OpenAI has announced three major scientific findings across mathematics, computer science, and physics, while introducing a framework to collaborate with external researchers to analyze the output. The announcement comes less than a month after the company published hundreds of manuscripts with potential solutions to long-standing mathematical problems.
New Discoveries Across Multiple Disciplines
In computer science, OpenAI reported a proof refuting the Exponential Time Hypothesis (ETH), a long-held belief in computational complexity. Verification of the result is being overseen by a panel including MIT professor Ryan Williams alongside Russell Impagliazzo and Ramamohan Paturi from UC San Diego, who originally proposed the hypothesis in 1999. The computer science team published the model’s write-up and a Lean-formalized proof on GitHub for public evaluation.
For the physics discovery, the company presented a mathematically rigorous existence proof regarding interacting quantum field theories in four dimensions, covering Quantum Chromodynamics and a unified completion of the Standard Model. David Gross will lead an expert team, including Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov, to examine the work, though full details have not yet been fully released.
In mathematics, the AI model generated a proof for the local smoothing conjecture regarding the Euclidean wave equation across all dimensions of two or greater. OpenAI stated this result also resolves both the Kakeya conjecture and Stein’s restriction conjecture in all dimensions. Notably, NYU professor Hong Wang was recently awarded a Fields Medal for solving the Kakeya conjecture specifically in three dimensions.
Academic Oversight and Community Debate
To evaluate these outputs, OpenAI is utilizing external academic experts, whom the company describes as “shepherds.” However, this advisory process has drawn criticism. The AGMAI panel of math advisers stated that “we recommend that AI companies do not adopt” the practice of providing select individuals with early access to AI research, pointing to issues such as unequal access, biased selection, and borrowed reputation. Instead, the panel advocated for releasing outputs directly to public scholarly repositories alongside Lean proofs, methodological disclosures, and research funding.
In response to broader questions about AI’s role in scientific research, the computer science working group published an open letter asking the academic community to help establish standards. The letter raised key questions regarding expectations for early access, how to fairly assign credit between initial discoveries and verification work, and methods to support researchers as AI tools transform scientific methodology.
