Anthropic Interpretability Team

circuits, sparse autoencoder, monosemanticity reports

Anthropic's mechanistic interpretability research group, known for the circuits thread, monosemanticity findings, and sparse autoencoder papers on neural network internals.