Where Did We Go Wrong? Localizing Errors to Repair Multi-Turn Conversations

Alessandro Morosini, Wei Sun, Georgia Perakis

Large language model (LLM) assistants are increasingly used in multi-turn conversations, where the user's intent is often unclear at first and revealed only gradually over time. Th…