OpenAI claims it solved an 80-year-old math problem — for real this time

OpenAI claims its caller reasoning exemplary has produced an archetypal mathematical impervious disproving a celebrated unsolved conjecture successful geometry, which was archetypal posed by Paul Erdős successful 1946.

If this sounds acquainted to you, it’s due to the fact that this isn’t the archetypal clip OpenAI has made specified a bold claim. Seven months ago, the AI giant’s former VP Kevil Weil posted connected X: “GPT-5 recovered solutions to 10 (!) antecedently unsolved Erdős problems and made advancement connected 11 others.”

It turns out, GPT-5 didn’t really lick those problems; it conscionable recovered existing solutions that already existed successful the literature. 

Taunts from rivals similar Yann LeCun and Google DeepMind CEO Demis Hassabis followed, and Weil promptly took down his premature post. Today, astatine least, it seems OpenAI didn’t marque the aforesaid mistake twice. Alongside the announcement, OpenAI published companion remarks successful enactment of the disproof from mathematicians similar Noga Alon, Melanie Wood, and Thomas Bloom, who maintains the Erdos Problems website, and antecedently called Weil’s station “a melodramatic misrepresentation.” 

“For astir 80 years, mathematicians believed the champion imaginable solutions looked astir similar quadrate grids,” OpenAI posted connected X. “An OpenAI exemplary has present disproved that belief, discovering an wholly caller household of constructions that performs better.”

The institution said this marks “the archetypal clip AI has autonomously solved a salient unfastened occupation cardinal to a tract of mathematics.” The proof, per OpenAI, came from a caller general-purpose reasoning model, not a strategy specifically designed to lick mathematics problems oregon adjacent this occupation successful particular. 

OpenAI says this is important due to the fact that it means AI systems are present much susceptible of holding unneurotic long, hard chains of reasoning and connecting ideas crossed fields successful ways researchers whitethorn not person antecedently explored. That has implications for biology, physics, engineering, and medicine.

“AI is helping america to much afloat research the cathedral of mathematics we person built implicit the centuries,” Bloom said successful a statement. “What different unseen wonders are waiting successful the wings?”