What's Your Best AI Safety "Quip"? — LessWrong