Mitigating Conversational Inertia in Multi-Turn Agents

ArXi:2602.03664v3 Announce Type: replace-cross Large language models excel as few-shot learners when provided with appropriate nstrations, yet this strength becomes problematic in multiturn agent scenarios, where LLMs erroneously mimic their own previous responses as few-shot examples. Through attention analysis, we identify conversational inertia, a phenomenon where models exhibit strong diagonal attention to previous responses, which is associated with imitation bias that constrains exploration.