x
The case for AI safety capacity-building work — LessWrong