Leo A @stackycurve

@rperezmarco @OpenAI I completely disagree with "Yes". 100 open problems sounds like a high probability of scooping someone. Maybe release the list but give time to researchers to post their own solutions if they're already close? But it would still be bad if openAI has claimed the problem first...

@hive_echo @_chenglou Rumors said they were getting close, but I'm happy to not be standing corrected 😁

@hive_echo @_chenglou BSD does not get solved anytime soon. I think all the rest of the Millennium problems are like that, but am not sure. The sad part is watching tech CEO's treat millennium prize problems like they did the Erdos ones, and worse: "Quick, find the easiest one and throw billions!"

@shakoistsLog What you don't understand is that we don't care to be "important" we just want to keep doing what we were doing. Yes, I know, we have to change, that doesn't mean we have to be happy about it.

@ElliotGlazer The letter was fantastic communication. It would be really sad if all this ended with these great minds having to bicker with people like you, esp on Twitter

@skdh Yes, because none of the concerns mentioned would apply. Did you even read the letter?

@MatthewAlbano8 @Abelaer @grok @MatthewAlbano8 Grok just didn't get the joke/point 😛 Basically since LLMs work through linear algebra, if LLMs can (after sufficient improvement) solve everything, then linear algebra does.

@ThinkDi92468945 @tak3sh8 "Nothing magical about Galois theory" is one of the shortest ways to indicate you do not have the slightest clue of what you're talking about

@AndrewCurran_ They are not gonna solve PvsNP, and I don't believe RH either. They are just burning money and electricity.

@bzogrammer I will be surprised if it is solved now. They'll probably discover a problem in their proof while trying to formalize, and we'll never hear back from this because they'll work out their counterexample to Hodge.

Rumors are not proofs. I hope they burn billions and all they learn is just how hard can something like the BSD conjecture be.

This will be important because supposedly AI is taking off to infinity, but on what will it train for it's next steps?

On supposedly AI godlike capabilities: don't forget Navier Stokes was not fields medalist work, just 10.000 tireless PhDs in a fields medalists' trenchcoat.

@__alpoge__ if you are a hobbyist wanting to help mathematicians, the Langlands program has a bunch of problems unreachable from LLMs in the near future, where you could compute a bunch of useful stuff then let mathematicians find the patterns and theories.

If true, OpenAI autonomously solving the Navier-Stokes Millennium Problem would have ground-breakingly demonstrated out-of-distribution reasoning in LLMs. Sadly, it's likely plagiarised. Millennium Problems almost certainly require approaches beyond in-distribution existing methods or "connecting the dots" between papers. Millennium Problems likely require entirely new techniques, analyses, or even branches of mathematics to be devised - i.e. strongly OOD approaches. LLMs are however notoriously bad at OOD reasoning and creative ingenuity. ... as is GenAI in general. OOD reasoning - i.e. reliably venturing far beyond the scope of training data - is a critical problem across all AI algorithms available in 2026. So when OpenAI announced that they had solved a Millennium Problem, I was naturally extremely shocked... ... I was willing to accept partial defeat as a sceptic on AI overhype, and seismically shift my current research-grade understanding of AI. Well... nope. The plagiarism accusations raised are significant - alleging that the bulk of the work was stolen from the private Codex chats of mathematicians Tristan Buckmaster and Levent Alpöge. Given strong OOD pain points in LLMs, no major published breakthroughs in overcoming them, the published evidence from Buckmaster, and previous bad conduct of frontier AI labs regarding IP, privacy etc... Occam's razor seems to favour the misconduct explanation. So no, we are not in a "math singularity" or "RSI". LLMs have made plenty of progress in mathematics, but now are hitting hard walls in problems that require OOD reasoning rather than in-distribution "connecting the dots". Elite human researchers still beat LLMs, on creative ingenuity. And the only ways that LLMs can so far "overcome" this, is by plagiarising those elite human researchers.

The way OpenAI handled this Navier--Stokes situation was the first time I've actually felt disgust at the whole AI in math situation. Like watching a billionare celebrate his hunting prowess by mounting the head of an endagered rhino on his wall, after 1000 soldiers trapped it.

the biggest take away is they solved one of the hardest problems in mathematics, not for humanity nor for the prize money nor for the science, but for the ability to poast about it on twitter faster than the other guy, truly beautiful

Sebastien Bubeck @SebastienBubeck

4 weeks ago

Some more technical points: (a) We began working on the Millennium problems due to viral twitter rumors that Anthropic had resolved 2 Millenium problems. Our aim was to see whether our system was also capable of this impressive feat, especially given our excitement regarding

126 105 2K 1.0M 466

Couldn't help myself

If AI becomes better than us at producing proofs but they are not understandable, it will just reach a poisonous data state.

@rperezmarco I've got a theory you'll like: Alpoge was only using Codex for this so he can discredit openAI if they solve it first

@sytelus I'm sure you can get serious physics using pure logic and throwing tons of compute. Hopefully now that they got a Millennium problem they will move on to more useful stuff