You must see the difference, right? The CFSG is a proof that humans produced first, they understood, and many experts understand and can explain. A verifiable formalization of an existing proof of that magnitude, that humans already understand, is clearly a net positive, to double check that there are no gaps in the 10K pages of proof (there were gaps!). What we have here is Lean-proof of a non existent natural language proof that no one understands since it does not even exist.
Is a "17M-LOC Lean proof" a proof? If no human was ever able to understand or check every step in the proof, is it a proof? We may see rough digestions with the conceptual birds-eye view of the proof, but is that plus a Lean proof... enough? What if BSD's proof is 1T-LOC and there is no digestion that is even close enough to a birds eye view that we can wrap our heads around? Would that be a proof?
ANT has adopted a clever approach to AI math drops (much smarter than OpenAI's), by letting Levent do it and claim them as his own. If he succeeds, Claude+ANT get praise and good press for being the model that made it possible, and if he gets flak for a slop drop, they can let him burn and get all the hate.
Some people are referring to the "RH-Lean-proof" as "Alpoge's theorem." At this point it is unclear what resources were at his disposal to accomplish this. Levent is clearly a very capable mathematician, but he also clearly has essentially unlimited computational resources from Anthropic that made this possible.
Remember, few people here actually know anything about how Lean works, or about what the Riemann Hypothesis says, or the maths used in the claimed proof.
Take everything you read with an appropriate amount of salt and scepticism - even (especially) if they sound very confident.
Yeah people keep pinning Levent as a non-expert AI bro because he (1) works at an AI company, and (2) posts on this website like a bro. But Levent doing cool things with Claude is actually a clear demonstration of what expertise in combination with these AI tools can mean.
Too many talented college students leave math programs over small financial emergencies. @maanow & @LastMileFund launched the National Math Talent Fund to help change that. Students in need: please apply! Math professors: maybe become an ambassador? https://t.co/T85Bm9sGsz
This (Ken's quote) is as far as I know (from my own sources) not accurate. The only update, if any, is from Vakil (image below). The document that was released was admittedly incomplete because it was only released in a hurry due to time pressure. Trying to assess the proof from that draft is premature and trying to use AI or Lean on it is misleading.
One of the most dangerous aspects of this new AI driven world is to give in to the false idea that we need to have a yes or no answer to everything instantaneously. Please let's be patient to see what comes out of Masullo's proposed proof. We don't need to decide today or tomorrow.
This (Ken's quote) is as far as I know (from my own sources) not accurate. The only update, if any, is from Vakil (image below). The document that was released was admittedly incomplete because it was only released in a hurry due to time pressure. Trying to assess the proof from that draft is premature and trying to use AI or Lean on it is misleading.
@0001Polo@KenOno691@skominers The document that was released was admittedly incomplete because it was only released in a hurry due to time pressure. Trying to assess the proof from that draft is premature and trying to use AI or Lean on it is misleading.