Altman acting smug. Another trillion dollar valuation soon.
>>110019258How did this homosexual jew spawn out of nowhere and become a ceo
he sound like a cult leader now
not so fast>>110018276
>>110019275>now
>>110019296It's not that the lean proof was wrong, it's that the model did some sketchy things to arrive at that lean proof that ultimately turned out to be correct. That has obvious implications for the current tranche of slop being released but in the specific case of the N-S blowup, no.
>>110019303It's that the lean proof is a different proof than the natural language proof. How do you even know what it's proving in that case?
>>110019258can he work on making astra actually good instead of shilling on twitter, thanks
>>110019296They've also already retracted three of the papers they posted yesterday and made updates to more than 10 others.
>>110019317Because the finite-time blowup counterexample being described is verifiably correct.
>>110019328Can you read lean code or something? How the fuck do you know? Just because there's some code and it is correct, doesn't mean it does what you think it does.
>>110019334>How the fuck do you knowBecause that part is no longer under dispute. Go look it up. What happened was that the model wrote an NL proof of N-S (probably incorrect, has not been verified), formalized it into Lean (incorrectly) and ended up with a verifiably correct Lean proof. Even this paper itself doesn't have an issue with the Lean proof>Disclaimer: We do not make claims about the correctness of OpenAI’s NL proof, we only make statements about mistranslations into Lean.
how many new eras have there been in the last three years?
>>110019345So it wrote an incorrect proof and then incorrectly translated it to lean and somehow two wrongs accidentally made a correct proof of the original claim? Bullshit.
>>110019265He got that "seed round" from Thiel.
>>110019317The NS informal proof was correct since it got reviewed by experts in the field but the lean proof wasn't one on one translation to the informal proof. It could be correct in different way or the model exploited lean bug. The matter is complete unsupervised lean propf should be taken with some salt. It could that your informal proof be correct or wrong but autoformalization might produce different result to fit the audit definition.
>>110019545really? I'm seeing that experts in the field say it doesn't prove anything interesting
>>110019265He made his bones, probably killed a child in some bohemian-grove-style sex ritual. He's a useful idiot.
Do any of their discoveries have practical applications?The underlying question on my mind is not even related to AI. I'm wondering if these problems were ever worth solving or if it's all just like purely theoretical or academic in terms of merit/value.More like a puzzle for fun than a true problem to solve.
>>110019628Yeah because if the lean slop is too big and too unreadable how are you going to extract anything else other than yes/no?
>>110019827There's matrix multiplication, integer multiplication and fourier transformation result. Others are trying to improve it now. Also the unique game conjecture proof confirms far more tighter approximation bound for many algorithms.
>>110019628Oh sorry you meant ns. Kinda since the ansatz they used was quite unphysical so only answer we get is that its false also the proof can't be extended to unforced case.
>>110019258>Acting smug when your product is not profitableLmao
>>110019265He didn't spawn out of nowhere, he has been scamming investors for money for a long time.He is a genuine sociopath specimen though lmao, who the fuck would wear two polo shirts like that
>>110019258man who says gay shit like that? is it le profound to sneed valley metrosexuals?fuck you and your small boat
>>110020460Why did you use AI on this image? How do I know you didn't just transplant his face? Retard
>>110019296openai already got their press, doesnt matter
>>11001935678
>>110019799I'd wager it's natural to him consider he raped his toddler sister when he was in his teens.
>>110019296I'm a nigger and I need someone to explain this to me. Does this mean that the lean proof is wrong? Or that the natural language one was wrong? Or both? What's going on?
>>110019829>>110019895I mean that since the proof deals with an ideal fluid that even if the proof is correct it doesn't apply to anything in the real world because the real world doesn't contain ideal fluids. And a human would realize that but I guess the LLM doesn't know or care.
>>110022572It means that the lean proof is not an accurate translation of the natural language proof, so even if the lean proof is valid, it doesn't prove the same thing, and maybe doesn't prove anything meaningful at all.Sort of like how you can ask an LLM to write a computer program and it might even output a valid (compiles with no errors and runs) program, that doesn't necessarily do anything, or doesn't do what you asked it to make it do.
>>110019265he networked with the right people (graham), at the right time (low interest startup era), and impressed them with his ruthlessness and business-oriented mindset (sociopathy).
>>110022883Unironically why do they even bother with the natural language proof? If it has to be ununderstandable slop anyway, might as well cut out the middleman and just do the lean proof directly (assuming it's correct).
It's 2026, we don't have to type in lower case to look quirky and hip anymore.
When it comes to clankers who really cares who got there first. If a human solves a hard problem you can't always know for sure that someone else could have solved it in the near future. Does anyone think that Anthropic couldn't do the same with their next model?
>>110023333Without connecting the lean code to something then it's meaningless. It would just be a bunch of disassociated logic. Like a computer program that runs without input or output. Yeah maybe it works but it doesn't do anything.
>>110019258He needs to be slapped.
>>110024008As a gay jewish faggot, he would like that.
>>110023996I mean couldn't they formalize just the problem statement in lean (which I don't think would be too hard even manually) and then let the ai brute force it directly in lean, skipping natural language entirely? This way they would be sure the ai proved what they asked for. If it helps the ai, maybe give it the natural language problem statement too, but I would have the lean formalization first. Why go through the larp of having something that resembles human reasoning?
>>110025790Evidently it's not that easy or they would have done that.
What the hell is going on in this thread. The Lean theorem was written manually by mathematicians. The AI wrote a lean a proof for that theorem. The proof is valid. The natural language translation doesn’t faithfully represent the lean proof. So it’s unclear how the valid proof works without reading the Lean code. Reading the Lean code directly for a long proof is reading extremely detailed logical steps without any context.
>>110026563meh, I guess we'll figure out a use for it, even the dude that discovered invincible radiation thought it would be useless, but then the tech caught up and now we have wifi
>>110026563The problem is LLM generated code is much more verbose than human written code.This isn't that much of a problem for software development but it is for mathematical proofs.Furthermore LLMs often don't rely on third party logic, be it libraries in software or reliance of other proofs in mathematics also they don't generate these abstractions.So the OpenAI proofs aren't useful for driving progress in mathematics.
>>110025790That's not how it work. For it to be possible you need to formalize majority of existing mathematics. If you could you wouldn't even need llm for proof search rather replace it with some simpler algorithm. But since vast majority of math is still informal it actually needs to reason in natural language. It only can formalize it after it does so.
>>110026629It’s certainly difficult to understand the Lean, but is it more difficult than solving the problem? It’s too early to tell if this is useful or not I think, and the technology still appears to be improving so the explanations may get better. That being said it is not clear the current iteration of math results is useful.>>110026674The theorem statement was written by humans. A good chunk of math has been formalized. Go check out mathlib.
>>110026729Whoops, meant to say: “It is unclear that the current iteration of math results is useful.”
>>110026729>A good chunk of math has been formalizedKek. Do you have any idea how many theorem lemma or definitions out there? Good chunk is laughable unless you meant undergraduate mathematics.
>>110026729>It’s certainly difficult to understand the Lean, but is it more difficult than solving the problem?In some cases it did help.For instance there's a proof that the bound for integer multiplication computational complexity is very slightly better than the currently known limit, by some ridiculously small amount like 10^-127Human mathematicians were then able to improve that through actual reasoning by several magnitudes.Still useless but it might be possible for them to keep improving the result, by human ingenuity to be actually useful and it might lead to slightly faster computation if that were to be the case.
>>110026563>The Lean theorem was written manually by mathematicians.Do you have a source for that claim?
>>110026729>The theorem statement was written by humansIn this case, yes.>A good chunk of math has been formalized. Go check out mathlibLol, no, not even close. Mathlib is missing a whole bunch of basic theories, so you can't even state a huge chunk of problems in Lean to begin with, let alone prove them. Take something like FLT, Anthropic's attempt at formalizing it took something like 13 million lines of lean. Formalizing mathematics is a shitshow and "lol just throw AI at it" isn't the answer either.https://leanprover-community.github.io/contribute/index.html#use-of-ai
>>110026857>In this case, yes.Why do you keep repeating that when it's not true?
>>110026876The math community has already moved on and are running with the proof and seeing if they can generalize it further. I'm not sure why you're so hung up on insisting that it's wrong.
>>110026843>>110026876https://github.com/openai/NavierStokesAndEuler/blob/main/ComparatorChallenges/README.md
>>110019258>yes yes well done...>HOWEVER
>>110020443>leftists suddenly care about the bottom line
>>110026910lol
>>110020495Mindbroken lmfao
>>110027186>Noooooo saaar do not redeem the negative opinion about AI noooooo
>>110026933Yeah that's a biggie. Sattellites alone need adjusting regularly just from cosmic radiation and whatever cuz it messes with electrons in their CPUs
>>110026857The lean proof was fully written by the clanker which is why it failed to give an exact translation
https://www.ft.com/content/205ef7a5-321f-46e3-8097-d055bf1f6b3b?syn-25a6b1a6=1
