Physix Frontier badgePhysix Frontier
PublishChinese | EN

Physix Frontier · News Briefing Card (enbrief · Oct 11, 2026)

OpenAI posts 722 AI math manuscripts, retracts 3 next day

KEY FACTS

  • OpenAI published 722 math manuscripts on GitHub on October 7, produced by an undisclosed internal model.
  • The model attempted roughly 4,000 unsolved math problems and returned results for 372 problem sets.
  • The manuscripts span 17 branches of mathematics; 3 were retracted the day after release and another 14 were patched.
  • As of October 10, about 42% of the results had passed Lean machine checking, while the rest may have issues.
  • A group of more than 800 mathematicians has called on peers to stop collaborating with OpenAI.

KEY DATA

722Manuscripts published
~4,000Problems attempted
372 problem setsResults returned
~42%Machine-check pass rate

PHYSIX OBSERVATION

AI mass-produces mathematical results that no one can digest, exposing a gap between capability and verifiability. The retractions and the unreadability are not minor blemishes; they shift the cost of verification onto the entire academic community. In the short term it looks like an efficiency spectacle, but in the long term, if it becomes standard practice, it could erode the mechanism by which the mathematics community transmits methods through reports and seminars - a trend worth watching.

Source: enbrief original report ↗

OpenAI posts 722 AI math manuscripts, retracts 3 next day | Physix Frontier Briefing