I feel something that is not mentioned enough in this whole "AI solving math problems" development is the role of Lean, which allows formalizing and checking proofs. It's the tool that lets you distinguish between "hallucination" and "correct result". The way these proofs are generated is essentially "AI throws stuff at lean, lean tells AI whether it's correct, repeat."
I think like half of the credit for all these findings should go to the Lean creators.
1 Boost
0 Likes
No replies yet
Be the first to share your thoughts.