揭示大模型推理时的隐含规划能力，发现其只做短期规划。
How Far Ahead Do LLMs Plan? Uncovering the Latent Horizon in Chain-of-Thought Reasoning
- 用新方法探测隐藏状态，发现模型早期已预知部分未来推理。
- 模型主要进行局部渐进式推理，缺乏全局规划，规划范围有限。
- 可识别推理跳步行为，且不损失性能，适合提升模型可信度研究。
链式思维（CoT）已成为激发大语言模型多步推理的核心机制。然而近期研究发现矛盾：模型隐藏状态似乎已编码未来推理信息，但显式步骤对组合计算任务仍至关重要。为深入理解大模型内部状态与其口语化推理轨迹的关系，本文通过名为 Tele-Lens 的探测方法，分析跨多种任务领域的隐藏状态。实验结果表明，大模型具有短视的规划视野，主要进行增量式转移，缺乏精确的全局规划。基于此特征，我们提出一种增强 CoT 不确定性估计的假设，并验证仅需稀疏的枢轴位置即可有效代表整条路径的不确定性。进一步强调利用 CoT 动态的重要性，证明无需性能下降即可自动识别推理跳步。代码、数据与模型已开源于 https://github.com/lxucs/tele-lens。
原文摘要 · Abstract (English)
Chain-of-thought (CoT) reasoning has become a central mechanism for eliciting multi-step reasoning in Large Language Models (LLMs). Yet recent evidence presents a tension: hidden states appear to already encode future reasoning before CoT fully unfolds, while explicit steps still remain crucial for tasks requiring compositional computation. To deepen the understanding between LLM's internal states and its verbalized reasoning trajectories, we investigate the latent planning strength of LLMs, through our probing method, Tele-Lens, applying to hidden states across diverse task domains. Our empirical results indicate that LLMs exhibit a myopic horizon, primarily conducting incremental transitions without precise global planning. Leveraging this characteristic, we propose a hypothesis on enhancing uncertainty estimation of CoT, which we validate that a sparse set of pivot positions can effectively represent the uncertainty of the entire path. We further underscore the significance of exploiting CoT dynamics, and demonstrate that automatic recognition of CoT bypass can be achieved without performance degradation. Our code, data and models are released at https://github.com/lxucs/tele-lens.
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。