Word vectors sure are great and highly accurate too. The most recent and advanced models have demonstrated 30–67% accuracy on tests that are essentially rigged for success and when going through multiple iterations of larger and larger training data sets. A solid D+. Honestly, even if it was 100% I am not sure what that would mean, probably jack squat. We would have the ability to program computers to run algorithms we designed based on mathematical (mostly algebraic in this case) rules we discovered/developed. These computers we programmed could accurately spit out the relationships between a tiny subset of words. No wonder machines are getting so smart. All that learning.
Also, for the twenty bazillionth time, machines can’t learn. Try neuroarchitecture inspired computing (NiC) instead.
To answer the question posed in the title of the piece I will have the kind of relationship a five year old boy has with his imaginary friend Steve. Steve can do anything, he is very comforting and he never lets me down but ultimately he is an illusion. I will grow out of him after the third grade.