Intellectual property isn't real. It's a fiction we constructed to try to control expression in order to allow extraction of profit from ideas. We had to keep exceptions like expiration and "fair use" to make it not absurd and obviously self-contradictory.
All LLMs are doing is shuffling around the roles to bring light to an underlying contradiction. Yes they are profiting off of unpaid labor, but what that actually means is the models themselves should be "property" of everyone just as the training data should be.
> Intellectual property isn't real. It's a fiction we constructed
Regardless of my opinion about IP in particular, argument "X isn't real, it's a fiction we constructed" is silly. We have "constructed" things like justice, duty, charity, mercy, and a lot of other social and moral constructs, and it's good that we did. They're also just as real as potential energy in physics: it's not a material object that you can see or touch, but it greatly affects what happens in reality.
Sure, but I would argue that those are concepts that have existed for millennia and has real material grounding in reality. Whereas intellectual property is entirely a fictional construction.
At its core, owning property involves the ability to use force to assert your control over it. This is completely impossible with ideas (and information more broadly) since they're non-physical, so it's not really property in the way real world property like land is.
So because it's not reflective of how the material world works, that's the heart of the contradiction I alluded to in my previous comment. There is no way to resolve the problem of LLMs from within the logical framework that doesn't lead to some further counterintuitive result. There has to be legislation around it if we want to keep the charade going, or ideally we'd want to drop it altogether.
If you share enough copies of a Disney movie, lots of men with big guns will come to your house and haul you away to a locked room. I fail to see how this isn't using force to control intellectual property, nor how it's impossible.
All LLMs are doing is shuffling around the roles to bring light to an underlying contradiction. Yes they are profiting off of unpaid labor, but what that actually means is the models themselves should be "property" of everyone just as the training data should be.