Full Breakdown
OpenAI Claims AI-Generated Proofs in Mathematics, Computer Science and Physics
By Drooid · · How we work
Core Event
OpenAI announced that its internal AI models produced three research results: a proof of the local-smoothing conjecture in harmonic analysis, a disproof of the Exponential Time Hypothesis (ETH) in computational complexity, and a mathematically rigorous existence proof for quantum chromodynamics (QCD) in quantum field theory. The outputs are presented as raw data that still require independent verification and interpretation.
Background & Context
The announcement follows OpenAI’s earlier “October 6, 2026” release of more than 700 mathematical manuscripts, in which the firm also made model-generated proofs publicly available on GitHub. In the current rollout, OpenAI adopted three distinct release strategies, consulting external “shepherd” groups to decide how each result would be shared. The mathematics and computer-science outputs were released immediately, while the physics result will be published after an external collaboration prepares its own write-up.
Timeline
- October 6, 2026 – OpenAI published a batch of AI-generated mathematical manuscripts and released both natural-language write-ups and Lean formalizations on GitHub.
- October 16, 2026 – The AI produced a natural-language disproof of ETH; the proof was formalized in Lean on October 17, 2026.
- October 18, 2026 – The local-smoothing proof was obtained; it was formalized in Lean the following day.
- October 26, 2026 – OpenAI announced the three results, described the differing release approaches, and invited feedback on the process.
Data & Statistics
- The local-smoothing proof resolves a conjecture first posed by Christopher Sogge in 1991 and, according to OpenAI, simultaneously settles the Kakeya and Stein restriction conjectures in all spatial dimensions.
- The ETH disproof claims a deterministic algorithm that counts satisfying assignments of fixed-width k-CNF formulas in sub-exponential time and polynomial space, with sharper bounds for formulas with *m* clauses (as documented in a commit dated October 26, 2026).
- OpenAI’s “shepherd” groups are led by Russell Impagliazzo, Ramamohan Paturi (UC San Diego) and Ryan Williams (MIT).
- The company noted a valuation of $852 billion and a prior $5 million contribution to the Simons Foundation to support workshops on AI-generated research.
Official Statements & Responses
OpenAI emphasized that the three approaches involve trade-offs: immediate public release enables unrestricted community examination, while staged releases give expert groups time to craft contextual explanations. OpenAI also pledged to fund a conference on the computer-science result and to turn over all raw data and artifacts to the collaborating groups.
Criticism & Opposition
Mathematician Terence Tao, a Fields Medalist at UCLA, described the mass release of AI-generated proofs as “toxic waste,” arguing that the outputs lack elegance and explanatory depth. Tao warned that the practice could pressure the community to prioritize benchmark-driven AI training over traditional scholarly values.
Verbatim Quotes
- “The conjecture was formulated by my PhD adviser, Chris Sogge, so it is near and dear to my heart, and the news was a shock at first,” — Alex Iosevich, University of Rochester
- “All of these approaches involve tradeoffs,” — An OpenAI representative