What changed
OpenAI has released three significant results from its internal AI models, spanning mathematics, computer science, and physics. The company claims these results include disproving a central assumption in computer science and proving part of one of the Millennium Prize Problems. OpenAI has now invited outside experts, including a Nobel laureate, to review and validate these findings.
What this means in practice
The results include a claim that the Exponential Time Hypothesis (ETH) in computer science is false. If correct, this would challenge the foundation of many computational complexity theories and could lead to new algorithmic approaches. OpenAI has also produced a proof for the local smoothing conjecture in all dimensions, which would automatically solve related conjectures in geometry and Fourier analysis. Additionally, the AI has claimed to prove the existence of certain quantum field theories in four dimensions, potentially advancing the mathematical foundation of particle physics.
Limitations and availability
OpenAI has made all raw data, including written proofs and Lean proofs, publicly available for review. However, the company acknowledges that Lean only checks whether the proof aligns with its internal logic, and experts must still determine if the statement accurately represents the intended problem. No dedicated group is currently overseeing the verification of the math result, and OpenAI has opted to publish the proof directly for researchers to evaluate independently. The practical implications of these findings remain uncertain, with experts cautioning that significant research would be needed to translate theoretical results into real-world applications.
