Xodarap comments on Discovering Latent Knowledge in Language Models Without Supervision

[Xodarap]

One of the authors has now posted about this here