Artificial Intelligence, Values and Alignment — LessWrong