While I personally wouldn’t want AI inserting trains into photos of my kids without my consent, many kids like trains, and they could add some whimsy to an otherwise uninteresting picture.
I know you’re making a joke, but this doesn’t really feel like the place to do it given the subject being discussed.
Yeah, robots looking at photos of kids that their parents voluntarily posted on the internet is no laughing matter. Way more serious than, say, violent crime. And nobody makes jokes about that, do they?
I personally don’t appreciate jokes about violence either, but whatever. I’m not policing the Internet.
“Current AI models cannot forget data they were trained on, even if the data was later removed from the training data set,” Han’s report said.
Bullshit. You delete the entire model and start again.
I find this wholly unsurprising.
All ai projects should be forced to show the entirety of their training data. I don’t give a flying fuck if they want to call it proprietary, they don’t own most of the data in the first place. Even if they bought it, it doesn’t belong to them, just like we don’t own digital movies we buy.
And if even a single piece of that training data doesn’t have proper licensing for that specific use for that specific model, or they are ever found to have withheld any of the data, the model as a whole should be immediately scrapped, along with everything even tangentially derived from it, and the company should be fined fully double whatever amount of money that model generated or one years revenue for the company as a whole, whichever is more (no I don’t care if this leads to bankruptcy, should have thought about that before you stole data), and like use if for affordable housing programs or public schools or something, whatever.
They can try again with clean data, also subject to review. One time. Second time they do the same shady shit, permanently banned from the entire sector.
But regardless, we need to stop rewarding them for this behavior. And we need the consequences to actually hurt or we can expect it to get worse, not better.
All ai projects should be forced to show the entirety of their training data.
Agreed—but note that in this case the information was only discovered because the organizations involved (Common Crawl and LAION) do show their data. We should assume that proprietary data sets have similar issues—but this case should be seen as an opportunity to improve one of the rare open data sets, not to penalize its openness and further entrench proprietary sources.
Hey parents, wanna let me hold your money for safe keeping? I sure you it’s safer with me.
Maybe instead, so ya wanna stop getting your fucking children involved with proprietary software and remote hosting ya fucking dunces!??!
This is such a bad headline. There is no “the AI”. There are lots of different people doing unrelated things.