Hacker News new | ask | show | jobs
by 1024core 888 days ago
From the paper:

> When producing full natural-language proofs on IMO-AG-30, however, GPT-4 has a success rate of 0% ...