自我验证困境：经验驱动的过度验证抑制在大语言模型推理中
Self-Verification Dilemma: Experience-Driven Suppression of Overused Checking in LLM Reasoning
AI总结：
本文提出了一种经验驱动的测试时框架，通过抑制过度的自我验证步骤，减少大语言模型推理中的token使用，同时保持或提升准确性。
摘要：
大型推理模型（LRMs）通过生成长推理轨迹并进行反思来实现强大的性能。通过大规模的实证分析，我们发现大量反思步骤由自我验证（重新检查）组成，这些步骤反复确认中间结果。这些重新检查在不同模型和基准测试中频繁出现，但大多数是确认性的而非纠正性的，很少发现错误并改变推理结果。这揭示了自我验证被激活的频率与其实际有用性之间的不匹配。受此启发，我们提出了一种新颖的经验驱动的测试时框架，以减少过度验证。我们的方法检测重新检查行为的激活，咨询一个离线经验池中的过去验证结果，并通过高效检索估计重新检查是否可能不需要。当历史经验表明不需要时，抑制信号会引导模型继续前进。在多个模型和基准测试中，我们的方法将token使用量减少了高达20.3%，同时保持准确性，在某些数据集中甚至实现了准确性提升。
英文摘要：
Large Reasoning Models (LRMs) achieve strong performance by generating long reasoning traces with reflection. Through a large-scale empirical analysis, we find that a substantial fraction of reflective steps consist of self-verification (recheck) that repeatedly confirm intermediate results. These rechecks occur frequently across models and benchmarks, yet the vast majority are confirmatory rather than corrective, rarely identifying errors and altering reasoning outcomes. This reveals a mismatch between how often self-verification is activated and how often it is actually useful. Motivated by this, we propose a novel, experience-driven test-time framework that reduces the overused verification. Our method detects the activation of recheck behavior, consults an offline experience pool of past verification outcomes, and estimates whether a recheck is likely unnecessary via efficient retrieval. When historical experience suggests unnecessary, a suppression signal redirects the model to proceed. Across multiple model and benchmarks, our approach reduces token usage up to 20.3% while maintaining the accuracy, and in some datasets even yields accuracy improvements.
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。
