OpenAI shares new science discoveries using external expert oversight
OpenAI has released three new research findings in mathematics, computer science, and physics generated by artificial intelligence.
2 min read·Tuesday, 27 October
OpenAI announces three new scientific discoveries generated by artificial intelligence.
OpenAI has announced three new scientific discoveries spanning mathematics, computer science, and physics. This release follows a previous distribution of hundreds of mathematics manuscripts and incorporates feedback from external researchers.
Overview of discoveries
The newly released work includes three distinct breakthroughs made public via GitHub. In mathematics, the model produced a proof addressing the local smoothing conjecture for the Euclidean wave equation across every spatial dimension. This output also addresses the Kakeya maximal and dimension conjectures, alongside Stein’s spherical restriction estimates.
The second breakthrough addresses the Exponential Time Hypothesis, or ETH — a core assumption in computer science regarding how long specific algorithmic problems take to solve. The model provided a deterministic algorithm that counts satisfying assignments of fixed-width CNF formulas in subexponential time and polynomial space, which refutes the hypothesis. An independent working group consisting of computer scientists Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams will examine and explain the proof.
The third discovery involves a mathematically rigorous existence proof of interacting quantum field theories containing gauge fields and fermions in four dimensions. This includes Quantum Chromodynamics and a unified completion of the Standard Model of particle physics. A group led by physicist David Gross, alongside Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov, will shepherd this work.
Release methods and academic response
For these releases, OpenAI utilized external academic experts to guide the process. For the computer science and physics topics, designated shepherds helped oversee the findings. For the mathematics problem, the raw model output was released directly for researchers to review.
However, these distribution practices face ongoing scrutiny. An advisory panel recently recommended that artificial intelligence laboratories avoid designating specific individuals for early access, pointing to concerns over unequal access and biased selection. Instead, the panel suggested utilizing scholarly repositories and formalized proof assistants to share findings with the broader research community.
The Ultimate 2026 Free AI API Engine: Top 10 Provider Blueprints & Setup GuideFree guide
Comments