Michael Lan

Measuring Sparse Autoencoder Feature Space Similarities Across Large Language Models

Sparse Autoencoders Reveal Universal Feature Spaces Across Large Language Models

Activation Space Interventions Can Be Transferred Between Large Language Models