Current neural networks are basically a set of highly sophisticated statistical rules. So unless the Facebook engineers have come up with some new breakthrough, they probably won't be able to build something much better than what's already around, as statistics can only take you so far.