My Blog

Enhancing Automated Interpretability Pipelines with Output-Centric Feature Descriptions

7 min read   ·   January 18, 2025

2025   ·   NLP   SAEs   Interpretability   AI   ML

Using Sparse Autoencoders for Knowledge Erasure

Can we leverage SAEs to effectively erase knowledge from LLMs in a targeted way?

7 min read   ·   January 17, 2025

2025   ·   NLP   SAEs   Interpretability   AI   ML