Well they (OpenAI) never made such a claim. And yes, LLMs have made unique solutions/contributions to a few erdos problems.