crayhippo comments on Dangers of Closed-Loop AI

[crayhippo]

Do you see possible dangers of closed-loop automated interpretability systems as well?