OpenSAE

non-profit

Building open foundations for understanding, evaluating, and intervening in neural networks.

Hunter26666666 updated a Space about 15 hours ago

Hunter26666666 updated a dataset about 15 hours ago

Hunter26666666 updated a Space about 18 hours ago

Behavioral Compositionality of Sparse Autoencoder Representations

OpenSAE 's Spaces 3

Explore recorded SAE features, interventions, and results.

Open research in mechanistic interpretability.

Five-minute tutorial on SAE behavioral compositionality.