scasper comments on Barriers to Mechanistic Interpretability for AGI Safety

[scasper]

Several people seem to be coming to similar conclusions recently (e.g., this recent post).

I’ll add that I have as well and wrote a sequence about it :)