I'm definitely assuming that they don't limit their training data to what is open source and crawlable.