No Global Plan in Chain-of-Thought: Uncover the Latent Planning Horizon of LLMs
This work provides insights into LLM reasoning dynamics, potentially improving efficiency and uncertainty estimation for tasks requiring multi-step reasoning, though it is incremental in nature.
The study found that large language models (LLMs) exhibit a myopic planning horizon in Chain-of-Thought reasoning, primarily making incremental transitions without global planning, and demonstrated that a small subset of CoT positions can effectively represent uncertainty, enabling CoT bypass without performance degradation.
This work stems from prior complementary observations on the dynamics of Chain-of-Thought (CoT): Large Language Models (LLMs) is shown latent planning of subsequent reasoning prior to CoT emergence, thereby diminishing the significance of explicit CoT; whereas CoT remains critical for tasks requiring multi-step reasoning. 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 small subset of CoT 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.