An OpenAI reasoning model disproved the Erdős unit distance conjecture, an open problem that had stood for roughly 80 years. Researchers Alexander Wei, Hongxun Wu, and Lijie Chen describe the moment the result became credible, what the verification process actually required, and why a general-purpose model, not a specialized math tool, produced the breakthrough.
The detail worth reading for is not the conclusion but the process: how the researchers checked the proof, what it revealed about the gap between generating a result and trusting it, and why general models matter more for discovery than narrow ones. The discussion of creativity versus tool use starting around the 15-minute mark is the sharpest part.
What comes next is the open question. The researchers address how mathematicians should be rethinking their workflows now, not after the next result lands. The conversation closes on cryptography and quantum computing, signaling where they think AI-assisted mathematical reasoning is heading next.
[WATCH ON YOUTUBE →]