Gunnar_Zarncke comments on Barriers to Mechanistic Interpretability for AGI Safety