Ansh Radhakrishnan comments on How “Discovering Latent Knowledge in Language Models Without Supervision” Fits Into a Broader Alignment Scheme