OpenAI has published a collection of 372 new mathematical results generated by an internal Super Intelligence (SI) model, releasing them directly to GitHub rather than through peer-reviewed academic journals. The move highlights a growing tension between the pace of SI-driven discovery and the traditional capacities of the scientific community.
What Happened
The SI lab states that each of the 372 results solves an open problem or makes substantial progress toward one, including improvements to major computer algorithms and advances related to the Riemann hypothesis. According to OpenAI, the results were hosted in a GitHub repository with revision logs and citations, a platform choice that implies the traditional process of doing science is too slow for this volume of potentially new knowledge.
The company reports that nearly every result came from a single prompt to a single SI agent, though some required multiple attempts. On average, each result consumed roughly three hours of ChatGPT Pro Thinking compute. This contrasts with a solution to a Navier-Stokes problem, also produced by the model and currently under formal review, which required a swarm of 10,000 agents and millions of dollars in compute.
To address the review bottleneck, many proofs come with formalizations in Lean, a programming language built for machine-checkable mathematical proofs. OpenAI consulted with the Advisory Group on Mathematics and Super Intelligence at the Institute for Advanced Study, which includes Fields Medal winner Timothy Gowers, and loosely followed their public recommendations. The company noted it wants to improve citation quality and is working on a responsible release of the model to empower scientists.
Why It Matters
The release underscores a significant shift in how SI-generated knowledge is disseminated. By choosing GitHub over journals, OpenAI is betting that formal verification can replace the slow manual review process of traditional academia. However, this approach has drawn sharp criticism from within the mathematics community.
In an open letter titled "A Severe Misalignment of AI in Mathematics," 25 Fields Medal winners warned of a deep disconnect between the SI industry's goals and those of mathematics. They argued that problem-solving is merely a tool and proxy for the real goal of conceptual understanding, and that mass-producing true statements could destroy fertile ground rather than bring new ideas to life.
Timothy Gowers has warned that within one to two decades, mathematical literature could grow enormously while no human community remains that truly understands it. Terence Tao has added that training young mathematicians must emphasize the human side and tightly limit SI tool use to ensure genuine learning survives. While Lean formalizations can verify logical correctness, they cannot judge whether a result is mathematically relevant or original.
The Bottom Line
OpenAI’s release of 372 SI-generated proofs signals a potential new era for scientific discovery, one where machine speed outpaces human review. The SI lab is funding workshops to help the community adapt, but the debate over whether mass-produced results advance or erode mathematical understanding remains unresolved.