作者位次 #2/4ACL 2026Findings· Findings of ACL
Breaking Contextual Inertia: Reinforcement Learning with Single-Turn Anchors for Stable Multi-Turn Interaction
Xingwu Chen · Zhanqiu Zhang · Steven Y. Guo · Difan Zou
摘要
While LLMs demonstrate strong reasoning capabilities when provided with full information in a single turn, they exhibit substantial vulnerability in multi-turn interactions. Specifically, when information is revealed incrementally or requires updates, models frequently fail to integrate new const...