If they have to ask it questions like this to argue that it can't reason, then to me that looks like some cope and that it can reason just fine it's just kind of dumb from its limited model size, data size, and cpu amount.