My understanding is that that's because they were trying to do a structurally homologous port from Zig to Rust, precisely to keep their mental model and not change "too much" at once, and then they plan to refactor to make it safe Rust later.
it's clear that as of the time of this merge, no human has read any appreciable fraction of current mainline bun, so it's not particularly clear how much of a "mental model" exists anymore.
@simonw explains how hilariously misguided that paper is in one of the top comments, and how it doesn't apply remotely to a real agent harness. Plus it's not even clearly relevant here, because the model isn't trying to regurgitate the original document, but generate a new one, and there are guardrails to put it back on track in the form or a compiler and tests. Also, the test suite is very thorough, and pre-existing, and the vast majority passes already. This is skepticism for the sake of it.
> The duo had jump-started the AI-for-Erdős craze late last year by prompting a free version of ChatGPT with open problems chosen at random from the Erdős problems website. (An AI researcher subsequently gifted them each a ChatGPT Pro subscription to encourage their “vibe mathing.”)
Wonder who the AI researcher worked for? Is a "craze" something which a for-profit company would want to encourage? Maybe they'd think the publicity would help keep people talking about their company and product as we are now doing?
> “There was kind of a standard sequence of moves that everyone who worked on the problem previously started by doing,” Tao says. The LLM took an entirely different route, using a formula that was well known in related parts of math, but which no one had thought to apply to this type of question.
Yep, I do remember this now. Everyone was yelling that this was definitely a sign of ground-breaking and creative work, citing the expert. What the expert actually said suggests that the solution was available in training data! That also suggests the math in TFA is harder in comparison, answering my other question.
Pleasure as always HN, thanks for voting me to the bottom of the thread for this
First, you ask for evidence of someone who isn't a VIP doing a similarly difficult problem using an LLM, to show that it isn't just VIPs being given special models. And then, when I provide that example, you say it doesn't count because the whole craze was started by researchers working for AI anyway.
Furthermore, you start out stating that the access being given to these VIPs is to insanely massive, impractical models no one will ever have access to, but then you point to them getting a free ChatGPT Pro sub as evidence of your point.
Finally, you look at the fact that the AI solved the problem by applying a technique that no one in sixty years had thought to apply to that situation, from a totally different field of mathematics, and you claim that that isn't sufficiently novel to "count" as being in the same ballpark of difficult as solving literally other easier Erdos problems, or this new problem, because the technique still existed previously, and so actually isn't hard enough to be comparable to all the other stuff that's been done.
If you are upset that you are being down-voted, I think you should do some introspection. It seems like it would be impossible to convince you, no matter how many non-VIPs solved difficult open math problems, as long as it wasn't literally the exact same level of difficulty or type of problem.
I'm grateful for the reference as context, but it simply does not settle the issue of transparency and it did reinforce the question. That's not controversial, nor is the statement that OpenAI avoids transparency, wants good publicity, will go to great lengths on both of these things like all corporations. It's also completely normal to think progress on Erdos problems is fascinating and inspirational, but aim for clarity on what the scope of the achievement really is (invention, composition, literature search) per the limited information available.
What I raised is a real question: the erdos folks are not affiliated with AI companies.. but is the AI company affiliated with them? Actually knowing the user/org accounts involved is optional because they could just divert resources to anyone prompting near the problem. Anecdotally.. I've noticed what appears to be token-discounts based on topic, for example more generosity for AI-related research than random stuff, but it's hard to know for sure. Wouldn't you promote interactions you could profitably train on?
So again, allocating resources to Erdos one way or another is just a clearly smart business decision for something people are talking about and which has become an unofficial competition among vendors, not a big scandalous accusation. Something like a reasoning-trace is the only way to settle it. This isn't conspiracy or nitpicking because this is the topic itself: the AI usage is more a matter of public interest than the actual problem solution. What's the argument against more transparency?
Honestly, I fully support the rewrite to Rust, but he should have just owned this from the start. I'm sure he knew in the back of his mind how dedicated he was to that branch as he had already spent the equivalent of thousands of dollars in tokens by that point.
That's why I said "the equivalent of". Additionally, time and cognitive effort are not free. The work spent on this branch was work that was not spent on other branches. Does that make sense?
6 days is also nothing when you're doing R&D on your company's dime. He could have spent a month trying a dozen different things and thrown away all the code at the end. As long as he ends that month with a clear picture of where to steer the company over the next 5 years, it's time well spent.
Not even the company is spending money. It’s their employee working on a rework of the code owned by the company that owns the infrastructure on which the rework is done. And that company is still yet to turn profit. This work is subsidised by everyone who pays for Claude.
Announcing the decision a week earlier wouldn't help anyone. Maybe he expected it to work (though he didn't say that), but there's no reason to make a final call before seeing that it did work.
Fair enough. I didn't say anything about a "final call". It just feels like there is a middle ground between that and telling people they are overreacting.
Yeah, but he obviously had enough confidence in this project to keep the agents working at it, didn't he? Given infinite time and money, if you prompt an LLM about something enough times, it will eventually work.
Insert something about monkeys, typewriters, and Shakespeare here.
> It is always Gowers, Tao and Lichtman (math.ínc startup) who are pushing these technologies.
In your mind does this mean that they are lying, or driven by motivated reasoning and cognitive bias, or whatever you'd like to say?
Because I feel like people bring up these facts as a way to discount everything that these people are saying, but whether or not they've chosen to align themselves with AI aligned venture capital funding or not. The question is really, did what they say is happening happen or not? Are these capabilities real or not?
To my mind, mathematics is pretty definitely, externally, objectively verifiable, so it would be easy to catch them in a lie. In the case of the Erdös problem that was recently solved in a novel and productive way, it wasn't even initiated by them and the chat GPT transcript is public for all to see. And the proof could easily be verified by other people, for instance.
In addition, I think it's unlikely that they're not explaining things as they honestly see them and also doing their due diligence to make sure that they are seeing them as close to correctly as possible. Because their positions with these organizations not to mention their entire reputation and life's work and passion depends on their reputation in academic mathematics. If they were to give that up by falsifying these claims or not verifying them sufficiently, they would lose everything.
I think it's also worth pointing out that it is totally possible for someone to align themselves with such organizations after the fact because they agree with them instead of being bought out by such organizations. Otherwise, it would be possible to dismiss the opinion of anyone working at any NGO dedicated to being against AI and denying AI's capabilities or whatever, as well by the same logic of their salary being paid by an organization dedicated to pushing those ideas.
Yeah. People (Gary Marcus) have been claiming that AI will hit a wall or is hitting a wall or already has hit a wall since 2023, basically. And yet every time they proclaim that the AI industry found new ways of training their AI's, new ways of integrating them with external tools and feedback loops, new architectures and more to keep the exponential growing. And sure enough if you look at literally every attempt to objectively rate and verify the capability of these models, including things like the METR time horizon autonomy index or the artificial analysis intelligence index, you see exponential or even greater than exponential growth, continuing smoothly through each of the points people claimed that it would begin to slow down, with no sinus slowing down or stopping at all. So yeah, I think at some point the onus has to lie on the ones that are making the claim that keeps being wrong and the continues to be wrong and it completely goes against the current tangent of the curve that we're seeing in all objective metrics. Especially when they can't give specific new reasons for progress to stop beyond the ones they gave last time. It didn't stop and really can't give specific reasons at all besides vague general points about stochastic parrots and S curves.
I really have to highlight the S-curve nonsense because, like, yes, I think this technology's improvement will follow an S-curve. It's absurd to think that it will just follow an exponential up towards infinity forever because nothing in the world really works like that. However, like everyone else in this thread is saying, we have no idea where on the S-curve we actually are, and it's impossible to know until it's already slowed down. So really all appeals to the S curve do are as function as a sort of non-specific, unfalsifiable prophecy that someday it will slow down, which doesn't really tell us anything useful, and also frees the person referencing the S curve from ever actually having to worry about being wrong. Just like the Singularity people, the slowdown of the S curve is always near. This is actually a known and well-established tactic of religions and other people that want to make prophecies without having to worry about turning out to be wrong — unfalseifiable vague prophecies with no actual timeline, and thus no clear import to the present so that they can never be shown to be wrong.
Projects like opencode are making the distinction between GUI and TUI almost meaningless. And that "only" downside is a massive, deal-breaking one. At this point I only have a browser besides the terminal, and I can see that going away soon for the most part thanks to LLMs.
