I Have Covered All the Bases Here: Interpreting Reasoning Features in Large Language Models via Sparse Autoencoders
Paper • 2503.18878 • Published • 121
Interpretable
-
-
Large Language Models are Locally Linear Mappings
Paper • 2505.24293 • Published • 14 -
Thought Anchors: Which LLM Reasoning Steps Matter?
Paper • 2506.19143 • Published • 14 -
Steered LLM Activations are Non-Surjective
Paper • 2604.09839 • Published • 15 -
Negligible in Size, Significant in Effect: On Scale Vectors in Large Language Models
Paper • 2605.26895 • Published • 18 -
LLM Explainability with Counterfactual Chains and Causal Graphs
Paper • 2606.05972 • Published • 19 -
A Geometric Account of Activation Steering through Angle-Norm Decomposition
Paper • 2606.06735 • Published • 27 -
ICA Lens: Interpreting Language Models Without Training Another Dictionary
Paper • 2606.11722 • Published • 17 -
Formalizing Latent Thoughts: Four Axioms of Thought Representation in LLMs
Paper • 2606.27378 • Published • 60 -
Towards Mechanistically Understanding Why Memorized Knowledge Fails to Generalize in Large Language Model Finetuning
Paper • 2607.08393 • Published • 14 -
Scaling Inherently Interpretable Language Models
Paper • 2608.07594 • Published • 21 -
U-Space: Uncovering When and Why Uncertainty Arises in Language Models
Paper • 2610.09087 • Published • 33