An OpenAI model has disproved a central conjecture in discrete geometry

OpenAI Model Solves 80-Year-Old Erdős Math Problem

OpenAI has announced that one of its internal models has resolved a central, decades-old question in discrete geometry. The model disproved a longstanding conjecture related to the planar unit distance problem, first posed by mathematician Paul Erdős in 1946. This marks a significant milestone, as it appears to be the first instance of an AI autonomously solving a prominent open problem that has stumped human mathematicians for nearly 80 years, with the proof being externally verified by leading experts in the field.

An Unexpected Proof from Number Theory

The conjecture, widely believed to be true, proposed an upper limit on the number of unit-distance pairs in a set of points, growing just slightly faster than linearly (n¹⁺ᵒ⁽¹⁾). The OpenAI model defied this by constructing a family of examples with a significantly higher polynomial growth rate (n¹⁺ᵟ). The method itself is as notable as the result. Instead of building on existing geometric approaches, the model's proof introduces sophisticated concepts from algebraic number theory, a different mathematical field, a connection that leading mathematicians have called surprising and elegant.

• Problem Solved: Erdős's planar unit distance problem (1946).
• Conjecture Disproved: The belief that the maximum number of unit-distance pairs was n¹⁺ᵒ⁽¹⁾.
• New Lower Bound: The AI proved the existence of point sets with at least n¹⁺ᵟ unit-distance pairs, a polynomial improvement.
• Key Technique: Application of advanced algebraic number theory, including concepts like infinite class field towers.

This achievement shifts the role of AI in scientific research from a helpful assistant to a potential collaborator capable of original ideation. According to number theorist Arul Shankar, the model demonstrated the ability to have 'original ingenious ideas, and then carrying them out to fruition.' The use of a general-purpose reasoning model, rather than a specialized system, suggests that advanced AI now possesses a depth of reasoning applicable to other complex, formal domains like physics, materials science, and engineering, potentially accelerating discovery across the board.

Strategic Takeaway: OpenAI's solution to the unit distance problem is less about mathematics and more about demonstrating a defensible moat in general-purpose reasoning; this result serves as a high-profile validation that its models can move beyond interpolation to genuine, verifiable innovation, setting a new competitive benchmark for AI's role in complex problem-solving.