Comments on Carlsmith's “Is power-seeking AI an existential risk?” — LessWrong