Well, what convenient timing... "Deepfake detectors can be defeated, computer scientists show for the first time"
(www.eurekalert.org)
BREAKING NEWS
You're viewing a single comment thread. View all comments, or full comment thread.
Comments (57)
sorted by:
I'm confused. I actually read the article, and it seems to say to me that this is not about having the ability to prove something is a deepfake with detectors we've already had in place. Rather, it says that there is now programming that interferes with the detector's ability to identify a deepfake. In other words, if the DS puts out a deepfake of Biden, the detectors would prove this and the whole thing is proven as fake. But if new software over rides or defeats the detector's software, then the deepfake of Biden will be shown to be "real." Do I have this right? Many seem to be reading this simply as we have the ability to identify deep fakes. It seems way more than that and quite scary given the potential ramifications if a deepfake is unable to be detected as such.
You're right. This is not about identifying DFs, it's about defeating automated detection.
I didn't watch it closely but it looked like there was some kind of intentional mistake near his nose. This might be the same thing as face detection defeating face paint. You paint simple black and white geometric shapes and then your face can't be detected but humans can easily still recognize.