通过分析熵值变化动态，识别大模型推理错误的内在模式。
EDIS: Diagnosing LLM Reasoning via Entropy Dynamics
- 用熵值随生成过程的动态变化替代静态置信度，捕捉推理过程中的不稳定性。
- 发现错误推理常出现持续上升或剧烈波动的熵值异常，正确推理则更平稳。
- 提出EDIS指标，可用于推理时筛选结果或训练时优化样本选择。
基于熵的置信度信号被广泛用于提升大语言模型（LLMs）的推理能力，但现有方法将置信度视为静态量——通常在所有标记上聚合。我们发现，置信度在生成过程中的时间演变包含比聚合统计更丰富的信息。分析标记级熵轨迹，我们识别出区分正确与错误推理的典型模式：错误解法表现出不稳定的动态特征，包括持续上升的爆发性峰值（不确定性持续增长）和峰谷型爆发（短暂信心后突然反弹）。这些模式在不同模型和训练阶段均存在，表明它们反映的是推理失败的本质属性而非表面噪声。为此，我们引入熵动态不稳定性评分（EDIS），一种量化熵演化不稳定性的事物级指标。EDIS可作为有效的推理期诊断信号，显著提升推理准确率，并为训练期样本筛选提供新方向。研究结果确立了熵动态作为理解与改进LLM推理的未充分探索但极具价值的新视角。
原文摘要 · Abstract (English)
Entropy-based confidence signals are increasingly leveraged to improve reasoning in large language models (LLMs), yet existing approaches treat confidence as a static quantity -- typically aggregated over tokens. We show that the \emph{temporal evolution} of confidence during generation carries richer information than aggregate statistics alone. Analyzing token-level entropy trajectories, we identify characteristic patterns distinguishing correct from incorrect reasoning: erroneous solutions exhibit unstable dynamics, including burst spikes (sustained uncertainty growth) and peak-valley spikes (sharp rebounds following transient confidence). These patterns persist across models and training stages, suggesting they reflect intrinsic properties of reasoning failure rather than superficial noise. To formalize this observation, we introduce the Entropy Dynamics Instability Score (\textbf{EDIS}), a trajectory-level metric quantifying instability in entropy evolution. EDIS serves as an effective diagnostic signal for inference-time selection, substantially improving reasoning accuracy, and offers a promising direction for training-time sample curation. Our findings establish entropy dynamics as an underexplored yet informative lens for understanding and improving LLM reasoning.
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。