OpenAI Praises Mathematicians' Work as a Possible Milestone for Recursive Self-Improvement

Key Info

OpenAI congratulated mathematicians Levent Alpöge and Tristan Buckmaster on their recent mathematical work, with some observers calling it the first successful demonstration of recursive self-improvement (RSI). OpenAI says it accessed no specific user data to solve the problem, though it cannot rule out that de-identified data from product usage helped improve its models.

Highlights

• OpenAI states that its researchers and agents never saw the mathematicians' work before it was released publicly, and no specific user data was accessed.
• The company acknowledges it cannot fully rule out that de-identified data derived from users' interactions with its products contributed to model improvements.
• The episode is described as a potential RSI loop: models help mathematicians solve hard problems, producing rare and difficult reasoning tokens that can then be used to train even better models.
• OpenAI notes that its proofs differ significantly from the mathematicians' results, rather than being a direct reproduction.