Researchers at Harvard Medical School found that a new open-source artificial intelligence tool is diagnosing patients as accurately as leading proprietary models — like OpenAI’s GPT-4 — for the first time.