Read news on transformer outputs with our app.
Read more in the app
Logit Prisms: Decomposing Transformer Outputs for Mechanistic Interpretability