Superintelligence 11: The treacherous turn — LessWrong