trying my best to shake off the Luddite accusations here hahaha

The wildest thing about those sorts of accusations is that the literal Luddites themselves were basically taking the position "it sucks when the rich bastards own all the machines and screw over all the skilled laborers, because otherwise these machines are actually pretty cool and if they were ours instead that would make all this much better" and then engaging in sabotage as a form of a class warfare. "The workers should take this for themselves and also OpenAI should be redacted in minecraft" is basically the modern equivalent of that.

couching it in 'nice', FOSS-adjacent, communist-adjacent language.

Yeah the reactionary bent in the FOSS scene is shitty and I feel like there's a broader point to be raised there that's related to a different point I bring up all the time, about these sort of libertarian chauvinists who are at odds with big business and other reactionary institutions because those are standing between them and things that they personally want, where their entire worldview just revolves around cynical self-interest and they just happen to be the (at least comparatively) little guy in that scenario.

And just as expected, a lot of the ideological "everyone should have free access to these tools" stuff is a shallow lie for the open source AI scene, where for all that there is absolutely a ton of work being done just for the sake of making better tools and sharing them there's also an entire ecosystem of circling grifters trying to monetize and enclose that work as much as they can while still blending in. There's also a huge chunk of grifters hoping to win big through getting some startup cash and maybe being able to sell out to a big tech company or win some big corporate contract.

but I think you'd agree that once AI is completely normalised, it's the AAA company which could hire an artist but wants to cut corners that will turn to AI, rather than the small dev who is more likely to stick to their ideological guns

Honestly I'd say it's a toss up: AAA companies have absurd budgets and employ small armies of artists and techs to the point that they can afford to be indulgent and try to compete with each other on quality as a prestige thing, but they can also just as easily cannibalize themselves and chase the minimum viable product they can get away with; similarly indie studios can be extremely dedicated and indulgent within their resources, or they can be running on a shoestring budget and trying to get by with stock assets anywhere they can't cover with their own personal labor. The reasons why or why not may be different, but I don't see the products of AI as meaningfully distinct from stock asset libraries there - better in some ways, potentially worse in others, and ultimately dependent on how and why they're used.

it still (as it stands) relies on the work of people outside those teams, who must necessarily have their work fed to the AI in order for it to be useful - we can talk about the degrees of modification here, and we can draw comparisons to things like collage and sampling, and perhaps in that intentionality is some sort of answer but i also can't lie and say it feels good to me. Maybe there's some path to genuinely ethical AI, maybe an actually novel AI comes through and this is just no longer a concern whatsoever, but from here there still seems to be a lot of work to do before we approach that point

The key thing there is to think about how that meshes with proprietary models trained on licensed material or material owned by the company in question: if openAI or google or adobe pay someone like reddit or imgur or deviantart or whoever for the right to train on content they host, does that make the end result more ethical? If Disney has a model trained on its own properties, is that an ethical generator? If a huge company were to hire a bunch of artists to produce enough material for it to be trained, would that machine be ethical? Of course not, because these would all be controlled by the corporations and put to the same ruinous ends.

And that's what the IP angle is for: it's a way for big media hosts and property holders to attach themselves to the bubble and extract wealth from it through licensing fees or other agreements, while at the same time laundering the effects. Because that's the conclusion of that that they're angling for: the idea that properly licensed training data does make the model ethical and ok regardless of its use. The labor crushing machine is fine and dandy as long as property rights are respected and the right people get to own it, that's what the whole media push about "AI stealing art" is for.

So I just reject it out of hand. The corporations are trying to enclose all of human culture and turn it into a neat little commodity to exploit, and that enclosure relies on property rights. Simply not taking part doesn't stop them, and seizing upon tools outside of their control to try to compete doesn't help them. If there is stolen surplus value embedded in the very being of those tools, I don't think that affects the ethicality of using them once they are already made. All that matters is how and why you are using it.