How do startups building to detect deepfakes even compete?

10 points·uma08··6 comments

With so many image models and video models like Sora or RunwayML, how can a startup building to detect/label/tag deepfakes compete?

View on HN

They could always be like IBM. IBM does very little well. We laugh about them here on Hacker News. But when the job is to provide cover if a project goes bad, actually delivering working projects isn't as important.

Startups could compete by assuring whoever the CEO is talking to that the problem is being addressed. Even if the solution barely works.

Plenty of these systems exist out there and make money. I am personally aware of two that claim to verify something about me that are absurdly incorrect.

thank you! I wonder and worry about a company like Google that's got access to so many personal images of users either via Drive or Android; and how they might be able to just feed all that into a multimode engine like Gemini and expose the API endpoint to it's users and put the product into the hands of potential bad actors in our society.

I mean, that was how things were a decade ago. Do you remember that period in the mid-2010s when you could just photograph a stranger on the subway, drag it into Google Images, and get a list of all their photos and social media accounts? Of course, Google since disabled it, and only does facial recognition of celebrities in their knowledge graph now. But it still works on Yandex and Pimeyes.

Yeah, but now with deepfakes just skyrocketing much more, I wonder who will - and if they even could - come out a winner on top for detecting it.

They aren't. It's snake oil to fool schools who have to buy them for cheating.

yeah, my thoughts exactly

Reply on news.ycombinator.com