OpenAI releases 719 AI math proofs, Terence Tao warns of 'proof indigestion'

OpenAI has published 719 AI-generated math solutions, covering 372 groups of math results and hundreds of open research problems. The company had consulted a math and AI advisory group, but public info shows the release did not fully meet the group's standards. The group suggested top AI labs stop testing hard math problems on closed models, and share model names, prompts, and reasoning steps. But OpenAI still used closed models, only 10 papers came with reasoning steps, and about 42% of the proofs were not formalized. Fields Medal winner Terence Tao says he is not against AI in research, but opposes using it to quickly crack famous problems as a product showcase. He believes AI creates proofs faster than humans can check and absorb them, causing 'proof indigestion' in the math world. Once a problem is publicly 'solved', it's hard to go back, which could hurt open exploration in math.