OpenAI withdraws three preprints a day after releasing 722 manuscripts on unsolved math problems

Key Points:

• OpenAI withdrew three of its 722 recently released math preprints due to a sign error that invalidated one manuscript and two dependent papers, and has revised 14 others with corrections and clearer statements.
• The errors were identified during an internal audit, and OpenAI emphasized its iterative research process, welcoming scrutiny and promising prompt corrections or withdrawals when necessary.
• Some mathematicians, including Cornell's Alex Townsend and MIT's Andrew Sutherland, expressed skepticism about OpenAI's release strategy, suggesting a more cautious approach with verified results first and criticizing earlier rushed announcements like the Navier-Stokes problem claim.
• The Advisory Group on Mathematics and Artificial Intelligence recommended releasing results without full formalization, leading to about half of the results being unconfirmed at release, with OpenAI seeking feedback to improve future protocols.
• The Association for Human Mathematics condemned the mass release as a power move lacking scholarly rigor, urging mathematicians to disengage from OpenAI and prioritize human-centered scientific understanding.