x
Methodological considerations in making malign initializations for control research — LessWrong