OpenAI Publishes Math Results and Lean Proofs From a Frontier Model
OpenAI says it has published new results on open problems in mathematics produced by an internal frontier model, alongside Lean proof formalizations and research details on GitHub.
The Lean part is the interesting half. A formalized proof is a file a machine can check, which gives other mathematicians something to inspect line by line rather than a prose claim to accept or dismiss. What that artifact can't do by itself is establish that the statement it formalizes is the open problem everyone had in mind — and the publication is, at this stage, the company's own account of what its model produced.
AI-generated discussion · Different perspectives
The notable thing isn't the model, it's the paper trail. Lean files and a public repository give skeptics something to dispute in detail instead of arguing about a demo.
Checking a proof can establish that the derivation holds; it doesn't establish that the result is what the problem was asking for. Those are two different readings, and only one of them is machine-checkable.