Building open foundations for understanding and evaluating LLMs.
Behavioral Compositionality of Sparse Autoencoder Representations