Researchers used a technique discovered by an OpenAI AI model to solve a mathematical problem that had remained open for 50 years [1].
This breakthrough demonstrates that artificial intelligence can move beyond simple computation to discover novel proof techniques. Such capabilities could allow mathematicians to resolve other longstanding conjectures that have resisted human effort for decades.
The discovery centers on a new approach to mathematical proofs. Reports said an internal OpenAI model first uncovered a technique that allowed researchers to crack the 50-year-old problem [1]. Other reports said the AI found a counterexample to an Erdős conjecture that had persisted for 80 years [2].
These developments occur amid a broader trend of AI-driven mathematical discovery. Separate reports said that Google DeepMind's AlphaProof Nexus solved nine Erdős problems [3]. Those specific problems had reportedly remained unsolved for 56 years [3].
The discrepancy between reports highlights a competitive landscape in AI research. While some sources credit OpenAI with the discovery of the core technique used for the 50-year-old problem [1], others focus on the volume of problems solved by Google DeepMind's systems [3].
Mathematicians are now exploring how to adapt these AI-discovered methods to other fields of study. The ability of a model to generate a counterexample or a novel proof structure suggests a shift in how theoretical mathematics is conducted, moving from human-led intuition to AI-assisted verification.
“AI can move beyond simple computation to discover novel proof techniques.”
The emergence of AI-discovered proof techniques marks a transition from using AI as a calculator to using it as a theoretical collaborator. By finding counterexamples to conjectures that stood for 80 years, AI is proving it can identify logical gaps that human intuition missed, potentially accelerating the resolution of the most difficult problems in mathematics.
