OpenAI says an internal model autonomously cracked an 80-year-old math problem, raising new questions about AI's research ...