Artificial Intelligence

OpenAI Reportedly Cracks 80-Year-Old Math Conjecture

Did OpenAI's AI achieve a true mathematical breakthrough or is it just an overhyped sleight of hand?

Published MAY 22

OpenAI Reportedly Cracks 80-Year-Old Math Conjecture

Story

Above: In this photo illustration, the OpenAI logo is reflected on the screen of a smartphone with the ChatGPT website displayed in Los Angeles, California on May 20. Image credit: Justin Sullivan/Getty Images

The Facts

OpenAI reported Thursday that its reasoning AI system has disproved a conjecture in combinatorial geometry first posed by Hungarian mathematician Paul Erdős in 1946, known as the planar unit distance problem, which asks how many pairs of points can be placed exactly one unit apart on a flat plane.
Openai
New Scientist
For nearly 80 years, mathematicians believed square-grid arrangements offered the best possible solution to the problem. The AI system identified an infinite family of point configurations that outperform those classical constructions, with Princeton mathematician Will Sawin later refining the result to show a fixed improvement exponent of 0.014.
Openai
Interesting Engineering
The proof drew on algebraic number theory, using tools such as infinite class field towers and Golod-Shafarevich theory to construct high-dimensional lattices with special symmetries, which were then mapped down to two dimensions.
Openai
Scientific American
OpenAI said the proof was generated by a general-purpose reasoning system rather than a geometry-specific theorem prover. Engineers also reportedly avoided creating specialized search procedures for the problem or fine-tuning the system solely on earlier work related to it.
MIT Sloan School of Management
New Scientist
The announcement drew comparisons to a prior OpenAI claim, made seven months earlier, in which a former company executive said GPT-5 had solved 10 previously unsolved Erdős problems. Mathematicians later determined the system had reproduced solutions already existing in the literature, and the post was deleted.
MIT Sloan School of Management
TechCrunch
This time, OpenAI released a companion paper co-authored by external mathematicians, including Thomas Bloom, who had previously criticized the earlier GPT-5 claims. Bloom wrote that while the AI proof was valid, it "was significantly improved by the human researchers at OpenAI and the many other mathematicians involved in the present paper."
New York Post
guardian

Sources Split

LEFT

CENTER

RIGHT

ESTAB
CRIT

CENTER

PRO
ESTAB

The Spin

Narrative A

OpenAI's breakthrough on an 80-year-old math problem marks a turning point for artificial intelligence research. Instead of relying on a narrowly trained theorem solver, a general reasoning system independently discovered a credible proof strategy. The achievement shows AI systems are beginning to handle abstract reasoning, once considered uniquely human. This is real proof that AI is unlocking new frontiers in mathematical discovery, science and education.

Haider. on X

Shakthi Vadakkepat on X

Openai on X

Narrative B

The OpenAI math result is being oversold — mathematicians had to extract and rewrite the proof from a long, messy transcript, and the system only succeeded by grinding through paths humans found too tedious to bother with. There's no denominator showing how many failed attempts preceded this one. Without independent peer review, generalizing this narrow win into broad AI brilliance is a stretch that serves marketing more than science.

Gary Marcus on Substack

Cynical narrative

OpenAI's reported breakthrough is less about mathematics than the emergence of AI-driven creative reasoning. The system reportedly connected algebraic number theory with discrete geometry in ways human specialists had not explored for decades, signaling a shift from narrow computation to cross-domain synthesis. For businesses, the implications are profound — competitive advantage may soon depend less on isolated expertise and more on deploying AI systems that can discover novel links across disciplines faster than institutions can govern or verify them.

Aiona Edge on X