I think that considering it's the V.1 of the first model of its kind, it did pretty well to tell a joke that almost works. Just the fact that it can create something like that on its own and not simply regurgitate jokes that it finds in its data set is pretty amazing.