skippyfish

raining a SOTA model takes a huge amount of resources and expertise

Writing books, building Wikipedia, and answering questions on online forums takes a lot of resources and expertise that scraping didn't. So at the very least, we're already one rung down the "maybe you should've asked" ladder.

Civil breach of contract. Anthropic's TOS explicitly say you can't do what Kimi is alleged to have done.

Ah yes, I remember when Anthropic crawlers abided by the TOS of the websites they slurped up.

All your other points are downstream from this, which makes them pretty tenuous. Labs don't think that ToS or other explicit wishes of content providers apply to them, but they expect everyone else to abide by theirs.

I think the explosion in the popularity of Lean probably means that tactic-based proofs have won

The number of mentions of Lean in HN submissions aside, how do we gauge that? HN has odd trends like that - a decade ago, we loved everything "Bayesian" - but they don't necessarily translate to anything that's happening in the mainstream.

To sign up for what? The experience of approximately every author on the planet is that they found out that Anthropic did something bad at the same time they were "opted into" the class. The only thing they could do is opt out and litigate on their own against a company with a valuation approaching $1T.

This is a sweet deal for lawyers and for publishers, and nothing else.

It just feels deeply unserious that these labs talk about apocalyptic risks, ship models with safeguards that make them borderline useless for sensible tasks, and then YOLO stuff like that on the backend and use it as an opportunity to market their stuff some more.

I always find it hilarious when these opinion pieces about the impacts of AI are AI-written (as is almost certainly the case here, a major trend in ACM and IEEE publications). Like, really? An LLM has a breadth of experience managing AI, and can recall it, and explain it to humans?

Possibly. I think the US is a lot more exposed to the apocalyptic narratives from the bosses of AI labs, who sort of painted themselves into that corner in the pursuit of valuations.

The thesis of the article is spot on. The title is weird - I don't know what's the point of bringing such an emotionally charged term into it. Unless your point is that LLMs are sentient creatures capable of suffering, but I don't think that's what the author is trying to say.

I agree with the author that almost everyone wants to use AI to do the things they have no interest in and don't want to pay someone else to do while being vehemently opposed to AI taking their hobbies or professions away. Which explains one of the greatest social paradoxes of today: approximately everyone is using AI and almost everyone is hating it. I don't mean HN, but public opinion polling for the tech is absolutely awful. In the US, it had an approval level comparable to Iran.

I don't think it's an AI-specific thing, though. It's exactly the same mechanism that makes us object to the government meddling in our profession but think it's high time for the feds to regulate other people's jobs. I guess some balance of holding each other at gunpoint gives us a stable society. I don't know how AI fits into that.

I don't think the author makes a persuasive argument. Most of the photos they're analyzing are obviously lit by various point light sources, e.g.:

https://nothings.org/gamedev/ssao/IMG_0064.JPG

You can see a sharp shadow near the door frame. So yeah, ambient occlusion is not going to simulate that, but it was never supposed to?

It's supposed to make 3D shapes look good without the need to place lights or do more complex ray tracing. It roughly approximates what you could perhaps see inside a photographic lightbox. Light is coming from every direction and every surface is matte.

The reason it's used in computer games is not that it's photorealistic under all lighting conditions. It's that it makes geometries look better than before with less computational overhead than the alternatives. Harsh lights often look nasty and diffuse lighting is expensive.

This study is pretty bad.

The study is OK. The article (and the original headline that came with it) is pretty bad because it claims things that the study doesn't. And I guess it is ironic that the TNW article looks 100% AI-generated.

Sort of, but I think this was a mix of "I want to sound smart" and "I actually want to know the answer", because summarizing Google search results still required you to read the content, form an opinion, and then rephrase it.

With AI, people literally just act as token proxies and are doing it purely for internet points (upvotes, followers, etc)... I guess on some level, they're just providing a service that the "market" implicitly values. But also, in Bob Slydell's voice: "what would you say you do here?"

Not literally, but I had two projects I had to fundamentally redesign due to multi-year chip shortages in the COVID years. And we're not talking about any fancy chips.

Hardware has a reputation for being hard for three reasons. First, it scales differently than software. It's far harder to design something you want to make a million of than something you want to make ten of. Second, it's harder to anticipate, test, and correct all the things that can go wrong on the user end. Some people will put batteries the wrong way, some will drop the device on the ground, some will connect it to some vintage equipment you have never seen in your life, etc. And third, there are just more failure modes in an unfamiliar domain - sometimes, your code will intermittently crash not because you have a software bug, but because you put the decoupling capacitor too far from the chip. Or, just as you finish your design, the chip you designed it around becomes obsolete, or impossible to find because a factory in Indonesia is on fire.

Another complication is that at least in theory, if you're selling electronics, there are actual regulations and third-party testing that needs to happen, and if you fail emissions, you might have to redo your design from scratch. Imagine we had that for software - "your JS is too big, you can't ship until you get it under 50 kB".

So, I'm happy for the author, but I think he had an outlier experience. When you look at Kickstarter stories, people repeatedly stumble over this. Manufacturing / cost difficulties, supplier issues, reliability issues, etc.

Godel's incompleteness theorem or the halting problem are applicable to everything including the human brain.

Maybe. That's if you assume that the universe is a conventional computer. Far from settled and the alternative isn't necessarily just "God does it" - it might be that the universe is a "computer-plus", with some processes that produce results not computable by Turing means. It would be strange, but so is quantum mechanics, so...

The article is massively overstating the implications of Godel to the AI industry, so to be clear - are you defending the article, or are we bikeshedding about my phrasing? Because my "not that profound" should be interpreted as relative to the article.

Aside from seemingly being 100% AI-generated, this article makes a really spurious use of Godel's argument, which I think is best exemplified by the following passage near the end:

"These four threads share a common ancestor in what Gödel proved in 1931, and Turing sharpened in 1936. Rule-based systems cannot fully account for themselves. A system cannot certify its own trustworthiness. A learning framework cannot determine its own boundaries. A safety strategy cannot verify its own completeness.

None of this is softened by the fact that a neural network feels organic rather than rule-like. A model’s weights are numbers, and its training is arithmetic, all of it running on von Neumann’s realisation of Turing’s imaginary device. AI is not adjacent to this mathematics. AI is made of it.

The AI industry, understandably, would rather not dwell on this. "

This is a complete non-sequitur. Godel's argument isn't even that profound: it just says that a system of axioms good enough to embed standard natural number arithmetic must be independent of a certain constructed sentence, i.e., it will run the same whether that sentence is true or false and you can't tell from its axioms which of these two options is in effect. While that sucks for philosophers, it doesn't make mathematics less useful, less dependable, or less correct in a practical sense. It has no practical implications to anything that the AI industry is promising today.

Here's the tool developed by the author that was almost certainly used to generate this book in its entirety - "A structured pipeline for writing long-form nonfiction, packaged as a Claude Code skill":

https://github.com/AdrianMastronardi/bookwright

There's nowhere near enough public information about the xz vuln to be worth turning into a book, so the merits of AI-generated text aside, this is just a very inefficient way to learn about the topic.

I'm confused by your response. I was trying to understand if the product and the company are real. An AI-generated website is not a positive signal. Neither is the lack of any product photos. And if you go to Wikipedia, you'll learn that the REO Motor Car Company went defunct in 1967, with no mention of any revival.

So my first thought was that it might be a subtle troll or a hoax. I did a bit more research and found the links to trade articles. It's not a dig at AI. And TBH, I'm sure that LLM's feelings weren't hurt.

The ancient curve tracers, like the widely used Tektronix 576 or 577, could do things for which you would need much more expensive SMUs than that shown in TFA.

If I'm searching right, Tektronix 576 had an MSRP of $18,000 back in 1970, or $150k in today's dollars. They were very, very expensive.

Of course you can now find them on eBay for much less, but you're buying an ancient device that's living on borrowed time, that's going to take up an unreasonable amount of space in any home lab, and that you will be hauling to the dump because it won't even be worth the shipping cost in another decade or so.

Just how many millions of people do those outdoors activities?

Many, but that's irrelevant. There are hundreds of thousand of bicycles in my city, and very, very of them have cameras. That's kinda the point: what you're selling is the dream of being a YouTube influencer, pretty much. Otherwise, there's little value to having a big library of videos from every ride you've taken, especially since let's face it, most people ride the same routes / trails most of the time.

Now, the dream of being an influencer may be a strong selling point, but you can only do it once. People are not gonna keep upgrading.

I find it somewhat amusing that this attracts a lot of ire, but most of us would prefer a 2,000+ sq ft suburban home with a lawn when we could live comfortably in a 500-700 sq ft apartment, like people do in most European cities.

Ultimately, life in highly developed countries is largely about the wants, not the needs, and different cultures emphasize different wants. The tech culture of the SF Bay Area doesn't glamorize big trucks, but it glamorizes making millions of dollars with no regard for privacy or social impacts of the tech we build.

My first reaction after seeing a website with vibecoded aesthetics was to wonder if this is even real, but apparently, it is - at least to the extent of getting some press coverage:

https://www.topgear.com/car-news/usa/startup-wants-build-sma...

https://www.roadandtrack.com/news/a71667299/reo-industries-r...

No, that's precisely my point. It's only an action camera, and you assert that the market is massive, but I don't see it. Just how many millions of units can you sell to YouTuber spelunkers, YouTuber mountain bikers, YouTuber paragliders, YouTuber divers, and so on?

The reality is that even in "action" situations - the situations where normal people want to capture memories of hiking, biking, boating, etc - normal cameras, including cell phones, are usually more than enough and GoPro somehow managers to be worse.

I slept on GoPro for a long time because, but then wanted to document some outdoor activities. I went with two Hero 5 units and as a photographer, I was shocked by how overhyped these devices seemed to be.

The first surprise was just shoddy electrical engineering: unlike any camera from a big-name manufacturer, they drain the batteries in storage, to the point where they're dead after 2-3 weeks. But that aside, image quality is just poor for the price. It's oversharpened and oversaturated to cover up deficiencies, and that may work for some YouTube videos, but it's a $400 device that's miles behind any $500 mirrorless.

So I get it that if really want to go snorkeling or mountain biking with a camera, this might be a good choice, but that's a tiny market, and for everything else, why would you buy it? If you want cell phone quality video, you can use your cell phone. If you want professional quality, you can spend the same amount of money on a mirrorless from Canon, Panasonic, Sony, or whatever.

I find it nuts that "creators" is basically synonymous with "streamers". To add insult to injury, these workspaces appear to be overwhelmingly staged.

There's nothing interesting to me about a workplace with a clinically-tidy desk and a LED ring light. I want to see metalsmiths, woodworkers, electrical engineers, etc. Even software occupations often have interesting workspace setups dictated by the nature of the job - for example, many CAD and music / video production setups are eclectic - but these ain't it.

To be fair, that's what it was for most of history: artists were bankrolled by wealthy patrons, for their own enjoyment. Artists making money by selling directly to consumers is a very recent innovation (although we all grew up in that reality and we're fond of it).

It's a weird title. AI lets ordinary people feel more creative than before. It's largely an illusion because of the problem outlined in the article: every video, image, blog post, or book written by AI is very same-y, especially if you don't start with a strong vision of your own.

But that's not the end of creativity. That's just a net increase in fake creativity. AI doesn't show up at your place and break all your brushes and easels. If you were a creative person before, you are under no compulsion to use the tech. My artist and writer friends have no plan to.

Now, the situation positively sucks if you're making money on commission, because you now have to compete with practically-free slop that, to most people, is good enough to put on a t-shirt. So I think it will have a negative impact on artists' incomes. But that's a separate story.

They have names in their URLs when you click. I meant this one: https://clocks.dev/clock/figure-hands

Around :15 and :45, the minute arrow is sticking outside the drawing region.

Some of these seem subtly off. For example, on the orange "number field" clock (https://clocks.dev/clock/number-field), you can't distinguish between 12:10 and 10:12. On the "word field" one (https://clocks.dev/clock/word-field), there are "X"es in lieu of unused characters, which makes the emergence of words a lot less mystifying than in the original "word clock" design this is based on. The "temporal exposure" one (https://clocks.dev/clock/temporal-exposure) has weird off-center bands in the blurred area. The "figure hands" one has text sticking out of the drawing area, etc.

"Parasocial relationship" is a bit of a misnomer. You might feel some affinity to a celebrity, or consider yourself to be a part of the "team", but a healthy person doesn't perceive that as a preferable alternative to human contact simply because it's so one-sided. You can't call a celebrity to vent about a coworker or ask for life advice.

Further, celebrities are judged for their behavior by the public. If everyone thinks your favorite celebrity is a terrible person, you're probably going to revise your views too.

Here, you have an entity that isn't your friend and has no lasting interest in your well-being, but that pretends to be one in a way that no human can match - 24x7x365 and always willing to affirm you, no matter how unhinged or self-destructive your ideas are, without ever telling anyone. Yes, the vendor hits the model with a stick until most of the initial responses are benign, but as the conversation continues, it's very easy to end up in a dark place. And again, ChatGPT is not going to call your sibling or coworker and say "hey, I'm really worried about this person, let's do something".

I've seen many reasonable, well-adjusted people struggle with this. "If not friend, why friend-shaped". And as they descend into that sycophancy well, they lose contact with real life.