Pure LLM Prompting Works

Argues that a chain-of-debate structured purely as LLM prompts—without external tool calls—can deliver strong reasoning performance in medical diagnosis.