Particle.news
Download on the App Store

OpenAI Model Produces Proof That Challenges Erdős Unit‑Distance Conjecture

The model-generated argument, now posted by OpenAI with external verification, shows point configurations grow faster than previously believed and raises questions about how AI will shape research.

Overview

  • OpenAI released a model-derived proof on Wednesday that reports constructions of n points in the plane with at least n^{1+δ} unit-distance pairs, contradicting the long‑standing conjecture that the count grows only nearly linearly.
  • The claimed result gives a fixed positive exponent δ and holds for infinitely many values of n, which changes the expected growth rate for the planar unit distance problem first posed by Paul Erdős.
  • OpenAI says the argument originated from a general-purpose internal reasoning model rather than a math‑specialized system and uses tools from algebraic number theory to build the geometric constructions.
  • Researchers named by OpenAI — including Mark Sellke, Mehtaab Sawhney, and Lijie Chen — converted the model output into a formal proof, external mathematicians have reportedly checked the work, and OpenAI published a public discussion titled “The Erdős Breakthrough.”
  • The episode highlights a new workflow for human‑AI collaboration in pure math and points to wider effects if the result holds under formal peer review, including faster discovery across fields and growing demand for rigorous verification systems.