2026-10-08 / Signal #6
OpenAI already withdrew three papers from the 722-math dump
“They solved hundreds of open problems. Then they had to take three back.”
The Story
Two days after dumping 722 AI-generated math manuscripts, OpenAI withdrew three of them after a sign error broke a key argument and two dependent papers. Mathematicians are arguing that industrial-scale solving “harvests” problems and leaves the field less fertile; Tao is calling for a Math 2.0 that values exposition and community over raw solutions.
Why It Matters
The proofs got cheap. The checking did not. An AI lab can flood GitHub overnight and still have to retract like a human journal — while the village of mathematicians tries to decide what the job even is now.
Evidence
OpenAI GitHub history.md (Oct 7–8 withdrawals), Dan Roberts/OpenAI, New Scientist, Scott Aaronson “Mathocalypse,” Terence Tao “Math 2.0” Mastodon thread, HN. Sign error invalidated a chain of three manuscripts.
Sources
Daily scan: 2026-10-08