发现大模型推理中存在‘表演式思维’，能提前识别并减少无效计算。
Reasoning Theater: Disentangling Model Beliefs from Chain-of-Thought
- 通过激活探针和监控器对比，发现模型答案早于思维过程就能预测。
- 在简单任务上可减少80%生成token，复杂任务也减少30%，准确率不变。
- 能捕捉真实推理中的困惑时刻，适合优化推理效率的场景。
我们揭示了推理模型中存在表现性链式思考（CoT）的现象：模型对最终答案高度自信，却仍持续生成文本而不暴露其内部信念。通过在两个大模型（DeepSeek-R1 671B 和 GPT-OSS 120B）上对比激活探针、早期强制作答与 CoT 监控器，发现任务难度相关的差异：对于简单的回忆类 MMLU 问题，模型最终答案的激活信号远早于监控器能检测到；而在困难的多跳推理 GPQA-Diamond 问题中则体现为真正的推理过程。尽管如此，拐点（如回溯、“顿悟”时刻）几乎仅出现在探针显示信念显著变化时，表明这些行为反映的是真实不确定性而非训练出的“表演”。此外，基于探针的早期退出机制在 MMLU 上最多可减少 80% 的生成 token，GPQA-Diamond 上减少 30%，且保持相似准确率，证明激活探针是识别表现性推理并实现自适应计算的有效工具。
原文摘要 · Abstract (English)
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.
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。