[craffel]

Agreed! The interesting thing is that basic unsupervised pre-training seems to produce a model which functions not only as a knowledge base but also an NLU system which can effectively query the knowledge base using natural text questions. This is exactly what our follow-up paper is on.