MAIN ARTICLE
A release that changed the scale of the conversation
On October 6, OpenAI announced an online collection of mathematical manuscripts produced by an unreleased internal system. The repository is continually updated. The catalogue cited in Le Monde contained 719 manuscripts in 372 families, whereas other snapshots listed 722 manuscripts. Neither figure should be mistaken for an equal number of independently established theorems; a family can contain a main result alongside companion arguments or alternative proofs.
The papers range across mathematical analysis, geometry, logic and theoretical computer science. Some come with Lean formalizations that allow computers to check the structure of a proof. Others have no such verification yet, and OpenAI acknowledges that unformalized results may contain errors. A striking volume of output therefore presents two separate stories: a genuine technological advance and a substantial validation burden for academic specialists.
For mathematicians, a research topic is also a life’s work
According to Le Monde, Fields Medalist Hugo Duminil-Copin spoke at an October 8 video event hosted by the Société mathématique de France about the personal shock of seeing 16 questions connected to his research among the release. His distress captures something numbers cannot: a field of research is not merely a collection of answers but years of intuition, collaboration and intellectual identity.















