Insights on Crosscoder Model Diffing

At the link above, we report some developing work from the Anthropic Interpretability team on Crosscoder Model Diffing, which might be of interest to researchers working actively in this space.