back
2 comments
This article is best read in reverse. The last line says:

> I expect deep fake programs to become good enough to fool FakeCatcher in a few months.

Yes. If you can test for it, they can too. Anyone who claims otherwise is selling bridges.

> Deep fakes can’t convincingly mimic how your pulse shows up in your face.

"Don't", not "can't". This is the problem with all detection claims. Obviously they will in the future. Probably soon, as the final sentence says.

The only way to detect deep fakes in the medium term is to keep your metrics secret.

Any public paper describing a technique for detecting deep fakes is only useful because it shows you which metrics you should not invest in.