揭示对话历史如何像几何陷阱一样困住大模型的生成行为
Old Habits Die Hard: How Conversational History Geometrically Traps LLMs
- 将对话历史建模为马尔可夫链，量化状态一致性
- 发现隐层表示轨迹在潜在空间中存在显著间隙
- 首次揭示行为持续性源于潜在空间的几何约束
大语言模型的对话历史如何影响其后续表现？近期研究发现，先前交互中的幻觉可能影响后续响应。本文提出History-Echoes框架，从概率和几何双视角分析对话历史对生成的偏差影响。概率上将对话建模为马尔可夫链以量化状态一致性；几何上测量连续隐藏表示的一致性。在三个模型家族和六个数据集上的分析显示两者存在强相关性。通过融合视角，揭示出行为持续性表现为一种几何陷阱：潜在空间中的间隙限制了模型轨迹的演化。代码已开源。
原文摘要 · Abstract (English)
How does the conversational past of large language models (LLMs) influence their future performance? Recent work suggests that LLMs are affected by their conversational history in unexpected ways. For instance, hallucinations in prior interactions may influence subsequent model responses. In this work, we introduce History-Echoes, a framework that investigates how conversational history biases subsequent generations. The framework explores this bias from two perspectives: probabilistically, we model conversations as Markov chains to quantify state consistency; geometrically, we measure the consistency of consecutive hidden representations. Across three model families and six datasets spanning diverse phenomena, our analysis reveals a strong correlation between the two perspectives. By bridging these perspectives, we demonstrate that behavioral persistence manifests as a geometric trap, where gaps in the latent space confine the model's trajectory. Code available at https://github.com/technion-cs-nlp/OldHabitsDieHard.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。