推理剧场：从思维链中分离模型信念
Reasoning Theater: Disentangling Model Beliefs from Chain-of-Thought
AI总结：
通过激活探针、早期强制回答和思维链监控器分析，发现推理模型存在表演性思维链现象，并利用探针引导的早期退出实现高效计算。
摘要：
我们提供了推理模型中表演性思维链（CoT）的证据，即模型对其最终答案变得非常自信，但继续生成令牌而不揭示其内部信念。我们的分析比较了两个大型模型（DeepSeek-R1 671B 和 GPT-OSS 120B）中的激活探针、早期强制回答和思维链监控器，并发现了任务难度特定的差异：模型的最终答案可以从思维链中远早于监控器能够判断的激活中解码，特别是对于基于回忆的简单MMLU问题。我们将此与困难的多跳GPQA-Diamond问题中的真正推理进行对比。尽管如此，转折点（例如回溯、“啊哈”时刻）几乎只出现在探针显示大信念转变的响应中，表明这些行为追踪的是真正的不确定性，而不是学到的“推理剧场”。最后，探针引导的早期退出在MMLU上减少了高达80%的令牌，在GPQA-Diamond上减少了30%，且准确率相似，将注意力探针定位为检测表演性推理和实现自适应计算的高效工具。
英文摘要：
We provide evidence of performative chain-of-thought (CoT) in reasoning models, where a model becomes strongly confident in its final answer, but continues generating tokens without revealing its internal belief. Our analysis compares activation probing, early forced answering, and a CoT monitor across two large models (DeepSeek-R1 671B & GPT-OSS 120B) and find task difficulty-specific differences: The model's final answer is decodable from activations far earlier in CoT than a monitor is able to say, especially for easy recall-based MMLU questions. We contrast this with genuine reasoning in difficult multihop GPQA-Diamond questions. Despite this, inflection points (e.g., backtracking, 'aha' moments) occur almost exclusively in responses where probes show large belief shifts, suggesting these behaviors track genuine uncertainty rather than learned "reasoning theater." Finally, probe-guided early exit reduces tokens by up to 80% on MMLU and 30% on GPQA-Diamond with similar accuracy, positioning attention probing as an efficient tool for detecting performative reasoning and enabling adaptive computation.
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。
