GPT-6.2 Astra formalizes Smoktunowicz’s gap theorem in Lean
An AI model named GPT-6.2 Astra successfully formalized a 2006 theorem by Agata Smoktunowicz regarding the growth of graded algebras using the Lean proof assistant. Operating within Autoform-Hard constraints, the model found a novel proof route avoiding three deep algebraic results absent from Mathlib by utilizing a capacity argument. The run cost $21.58 and took 94 minutes. Open article →