x
Crazy ideas in AI Safety part 1: Narrow-Defined Communication — LessWrong