Mechanistic Interpretability Benchmark

AI & ML interests

Principled evaluation of mechanistic interpretability methods.

Recent Activity

hij  authored a paper 23 days ago

AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoders

hij  authored a paper 23 days ago

LLMs Encode Harmfulness and Refusal Separately

hij  authored a paper 23 days ago

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

View all activity

Team members 19