Jing Huang

Eliciting Behaviors in Multi-Turn Conversations

GIM: Improved Interpretability for Large Language Models

AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoders

Blackbox Model Provenance via Palimpsestic Membership Inference

The Unlearning Mirage: A Dynamic Framework for Evaluating LLM Unlearning

Internal Causal Mechanisms Robustly Predict Language Model Out-of-Distribution Behaviors

LLMs Encode Harmfulness and Refusal Separately

HyperDAS: Towards Automating Mechanistic Interpretability with Hypernetworks

MIB: A Mechanistic Interpretability Benchmark