Some Lessons from the OpenAI FrontierMath Debacle