Such an entity would not be hyper-intelligent. It would be idiotic. One huge hole for me in the paperclip argument is that an AI capable of that kind of power would not be stupid enough to misinterpret a command - it would be intelligent enough to infer human desires.