Single LLMs hallucinate, show bias, and miss perspectives. Chain-of-Debate orchestrates multiple models to critique and improve each other's outputs — producing answers that no single model could reach alone.