MODELS WITH BUILT-IN INTERPRETABILITY

Summary from the TLDR AI newsletter

"B-cosification" is a method that adapts to existing pre-trained models to offer highly interpretable explanations of their predictions.