Jul 3, 2026, 6:15 PMThe cure for most of the problems of Large Language Model generative AI systems is making the firms absolutely liable for what their AI models say and do. Financially, Legally, and Criminally. The firms can then decide which of their AI services they wish to continue deploying to the public under that set of constraints. It will be completely up to them!
@lauren I feel like this goes beyond LLMs. There’s a shocking lack of liability when it comes to autonomous vehicles. The model they seem to be pushing is zero liability for anything they create and they are getting away with it. Let’s hope they don’t try to start an airline.
@lauren On a bright note, none are making any money. If we're lucky, things will sort themselves out. Pop!
Jul 3, 2026, 6:48 PM@WildEyedBoyFromFreecloud Well, the idea is to get this situation under control ideally without tanking the entire economy in the process.
@lauren Agreed.
Although, it reminds me of someone I worked with years ago. He would make tweaks on a daily basis on prod servers. It was a disaster, obviously. My boss was worried about letting him go. I said the longer we wait, the worse it will be.
@lauren And requiring them to pay compensation for everything they stole to train their models.
I would also like to see recommendation systems qualify as liability generating "AI" in this case.
Monetization too, IMO, should be taken as a liability-generating endorsement of a piece of "content", at least in a for-profit entity.
Jul 3, 2026, 7:26 PM@TallSimon "Recommendation" is generally a different category than AI. The key issue with AI is that "original" content is being created, not simply references to third party content. This is a crucial difference which pushes Section 230 protections out of scope. The same issue applies to monetization. The crucial issue is creation of new content.
@lauren @TallSimon And broadly speaking that distinction probably should be made (i.e. having an "algorithm" other than a chronological subscription feed shouldn't be completely outlawed).
Though some specific cases should see more scrutiny. Like investigating X to see how much it is manipulating things to promote what Elon Musk agrees with.
At some point, that starts to look more like a publisher with its own editorial positons, and not merely a neutral platform.
Jul 3, 2026, 8:49 PM@ids1024 @TallSimon 230 is pretty explicit. Attempting to fold other stuff into the same category as AI (absent actual changes to 230 itself) is not a good idea.