Fetching the paper…

Interpretability as Compression: Reconsidering SAE Explanations of Neural Activations with MDL-SAEs · Around