I'm very skeptical that this "model poisoning" approach will work in practice. To pull it off would require a very high level of coordination among disparate people generating the training data (the images/text). I just can't imagine it happening. Add to that: big tech has A LOT of resources to play this cat & mouse game.
I hope I'm wrong, but I predict big tech wins here.
I hear what you're saying and it makes sense. But I do not believe that my particular case was one of priming/manipulation/freq bias. The topic of conversation was too uncharacteristic, too random, and there were too many similar ads within moments. It was either a colossal coincidence or a breach in privacy.
I suppose it doesn't matter. My phone is much more locked down now.
Something on stock Android phones is always listening though. I had a similar experience as OP where I had an IRL conversation once with my son about a product I don't normally talk about. My phone was unused & "asleep" nearby. An hour later at work I was inundated with ads for said product all over the internet in my Chrome browser on my work computer. It was way too heavy handed to be a coincidence. The phone had listened to our conversation.
(That day marked the first on my journey to de-Google and take serious steps preserve my privacy online)
Surprised the rates of adblocking is so high! I thought it was a little more niche.
Also surprised that the article didn't mention manifest v3 rolling out later this year to Chrimium-based browsers - which will effectively end adblocking in all browsers except Firefox.
Google isn't stupid, they know that ad blocking undermines their business. And Google controls Chromium: the backbone of almost all browsers. So of course they're going to engineer it to prevent ad blocking. It was only a matter of time.
Dude, Turbo Killer DID get turned into a movie!! Carpenter Brut & Seth Ickerman did a ~30 min followup to Turbo Killer called Blood Machines. The plot and acting are awful. But the aesthetics are really good.
I don't have a good response here. My existing case is transparent, so I could easily put some art in there. I guess I just didn't think of it! Thanks for the suggestion!
I'm very skeptical that this "model poisoning" approach will work in practice. To pull it off would require a very high level of coordination among disparate people generating the training data (the images/text). I just can't imagine it happening. Add to that: big tech has A LOT of resources to play this cat & mouse game.
I hope I'm wrong, but I predict big tech wins here.