我已覆盖所有基础：通过稀疏自编码器解读大型语言模型中的推理特征
I Have Covered All the Bases Here: Interpreting Reasoning Features in Large Language Models via Sparse Autoencoders
AI总结：
本文通过稀疏自编码器识别并放大与不确定性、探索和反思相关的推理特征，提升LLM推理性能（+2.2%）并延长推理轨迹（+20.5%），为机制性理解推理提供初步证据。
摘要：
近期如DeepSeek-R1等大型语言模型通过在生成过程中整合深度思考与复杂推理，展现出最先进的性能。然而，这些推理过程背后的内部机制仍未得到探索。我们观察到推理型语言模型始终使用与人类推理过程相关的词汇。我们假设这些词汇对应于模型内部机制中的特定推理时刻。为验证这一假设，我们采用稀疏自编码器（SAEs），这是一种将神经网络激活稀疏分解为人类可解释特征的技术。我们引入了ReasonScore，一种自动度量方法，用于识别这些推理时刻中活跃的SAE特征。我们对我们度量检测到的特征进行了人工和自动解释，发现那些激活模式与不确定性、探索性思维和反思相匹配的特征。通过引导实验，我们证明放大这些特征能提升推理密集型基准的性能（+2.2%），同时产生更长的推理轨迹（+20.5%）。利用模型差分技术，我们提供了证据表明这些特征仅存在于具有推理能力的模型中。我们的工作为理解大型语言模型中的推理机制迈出了第一步。代码可在https://github.com/AIRI-Institute/SAE-Reasoning获取。
英文摘要：
Recent LLMs like DeepSeek-R1 have demonstrated state-of-the-art performance by integrating deep thinking and complex reasoning during generation. However, the internal mechanisms behind these reasoning processes remain unexplored. We observe reasoning LLMs consistently use vocabulary associated with human reasoning processes. We hypothesize these words correspond to specific reasoning moments within the models' internal mechanisms. To test this hypothesis, we employ Sparse Autoencoders (SAEs), a technique for sparse decomposition of neural network activations into human-interpretable features. We introduce ReasonScore, an automatic metric to identify active SAE features during these reasoning moments. We perform manual and automatic interpretation of the features detected by our metric, and find those with activation patterns matching uncertainty, exploratory thinking, and reflection. Through steering experiments, we demonstrate that amplifying these features increases performance on reasoning-intensive benchmarks (+2.2%) while producing longer reasoning traces (+20.5%). Using the model diffing technique, we provide evidence that these features are present only in models with reasoning capabilities. Our work provides the first step towards a mechanistic understanding of reasoning in LLMs. Code available at https://github.com/AIRI-Institute/SAE-Reasoning
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。
