Mechanistic Interpretability

Autointerp simulation scoring is used as a quality signal for sparse autoencoder latent descriptions, a use that presumes the score …