The LLM+Lean approach to proving mathematical theorems and hypotheses seems to be a genuine new phase for some important part of mathematics. Currently it is full of credibility issues and "AI bro" dodgyness, but intrinsically it is simply a reflection of the fact that a good chunk of a certain type of pure mathematics has a combinatorial logic that is amenable to an algorithmic approach. Its not the part of mathematics that is most creative in connection with reality, but thats just me 🖖
0 Boosts
0 Likes
@jacques it's indeed probably the case. The "AI" bros wouldnt be the ones pioneering elegant or efficient solutions 😆. Right now the overwhelming monetary incentives create an intense pressure to try to use LLM's for anything that can create marketing buzz, whether it's suitable or not. But to the degree these types of results are actually valid / validatable, they do suggest new ways of working, for at least some class of mathematician. Not holding my breath for any sort real creativity though.
@openrisk sure but this is a dumb way to do it. I suspect genetic programming ie ant colony optimization would roflstomp an LLM. Hell any old fuzzer with a trillion dollar budget would do better