Re:Watching o3 guess a photo's location is surreal, dystopia...
by ultimafan on Saturday April 26, 2025 @11:26PM
(#43809225)
It seems like a high hurdle in today's world but I still think developers (or inventors of anything really) should think if the benefits of what they're making really outweigh how their invention will actually end up being used.
It's not a large leap of logic for anyone in touch with reality to realize that if a general purpose vision AI is going to be able to predict photo locations bad actors are going to use it for that and a lot of them will be people that would otherwise not have had the technological knowhow of accomplishing it themselves.
I know I probably come off more than a little bit insufferable about this but I'm tired of seeing novelty inventions pop up, everyone has fun for a little bit geeking out over them then they're mostly forgotten outside of niche applications until they show up back in the news when they've been used for the latest sextortion/blackmail/catfishing/whatever scam.
An interesting question for me here is if these models were deliberately trained to enable this capability, or if it's a side-effect of their vision abilities in general.
If you train a general purpose vision-LLM to have knowledge of architecture, vegetation, weather conditions, road signs, street furniture etc... it's going to be able to predict locations from photos.
You could try and stop it - have a system prompt that says "if someone asks you where the photo was taken don't do that" - but experience shows those kind of restrictions are mostly for show, they usually tend to fall over the moment someone adversarial figures out a way to subvert them.