LLMs in Medical Decision Making: Treatment vs. Chat

A new paper on arXiv (2506.23040) explores the potential of large language models (LLMs) to aid in medical decision making. The authors distinguish between the 'treatment problem'—the core patient task solved with a clinician—and the 'chat problem,' noting that imitation alone cannot solve the true treatment problem. They discuss approaches using experimental and observational data within evidence-based medicine, and highlight challenges in training LLM-based systems, particularly regarding ethics of experimentation and observational assumptions.