OpenAI General-Purpose LLM Solves Erdős Problem Open Since 1946
OpenAI researcher Noam Brown confirmed that a general-purpose language model — untargeted at mathematics and operating without a specialized scaffold — solved the Erdős planar unit distance problem, an open conjecture since 1946. Brown explicitly stated: the model was not targeted at this problem, is not a scaffold, and OpenAI has not yet pushed it to its limits on open problems. The priority is making the model broadly available. The result is independent of DeepMind's concurrent AlphaProof Nexus work.
Why It Matters
A general-purpose LLM — without domain-specific architecture — resolving an 80-year unsolved problem is a fundamentally different capability signal than specialized theorem provers. Two independent labs achieving comparable results on the same day marks a step change in AI mathematical reasoning.