Hacker News Reader

Won't there always be a better dataset for deep-fake detectors than there will be for generators?

Other way around, no?

The problem with the other way around is that there is no definitive measure of "convincingness."