OpenAI claims it solved an 80-year-old math problem — for real this time