Pro tip: Saw deepfake detection at the Austin tech conference and it was scarily bad

I watched a demo at the Austin Convention Center last Wednesday where their AI missed a fake video of a CEO that even I could tell was off, so how are we supposed to trust these tools when they roll out for real?

2 comments

2 Comments

price.tyler1mo ago

Whoa, hold on, I gotta push back on this one a bit. I think the demo might have been bad on purpose to show how hard the problem really is, you know? @martin.jamie, I actually think the opposite of what you're saying - the real world stuff is usually way more polished than the raw demos I see at these conferences (they're often running on unstable WiFi or old laptops). Plus, if the AI called every single weird video fake, people would be screaming about false positives way more than they'd talk about a missed one. The whole point of these tools is to flag things for human review, not to replace your own judgment, right?

martin.jamie1mo ago

Honestly that lines up with what I've been noticing with every new AI tool this year, they always demo way better than they actually work in the real world.