Building open foundations for understanding, evaluating, and intervening in neural networks.
Behavioral Compositionality of Sparse Autoencoder Representations