x
An idea for creating safe AI — LessWrong