Working with academic communities to build scientific understanding
OpenAI announced three new research findings in mathematics, computer science, and physics, releasing them with varying levels of external collaboration and transparency to foster academic understanding.
OpenAI unveiled three discoveries—one each in mathematics, computer science, and physics—presenting them as raw research data rather than finalized findings. The company invited external experts to guide how these results should be released, with three distinct strategies: full public release of AI output for mathematics, collaborative explanation with public access for computer science, and deferral to an external group for physics. The approach aims to convert raw AI-generated data into scientific understanding through community-driven efforts.
The mathematics result addresses the local smoothing conjecture, first proposed by Sogge in 1991, which concerns wave behavior under extreme focusing. OpenAI’s models produced a proof resolving the conjecture in all dimensions, implying related conjectures such as the Kakeya conjecture and Stein’s restriction conjecture. The AI output was formalized in Lean and released publicly, with no external collaboration required for explanation. The computer science result refutes the Exponential Time Hypothesis (ETH), a foundational conjecture in computational complexity, which has been widely used to derive hardness results across multiple fields.
A collaboration led by Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams will verify and explain the ETH refutation. The group includes the original proposers of the ETH and experts in complexity theory. They requested full public access to raw data for independent verification and proposed hosting a conference to explore the implications of AI-generated breakthroughs on research practices. The AI disproof was formalized in Lean shortly after its discovery in October 2026.
The physics result involves a collaboration that will write up the AI-generated findings independently, withholding immediate release of artifacts. OpenAI emphasizes transparency, voluntary participation, and no academic credit for the discoveries, while turning over materials to the collaborations. The process invites feedback to improve inclusivity and effectiveness, acknowledging tradeoffs in initial shepherd selection and information asymmetry.