“OpenAI created a benchmark called HealthBench, trained with input from over 250 physicians, to measure an LLM's proficiency at answering medical questions.”