Twitxy Saved

Federico Ulfo πŸ’š

Verified account This account is verified. Learn more

@feulf

New York, NY Joined October 2009
4K Following 3.5K Followers 19K Posts

Latest posts 5

Federico Ulfo πŸ’š Verified account @feulf
how do we even solve this?

The "is AI" solution doesn't work reliably as model gets better. One solution is time-stamping media content and have ffmpeg and other tools to resize and cut to start signing updates with ZK proof.

At this point is impossible to define what's real, but we can give at least a confidence score, the older is a video/photo/audio, the more likely is to be real.

Another problem is provenance, how do you prove a video came from Facebook, Youtube, and from which specific device (IME). New media will need to be signed, always.

This to me sounds like a coordination problem more than it is an engineering one, we basically need an authority that define standards.

I recall @pmarca mentioning this problem 2 years ago. It's not something I want to work on, but it happens that by working in crypto and dealing with digital identity proof in my blockchain healthcare startup in 2017, I've had to thinker around this for too long.

Again I won't take a bite at it because it's a regulation / authority problem. You don't solve this by building a tool or adding another protocol.
Federico Ulfo πŸ’š Verified account @feulf 6d
@NatureUnedited there's a new problem with videos, if you haven't seen a video 5 years ago, chances are, is fake

new videos are all AI generated
View quoted post on X