x
Counterfactual control incentives — LessWrong