Practically speaking, who is going to enforce such a regime? Do you really want to give Chinese companies such a huge competitive advantage, that they aren't subject to the same costs as western companies? How do you even sort out which "creators" are owed, and how much? It's next to impossible, and would drown the legal system in litigation; it would likely cause more problems than it solves. On top of which you can find open weights for most, if not all, of the scraped material already. If you make those illegal to use, or prohibitively expensive, you just destroyed local LLM legality, and put the technology firmly in the hands of only the monopolists.
If models are trained on the collective whole, they must be owned by the collective whole. If you believe funding creators for the training of private models is too slow, inconvenient, or creates a global disadvantage, then embrace collective ownership.
Sure, I wish everything was perfectly fair too. But how do you practically and REALISTICALLY proceed, and ensure you don't end up doing more damage than benefit? The road to hell is paved with good intentions. Everyone seems much more focused on complaining, and talking about "what's fair", than actually proposing concrete steps that would lead to a better world, without a significant risk of creating a worse one.
> concrete steps
Start by legally compelling companies that trained on unlicensed data to either (1) license the data, (2) publish their model, or (3) destroy their model.
> Start by legally compelling companies that trained on unlicensed data to either (1) license the data, (2) publish their model, or (3) destroy their model.
You are lost in an imaginary world where everything is simple and has no negative consequences. First off, there is NOBODY who has that power over all the companies in the world. So immediately you are creating an imbalance between companies and potentially destroying your domestic industry; with long term negative consequences for the people you're supposed to be protecting. Secondly, you might be creating a situation where it's impossible to ever create a competitor to those companies who are already entrenched monopolists, potentially even making it impossible to ever run self-trained or local LLM's. Also, you just unilaterally made it legal to publish all copyrighted work (since that's what you believe their model to be) to the general public, presumably in a way that can be used by everyone; further eroding copyright law in one fell swoop. You've completely disregarded the legal issues around what constitutes "unlicensed data", and how much is required before triggering your new law, and what that would mean for the legal system potentially being inundated. You're reacting way too emotionally and flippantly, with no apparent thought about what harm you are doing and how you might actually be making things worse, not better.