Xuansheng Wu

Interpreting and Steering LLMs with Mutual Information-based Explanations on Sparse Autoencoders

Investigating CoT Monitorability in Large Reasoning Models

Soundness-Aware Level: A Microscopic Signature that Predicts LLM Reasoning Potential

Enhancing LLM Steering through Sparse Autoencoder-Based Vector Refinement

Concept-Centric Token Interpretation for Vector-Quantized Generative Models

Interpreting and Steering LLM Representations with Mutual Information-based Explanations on Sparse Autoencoders

CORTEX: Concept-Oriented Token Explanation in Vector-Quantized Generative Model