It's the 'conversations' of some mathematicians with OpenAI. So the question is: did anyone feed the conversation(s) back as training data.