emergency-medicine

HealthCraft is a reinforcement learning environment designed to evaluate LLM safety in emergency medicine, revealing that frontier models like Claude Opus 4.6 and GPT-5.4 perform poorly on real-world multi-step tasks, with high safety-failure rates.

This study evaluates how interactive dialogue with an LLM (via the MedSyn system) improves diagnostic accuracy for physicians in emergency care settings, showing significant gains for residents on difficult cases.