x
Stabilizing logical counterfactuals by pseudorandomization — AI Alignment Forum