I Have Covered All the Bases Here: Interpreting Reasoning Features in Large Language Models via Sparse Autoencoders
Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel pith:VYV7BBXJrecord.jsonopen to challenge →
read the original abstract
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
This paper has not been read by Pith yet.
Forward citations
Cited by 7 Pith papers
-
The Tell-Tale Norm: $\ell_2$ Magnitude as a Signal for Reasoning Dynamics in Large Language Models
The L2 norm of LLM hidden states signals reasoning intensity, with a theoretical bound on SAE feature activations, enabling three new test-time scaling techniques that boost performance.
-
Why Does Reinforcement Learning Generalize? A Feature-Level Mechanistic Study of Post-Training in Large Language Models
RL generalizes better than SFT by preserving and slowly evolving a compact set of task-agnostic features from the base model rather than introducing many specialized ones.
-
Contrastive Attribution in the Wild: An Interpretability Analysis of LLM Failures on Realistic Benchmarks
Token-level contrastive attribution yields informative signals for some LLM benchmark failures but is not universally applicable across datasets and models.
-
Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces
Reasoning in large output spaces proceeds via shortlisting then fine-grained reasoning; this characterization enables a mechanistic distillation strategy that outperforms standard distillation.
-
Steered Generation via Gradient-Based Optimization on Sparse Query Features
Prototype-Based Sparse Steering decomposes query activations with SAEs and optimizes sparse features via gradients to steer LLM outputs toward specific behaviors.
-
Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models
The survey organizes mechanistic interpretability techniques into a Locate-Steer-Improve framework to enable actionable improvements in LLM alignment, capability, and efficiency.
-
The Periodic Table of LLM Reasoning: A Structured Survey of Reasoning Paradigms, Methods, and Failure Modes
A literature survey that introduces a taxonomy for LLM reasoning paradigms, analyzes methodological trends, and synthesizes failure modes from over 300 papers.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.