What if "friendly/unfriendly" GAI isn't a thing? — LessWrong