[r/MachineLearning] OpenAI claims a general-purpose reasoning model found a counterexample to Erdos's unit-distance bound [D]