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

Open in new window