This, their training data doesn't even seem to be open either. So it's literally impossible to replicate their model. This makes me highly skeptical.