Overview
- OpenAI announced Wednesday that an internal general-purpose reasoning model autonomously produced a new infinite family of point configurations that contradicts the decades-old Erdős conjecture about unit distances in the plane.
- Princeton mathematician Will Sawin and others quantified the improvement at about n^(1+0.014), meaning the number of unit-distance pairs grows slightly faster than linear for infinitely many n.
- The proof has been formalized and checked in the Lean proof assistant and reviewed publicly by prominent mathematicians including Tim Gowers and Will Sawin, giving the result strong machine‑checkable and expert validation.
- The model’s argument links planar geometry to deep algebraic number theory—using tools such as infinite class field towers and Golod–Shafarevich ideas—which produced a construction human researchers had not previously found.
- The announcement raises practical and epistemic questions: OpenAI’s prior overstated Erdős claims increase scrutiny, formal peer review and community vetting are still pending, and the result could affect formal verification and cryptography if the techniques generalize.