Could artificial intelligence solve theoretical problems that have resisted the world's leading researchers for decades? OpenAI claims its models have achieved breakthrough results across mathematics, theoretical computer science, and quantum physics, publishing automated proofs and algorithms on GitHub. Yet the true test of these assertions is only beginning, as outside specialists face the demanding task of scrutinizing machine-generated solutions and determining how much they genuinely advance human knowledge.
In computer science, OpenAI published a paper claiming to refute the Exponential Time Hypothesis, a foundational conjecture regarding computational limits that has stood since 1999. To assess the result, an independent working group led by MIT professor Ryan Williams alongside Russell Impagliazzo and Ramamohan Paturi—two of the theorists who originally formulated the hypothesis—has agreed to examine the automated proof and counting algorithm. In mathematics, the company posted model outputs claiming to establish Sogge’s local smoothing conjecture in all spatial dimensions, along with related problems in geometric measure theory where partial progress earned NYU professor Hong Wang a Fields Medal earlier this year.
The company took a different approach in physics, where its unreleased work claims an existence proof for interacting quantum field theories and a completion of the Standard Model. Nobel laureate David Gross will lead a group of prominent physicists to shepherd and evaluate the findings. However, this strategy of appointing hand-picked experts has already sparked criticism. An advisory panel formed by OpenAI warned against giving selected academics privileged early access, arguing that the practice risks borrowed reputation, unequal access, and biased evaluation. The panel recommended that laboratories instead publish code and formalized proofs directly to open repositories, accompanied by funding for independent scrutiny.
The disclosures have ignited an urgent debate over how science should manage machine intelligence. In an open letter accompanying the computer science paper, researchers asked how the academic community should divide credit between AI generation and the painstaking human effort needed to audit, fix, and explain raw automated proofs. As scholars begin parsing thousands of lines of machine reasoning, the lasting significance of OpenAI's claims will depend not on the algorithms that produced them, but on whether human scientists can independently validate, understand, and build upon the work.