It’s the same entitlement that determined one could just download all the content available online to train your models against.

That's hoarding. The content is downloaded and then used to train proprietary models at no benefit to greater humanity. Thus some few corporations are robbing the commons and trying to rent it back to us.

People would have far fewer problems with that if the resulting models were also released back to the general public.

Weird to call it entitlement when the natural state of information is to be free. What's entitled is asking the government to enforce arbitrary restrictions on other people making use of some information that you somehow intangibly "own."

(Of course, it's fucked up that corporations can siphon up all this content and then try to twist the law to carve out an exception for their extra special use case. Information still isn't free unless you're an AI company, I guess.)

There's a difference between "infringing IP[1]", "stealing IP", and whatever we should call AI training. And it turns out the worse the behavior gets, the less likely the law is going to recognize it as bad.

IP infringement is what we're used to talking about. This is when I go and give a stranger a copy of some music I don't own. Or when some sketchy ass guy resells IPTV services to an entire island in Greece or whatever. They're not saying it's their work, they're just refusing to pay the appropriate licensing fee for it. And sometimes we might even agree that a license fee shouldn't have to be paid. What the Linux video driver people want is for the HDMI people to say "yes, you can tell people how to light up this video card in such a way that it successfully negotiates a connection at HDMI 2.1 bitrates", which shouldn't even be infringement at all, but here we are.

What China does is wholesale IP theft. They don't just make their own version of someone else's thing, or just do industrial espionage, they actively make an attempt to deny the original creator of their own work. This can include things like forcing foreign entities to go through a JV, or playing games with trademark law to allow domestic companies to actually take legal ownership over foreign works. This is why a lot of American companies spent time and money carrying water for Xi Jinping, despite it going against everything they claimed to stand for.

AI training doesn't fit in either mold. It's more like rugpulling human labor by turning know-how and creativity into ownable capital distinct from that of traditional copyright and patents. Copyright gives you ownership over your own work, but says nothing about having your entire craft being automated away by a robot that can turn your work into legally distinct knockoffs of it[0]. So we have an entirely new form of enclosure of the commons, where if you ever do a thing, someone else can turn that thing into their own property that everyone else can pay to rent. Like, to be clear: AI is not Napster. AI is the opposite of Napster. AI is the apotheosis of "you will own nothing and be happy".

[0] The only way that copyright claims on AI even sort of fit into recognizable harms is the fact that at some point a Facebook engineer pointed LLaMA's crawler at a torrent site. In fact, I kinda hate how this is sort of saying "well actually fair use only applies if you bought the book first". Which is a problem, because the condition of sale can be "don't make a fair use of it", and the only way to avoid that was to pirate the work and then make your fair use.

[1] As Cory Doctorow said, paraphrasing: Intellectual property is the laws that allow you to dictate the conduct of your competitors.