Should you start a for-profit AI safety org? — LessWrong