Summary
OpenAI has released hundreds of AI-generated mathematical results, a collection of 722 papers, and this Two Minute Papers episode looks at what the work actually says rather than the surrounding controversy. Dr. Károly Zsolnai-Fehér highlights two results: a claimed proof that almost every ray in a triangular mirror room with an irrational angle visits every spot equally, and a claimed proof that a conformal transformation can always morph a region full of holes into perfect circles or points.
The video then turns to a debate dividing mathematicians. The proofs are described as far from elegant, and often need another AI to interpret. Some researchers just want correct answers, while others see mathematics as a way to gain understanding, where each new proof teaches something different. The presenter says he sympathizes with both camps.
He also traces how quickly AI math ability has progressed, from struggling with addition with GPT-3 to Olympiad gold medals and a reported solution to a Navier-Stokes millennium problem. Each result reportedly took about three hours of compute on average. Since the model is available only to a handful of people at frontier labs, the video closes by arguing for open-weights models that anyone can run, own, and contribute to.
📺 Source: Two Minute Papers · Published October 13, 2026
🏷️ Format: News Analysis
