OpenAI put together an advisory group on mathematics and AI, with an eye toward scoring open problems and making progress that can actually be checked. The interesting bit is less the press release and more whether rigorous grading sticks when models start claiming theorem-level wins.