reallyeli comments on The goal-guarding hypothesis (Section 2.3.1.1 of “Scheming AIs”)