Model Directions, Not Words: Mechanistic Topic Models Using Sparse Autoencoders

Open in new window