p-e-w

If you consider supporting a Russian company equivalent to “supporting Russia”, then I assume you are happy to be supporting the United States by using US services… like the one you are on right now.

a human prompting with deep math expertise

The original tweet implied that the whole thing was done while the author was watching the World Cup final.

I know it’s tempting to hope that a human did the “real” work here, but if some special insight was put into prompting, the author kept it to himself, and there is no reason why they would hide this since it would elevate their own status.

Just because a company wants to know who you are doesn’t mean they are entitled to a scan of a sensitive government document that can be used to commit identity theft.

I’m amazed that it is even legal for businesses to ask for that at all, outside of a few exceptional cases of which “I want to rent a server” most certainly isn’t one.

Wait, Thunderbird is getting more donations than Blender?

I was under the impression that Thunderbird is on life support, while Blender is everywhere.

And they can’t stop because they want to IPO soon, and stopping this madness would signal loudly that they are bleeding cash, don’t really have a business model, and might not be worth $2 trillion after all.

It’s safe to assume that after less than a decade of LLM development, we’re nowhere close to the limit yet. In fact, progress still seems to be accelerating at the moment.

GitHub's moat is not code hosting

Of course not, it’s the number of people who are already signed up.

Instagram’s moat also most certainly isn’t a scrollable photo timeline.

Unless you’re claiming that AIs will suddenly (and very soon) stop improving, they are obviously a threat to everyone’s job.

Calling notable conjectures that have been open for decades “low-hanging fruit” is an act of desperation. Most professional mathematicians couldn’t have proved those conjectures if their lives depended on it.

I’d be interested too. Besides the fact that the company appears to be registered in Finland, I haven’t been able to find any information on who’s behind this, how they are funded, etc.

But the model recognized that they apply here. That’s absolutely non-trivial. It could have easily mistaken the line pattern for an autostereogram and told the user to cross their eyes instead.

It’s absolutely incredible that the model can deduce what the human needs to do in order to more clearly see the text.

The single point of failure is always the biggest instances or communities, which impose their rules on everyone else under threat of refusing to communicate with them.

This has happened with every single federated platform, and it can’t be solved with technology because it’s fundamentally not a technology problem. People want to be where other people are, so once the community reaches a critical mass it inevitably degrades into the same sump of mediocrities and their rules as everything else.

Nowadays, with the focus on agentic use and coding, it seems models have all been RLHF’d to death

I don’t get it. If nobody likes this writing style, how can it be the result of human feedback? Something else is going on.

So if things continue as they are today, I think in the near future, being a software developer is going to be more analogous to the medical field, where in the medical field you have different levels of professional expertise.

The medical field is also going to change though. Massively. Because people are going to realize you don’t need to pay someone $400k per year to hand out advice about moderate exercise and which antibiotic is appropriate for a sneeze-cough with yellow mucus.

Regulation isn’t going to prevent this. AI is already way too easily accessible to ever rein it in again. Not to mention that the US now has serious competition from a hostile country, so they can regulate their own AIs all they want without it making a difference in practice.

This happens to me every time I take a quick look at the “high culture” world, be it literature, art, music, or general intellectualism. I have a minor in philosophy and consider myself reasonably well-read, but today’s avantgarde is nothing like that of the past, and seems above all else designed to be exclusionary.

I think the Zig people are really just concerned that maybe Zig itself is a DOA language because it doesn't offer enough over C for any serious use and their flagship project has now abandoned it.

You hit the nail on the head there. Zig is 10 years old now and it’s pretty clear that the industry isn’t biting, compared to the behemoth that is Rust. Between Rust, C, and C++ there is very little room for another language with a woefully incomplete library ecosystem to establish itself.

A true competitor would need to offer genuine extra value, such as dependent types or other formal verification features, to carve out a niche.

Which of these papers “have been considered extremely harmful to society” by a significant number of people who wouldn’t be classified as cranks?

I’m asking specifically about claims that those papers have harmed society. Not cop-outs like “the author does things I don’t like”.

If they indeed made search worse “on purpose”, there should be plenty of alternatives with better results. I’ve tried about a dozen of them, including paid ones like Kagi, and found the quality of results to be universally worse than Google’s, especially for niche searches where it matters the most.

Don’t bother. This is the third or fourth time AI has solved an open conjecture, and once again the comment sections everywhere are full of people explaining how this doesn’t really matter at all, how it’s an irrelevant, obscure problem, how any mediocre grad student could probably have solved it if only they had bothered to try, and how of course, human mathematicians will still be vastly superior to machines 100 years from now because they have that magic spark for which nobody can say what it supposedly consists of.

These discussions have nothing to do with mathematics, and everything to do with ego and fear. I’ve never been less impressed with humans than since AI started challenging them.

Just because they published it doesn't mean that others haven't already discovered it.

That’s irrelevant for the moral evaluation. Everyone is responsible for their own actions. If you choose not to kill anyone it won’t stop others from killing. But the fact that this is so doesn’t give you a license to kill.

This does indeed seem comically evil.

And I have yet to see a single paper like this where a researcher bails out and publicly says they refuse to work on such projects. Not one.

The most benign interpretation of this observation is that science is filled with spineless opportunists who don’t care who they hurt with what they create. A slightly less benign interpretation might be that many of these people are doing this deliberately, and getting off on the sense of power it gives them.

The difference between today and 30 years ago is that if you are an individual developer, you are at the complete mercy of a single company (Valve), who can force you to do essentially anything they want for the privilege of publishing on their platform, with no meaningful alternative, and zero recourse if they say “no”.

You could (e.g. by replacing residual-dependent expert routing with hardcoded logic), but quality will suffer dramatically. It’s far better to use a similar-sized dense model then.

No, I’m talking about an article I saw maybe 10-12 months ago on a major news site, perhaps The Guardian.

Individual reports of scientists leaving are meaningless without overall statistics. If there is a net outflow of scientists from the US, I’m not aware of it, and I certainly haven’t seen any actual statistical evidence for it.

If you're talking about the article featured on HN just day(s) ago

I’m not.

I’ve seen titles like “Why top scientists are leaving the United States” where the article itself talked about A SINGLE RESEARCHER relocating to France.

It both works and fails, like many other things. But if you hold the goal that it must never fail in sufficiently high esteem, you invariably end up with a system like the one we have now.

It’s not accurate, it’s absurd hyperbole no different from the kind the people who peddle it have arrogantly ridiculed their entire lives.

A mentally unstable person being “made” to do something by a chatbot is no different from other mentally unstable people doing bad things because they saw them in a TV show.

This doesn’t work on iOS, where you can’t selectively disable automatic updates for a single app.

Isn't this conflating encryption with trust?

Absolutely, and the claim is somewhere between nonsense and pedantry bordering on nonsense.

The exact same thing is true for, say, Signal. The provider delivers the client, and they aggressively block non-official clients from participating. So the “ends” in end-to-end are ultimately controlled by Signal. But as long as you trust the Signal company not to insert a backdoor into your client, it’s still true that the company can’t read your texts.