“OpenAI's latest models are outperforming the human doctors who helped train them at the specific task of evaluating AI-generated medical outputs.”