The Ghost in the Latent Space

Using sparse autoencoders to map hidden concepts inside large language models.

read full log arrow_right_alt