JackM comments on Neel Nanda on Mechanistic Interpretability: Progress, Limits, and Paths to Safer AI (part 2)