David Bau

LLMs Process Lists With General Filter Heads

Discovering Forbidden Topics in Language Models

In-Context Algebra

Language Models Use Lookbacks to Track Beliefs

Eliciting and evaluating generalizable explanations from large reasoning models

In-Context Learning Without Copying

Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models

LLMs Encode Harmfulness and Refusal Separately

One-Step is Enough: Sparse Autoencoders for Text-to-Image Diffusion Models

The Dual-Route Model of Induction

NNsight and NDIF: Democratizing Access to Open-Weight Foundation Model Internals

Erasing Conceptual Knowledge from Language Models

MIB: A Mechanistic Interpretability Benchmark

When Are Concepts Erased From Diffusion Models?

Art-Free Generative Models: Exploring Art Creation Without Prior Artistic Knowledge

Erasing Conceptual Knowledge from Language Models