In mid‑May OpenAI revealed that an internal large‑language model had disproved the Erdős unit‑distance conjecture, a discrete‑geometry problem that has resisted proof for eight decades. The breakthrough reached a handful of mathematicians, including Fields Medalist Tim Gowers, who called the result a milestone for AI‑driven mathematics. University of Toronto’s Daniel Litt praised it as the first autonomous AI theorem that excites on its own merit.

The AI’s proof stitched together known techniques from several mathematical subfields rather than inventing brand‑new methods. Human collaborators subsequently cleaned up the argument and extended it, demonstrating that AI can supply a full proof while humans refine and verify it. This mirrors a recent trend where language models progressed from simple arithmetic to solving in high‑school contest problems.

While the result still does not overturn the long‑term trajectory of mathematical research, it signals a shift toward human‑AI collaboration on deep problems. AI’s encyclopedic recall and willingness to grind through combinatorial cases complement mathematicians’ intuition and creativity. As models improve, the balance of contribution may tilt, but for now the proof stands as a joint accomplishment.