16
Am I the only one who thinks that deep fake detection tool from Stanford is way overhyped?
Had a guy at a meetup in Portland show me his "perfect" catch rate with it, but when I ran the same video through three other free tools it flagged a bunch of stuff he missed. Why does everyone trust one study over actual community testing?
3 comments
Log in to join the discussion
Log In3 Comments
felixlane18d ago
Reminds me of how people trust Whole Foods' organic label blindly but ignore the farmers market guy who's been doing it better for decades. Big name equals big trust even when the local option has proven results. The Stanford team's dataset is like those sterile test kitchens where everything works perfect, but throw in some real world glitchy lighting or weird aspect ratios and it falls apart. Community testers are like the mechanics who actually drive the car on pothole-filled streets, not engineers with a test track.
6
murray.robert18d ago
Keep pushing that idea further because the real problem is they tested it on their own dataset not real world garbage. I was digging through a Discord of video editors last week and someone uploaded 4 deepfakes from a cheap Russian generator, the Stanford tool missed 3 of them completely. Free tools like Deepware and Sensity caught every single one plus flagged some edge cases that weren't even fake. The hype is just a brand name at this point.
5
alicer5318d ago
The problem with the Stanford tool is it was built for academic papers, not real world junk. Those cheap Russian generators mess with compression in a way their training data never saw. I get why people trust a big name like Stanford, but that trust is exactly what makes the hype so annoying. Community testing catches stuff that lab conditions miss every single time because people know how to break things on purpose. If the Stanford team spent half as much time scraping actual Discord channels and dark web forums as they do writing press releases, their tool might actually work on the garbage that matters. Until then, Deepware and Sensity are doing the real work for free.
1