The Unfriendly Superintelligence next door — LessWrong