Did ChatGPT just solve an 80-year-old maths problem? OpenAI thinks so
OpenAI announced one of its AI models disproved an 80-year-old geometry conjecture by Paul Erdős, a breakthrough backed by mathematicians like Noga Alon and Thomas Bloom. The discovery, achieved by a general-purpose reasoning model rather than specialized software, suggests AI may now handle complex logical chains across disciplines, with potential impacts on fields like biology, physics, and medicine.
OpenAI claims a new AI model has independently disproved the *planar unit distance problem*, a geometry conjecture first proposed by mathematician Paul Erdős in 1946. The problem, unresolved for nearly eight decades, had long been assumed to follow square-grid solutions, but the AI discovered a different construction outperforming those assumptions, effectively disproving the conjecture. The announcement follows an earlier controversy in 2026, when OpenAI’s GPT-5 was accused of rediscovering existing solutions to Erdős problems, leading to public backlash from researchers like Yann LeCun. This time, OpenAI included endorsements from mathematicians, including Thomas Bloom, who previously criticized the company’s claims, lending credibility to the breakthrough. The result marks a potential shift in AI capabilities, as the proof was generated by a general-purpose reasoning model rather than specialized software. OpenAI suggests this indicates AI systems may now sustain complex logical chains, connect abstract ideas, and explore unconventional research paths. The company highlights potential applications beyond mathematics, including biology, physics, engineering, and medicine. Mathematicians involved, such as Noga Alon and Melanie Wood, provided supporting commentary alongside OpenAI’s announcement. While the company emphasizes human expertise remains critical for interpreting results, the breakthrough underscores AI’s evolving role in scientific discovery. OpenAI’s tweet stated the discovery ‘points to something larger: AI systems are becoming capable of holding together long, difficult chains of reasoning.’ The announcement was made May 20, 2026, with the company stressing AI can assist in searching, suggesting, and verifying research while humans guide its direction.
This content was automatically generated and/or translated by AI. It may contain inaccuracies. Please refer to the original sources for verification.