引用hbbio · 2025-05
“This goes well with Microsoft paper "LLMs Get Lost In Multi-Turn Conversation " that was published Friday [1].”
指出与微软这篇论文吻合:LLM 在多轮对话中迷失
多轮对话下主流 LLM 平均性能下降 39%,走错一步就难以恢复
论文用大规模模拟实验比较 LLM 在单轮与多轮对话中的表现。测试的各大开源和闭源顶级模型在六类生成任务上,多轮表现平均下降 39%。对 20 万以上模拟对话的分析把退化拆成两部分:能力(aptitude)小幅下降,可靠性(reliability)大幅变差。LLM 常在早期轮次做出假设并过早给出最终方案,之后过度依赖它:一旦走错,就会迷失而无法恢复。
评论者引用它作为多轮对话效果更差的证据:逐步补充信息不如发现缺失后用修订后的提示词重新开始。
“This goes well with Microsoft paper "LLMs Get Lost In Multi-Turn Conversation " that was published Friday [1].”
“Microsoft/Salesforce found 39% accuracy drop in multi-turn conversations”
“There a recent(ish: May 2025) paper about how drip-feeding information is worse than restarting with a revised prompt once you realize details are missing.”