Finding the Wisdom to Build Safe AI — LessWrong