Remix.run Logo
▲ mxkopy 2 hours ago

Even in principle I’d reject this idea. Humans already break large pieces of math up into pieces (see the complete classification of finite simple groups and basically any other large work with multiple authors), so in a sense we’ve already been in this regime where things are too big to wrap our heads around for a long time. It’s not that tectonic of a shift. If anything it’s kind of an insult because the relationship implied here is that mathematicians should spend their time trudging through 200,000 page proofs instead of collaborating with other humans, which is a complete waste unless we know the AI can output the most succinct and powerful version of the proof (which we’re pretty sure it can’t, see the other comment about another GRH manuscript - at 1000 pages it’s still long, but makes the LLM one look suspiciously like LLM slop)