Safe training procedures for human-imitators - History — LessWrong