“Interpretability Will Not Reliably Find Deceptive AI” by Neel Nanda

od LessWrong (Curated & Popular)