Another breakthrough:

OpenAI revealed that an internal model, running with minimal human supervision, tackled the “First Proof” challenge and produced promising solutions to most of the ten frontier research problems, with at least six believed likely correct.

The results came from a one-week side sprint, highlighting how next-generation models are beginning to engage directly with unsolved, expert-level mathematics on their own.

https://cdn.openai.com/pdf...

aipost 🏴