OpenAI Claims AI Discoveries in Math, CS, and Physics Amid Academic Debate
OpenAI reports three major AI-generated breakthroughs in math, physics, and computer science while navigating criticism over its academic reviewer model.
OpenAI has announced three new scientific discoveries across computer science, physics, and mathematics, alongside an updated approach to collaborating with academic researchers. The announcement comes less than a month after the organization released hundreds of manuscripts containing proposed solutions to long-standing mathematical problems on October 6. Unlike previous releases, OpenAI is involving external academic experts—referred to as "shepherds"—to guide the evaluation and public presentation of these AI-generated outputs.
Computer science, physics, and mathematics breakthroughs
In computer science, OpenAI claims its model has produced a proof showing that the Exponential Time Hypothesis (ETH) is false. First proposed in 1999 by Russell Impagliazzo and Ramamohan Paturi, the hypothesis has been a foundational assumption in computational complexity theory. OpenAI’s material presents a deterministic algorithm designed to exactly count satisfying assignments of fixed-width CNF formulas in subexponential time and polynomial space, specifically refuting the hypothesis in the case of 3-SAT. To examine this claim, an independent working group led by Impagliazzo, Paturi, and MIT professor Ryan Williams requested immediate public release of the write-up and a Lean-formalized proof on GitHub. The group also published an open letter inviting the community to deliberate on norms for AI-driven discoveries, including credit allocation, conflict disclosures, and early access protocols.
In mathematics, OpenAI published model output on GitHub addressing local smoothing for the Euclidean wave equation. The proof utilizes an auxiliary partial differential equation to establish Sogge’s local smoothing conjecture for waves across all spatial dimensions of two or greater. According to the release, the proof also yields the Kakeya maximal and dimension conjectures, as well as Stein’s spherical restriction estimates throughout the interior of the conjectured range. The result expands on recent milestone achievements in the field, such as NYU professor Hong Wang’s Fields Medal-winning resolution of the Kakeya conjecture in three dimensions.
In physics, OpenAI reported a mathematically rigorous existence proof regarding certain interacting quantum field theories containing gauge fields and fermions in four dimensions. This includes Quantum Chromodynamics (QCD) and a unified completion of the Standard Model of particle physics. Although the full physics documentation has not been publicly released yet, David Gross will lead an expert evaluation group including Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov to shepherd the work.
Shepherding and the academic debate over AI releases
OpenAI’s decision to assign selective academic "shepherds" to evaluate results before full public release has sparked controversy within the research community. A day prior to the announcement, OpenAI’s AGMAI panel of math advisers issued a statement saying "we recommend that AI companies do not adopt" the practice of designating early reviewers, citing concerns regarding unequal access, borrowed reputation, and biased selection.
Instead of private early access, the advisory panel recommended that AI laboratories publish all outputs directly to open scholarly repositories. They urged companies to include Lean-formalized proofs, documentation on how problems were chosen, and dedicated funding for independent researchers working to verify, explain, and contextualize the outputs. OpenAI acknowledged these procedural challenges, noting that its decision to publish the math output directly mirrors its earlier October release strategy.
What it means for developers
For software engineers and system architects, OpenAI's latest releases highlight the expanding role of AI in algorithm design and computational complexity theory. If the refutation of the Exponential Time Hypothesis holds up under community scrutiny, it could alter theoretical assumptions around NP-complete problems and boolean satisfiability (SAT) solvers, which underpin logic optimization engines and verification tools used in production environments.
Furthermore, the heavy emphasis on formal verification tools like Lean marks a shift toward machine-checkable software engineering and automated theorem proving. Rather than relying solely on raw probabilistic model output, developers can see how combining large language models with formal proof assistants creates reproducible, verifiable logic. As AI models increasingly assist with complex logic and algorithmic research, developers looking to test or integrate top AI models cheaply through one API can visit https://apixoai.online.
As academic communities establish formal guidelines around credit, verification, and early disclosures, developers can expect more structured workflows where AI models generate raw algorithmic strategies while automated provers and human domain experts perform rigorous audit passes.
Source: OpenAI announces three big science discoveries and a new way to work with academics — The Verge AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
