arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2601.00514cs.AIcs.CL
收藏

推理模型中的“顿悟”幻觉

The Illusion of Insight in Reasoning Models

Princeton University（普林斯顿大学）

Liv G. d'Aliberti, Manoel Horta Ribeiro

阅读 PDF
arXiv 页面 ↗HTML 原文 ↗

AI总结：

研究探讨推理模型在推理过程中是否出现突然的顿悟现象，发现此类现象罕见且不提升准确性，但高不确定性下人工触发可提高性能。

摘要：

推理模型是否具有“顿悟”时刻？先前研究表明，如DeepSeek-R1-Zero等模型在推理过程中会出现突然的中间阶段认知转变，导致准确输出，暗示其具备内在自我修正能力。然而，这种推理策略的内在变化是否真正提升性能仍不清楚。本文通过研究中间推理转变并仪器化训练运行来检测此类现象。我们的分析覆盖了100万+推理轨迹、数百个训练检查点、三个推理领域以及多种解码温度和模型架构。我们发现，推理转变极为罕见，不随训练频率增加而增多，且很少提升准确性，表明它们并不对应先前对模型洞察力的感知。然而，其影响随模型不确定性而变化。基于此发现，我们展示在高熵下人工触发外在转变可可靠提高准确性。我们的结果表明，中间推理转变是不稳定推理行为的症状，而非内在的自我修正机制。

英文摘要：

Do reasoning models have "Aha!" moments? Prior work suggests that models like DeepSeek-R1-Zero undergo sudden mid-trace realizations that lead to accurate outputs, implying an intrinsic capacity for self-correction. Yet, it remains unclear whether such intrinsic shifts in reasoning strategy actually improve performance. Here, we study mid-reasoning shifts and instrument training runs to detect them. Our analysis spans 1M+ reasoning traces, hundreds of training checkpoints, three reasoning domains, and multiple decoding temperatures and model architectures. We find that reasoning shifts are rare, do not become more frequent with training, and seldom improve accuracy, indicating that they do not correspond to prior perceptions of model insight. However, their effect varies with model uncertainty. Building on this finding, we show that artificially triggering extrinsic shifts under high entropy reliably improves accuracy. Our results show that mid-reasoning shifts are symptoms of unstable inference behavior rather than an intrinsic mechanism for self-correction.

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。

↑