Think about the change we saw in combat death tolls when things went from flintlock muskets to machine guns, or when battleships gave way to aircraft, and how many people died unnecessarily due to the generals who were slow to update their tactics. Deepfakes are like that because they lower the cost and improve the success rates enough to be transformative and they cause harm which can’t easily be countered. We’re not going to instantly train society to be better at media literacy and the police can’t just ignore reports of sex crimes, so we’re just having to accept that it’s easier to hurt people than it used to be.
What if you get a model that is 99% similar to your “target” - what we do with that?