but feels frustrating for us ordinary folk. Nixpkgs is kind of a critical part of “ordinary folk” and with the core team disbanding, I feel like my personal moves away from Nix (for projects) are proving correct.

↳ reply 1 — mikepurvis:

I was lead on a Nix adoption effort for a few years that was kind of like that: it solved real problems, unlocked far faster, smaller, and cheaper builds than would have been possible any other way, and let us ship delta updates over crappy wifi connections to Linux computers on robots. Flakes were a perfect fit for our model, and we were just in time for stuff like up to date versions of cuda and tensorflow to be delivered via nixpkgs.

In my mind, Nix was unstoppable, and I particularly loved how empowered I imagined developers would feel. No mystery-meat CI scripts pushing packages to distant infrastructure that no one understands or even has the permissions to interact with, just the entire build system in one repo, trivially cloneable and hackable... add patches or build steps to anything and it's the same build as always, build it locally or send it to Hydra, it doesn't matter.

A devs "got" it and did leverage those things, used PRs to do safe evaluation of bumps to core dependencies, but I think on the whole it was regarded as cool but not tractable, and a few years after leaving, it sounds like plans are being laid to replace it all with something containers or whatever.

Very frustrating, particularly in a world where it should be trivial to identify the 5-10 typical tasks that people want to do with the nix code, and set up Claude skills to handle those. Given how easy and self-contained the build-test loop is, it feels like an almost perfect fit for agent-led development.

↳↳ reply — rossng:

We're using it in the same way - running NixOS on robots.

Overall I think it is succeeding, though it definitely doesn't 'just work' out of the box. We've invested a fair bit of effort into things like:

- containerised services[1]

- UI for branch deployment/service management

- delta patching (at the byte level, to save bandwidth)

The main problem was always that nixlang/nixpkgs are arcane and have a steep learning curve. The majority of your colleagues (in any workplace) don't care about build systems and just want things to work. I think that has been more or less solved by LLMs, and it's now more viable than ever to adopt Nix.

[1] https://bou.ke/blog/nixos-containers/

↳ reply 2 — haswell:

Can you share more about what made you move away?

I’ve been running a NixOS based homelab for awhile now with 5 physical hosts and about 30 NixOS containers/VMs in an Incus cluster and I can’t imagine moving away from my central Nix repo and ability to rebuild/upgrade the entire fleet in one command and feel confident that things will work.

While I’m concerned about the disbanding, it would take quite a lot to make me look elsewhere, and there’s enough critical mass that I feel confident others will step up.

→ HN comments  ·  199 points  ·  66 comments

Top #1 — firasd:

Just realized that there are basically no American open models right now ever since the Llama series was abandoned. Basically Gemma and GPT-OSS I guess?

Ah but Mira Murati's new Inkling is Apache 2.0

But it makes sense that if you're a university researcher you are thinking about what's a model that will be open weight and developed over the long term and doesn't raise 'Chyna' concerns in Washington DC

↳ reply 1 — ipsum2:

There's a bunch of American open models. Inkling, Nemotron, Trinity come to mind, but I'm sure there's others.

↳↳ reply — firasd:

Just looked into some Nemotron stats

Looks like on <https://arena.ai> agent arena (grouped by lab) Nvidia is 15/15 (much worse than Thinky and Mistral) and on text arena it's 18/27

On <https://openrouter.ai/models?order=most-popular> I definitely see usage though (probably mostly cause Nemotron 3 Ultra is free) the grouped order is DeepSeek, Tencent, Xiaomi, OpenAI, Z.ai, Nvidia

↳↳↳ reply — coder543:

I think glancing at a random snapshot from today misses all the context. Nemotron 3 is far more significant than you're giving it credit for.

At this point, Nemotron 3 is really an 8 month old model series. That's when Nemotron 3 Nano was released, and the Nemotron 3 Super/Ultra models this year are obviously based on that recipe, mostly just bigger with a few tweaks here and there. Against today's models, no, not that interesting. Each of the Nemotron 3 models were briefly competitive when they launched, but never exceptional, and less competitive with each scale up. The fact that it took so long for Nemotron 3 Ultra to launch really hampered its competitiveness.

The Nemotron 3 series is extremely open about training recipes and training data, far more open than most open weight models, and that is valuable.

Before Nemotron 3, Nvidia had never released a single LLM that I would consider interesting at all, so Nemotron 3 was a big step up. The closest thing was Mistral NeMo, but a significant part of the credit there goes to the Mistral team, not Nvidia.

Given how much Nemotron 3 improved, I'm curious to see if Nemotron 4 will take them to a leading edge level instead of just briefly competitive.

(Nvidia released a Nemotron 3 and a Nemotron 4 like 3 years ago... this year's Nemotron 3 is entirely unrelated. Nvidia's naming schemes leave a little bit to be desired.)

↳ reply 2 — loeg:

I would not be shocked if another open model eventually shakes out of Facebook (based on Zuckerberg's public remarks).

↳ reply 3 — wmf:

Also Nemotron and Arcee.

Top #2 — Aeroi:

https://science.osti.gov/-/media/grants/pdf/foas/2026/DE-FOA...

Top #3 — andsoitis:

Does Europe have an equivalent program?

↳ reply — shakna:

As part of a much larger series of initiatives towards digital sovereignty, yes. [0]

[0] https://commission.europa.eu/news-and-media/news/strengtheni...

↳↳ reply — andsoitis:

Oh. Being buried in hierarchy does not inspire hope.

→ HN comments  ·  162 points  ·  112 comments

Top #1 — aliasxneo:

Part of my career encapsulates a period where I was a PLC programmer, installer, commissioner, and troubleshooter for massive build outs (sky scrapers, data centers, laboratories, factories). Interestingly, this was after years of teaching myself software engineering, eventually participating in large open source projects. The clash of entering the PLC world was _extremely_ harsh.

Let me give an example: I once worked with an integrator who was working on an AHU feeding an extremely critical portion of a datacenter (I was a lead by this point and mostly played babysitter). During certain points of the day you couldn't open the door to this room due to negative pressure because the logic was over-ramping the exhaust fans. As I watched this contractor work, I saw him open his laptop, with Windows on it (because Microsoft has had a death grip on this industry for decades now), and proceed to backup the PLC program into a massive folder with God knows how many other "customer projects" he was carrying around in this thing. He then proceeded to go do some physical checks in the field, came back, and prepared to upload the fixed program. As I watched, I noticed he _grabbed a backup from ANOTHER customer_ and I immediately had to intervene. Who knows what untold damage I saved from that single move.

I tell this story to demonstrate just how far into the dark ages this industry is. I vividly recall coming into the data center for a fortune 50 company, one everyone here would know, and being astonished that they had never heard of Network Attached Storage or RAID and why they might want to consider a disaster recovery plan for their multi-million dollar mechanical plant.

This industry is in _desperate_ need of strong technical help, but unfortunately the "higher ups" tend to be the same people who are "comfortable" with the way thing are and refuse to move. I literally tried for a decade before giving up and moving into software engineering proper.

So, anyways, just imagine the most archaic and barbaric set of IT software, controls, and procedures, dumb that down even further, and you've landed on the infrastructure/teams that operate probably half of critical infrastructure.

↳ reply 1 — amluto:

Another odd observation about the industry. For better or for worse, I inherited a little system controlled by a Fanuc PLC. It had a bug and would occasionally get stuck in a bad state. By some minor miracle, I managed to track down the person who had originally programmed it (now retired and moved out of state), and he emailed me a file and introduced me to his apprentice (who lives in a third state). I found a trial version of the design software, opened the file and found the bug (despite never having seen this style of programming before). Since I had no way to upload a new file (and since it wasn’t clear that uploading it without a matching outdated and not easily available copy of the programming environment would be wise), I convinced the apprentice to come visit while we was nearby for another client. He did not see the bug even when I pointed it out, but I convinced him to make the change and upload the modified file. Now the system works. (He had a Windows laptop with a giant folder of customer files, of course.)

While he was on site, I asked him why there was nothing on the PLC or its enclosure to identify his old boss or give any contact information, and he seemed surprised and told me that no one ever does that. I asked how a new owner is supposed to get support, and he shrugged.

If this thing ever fails, I’ll probably replace it with an Arduino or an ESP32 or something along those lines. Or I’ll just find something off the shelf to replace the entire system.

↳↳ reply — spauldo:

A microcontroller based solution is rarely a good replacement for a PLC. You can find integrators willing to come out and troubleshoot your device in the field a lot more easily than embedded progammers and vendor support is usually excellent. When your plant is down, that sort of thing matters. PLCs handle harsh environments for decades, whereas with a microcontroller all the thermal issues and power conditioning are all on you. When something fails, it's usually just a matter of pulling a spare module off the shelf and plugging it in.

If that's an old GE 90-30 (a common FANUC system), it's time to replace it. You're currently in that period when the CPU has reached end of life but the modules are still supported and the vendor has an easy upgrade path. Emerson owns the brand now, and can sell you a new unit that'll run your old program with minimal changes. You can call them and get a list of integrators in your area. It'll be pricey, but you won't need to do it again for another quarter century.

↳ reply 2 — steve_adams_86:

Everything you're describing mirrors my experience with remote scientific deployments and lab equipment (wet labs in particular). I support a team which manages dozens of these, and while it's shocking how inefficient and unsafe their practices are, it's similarly shocking how poorly the industry has kept up with or even tried to establish better practices. We get brand new equipment that costs in the realm of $10k for a tiny logger, and the thing is barely capable of remotely managed deployment of version controlled firmware. There are no tools in place to allow proper testing of the firmware unless you have the cash to buy extras and build test harnesses from them. The firmware itself is often in languages that don't have proper testing frameworks, type systems, IDEs, etc.

It's HARD to do the right thing. Dragging and dropping files into proprietary hardware management programs is the de facto standard.

Then you layer on top the unfortunate reality that sometimes electricity does weird stuff, people design weird circuits or wire the wrong components in, and firmware tends to have to deal with non-deterministic inputs a lot more often than, say, an API on the web. It's rough.

The pay is also so much worse in my experience.

↳ reply 3 — technion:

Im supposed to be doing endpoint work with people working in this field amd basically have to convince compliance they'll need to be local administrators to do their job. They get onsite and dont know what app they'll need until they look at a plc model. They'll download something from a .tw ftp server and its an unsigned executable that absolutely must run as admin to program these things.

It is amazing how much of the sysadmin community just doesnt believe this is a thing you need to work with, everyone insisting its just security people being lazy and so on.

Top #2 — clbrmbr:

There are many wireless pump-and-reservoir systems that while not internet connected, use insecure RF links. These local RF (and casting a wider net, Bluetooth) interfaces are also ripe for abuse.

↳ reply 1 — barbazoo:

If that means you need to at least be physically present then I'd say that's a lot of protection already. Means someone in a foreign country can't simply get lucky fuzzing.

↳↳ reply — judge2020:

Alternatively, small drones exist, and so do low power devices you could slingshot/shoot into a secure area and set to auto wipe after they've done their job. Maybe could even build it out of biodegradable material so it'll clean up in short time.

↳↳↳ reply — closeparen:

If you've got secret agents on enemy soil who can get near targets with drones, you're probably just bombing them.

↳ reply 2 — Wowfunhappy:

Wouldn't the physical facilities themselves have security?

Top #3 — chmod775:

At modern population densities, basic infrastructure breaking on down on a large scale can kill millions in a matter of weeks.

The largest threat isn't bombs falling on our heads, it's incompetent fools leaving the door open to their enemies.

These aren't mistakes that can be excused. Failing in one's duty to steward important infrastructure must mean immediate replacement of leadership.

→ HN comments  ·  126 points  ·  113 comments

Top #1 — gitowiec:

I have a bad Ethernet cables through unused chimney from my apartament to the basement. The workers I asked to pull it just did it brute force and 2 of pairs are broken. When I was chatting with Claude to see if I can use it somehow to send Ethernet packets one of the Claude responses was to use broken cables as a pilot to pull fiber channel!

↳ reply — SOLAR_FIELDS:

I tried this in my house that has ancient copper telephone lines. I was trying to get some level of walled ethernet. I wanted to use the existing wires to pull new ethernet as is discussed here, sadly the builders had stapled them to the interior studs when they ran the telephone lines back in the 70's.

Top #2 — amelius:

Somehow, on HN, suggesting to use an LLM to solve some problem gives you negative responses.

But then ... posting to HN that you've used an LLM to solve a problem is OK?

↳ reply 1 — profsummergig:

Seems like the divide between the voters who vote in the Primaries and the voters who vote in the actual Election. Vocal minority vs. Silent majority.

↳ reply 2 — zamadatix:

I can't say I've seen the same universally, but, if it is s trend, it would seem one invites discussion between humans on the site while the other seeks to minimize it.

Top #3 — aidos:

Yesterday my 8yo prompted (using voice recognition) Claude to make a game where it would play her a song (say Twinkle) and she had to play it back and get scored. The UI was a nice piano with visual feedback. The laptop key served as the keys until I prompted for midi support so she could use the electric piano. The whole thing took about 15 minutes.

Meanwhile, one peak at the code and you can already see the state starting to become a bit of a spaghetti mess.

It’s a funny time to live through. A lot of code is being written and a lot of it is going to be a real future burden.

↳ reply 1 — bdcravens:

> A lot of code is being written and a lot of it is going to be a real future burden.

This assumes that the models of the future won't find it easier to just throw the code away and rebuild it

This also assumes that the same application build by humans wouldn't become a "spaghetti mess".

↳↳ reply — skywal_l:

Reminds me of Niven/Pournelle's "The Mote in God’s Eye" where the aliens have a very pragmatic/Jury-rigged approach to technology and everything is more or less improvised.

I have noticed this with co-workers also, when you have the ability to read/write/understand things very quickly, you tend to spend much less time on making things tidy, clear and maintainable.

↳ reply 2 — pjmlp:

Those of us involved in offshore projects already know the feeling.

→ HN comments  ·  291 points  ·  172 comments

Top #1 — DrJokepu:

> But that’s not what happened at all. Godier proceeded through a series of escalations up to the App Review Board and the Review Board responded that they determined the original rejection was valid because, I shit you not, “We understand that the app includes a live tarot reading feature.”

> It has no tarot function, no horoscopes, and nothing that I, or anyone else I’ve asked, would associate with astrology.

I have dealt with the Byzantine App Store bureaucracy before. What you do here, is you submit a new version, pretty much identical to the previous one, with the comment “tarot function removed”.

↳ reply 1 — bdangubic:

this often works even if you did in fact have “tarot function” and didnt remove it

↳ reply 2 — rkunde:

Ah yes, the Herobrine gambit

↳ reply 3 — echelon:

Is Gruber starting to realize Apple's monopoly is bad?

App stores should not exist in a world where only two phone operating systems exist. All apps MUST be installable via web.

Come on, EU. Help us out here. Free us from distribution tyranny. Do it before Google morphs into the same hell beast.

Soon you won't be able to read unapproved, unattested bytes. It's coming. We need as many freedoms as we can get.

Apple can be a great company without the app store monopoly. So can Google.

Top #2 — aliasxneo:

In my last SRE role I was the unfortunate victim of having to maintain a mobile app submitted to both the Android and iOS app stores. I would never wish it on my worst enemy. It was the one thing that I could absolutely _not_ guarantee any sort of reliability with. It was all up to the random person who happened to review the current release and the release process could be as quick as a day in the best circumstances or up to two weeks if we got a particularly difficult reviewer. They seem to have absolutely no context sharing between reviews.

Top #3 — duxup:

I’ve developed iOS and Android apps, store approvals are always a crap shoot, and it’s not just approvals, it’s paperwork… sudden rejections over something that was fine months earlier and getting pointed to a document that could mean ANYTHING.

I got rejected with a screenshot from an app that doesn’t have anything like the screenshot, it took me several rejections to realize someone reviewer watched some old YouTube video I submitted ages ago and didn’t like something that was fine before…. They have a login where they can see the first page , I gave it to them , but they chose to reject because of an old video.

I recently signed up for a personal dev account for the play store and I have to get 12 friends with android devices to “test” it before they’ll consider my app at all…

That is to say I have to go online and horse trade with 12 people (hell if I know who these people are or what their unapproved apps do…) and get them to test mine if I test theirs. That alone is suspicious kinda behavior but google all but requires it now.

Meanwhile on my work account I’m publishing just fine.

→ HN comments  ·  166 points  ·  55 comments

Top #1 — laichzeit0:

Interesting project. I would love if you could switch fonts to something like New Athena Unicode.

I built something similar to this by cloning the Diogenes repo and getting Claude to re-implement it in Python (it’s a very old battle tested Perl code base, so a great reference implementation) and using the TLG database for the Greek and Latin texts. You can take this even further by integrating it with the Barrington Atlas (there are scans on Anna’s Archive) for looking up ancient place names, so you have dictionary + map lookups. If you’ve read any of the Landmark series books you’d known what I mean.

Also better if you can generate chapter by chapter critical apparatus on difficult grammar and Anki decks. It’s an annoying part of learning these languages to have to stop and look up words, I like spending a few days learning vocab before reading and it makes it so much more pleasant than having to stop and look things up the whole time.

Obviously all this stuff is copyright so it can never be shared, but I don’t care it’s for my own personal use. I also bought like 6 hours worth of Ionnis Strattakis’ recordings where he reads a bunch of Ancient Greek in reconstructed Attic pronunciation and fine-tuned a text to speech model (styletts2) with full accent markings and breathings. It’s extremely natural sounding. My long term goal is to have a personal tutor that I can speak Attic to and basically have lessons with everyday (speech to text -> llm-> text to speech). All the pieces are there to actually do this.

LLMs have been an absolute game changer for me who is a hobby classicist but also knows how to build software. They are so amazingly good at Attic Greek and Latin, it’s like having the best teachers in the word at your fingertips for some really niche topics that it wouldn’t be possible with otherwise. Also extremely good at managing, building, cleaning up, deduplicating Anki decks.

↳ reply — spudlyo:

> LLMs have been an absolute game changer for me who is a hobby classicist but also knows how to build software.

Definitely! I myself have been having a lot of fun[0] recently using LLMs to enrich one of my favorite beginner Latin readers[1] with audio forced-alignment, POS tagging, morphology, definitions, and other niceties that learners might appreciate.

[0]: https://hercules.hookbangsplat.com

[1]: https://archive.org/details/p1fablesoforbili00godl/mode/2up

Top #2 — tmshapland:

I'm always surprised to see that a large enough portion of the HN community are interested in classics that posts like this make it to the front page. Who are you all? What are you doing here?

My undergrad was in classics and viticulture, then did grad school in atmospheric science, which led me to tech and hackernews.

How did you get here?

↳ reply 1 — spudlyo:

I'm a recently retired software nerd who got into reading 19th century literature. I was greatly inspired by characters in these novels, most of whom had the advantages of a classical education, and I lamented how poor my own was. One day I idly wondered how hard it would be to learn a bit of Latin (as I was constantly having to look up Latin references) and once I discovered LLPSI[0] I was hooked.

[0]: https://en.wikipedia.org/wiki/Lingua_Latina_per_se_illustrat...

↳ reply 2 — sturakov:

I grew up with an awareness of Greek and Latin as part of western heritage education.

A lot of role playing games, especially from Square Enix,leverage a lot of Latin and classical education.

Top #3 — lsb:

Neat! Similar to my graduate thesis, NoDictionaries: https://nodictionaries.com/cato/de-agri-cultura/156

↳ reply 1 — lr4444lr:

Nice.

May I point out that in "Crudam si edes, in acetum intinguito", that "edes" is more likely to be the future of edere/esse "to eat"? (Just guessing by context.)

↳↳ reply — lsb:

Fixed!

↳ reply 2 — svat:

I remember encountering NoDictionaries years ago, and love it. Thank you for building it!

→ HN comments  ·  180 points  ·  179 comments

Top #1 — extr:

I would be really curious to hear from devs at Databricks what the experience of development is like internally. I work at a small startup with essentially unlimited AI spend budget - the entire point is that I should be turning to it at every opportunity since our human labor is so expensive relative to tokens. So generally it's like:

- Spend most time prioritizing/discussing what to do.

- Once that's agreed, use Fable 5 High + 5.6 Sol XHigh come up with a design + plan. Agree on the high level plan. (Usually this just comes down to choosing where the change belongs on the spectrum between minimal patch <-> full redesign)

- Use Opus 5 or Sol Med to execute

- Auto-fix bugs and CI until green + thermonuclear review skill x3.

- Manual interrogation of change/nits

- Come up with QA plan and have Codex Computer Use execute on it

- Manually spot check the final result (usually a sizable diff, thousands of lines, complete feature E2E, etc)

I probably spend like $80 a day at least but I produce the output of 3 or 4 2022 engineers and probably at better quality. So it's easily worth it. Would I save money by switching to GLM 5.2 and such...perhaps? IDK. At our scale it's not worth the time spent building the eval harness to actually understand the performance tradeoff.

↳ reply 1 — pizza234:

In our team's experience, the product of agents is generally The Homer (1). It does work, but it's vastly overengineered.

When I personally want tight code, I have to spend a considerable amount of time adjusting it manually:

- It needs to be trimmed down. In my experience, at least one agent I use struggles to produce minimalist designs, and it's very frustrating

- I need to consider whether there are solutions based on higher-level assumptions, that AIs typically miss

- I need to check whether there are off-the-shelf solutions - AIs like to reinvent the wheel

IMO, software production has become a mass-produced commodity in every sense - it's much more expensive to produce software manually, but the quality is not the same.

(1) https://simpsons.fandom.com/wiki/The_Homer

↳↳ reply — dieselgate:

It took me a few seconds of deliberating if The Homer was a reference to baseball or "The Odyssey" and then realized there was a footnote

> software production has become a mass-produced commodity in every sense - it's much more expensive to produce software manually, but the quality is not the same.

Agreed.

↳↳↳ reply — GoofGarage:

I believe “The Homer” is a reference to Season 2, Episode 15 of The Simpsons, “Oh Brother, Where Art Thou?”

Homer reunites with his long lost brother, who runs car company Powell Motors. Homer is ultimately tasked by his brother with helping to design a car for the “average man” that ultimately bankrupts the company for being wildly overengineered and costing too much ($82,000 in 1991-money).

Here’s the car: https://simpsons.fandom.com/wiki/The_Homer

↳ reply 2 — reqo:

IME this works until it does not. This approach works well at the beginning of a greenfield project, but at the same time because it is so easy to add features, you will likely ship something that is way too over engineered. And that complexity will not amortize over next increments and will more likely lead to the entire project being a black box only fully understood by AI. However a more careful use of AI for targeted surgical changes is far more ”productive” in the long term IMO.

↳ reply 3 — samesense:

You have an unlimited budget, and you only spend $80/day? I’m up to $3k/week, and still expanding.

Top #2 — lbriner:

There are a surprising number of articles like this along the lines of, "we started using AI tools and ended up spending millions per year".

On what planet do people start paying for things without keeping an eye on the costs and no-one notices until you have spent a crazy amount? I don't understand. You are either paying a fixed amount which you are happy about in-advance or you are PAYG in which case you would ballpark how much it costs.

Otherwise it reads a bit like a fake problem, because it didn't really happen, you just foresaw it (as you should) and added a few guide rails.

↳ reply 1 — habosa:

There really has never been another product priced like AI is being priced right now. Each of these things has been done before, but all of them together is new.

1. Insanely discounted starter plans. Claude $200/mo plan is like $5k-$8k of API rate usage.

2. Very limited cost visibility, they make it hard to figure out where you spent money (unless you're on the enterprise plan which is for people with unlimited money).

3. Nobody, not even the model provider, knows what your request will cost before it returns. You're writing a blank check every time you hit enter.

4. When you run out you run out very suddenly and disruptively. It's very hard to tell a developer on the 28th of the month "sorry, code by hand until the 1st of next month" so you tend to grant exceptions.

5. The price is changing all the time. New models come in, old models come out, prices change, caching behavior changes, harnesses change, etc. The cost of doing a single task is not predictable even if the task does not change.

6. Basically no volume discounting. Anthropic offered us 2% off for committing to $1M+ per year at API rates.

I manage AI spend for my team at work and I try really hard to keep costs under control but it's absolutely herding cats. Much harder than any other spending I've ever had to manage at work.

↳ reply 2 — ankitmathur:

Something underlying a lot of this is that pricing models for enterprise coding tools have changed from seat-based to consumption-based pretty quickly, as AI usage has exploded. For months, engineers were able to use unlimited AI for no marginal cost, but that's changed quickly.

In addition, we're seeing people applying AI to more and more use cases, so token growth is very significant. Paired with consumption pricing, it's brought this problem to the forefront very quickly for lots of companies.

Top #3 — sashank_1509:

I suspect that when it comes to hard complex software products, you’re better off ignoring agents and doing “trad coding”. What you lose in short term speed you gain in manageable complex codebases.

If you have a 500k line codebase and even > 50% is written by agents, you are in a world of pain that won’t justify the costs longer term.

Now of course, there are products that just involve lots of code but are not actually complex. This is generally the project with like hundreds or thousands of features but most of the features are separate and don’t actually interact in complex ways. Think a task management app with hundreds of features like calendar, email integration etc. there I think agents gives you more bang for the buck. Just my thought, using agents at work.

↳ reply — SOLAR_FIELDS:

> if you have a 500k line codebase and even > 50% is written by agents, you are in a world of pain that won’t justify the costs longer term

A bold claim to make with little to no supporting evidence

→ HN comments  ·  138 points  ·  73 comments

Top #1 — AnEro:

I've personally completely avoided the USA because of ICE, not because of principle of border policies purely from unpredictable treatment of people in custody with no recorse. Even if mistreatment are edge cases I've got no faith in the system to correct the wrong within itself. I've seen too many stories of legal americans and travelers not even getting bare minimum due process, or just being locked away for days no contact. I present and am read as a woman, if I get sent to male holding facility I'm just chum in the water. Which that decision could be made completely at the whim of the officer and the areas commanding officer, regardless of local laws.

↳ reply 1 — jancsika:

> if I get sent to male holding facility I'm just chum in the water.

https://en.wikipedia.org/wiki/LGBTQ_people_in_prison

↳↳ reply — AnEro:

>V-coding is the practice of assigning trans women placed in men's prisons to cells with aggressive cisgender male cellmates as both a reward and a means of placation for said cellmates, so as to maintain social control and to, as one inmate described it, "keep the violence rate down".[170] Trans women used in this manner are often raped daily; and this process has been described as so common that it is effectively "a central part of a trans woman's sentence"

↳↳↳ reply — NDlurker:

Oh my God. That's a blatant rights violation.

↳ reply 2 — xyzelement:

Where are you from?

Top #2 — pbnjeh:

Jail, jail, jail. Nothing less will stop these Gestapo-esque thugs.

And if you’re waiting for the mid-terms, you’re already acquiescing to their plans.

Also, it says something — to me, at least — that I have to think seriously before posting this. All I’m calling for is timely criminal prosecution of what I think and believe are crimes. I.e. for a lawful process. Yet doing so puts me at risk of an overbearing, unaccountable response?

This is some serious sh-t.

↳ reply — the__alchemist:

All enemies, foreign and domestic.

Top #3 — MertsA:

How does this square with FOIA obligations? I get ICE sidestepping transparency by avoiding creating the responsive footage in the first place, but once the body cam footage exists, what basis could they use to withhold it other than redacting footage for the typical reasons?

↳ reply 1 — jjav:

> How does this square with FOIA obligations?

ICE doesn't exactly have a track record of following any laws or courts in general. So I wouldn't expect them to answer any FOIA obligation.

This is the fundamental flaw uncovered in the US system by the current administration. Turns out that if one agency decides to go rogue and declare that no laws apply to them, there is no mechanism in place to prevent that.

↳ reply 2 — orthecreedence:

> what basis could they use to withhold it other than redacting footage for the typical reasons?

The basis is "come and take it." A court order doesn't work on the brownshirts.

→ HN comments  ·  256 points  ·  58 comments

Top #1 — Retr0id:

Related, and linked in the readme: https://github.com/xoreaxeaxeax/smiiiiiiiiiiiiiiii (using the slow instructions to break SMI)

↳ reply — jonathrg:

I wish they would just explain it in normal terms instead of this nasty LLM "engaging blog post" style

↳↳ reply — twothreeone:

Chris usually takes an educational angle, I don't think this is LLM-generated content at all it's just his style. I highly encourage watching some of his DefCon or BlackHat talks, they're fun!

Top #2 — monocasa:

It says in the rules

> Trapped/emulated/virtualized instructions may only time the trap, not the handler.

But I feel like that 12ms write to an ACPI IO port at current leaderboard position 8 is probably trapping to SMM and being handled there.

Top #3 — layer8:

Nop should be #1, because it is infinitely slow for what it does. ;)

↳ reply — mito88:

Strategy: nop does nothing. It opens the leaderboard accordingly.

Score: 1 cycles Time: 0 nanoseconds

↳↳ reply — layer8:

It opens the leaderboard as #27, so in the last place.

→ HN comments  ·  471 points  ·  282 comments

Top #1 — LaurensBER:

I've been using it extensively since the release and the best summary I can give is that it's good enough to use it for (almost) everything and cheap enough that the cost are irrelevant. I'm running it in Oh My Pi with a second instance running as "advisor" and even with 5-6 active sessions (effectively 12 streams) I'm struggling to spend more than 5 bucks per day.

OpenCode Go even has double limits temporarily so for 10 USD you effectively get 140 USD of tokens to spend. It would impress me if someone could burn that amount with "normal" usage. Even when running multiple sessions.

I have a Claude Max subscription but I've barely touched it, it just feels like a step back to have to think about limits and usage even though the models are stronger.

The beauty of intelligence at this cost (even if it's not SOTA) is that it opens a whole bunch of new use cases. Test failure in CI? Have the bot automatically propose a fix, its cheap enough that you can discard it w/h issues. Test coverage too low? Auto generate tests on CI for every pull-requests! Monitoring server logs, continuous security audits and investigating every received exception now becomes possible.

I'm thinking about having it automatically filter and re-rank my social media feeds so I can steer the algorithm instead of the other way around.

Perhaps other people (with enormous budgets) were already doing all of the above but for us this is a really exciting release!

↳ reply 1 — abixb:

If what you're saying is true and accurate, then US-based AI labs are in big trouble. The only saving grace might be some sort of a 'national security' proclamation banning the use of state-of-the-art Chinese (and non-US) models across US federal and state governments and large enterprises (especially ones with federal government contracts), but even still, US AI labs will probably lose out massively on international market if a smaller model can match SOTA of just a few months ago.

There's no way large companies outside the US will pay the "US AI lab" premium if they can get the same workloads done at a fraction of the cost using open-weight models that they can self-host and optimize/fine-tune on.

↳↳ reply — FernandoTN:

I think you're overlooking the fact that for long-horizon tasks, even small errors compound over time and can lead to catastrophic outcomes.

For simple queries, we have reached the threshold since the beginning of the year, and models are good enough from every provider to make a meaningful difference between one another. (ChatGPT, Claude, Gemini, Grok, MuseSpark, Kimi, DeepSeek, GLM...)

The real unlock will be, and you can already see it with GPT-5.6 and Fable-5, to delegate complex enough tasks that will take more than 24 hours to get done and they will not lose track. I'm not talking about a loop, but the actual intelligence to recover from these compounding errors that accumulate in dumber models.

We're still a long way from the intelligence needed to let one of these agents go ahead and supervise multiple layers of sub-agents underneath to do complex orchestration. The future looks very promising and exciting. Imagine having the possibility of a Frontier model orchestrating as many sub-agents as needed that are running on cheaper models like DeepSeek.

↳↳↳ reply — Cookingboy:

That doesn’t make sense. It’s not like SOTA models are error free, yet we still use them.

You use Fable 5 right? If that’s good enough for you now, why wouldn’t a Chinese model that’s as good as Fable 5 but at 10% the cost be good enough in 6 months?

↳ reply 2 — paxys:

How is $5/day irrelevant? In the $150/mo range you can get effectively unlimited usage of GPT 5.6 Sol (Pro plan). Why use a much weaker model for the same price?

↳ reply 3 — Aeolun:

But DeepSeek now has a warning they’re going to sharply increase their API pricing sometime in the future.

Top #2 — NoboruWataya:

My Claude account was banned the other day. The only possible cause I can think of is that I tried to authenticate from the AI assistant in a JetBrains IDE and, not thinking, entered the details for my regular subscription rather than an API account. As soon as it became apparent that I needed an API account rather than a subscription, I just closed out of the tab. Nevertheless, about 20 minutes later I got an email saying my account was banned for a violation of the usage policy, and my appeal was rejected.

My initial thought was to sign up for ChatGPT, but I had $20 in OpenRouter so I've been trying out DeepSeek V4 Pro with Pi for the last few days and I gotta say, it's good enough for my use case. And even with paying for API usage rather than Claude's subsidised subscription, and with OpenRouter taking their cut, I will probably end up paying significantly less overall. And I really like the flexibility of being able to use whatever minimalist open source harness I want (and being able to switch providers easily, too).

(My demands probably aren't as high as many others' - I mostly use it for help with some hobbyist coding projects, and I tend to ask it questions about how to approach problems rather than just telling it to go off and code stuff for me.)

↳ reply 1 — nodja:

I'm the same way, I have a very low/sporadic usage of any subscription I've tried. I now just use openrouter with DS4 pro/flash. It also gets rid of usage anxiety where I would try to justify the $20/month by forcing myself to use the tokens for projects as the weekly limit deadline neared.

↳ reply 2 — andai:

I've been asking about psyops and bioweapons and I'm still going strong. I did get a Sonnet session shut down the other day though which feels like some kind of achievement.

Top #3 — nylonstrung:

Compared to the last Deepseek V4 Flash version I've had tons of issues with it getting in infinite loops and talking to itself without executing tool calls, wasting tons of tokens

This is on Pi agent, nothing fancy at all about my prompts or use case. Anyone else experiencing this?

I've also had it randomly go from talking about Rust to talking about the electric chair, controversies about D&D rules (both irrelevant and something I've never discussed) and it's completely blind to it in future prompts even when its pointed out and referenced directly

All this said its still worth it but the agentic performance has degraded in my experience at least

↳ reply 1 — natrys:

I think this one requires a bit of strong prompting. I am also normally a Pi user, but my experience in OpenCode with this model has been drastically better than in Pi, where it overthinks a lot and gets distracted by random things.

It might be even better in Codex or Oh My Pi according to this bench I saw earlier: https://nitter.net/composio/status/2085330847951970801

↳ reply 2 — the_duke:

Yeah, I saw the same thing - quite annoying. It can be mitigated through the prompt.

→ HN comments  ·  393 points  ·  259 comments

Top #1 — jerf:

Oracle, the law firm with a tech business attached, probably wants to retain the option to sue other people for AI-washing their proprietary code, and that doesn't work if they're also publicly accepting AI contributions to their code with no apparent concern for the provenance of it. The tech business would probably tell them this isn't going to be a needle they are going to be able to successfully thread, but the law firm is in the driver's seat.

↳ reply 1 — victorbjorklund:

Not even a conspiracy theory. Just the most likely explanation.

↳↳ reply — TimTheTinker:

"Do not fall into the trap of anthropomorphizing Larry Ellison. You need to think of Larry Ellison the way you think of a lawnmower. You don't anthropomorphize your lawnmower, the lawnmower just mows the lawn - you stick your hand in there and it'll chop it off, the end. You don't think 'oh, the lawnmower hates me' - lawnmower doesn't give a * about you, the lawnmower can't hate you. Don't anthropomorphize the lawnmower. Don't fall into that trap about Oracle." --Bryan Cantrill

↳↳↳ reply — AlexErrant:

This quote is why I have my Claude Code "done" notification sound be the sound of a lawnmower revving.

The lawnmower can't hate you. Don't anthropomorphize the lawnmower, no matter how clever it sounds. (Actually not that hard with Opus5)

↳ reply 2 — phendrenad2:

Legal departments are ironically very much like LLMs.

They are trained on a narrow set of data and don't really understand the real world.

If you let them run wild with no supervision they'll turn your company into slop.

Someone needs to be there to reign them in, asking the all-important questions like "is suing all of our customers and making them angry REALLY the best option?" (to which they will reply, "You're right to push back on this").

This is typically the role of the CEO, however, as we know, most CEOs of large successful companies are too busy to actually do that. They're mostly training Brazilian Jiu Jitsu, shitposting on their own social media platform, trying to manipulate international politics, or helping run their family nonprofit.

This is why I believe that legal departments need to be demoted. The legal department should not report to the CEO, they should report to a new role that is above both PR and legal. And that person's job should be to force legal and public relations (which we know are natural enemies) to work together. Every press release goes through legal now, and it's only logical that ever legal action should go through public relations review as well.

The fact that public relations at most large companies has atrophied from disuse in our public-equity-owned unicorporate cooperation-over-competition chaebol/zaibatsu/conglomerate world is a problem for another day.

↳ reply 3 — shevy-java:

Makes sense, but they take an anti-AI stand, which I find good no matter for which reason they do that.

Top #2 — fancyfredbot:

The link is to a really poor summary of a better article by the register.

https://www.theregister.com/ai-and-ml/2026/08/03/as-larry-el...

The register article is about this post:

https://openjdk.org/legal/ai

Top #3 — flakiness:

The post in question: https://openjdk.org/legal/ai

It is 'OpenJDK Interim Policy on Generative AI' and their lawyers are writing the final version, according to this page.

It sounds like a sensible action, given past scars around Java and copyright, plus this is from a big old corp.

That said I personally don't expect that final proposal will end up any better.

↳ reply — saghm:

> given past scars around Java and copyright

How many of these were self-inflicted?

↳↳ reply — rwmj:

Intentional rather than self-inflicted.

↳↳↳ reply — saghm:

Not sure how common this understanding of the terms is, but to me, those terms don't seem conflicting.

→ HN comments  ·  173 points  ·  180 comments

Top #1 — 1997cui:

The current issue I feel with taxing is that, at least in USA, it mostly taxes based on Salary on Employees, but never tax on the capital or the actual wealth.

Tax the rich means: Increase W2 tax and property tax which, takes money away from me to feed lower middle class

Tax cut means: Reduce the estate tax and cooperate tax, keep the W2 tax flat, and cut the benefits that most critical to the bottom: Food stamps while keep other unnecessary programs unchanged: Discounted Toll for low income when driving in HOV lanes.

Top #2 — drowntoge:

> The average citizen seems to be fairly poorly informed that taxes on the rich have fallen really dramatically in the past 40 years.

There's an observation.

Top #3 — rahimnathwani:

The paper talks about economic growth, income distribution and unemployment.

I'm mostly interested in economic growth, so looked at what the paper claimed about that.

It found that major, sudden reductions in taxes on the rich did not have any statistically significant effect on the trajectory of economic growth over the following five years.

But:

- Their sample is small. They only looked at relatively large, discrete declines in their home-grown measure of taxes on the rich. They did not look at all tax-rate changes.

- They did not look at effects beyond the five-year horizon, which means it would probably exclude the impact on people starting startups, as the successful ones usually take more than 5 years to start making serious money. (or did during the period the paper considered, even if timelines have subsequently accelerated.)

- Big sudden tax cuts don't happen in a vacuum, and I don't see a way to control for confounding.

↳ reply — wang_li:

The question I have is that at some point X taxes are collected and the economy is growing at Y, then the policies change and now X-5 taxes are collected and the economy grows at Y. What was the function of the 5 under the first regime?

→ HN comments  ·  155 points  ·  167 comments

Top #1 — Tiberium:

In my personal experience Sol with cyber verification is extremely capable of finding vulnerabilities, and it works even with binaries if you have some kind of IDA/Ghidra CLI access. Of course, unless the binary is protected with Denuvo/VMProtect/etc.

It sounds absurd, but in the last few weeks I've had a few cases where Sol found an RCE in self-hosted web applications in literal minutes just from reading the code (I prefer when it tries to reason statically instead of spamming runtime probes at first).

In another case it found an arbitrary file write in multiplayer in an old game by reverse engineering the binary - any other player in a match could just send you files to anywhere on your system.

I do these things for pure entertainment and curiosity, not for money from bug bounties, so if Sol can find those with a trivial prompt in tens of minutes for me, then what can focused companies/actors find in days or weeks?

Although I think most vulnerabilities are going to be closed in popular software by mid 2027, except in niche old or abandoned projects.

↳ reply 1 — matheusmoreira:

> it found an arbitrary file write in multiplayer in an old game by reverse engineering the binary

Video games are now ruined for me. I don't think I will ever feel safe playing online again.

> I do these things for pure entertainment and curiosity, not for money from bug bounties

Me too... Was it easy to get TAC access? My account isn't even launching the Persona verification, says I'm not eligible.

↳↳ reply — rapind:

> Video games are now ruined for me. I don't think I will ever feel safe playing online again.

Agreed, also WordPress powers around 43% of all websites on the internet. https://patchstack.com/whitepaper/state-of-wordpress-securit...

WCGW?

↳ reply 2 — TacticalCoder:

> Although I think most vulnerabilities are going to be closed in popular software by mid 2027, ...

If we're to believe how good those AI are at CTF and at escapes of all kind, then the only logical conclusion is that, by very far, most vulnerabilities were already closed, even before AI.

Otherwise we'd already be in deep shit since months if not years.

↳ reply 3 — LoganDark:

Is cyber verification a thing they're actually doing now? I thought they only reached out to really incredibly famous people and that there's no way to get access as a normal person.

Top #2 — thisisauserid:

So they finally found a business model: the cause of, and solution to, cyber security problems.

↳ reply 1 — kridsdale1:

People have been saying Tokens are the new Oil.

Turns out, it’s the new Alcohol. The cause of, and solution to, life’s problems!

↳ reply 2 — andriy_koval:

the question is if that market is large enough to justify their appetites.

Top #3 — Terr_:

It seems like we've got chaos-machines that are just orderly enough to to try all sorts of branching script-kiddie / guess-n-check / research-paper-suggestions in a tireless and parallel way, and it's finding all the things we kinda knew were always potential issues but it was never economical to do things differently.

I'm scared that the "solution" will be constantly the same tools in reverse as an army of junior devs doing counter-hacks, at the expense of changing something more fundamental about how we make systems and what constitutes "good enough." (Kind of like if fuzz-testing was the be-all-end-all of memory safety.)

→ HN comments  ·  130 points  ·  33 comments

Top #1 — xioxox:

We also released the 2nd half sky catalogue of the eROSITA X-ray survey at the same time, for 1.5 years of operations, in collaboration with SDSS. Here is a sky map and 3D plot with redshift: https://www.mpe.mpg.de/8215311/news20260731

This almost doubled the number of known X-ray sources to 2 million.

↳ reply — chrisweekly:

> "almost doubled the number of known X-ray sources"

That's a fantastic achievement!

Top #2 — epistasis:

The number of massive maps of the universe in recent years are absoultely fascinating, and have reignited my childhood interest in astronomy and cosmology.

These maps consist of a large number of statistical observations derived from images, and that lines up with all the data analysis I've used in the past decades in genomics. (Most DNA sequencing data comes from image analysis, coincidentally!). And the plots I see in these papers are looking increasingly like the plots I've produced over the past 20 years, so I'm thinking maybe it would be fun to download a few TB of atlas and see if I can reproduce their plots...

Thanks for everyone who works in this field, it's fantastic work that's reshaping our very understanding of reality.

Top #3 — RagnarD:

I tend to assume that the obvious gridded point pattern is a sky sampling artifact - but it would be far more interesting if it's a real thing.

→ HN comments  ·  198 points  ·  50 comments

Top #1 — atbpaca:

Beautiful and scary at the same time. Almost like the iceberg was a planet on its own with mountains, valleys, crevices. Then everything goes haywire. The fall of a majestic titan. The destruction of a world.

↳ reply 1 — baubino:

There’s also something very organic about it. I had to keep reminding myself that it’s “just ice” and not a living being.

↳↳ reply — dotancohen:

Especially the spine. If you had shown me a portion of the video with that thing moving as such, I would have believed it to be some leviathan.

Which brings us to the video itself - horrible. Presumably the original sound was recorded, because we hear it under the music. Just let us hear the sound! The people watching this video are people who appreciate nature, not people who need a dramatic soundtrack to be entertained.

And what is with changing the speed of the video every 60 seconds? Either show it at real speed, or speed it up, but not both intermittently! At first I thought there was an issue with the video playback hanging.

I would absolutely love to watch the original, unedited source.

↳↳↳ reply — artine:

> I would absolutely love to watch the original, unedited source.

From the linked video's info: https://www.youtube.com/watch?v=CJ5PhjFYRSo

↳ reply 2 — GuB-42:

In short, a titanic event...

Top #2 — joshuat:

Massive appreciation for whoever adjusted the position of the camera at 7:23 (3:18 in video)

Top #3 — raphinou:

I would have thought it would generate enormous wave, but that's clearly not the case. I think some of the biggest waves was generated by a mountain sliding or collapsing in the sea. I expected something similar here. Anyone know why not?

↳ reply 1 — GuB-42:

It is not a very energetic event for its size. The iceberg is floating, it can't accumulate a lot of potential energy before flipping, so when it happens, there is not enough to make big waves. The center of gravity of the iceberg barely moves.

A mountain sliding into the sea is different, these higher up rocks have a lot of potential energy, that convert into kinetic energy as they fall, which is transferred to the water as they splash down, making big waves.

↳↳ reply — tlavoie:

Here's an excellent example of the latter, from a short ways up the BC coast from me. Not caught on film, but found later by heli pilots, and on seismic data: https://hakai.org/stories/the-big-slide

↳ reply 2 — javier2:

i suppose its not actually displacing a lot of water volume, unlike a landslide.

→ HN comments  ·  355 points  ·  340 comments

Top #1 — jwr:

The worrying thing here is that so many people accepted outsourcing the decision on who can see their website to a large company (Cloudflare). If the company decides that a certain user should not see the website, the user will not see the website, and no one will know about it, and the user will have no recourse. That is not the open web that I would like to see.

A second side effect of a knee-jerk reaction to bots crawling websites is that if you try to fight all bots, you also end up hurting real users that use "bots". If I run a local script or an LLM that needs to fetch a web page from your site and you block that script or LLM as being a bot, you hurt me, the user. I wanted the information and I have no way of getting at it without investing my own time and effort in going to your website. That might or might not be what you expected, but it's worth taking into account.

And finally, something worth noting is that there are so many websites whose owners complain about bots, but the real problem is that the website is poorly built and should be improved anyway. Bot traffic is not necessarily bad.

↳ reply 1 — hk__2:

> If I run a local script or an LLM that needs to fetch a web page from your site and you block that script or LLM as being a bot, you hurt me, the user.

No; in this case you are not a user, you are a bot user.

↳↳ reply — matsemann:

I don't disagree, but there is a sliding scale here.

For instance, I wanted to buy a piece of equipment the other day from a local company for a specific usecase. I wanted to find a specific price/weight/specs ratio, and asked an llm to loop through the 20 or so items, fetch their page and calculate and present some values for each. This then led to me going and buying the one I found. So the llm was mainly just an extension of me clicking into every page and making a spreadsheet myself.

However, if it were to continuously poll, or just scrape or something with no intention of buying, I would be no better than a bot.

↳↳↳ reply — weego:

No, there isn't, it's a bot. Justifying it with a strawman case study doesn't change it.

This is giving "why is this plane full? What in doing is important but what is everyone else doing here?"

↳ reply 2 — m463:

I am denied by cloudflare CONSTANTLY on one system.

I have an old os (macos 10.11), running the highest firefox esr I can run, and I get denied by cloudflare.

But not always immediately - I get to enable javascript/cookies sometimes just to be denied.

they are not the folks we want gatekeeping the internet, they are opportunists increasing their OKRs

↳ reply 3 — autoexec:

> If I run a local script or an LLM that needs to fetch a web page from your site and you block that script or LLM as being a bot, you hurt me, the user.

People trying to block bots end up keeping out all kinds of users. I get blocked frequently for using a regular browser, just with JS disabled. 99% of the time, I just close the browser tab and move on with my life.

Top #2 — johnorourke:

Anubis[1] is a superb fix for sites not behind Cloudflare/Fastly/Bunny etc. We had millions of bot requests, on a site serving all countries so we couldn't block by country, with fake user-agents so we couldn't block using that. It uses 'proof of work' to detect real browser software.

[1] https://anubis.techaro.lol/

↳ reply 1 — RattlesnakeJake:

I wish they'd ease up on the whole "don't change the logo without paying us" thing. The furry anime character is a turnoff for anyone with a brand or personal image that doesn't mesh with those subcultures.

↳↳ reply — altairprime:

> is a turnoff for anyone with a brand or personal image that doesn't mesh with those subcultures.

Most people either don’t know or don’t care about “those subcultures”. I bet a lot of older people think it’s a cartoon figure of Betty Boop (nurse) and miss the furry bit since it appears and disappears quickly. Most people also don’t have a brand.

So it seems like you’re describing a concern that only affects a tiny fraction of people:

- Not interested in paying for custom branding, so obviously not a corporation or influencer

- Dislikes cartoons

- Aware of, and hostile towards, “furry” subculture

That has to be an exceedingly small fraction of potential users of Anubis, and given how much businesses and branders will pay to custom-brand something, I’d counsel them to stay the course. Sure, a few never-payers will never pay, but they wouldn’t have anyways, so they can cope with Nurse Betty or look elsewhere for a competing free product.

If you think about this in physical market square terms — in other words, a bazaar — it seems horrendously rude to complain about a shop logo sticker on a free product handed out to anyone that walks up and asks for it. If you want it white-labeled so you can write your own name/logo on it, you pay for the privilege of displacing their name with yours. But you don’t stand there and loudly complain that their shop mascot has dog ears while holding a freebie bag of product, without losing the respect of everyone who hears you doing so.

↳ reply 2 — marklar423:

I'm assuming a bot running a headless browser instance can still get past it?

It's still valuable to raise the cost of scraping of course. I don't think anything can really stop a determined scraper from impersonating a human. I wonder though if a system similar to Anubis but mining some crypto would make bots _welcome_ - since they're paying for their traffic.

Top #3 — tarr11:

> My normal bill for running this whole site is around $90 a month. During one bad spike month, it jumped about 500%.

This is D1 - which has very surprising costs. you may just want to drop D1 and move to a static site. There’s no reason your site should cost this much.

↳ reply — nickgray:

Thank you! I need to tighten up my KV compression, which is actually carrying a lot of D1's load otherwise. We also had some bad queries some months, as the pages and database grew, that were counting the wrong things (or extremely inefficiently) and those have since been fixed.

→ HN comments  ·  180 points  ·  70 comments

Top #1 — JSR_FDED:

I lived in the US many years ago. USA Today was then already a bit of a joke - always a color pie chart on the front page, ludicrously generic news and sports, and on the last page 1 paragraph of news per state for an entire state (if something newsworthy happened according to the editor I guess). Let’s just say very much not a highbrow paper.

I visited the US again last year and got the USA Today in my hotel. I thought I was being pranked. The writing, tone, complexity had all plummeted to the level of a child. That paper is basically for people who can’t figure out how to get their news from Yahoo. The Same people who get value from TV Guide.

Let me guess what Palantir will tell them: your audience is dying and/or illiterate. Add cartoons. You’re welcome.

↳ reply 1 — lkbm:

USA Today also owns a huge number of local papers -- possibly most. The main local papers in the three last places I've lived (Austin TX, South Bend IN, Asheville NC area) were all USA Today Co.-owned. (Used to be called Gannett.)

It's not dying. Or, at least, it's doing better than anyone else.

↳↳ reply — ceejayoz:

My local USA Today-owned paper moved from a giant historic building (800 people when I worked there), to a new much smaller office (about 150 people at that point), to announcing they're moving to a coworking space.

They're dying, it's just a slow bleed-out.

↳ reply 2 — RGamma:

Stories/comments like these make me want to dedollarise.

↳ reply 3 — jihadjihad:

> Let me guess what Palantir will tell them: your audience is dying and/or illiterate. Add cartoons. You’re welcome.

And USA Today will be billed $50M for the privilege.

Top #2 — eitally:

My question is this: why is Palantir required for this kind of fairly basic data analysis? Couldn't they use much cheaper, probably open platforms for data science & analytics?

↳ reply — jkestner:

I don't see where it says who's paying whom. Maybe the deal is that Palantir gets more personal data to throw in a national profile database.

Top #3 — waltbosz:

I dislike USA Today. It feels like lowest common denominator fluffy news.

Also, I remember staying at a hotel once where they delivered one to every guest room.

↳ reply 1 — SoftTalker:

USA Today is news written for 5th graders. It has always been a waste of time and paper for anyone who really wants to know what's going on.

↳↳ reply — micromacrofoot:

The people who can only read at a 5th grade level are precisely the people who need to be more informed about what's going on

↳↳↳ reply — SoftTalker:

Why? They mostly don't do anything very impactful. Voting and other civic engagement is very low amongst the illiterate.

↳ reply 2 — snickerbockers:

Im pretty sure I've only ever seen people read usa today when hotels give it away for free with the continental breakfast.

→ HN comments  ·  156 points  ·  99 comments

Top #1 — swader999:

I'm seeing people argue and debate with AI. One will cut and paste AI reasoning, another will rebut with the same approach. I need a popcorn machine in my office.

↳ reply 1 — dgellow:

Get yourself one of those popcorn microwave bowls: https://m.media-amazon.com/images/I/71mA19sd51L.jpg

You get hot popcorn in ~2min without the need for a whole machine you will have to clean and maintain. I got one as a gimmick a few years ago and it has become one of our favorite evening snack

↳↳ reply — tosh:

does this work better than putting a bag of popcorn into the microwave?

↳↳↳ reply — ndriscoll:

At least in the US, bagged popcorn uses synthetic "butter" flavoring that can cause lung damage. Look up popcorn lung.

↳ reply 2 — champagnepapi:

Don't be a meat proxy! See this all over the place, it's very frustrating and i've advocated for stopping this practice at my company!

- https://news.ycombinator.com/item?id=49151933 - https://gruhn.me/blog/2026-08-03/

↳ reply 3 — Zenst:

Having experienced a customer support AI today that whatever option you picked initially led to the AI, which had the rigid helpfulness of a McDonalds script of "Would you like fries with that", having already said you do not want any fries, I feel a popcorn machine is a good investment. Be the stage that office interactions will be reduced to, I'll get my AI to get onto your AI's suggestion and get back to you with an answer. I did have a good chat with an AI a few weeks back (Googles) were I challanged by asking it to do a picture of its best dream and then its worst nightmare, then used those to pull it apart as it went from some extreme to another, in the end it ended up with its ideal being an AI only world, and picked Europa as its home - scary on many levels, yet a perfectly acceptable and understandable response.

Top #2 — bonoboTP:

> In fact, the researchers found that almost half of employees who receive AI-drafted emails see senders as less trustworthy as a result of their sending it, and over a third see them as less intelligent. So much for AI enhancing us.

Maybe it did enhance them, and maybe the assessment was correct at the same time.

There is no single effect of AI on work quality. Depends on use and effort. Use it well, it will be good, use it with close to zero effort and you'll get something distinctly better than what you could smash together manually with that much effort, but people will see that you indeed put zero effort in it, and also they will notice that you spent no effort in cleaning up AI tells because you likely can't even notice or have no idea about anything and will indeed correctly conclude you're less intelligent.

There may also be a miscalibration that's worse for people with less taste / judgment / intelligence / experience, in that it's indeed easier to produce things, but while the effort required is reduced, people expect and believe it to be reduced more than how much it is truly reduced so they overshoot and don't notice. Meanwhile, the right way to use it is to keep your usual effort level or drop it only a little, and simply produce better things.

↳ reply — lo_zamoyski:

> There is no single effect of AI on work quality. Depends on use and effort.

Quoting the article...

"Nearly half of senior leaders say they would trust a chatbot’s judgment over their own, on questions they were hired to answer."

If they're using it like an oracle, then that is worrying, but how worrying - practically speaking - would depend on the baseline prior to the use of AI and how that compares with what the AI is spitting out.

The first big hazard of this AI worship is, of course, the atrophy of reasoning skills. I don't think there's an especially distinct skill needed here. A rational person is able to evaluate claims made by others responsibly based on a combination of proxy signals like an earned reputation of trustworthiness, integrity, and authority, the logical coherence of the claim, and plausibility and the absence of contradiction between the claims and more certain baseline knowledge.

So, psychologically speaking, the ease with which we obtain credible-sounding answers is a major motivator, but two other cultural factors support this credulity, namely, magical & superstitious and the broad acceptance that bullshitting is an okay thing to do. Both weaken the ability to apprehend and reason about reality in different ways (bullshitting, for instance, reinforces a carelessness about and indifference to the truth; magical thinking more directly sabotages reason; both undermine confidence in reason and place stock in the sub-rational instead).

↳↳ reply — bonoboTP:

If you were gullible before, you will keep on being gullible. It's a moral panic. So what now? Karpathy, Carmack, Terence Tao shouldn't use AI? Or you want to require a license before you can use it?

On bullshitting - yes, it was already a thing. Where nobody cares about truth anyway and it's a low trust environment where everyone just wants to take but not give, and then move on, yeah this will be the result. Those places then should indeed crumble down under the weight of this.

↳↳↳ reply — blooalien:

> If you were gullible before, you will keep on being gullible.

AI (supposedly) being a "10x multiplier"-type tool, if you were gullible before, then with AI assistance, you're at least 10x more gullible. ;)

Top #3 — JSR_FDED:

It’s just amazing we keep using the word “leaders” when they’re clearly not.

↳ reply 1 — deepsquirrelnet:

Around 2015, my boss got permission from corporate to take me and a few of my coworkers to a leadership conference for a day. It was setup at a church and had religious theming around it, and a lot of "leadership values" were communicated. It was the usual fare of responsibility, accountability, truth and trust you'd expect, mixed with the religious overtones you'd also expect. But what I didn't expect is how trite this fictional narrative would prove to be.

I think about that a lot now, and how my religious upbringing was bombarded with this messaging (in a good way). And how much of the same world that previously claimed espouse it has complete turned their backs on it in such a short time.

Is it dead or did it ever exist at all? At least we don't seem to pretend to have higher values anymore. If we even have leaders at all, they are not measured by the yardsticks we used to make.

↳ reply 2 — fzeindl:

Since they hire, command work and fire, they kinda are.

→ HN comments  ·  132 points  ·  94 comments

Top #1 — hamilyon2:

Many essential metabolic enzymes require metal ions for catalysis, structure, or substrate binding. So in some sense the dependence is still there, we just learned to take bits of substrate with us while travelling.

Top #2 — andrewflnr:

The headline is a little confusing. They're still talking about a "split" between bacteria and archaea, so even under this theory, the very roots of metabolism and (I'm guessing) genetic code etc still have a single origin.

Anyway, it does provide a nice reason that bacterial and archaeal cell membranes are different.

Top #3 — MarkusQ:

Clickbait headline.

Should be "Life left mineral substrate at least twice" or some such. The research is interesting, but they're playing word games in the telling of it by saying the most recent common ancestor of all life wasn't alive (because it was dependent on a mineral surface), but two of it's children gave rise to lineages that were.

→ HN comments  ·  332 points  ·  331 comments

Top #1 — vannevar:

I would hardly call it a sudden reversal, looking at the downward trend in the numbers over the last six months or so.

https://tradingeconomics.com/united-states/non-farm-payrolls

↳ reply 1 — ismailmaj:

Given the trend of revisions, I wouldn't be surprised if the real loss is around 6 figures, May 2026 revision went from 172k to 63k.

↳ reply 2 — tetris11:

What happened between February and March?

Top #2 — nunez:

> The BLS said employment contracted the most in “local government education,” which declined by 50,000 roles, likely reflecting teachers during summer break.

This is not entirely accurate. AFAIK, teachers usually remain on payroll throughout the summer. This stat _might_ be talking about contractual teaching roles, like adjuncts or subs, but it's hard to say from that BLS category.

↳ reply 1 — JeremyNT:

Not to mention, if it's seasonal for any reason then it's something that should be baked into the cyclical expectations already.

It's like reporting that Spirit of Halloween employees accounted for the biggest retailer layoffs in November. It's a totally useless statistic that tells you nothing about what might really be going on that month that's abnormal compared to baseline.

↳ reply 2 — Jcampuzano2:

This is correct, teachers don't just stop getting paid and lose all benefits every summer (my best friend is a teacher and so is his brother), but at the same time it seems tons of schools now prefer (or are only able to find positions for) short contracts, more subs and temporary positions etc.

Especially if its not in a well funded area.

Top #3 — sp1982:

I have recently started an experiment tracking a sample of 5000 employers in US and I see a bounce last week after a brief lull in july. I am not sure if there is a strong downward trend. https://corvi.careers/reports/usa-weekly-job-openings-report...

↳ reply 1 — mixdup:

It's tough to trust job openings. It is a fact that some number of job postings are ghost jobs that don't exist. Either they're scams from impersonators or the companies are trying to keep a good public face, but many openings sit for MONTHS which just does not line up with the anecdotal evidence of people going 6-12 months without an interview much less an offer

↳↳ reply — gmueckl:

Some of these offers may be part of visa or green card processes. Some require a labor market test in the form of a job ad and recorded response to it. If a suitable candidate applies, the visa/immigration process fails. Still, most companies don't intend to actually hire in this case as far as I understand it.

↳ reply 2 — VirusNewbie:

Google will put one job posting per PA per level regardless of one team hiring or twenty.

I have to assume they aren't the only employer to do this, so I think job openings are meaningless here.

→ HN comments  ·  186 points  ·  296 comments

Top #1 — Animats:

Addressing the question in the title, "What happens if an entire class of workers loses faith in their careers", look at what happened to printers. That was a good skilled trade for centuries. Then it just went away. Phototypesetting took the high volume end. Desktop publishing took the low end. Computer-controlled presses took press operators. And then the newspaper business died.

I've heard homeless former printers talking about the good days at the Burger King near the cable car turntable at Powell and Market. They thought they had careers. Now they're struggling to survive.

Knitting is not the answer. This is a survival issue. Somehow I suspect the author inherited money.

↳ reply 1 — Rendello:

Farewell, Etaoin Shrdlu documents the last day of hot type at The New York Times before the digital changeover. It's one of my favourite documentaries, everything feels so vivid and alive:

https://www.nytimes.com/video/insider/100000004687429/farewe...

↳ reply 2 — pj_mukh:

Also the headline is about tech workers but the article is about Knowledge workers. I know a class of tech workers absolutely fired up about the new Industrial Revolution, they're just not on this forum or featured in lengthy navel-gazing articles like this one.

↳ reply 3 — RealityVoid:

I sense there is this air around of "demand our jobs keep being relevant" - I personally feel this kind of thing holds the world backwards if it works. I think if our jobs were to become obsolete, embracing it and re-orienting oneself is the one thing that is personally effective and is morally right. But what do I know, I'm just some doofus on the internet.

The irony is not lost on me on all those years where people were advised to "just learn to code, bro" in order to escape the personally threatening situation of obsolete jobs. Well, now people feel it's our turn so, "just learn to _something_ dudes"?

This is all, assuming of course, that the end is nigh and our profession really is close to extinction. I personally don't buy it, but if it is, so be it.

Top #2 — marginalia_nu:

Saw a meme on mastodon the other day contrasting people of the '90s going online to escape from offline reality, and people of the '20s going offline to escape from online reality. There's something to that.

I think a big part of it is just how incredibly toxic the web has become. If you spend any time online, as most tech workers do, you have exceptional resiliency if you're able to stay sane and level headed.

Going online it's like everyone is really angry and tired and sad all the time, and it's getting worse every year. This really took a nosedive during the pandemic, but looking back this had been going on for several years before. Every election cycle makes things worse too.

It's just been spiraling in a very dark direction for a long time. You have these blackpill depression cirklejerks everywhere. Astroturfing is getting worse every year. Like what Bannon did with gamer gate is an amateurish precursor to what everyone with is doing everywhere now. With AI you can even run these campaigns with a very small budget.

All the news is just outrage porn and non-stop predictions of impending doom. Used to be more tabloid territory but now I don't know if anywhere is spared. A lot of the focus now is on AI, but when it isn't the world is burning in some other fashion. Reading the news, we've been in a permanent state of crisis for years now.

This information diet just isn't good for your mental health. It isn't just tech workers, but they're probably more exposed than most.

Candide's closing remarks spring to mind. Tend your garden.

↳ reply 1 — m463:

> '90s going online ... '20s going offline

It feels to me like the change from:

- old: going downtown to the restaurants and shops of your small village to enjoy a relaxing evening.

- new: downtown is a crowded confusing mishmash of unregulated blaring music, over-the-top blinking flashing signs, intrusive people accosting you and dangerous thieves lurking everywhere. don't go.

there's no "regulated" that makes thing better for anyone, it just gets morphed into oppression of the vistors making them more vulnerable without fixing any problems.

↳ reply 2 — arjie:

I live in San Francisco and life is quite nice. There are many obvious ways that this city is shitty that I’ve noticed other civilizations don’t have a problem with but also I’d say the overall experience is positive. Neither offline nor online reality is defacto unpleasant.

As far as the disillusionment sense, I think that’s an active selection bias. I think people (including me) like to feel angry and so we go out there looking for and finding people who will give us that. The key, for me, is to curate one’s feed. On HN, for instance, I usually block the majority of repetitive boring posters using a chrome extension. The majority of these people have very low context sensitivity which annoys me.

There are also abrasive people or generally sad people, both groups of which are safely blockable because they have a comment style where P(token|context) = P(token). There is no surfeit of humans on the Internet and no surfeit of interesting information. Almost all humans can be safely excluded from my feed and I can still get information.

By the way, I know you from your search engine. This might be an out there question but do you offer your search index as a downloadable structure? If so, how much does that cost?

Top #3 — dec0dedab0de:

This dropped off the front page in the time I read it, I assume because of the controversial gate that favors upvotes to comments (which I am making worse by replying to others). It is a shame, because this really resonated with me.

I've been working in tech for over 20 years now in various roles, and this is the least I have cared about it. I used to want to get better, and was constantly excited about learning new things just for the sake of it. Now I daydream about being homeless. I look around at all the meaningless crap I have accumulated, and feel shame that I didn't use that money to retire early.

I am surprised that remote work didn't get a bigger mention in the article. In a few weeks I will have been working from home for 12 years, and I believe it is the cause of most of most of my angst. At first, the convivence was such a relief, but it has become a trap. The effort of the commute, forcing me to go out every day and talk to human beings face to face, even just little things like making sure I was showered and dressed every day, they all felt like a burden at the time but they were good for me and I didn't realize it. Not to mention that it makes my house feel like work, and I now I hate it here.

I will say as an anecdote counter to the AI examples given in the article. I have a young co-worker who is very excited about AI and is using it way more effectively than I am. Talking to him gives me hope for the future, the kids will be alright.

But to the point of the article: I have many friends my age that are making less than half of what I do working objectively harder jobs for way longer hours, but they all seem so much happier than I am.

↳ reply 1 — cheschire:

I did the remote thing before Covid and rejected it for many of the same reasons. Then Covid happened and so many people got to experience it and loved it. But I still hated it while I was forced to.

Every time I told anyone that I personally didn’t like working from home, and was excited about the return to office that would inevitably come, I got SHOUTED at. Absolutely flayed for not being 110% in support of this thing that made their lives worth living or whatever.

When it first happened I thought it was a weird interpersonal conflict situation. Maybe I had broached it wrong. But it kept happening! So many people were so upset when megacorp said they had to return or find other employment.

But I’m an extrovert and working from home was suffocating.

↳↳ reply — ang_cire:

The problem isn't that WFO-preferenced people don't exist, it's that it aligns with the narratives of the extroverted management (who approve their own PTO) that it's better, and people who aren't extroverts dread being forced back. No one who prefers WFH is actually against people who prefer offices being allowed to go there, they're opposed to people publicly claiming they prefer that- unqualified- because they see it as a threat; a statement that will be weaponized to support management's RTO initiatives.

↳ reply 2 — THX1137:

> forcing me to go out every day and talk to human beings face to face, even just little things like making sure I was showered and dressed every day

This part is strange to me. I've worked from home for about 10 years and I've never once had an issue with twice-daily showering, meeting people face-to-face, going out for coffee, going outdoors, going to the gym, etc. The fact I can just go sit in a park, on a hiking trail, or whatever and work there is a genuine "killer app" for me. I'm surprised at people who tell me that they are isolated by home work when there are so many opportunities for social and natural connection everywhere...

→ HN comments  ·  153 points  ·  90 comments

Top #1 — jerf:

I love the ambition and the fact that this is not just another "state of the art in 2015" language like I see so often. It's trying to do something genuinely different. The field of "taking stuff out of academia and making it work" is a rich and underharvested one.

However, your light is hidden under a basket, to use an old metaphor. I'm having to go digging to find the genuinely new things going on. I suggest recalibrating the entire documentation with a focus on the new stuff. People on HN often complain about not having syntax examples front and center but I would say for you, the very first thing I you should be hitting your new visitors with is the choreography idea.

Set up a simple example of doing something like a concurrent remote counter that is atomically safe by the construction of your language and immediately dive in to what that means. Forget even educating us on the rest of the mundane syntax of the language, immediately dive in to what that is and what that means. I see in the docs/ dir that it probably hurts your programmer mind to cover the choreography before covering sections 1-8, but you can safely assume that if you intrigue with the choreography that they'll hang around to learn about the rest, whereas you can't safely assume that a new reader will wade through all the rest of the relatively mundane details to get to the really interesting stuff.

A modern language with a modern take on compiling a "program" that takes a unified view of the world at the programming language level, and then emits a "server" and a "client" (and perhaps other roles) as separate executables is a pretty nifty idea. Hit it early and hit it hard.

↳ reply — bitwizeshift:

Thank you for this comment. I had mostly shrugged off this project as “neat” but didn’t go digging until I read this.

The choreography stuff is actually really cool! Definitely agree this should be front and center

Top #2 — hyperhello:

First of all, the syntax is very generic and conservative. I’m extremely positive about that. It just looks like C or Typescript or Java to me and I don’t see mysterious diacritical marks.

The next thing is, I need more examples. Read me documents can scroll forever and that’s fine. Add examples for every concept your language wants to cover to it. This is your chance to think things through and make the read me and the language astonishing.

And your AI disclaimer (or your AI’s disclaimer) makes sense.

Top #3 — vlovich123:

One thing I don’t understand is how you can guarantee the lack of a distributed deadlock. I’m sure it’s covered in the underlying research, but just conceptually it’s hard to picture.

What stops a choreography where Claire send a message to Bob but Bob is waiting for Alice and Alice is waiting for Claire?

Is it like Rust memory safety where not all valid programs are accepted but all invalid programs are rejected?

I think some examples of the distributed code in action on a trivial and non trivial distributed example is more compelling than a 3d donut render

↳ reply 1 — minraws:

You can think about it as a protocol.

The protocol doesn't describe "Claire sends" and "Bob receives" as two independent actions that wait for each other, it describes them as a single communication in the global program/state.

When it gets executed into code, every send already has a corresponding receive by construction, so you can't write something like "Claire sends to Bob while Bob is actually waiting for Alice" unless the protocol itself allowed that execution.

So the cycle you're describing can't just accidentally appear because of sync issues, because the assumption is a communication is represented correctly at each turn.

So yeah it kinda does limit the set of all possible programs since I would expect not everything can be encoded this way also seems hard to resolve this in practice without only allowing communication b/w Wyzer systems.

I am not very familiar with this topic in practice so OP would be the best person to answer this, I am quite intrigued by how it works in practice as well.

↳ reply 2 — pgt:

hmm, I suspect distributed systems can be modelled with a form of probabilistic borrow-checking as long as you have control all the connected systems.

→ HN comments  ·  134 points  ·  67 comments

Top #1 — bix6:

And all you techno futurist republicans voted for this. And you’ll do it again cuz money or whatever.

↳ reply 1 — classified:

But now they have "startup experience" and can fuck up even worse the next time.

↳ reply 2 — xgulfie:

The right's platform has devolved into, essentially, "whatever owns the libs"

Top #2 — blitzar:

You mean they didn't save 25 trillion dollars a year from the 7 trillion annual spend?

I cant believe they would lie about this ... i am sure that this is a totally isolated case and they have been straight forward and honest about everything else.

The DOGE refund cheque is due to arrive any day now.

↳ reply — jasonvorhe:

Who ever claimed these numbers? Wtf.

Top #3 — chevman:

They'll just extend the forecast horizon when questioned:

"Oh, those projections were over the life the expected costs,forecast out 10/15/20 years into the future. Everyone knows this is how you calculate such savings..."

→ HN comments  ·  135 points  ·  160 comments

Top #1 — yorwba:

I wondered where this article got the idea that "half the planet" would die "in 26 years or less" with 3°C of warming. It appears to be a misreading of the table in https://actuaries.org.uk/media/ni4erlna/planetary-solvency.p... "Figure 12: Planetary solvency risk impact and likelihood definitions (illustrative)" qualified by "A risk impact matrix is a methodology for ranking potential impacts, not a prediction or central scenario."

In particular, they rate ≥50% mortality as Extreme and also rate warming by 3°C or more by 2050 as Extreme (this is relevant for the colorful severity/likelihood grids that are the focus of the paper) and somebody appears to have made the incorrect connection that an Extreme climate would cause Extreme mortality, and also by the time the Extreme climate is reached and not afterwards. Elsewhere, the paper mentions that "We have not shown an assessment for mortality in this illustration, as very limited research has been carried out on the potential for large-scale loss of life in relation to these interconnected risks on which to base an assessment."

Unfortunately, that makes the article title highly ironic.

↳ reply 1 — tejohnso:

The report does include a >50% human mortality extreme category and the author of the article is referring to a separate report suggesting 3°C is much more likely than expected. I think it's pretty reasonable to think that an extreme climate would cause extreme mortality. The report is pretty confusing to digest.

But because you think that particular consequence is overstated in the article, you're suggesting that the article itself is a lie? The title refers to the lie that we still have more time and that we don't need to get to work immediately to address climate change. It's about the overall lackadaisical attitude toward the climate emergency.

You seem to be suggesting a dismissal of the article's core message.

So, are you suggesting that we actually have plenty of time to deal with climate change, and that things like "Net Zero by 2050" correctly convey the severity of the situation?

What is your assessment of the current year number of deaths attributable to climate change, and expected number of deaths at a 3°C above baseline global average temperature?

↳↳ reply — square_usual:

> I think it's pretty reasonable to think that an extreme climate would cause extreme mortality.

No, it's not reasonable. They're using the same label, but there's no evidence showing that.

↳ reply 2 — ashirviskas:

Thank you for putting this here. While I do believe in climate change, I could not finish reading the article as it seemed like the authors opinions are led more by emotions rather than facts. And gourds.

↳ reply 3 — alexk307:

Not to mention that if we limited access to fossil fuels today, billions of people would be dead within months. If we turned off the spigot, billions would starve and civilization would collapse quicker than you could say climate change

Top #2 — Synaesthesia:

It's quite striking to note how media coverage of climate change and pollution have gone down so much. I remember in the 90's it was very much in the popular mileu.

What's more we are living through an actual green revolution. Solar power and batteries have never been so cheap - and still promising to get cheaper and better with sodium ion batteries.

↳ reply 1 — losvedir:

Well "climate change" as understood today wasn't really talked about much in the 90s, as I remember it. Some broad, apocalyptic concern about creeping temperatures and knock on effects over many decades.

But I do remember quite a bit about "pollution" and like general acute environmental issues. Smog in LA, the hole in the ozone, etc. Those were both more understandable and visible to your average person and more tractable. And there was the obvious hook about energy independence after the fairly recent (at that time) oil embargo and such.

The Green party (are they still around?) talked about more vague issues like ocean plastic and six pack wrappers on birds, but I don't think we made a ton of progress there, but then that was less consequential anyway.

I feel like we went down the worst timeline since then when it became very political and people burn gas for the hell of it, and others hate people who do so. Solar and nuclear (or hell even just solar) could give us clean, cheap energy abundance, but the people championing solar tend not to do it from the abundance perspective, but in tandem with cutting energy use which is a hard sell and not necessary anyway, in my opinion.

↳↳ reply — roryirvine:

Did you forget the Earth Summit and its Climate Change Convention? Or the Kyoto Protocol?

Both sat at the top of the news agenda for months! Not to mention years of features, long-form journalism, and documentaries. And then the denialist backlash began to kick off in the late 90s, which (initially) added to the media storm.

↳ reply 2 — marcyb5st:

I was reading somewhere that we are already fucked even if we were to completely halt greenhouse emissions today. Apparently we kickstarted some positive feedback loop (I think it was the thawing of the permafrost in Siberia/Canada) which releases enormous quantities of Methane and that with the higher temperatures we already reached we decreased Earth albedo as there is less snow/ice reflecting light into space.

I really hope that particular article was wrong, because the alternative is that we have to resort to geo engineering. Considering that the governments can't really coordinate in preventing this, I don't see them coordinate in fixing it, especially because the scale and costs of the "fixes" are astronomical.

Top #3 — rendx:

Just in case you miss it: The author of the post, Adam McKay, "is the Academy Award-winning director of The Big Short, Vice, and Don’t Look Up and the founder of Yellow Dot Studios."