重新审视大语言模型推理中的拟人化反思标记
Revisiting Anthropomorphic Reflection Markers in Large Language Model Reasoning
AI总结：
本文通过提示级和令牌级干预抑制拟人化反思标记，发现这些标记并非推理性能的必要条件，且抑制后模型仍能进行无标记验证，表明它们更多是表面线索而非可靠反思代理。
摘要：
大语言模型（LLMs）在复杂推理过程中经常产生显式的反思痕迹，并伴随有拟人化标记，如“wait”、“hmm”和“alternatively”。尽管这些标记通常被用作反思的可见指标，但其机制仍不清楚，这带来了与冗余和重复反思标记相关的过度思考风险。在这项工作中，我们重新审视了拟人化反思标记，考察了它们对推理的必要性以及在反思中的作用。我们通过提示级和令牌级干预抑制这些标记，并分析了它们对四个基准测试和两种模型规模的任务性能的影响。我们的结果表明，拟人化标记对于推理性能并非普遍必要：抑制它们可以在多种设置下保持或提高性能，尤其是在较大的采样预算下。同时，标记抑制并不一定消除反思行为，因为模型仍然可以进行无标记验证。这些结果表明，拟人化标记更倾向于表面线索，而不是反思本身的可靠代理，并激励未来在显式标记模式之外对推理机制进行研究。
英文摘要：
Large Language Models (LLMs) often produce explicit reflective traces during complex reasoning, accompanied by anthropomorphic markers such as wait, hmm, and alternatively. Although these markers are commonly used as visible indicators of reflection, their mechanisms remain unclear, which leaves the risk of overthinking associated with redundant and repetitive reflection markers. In this work, we revisit anthropomorphic reflection markers, examining their necessity for reasoning and role in the reflection. We suppress these markers through prompt-level and token-level interventions, and analyze their effects on task performance across four benchmarks and two model scales. Our results show that anthropomorphic markers are not uniformly necessary for reasoning performance: suppressing them can preserve or improve performance in several settings, especially under larger sampling budgets. Meanwhile, marker suppression does not necessarily remove reflection behavior, as models can still perform marker-free verification. These suggest that anthropomorphic markers tend to be surface cues rather than reliable proxies for reflection itself, and motivate future research on reasoning mechanisms beyond explicit marker patterns.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
