Neel Nanda comments on Interpretability Will Not Reliably Find Deceptive AI

[Neel Nanda]

I’m not trying to comment on other theories of change in this post, so no disagreement there