Alignment can be the ‘clean energy’ of AI — LessWrong