Some reasons to start a project to stop harmful AI — LessWrong