I think this will be isolated to highly specialized fields where training data will need to be selectively curated.