Any company that wants to stay arround has to charge at least the cost of doing buisness.

Not necessarily. Open source developers, non-profits, and independent researchers aren’t “companies.” There are operational costs, but sometimes those are covered by the university they work for, through grant funding, etc.

If we force big companies to pay licencing costs, they won’t be able to continue selling LLMs.

Of course they will. They’ll just make deals with large publishing platforms that will pay the big authors a lot of money and smaller authors pennies or nothing, similar to music streaming services or ASCAP. They just don’t want to because it means less profits.

The other part is that they can operate in a gray area where if they don’t disclose the source of their training data, you’d have to find your own evidence to prove that they used your works. Since the training data doesn’t exist in the models themselves, you’d need a disclosure of some kind.

Speaking of licencing costs: this only matters if copyright holders say “Yes”. If they say “No”, as they have a legal right to do so, then LLM companies start having an input problem.

Many publishers reserve rights to make licensing decisions for authors, so it won’t necessarily be the millions of authors. It’ll be Jeff Bezos saying, “sure, pay me and it’s yours.” There might be fallout, but Amazon has a bulk of the publish market already and is already boycotted by people who don’t like them and it hasn’t hurt their bottom line much.

Licencing costs also don’t just matter during training. It’s not at all unusual for licences to include royalties, meaning licence fees need to be paid out for the lifetime of the model, not just the training period.

No, that wouldn’t be something the LLMs would agree to and if the publishers or the authors want to get paid, they’ll have to compromise. There will be enough copyright holders who will be fine with it, so the LLM companies can just exclude the ones who have higher expectations. They might do it for a big name author, but they’re not paying royalties to an author who sold ten copies of a book to his friends and family members. They want to ingest as much as possible for better models, so the price point per each work or author will be minimal and there will be enough willing to get paid anything at all that there won’t be as much of a bargaining position. The big LLM companies are opportunistic tech bro hustler assholes. They’re not going to agree to terms that are too expensive for them. They’ll treat 99.99% of authors the same way they do gig economy workers.

SO lets say we started enforcing the law here:

While there’s an argument to be made for criminal copyright infringement, it’s more commonly a civil lawsuit, so we wouldn’t start enforcing the law, the affected plaintiffs would have to lawyer up and sue, which means they’re already out money unless the lawyers see enough certainty and dollar signs.

Enfringing models would have to be retired, since many of the copyright holders would refuse to authorize use of their content.

You’d have to prove which works were used first. You’d have to subpoena and get the training data, which may already be gone in some scenarios. And the companies know how to train models, they’ll just train them without the plaintiff’s work.

LLMs get even more expensive, and companies do the math. They replace the LLMs with humans and things can chill some.

Except the companies pushing LLMs don’t want the humans. They’re not looking for humans as a solution. They embrace LLMs because they perceive that humans are more expensive individually in the long term and more problematic. They want max profits. They don’t care about environmental impacts or climate change. Everything is an opportunity and every employee is a financial expenditure and liability. If you ban LLMs, they’ll just rename them “advance database scripts” and keep going until the next lawsuit they have to pay off.

Ethical Developers no longer need to compete with GROK, since GROK’s enfringement has caused it to be retired by legal means

This bill is only in California. If it becomes law and isn’t shot down in the courts (and it will most definitely be challenged by big money), developers will just move out of California. And this only hurts US firms. China won’t stop its development. Foreign developers can violate copyright all they want as long as their country’s government isn’t overly deferential to the US (and who is to say if Trump would worry about it unless he could just extract a bribe to let it go…).

The list of copywritten works allows work authors to skip proving their work was used in trainig, making their cases easier to build.

That’s ignoring how difficult it is to attribute works, which is covered in the article.

Try attributing my comments. You don’t know my full name. Maybe I’m quoting someone else and the copyright belongs to them. Maybe I’m repeating myself and the copyrighted content has a different date than you thought. What’s the name of the work? How would you find it in a list? List? You’d need a database with a really good search function. Who is hosting the database? Who is paying the costs of hosting the database? Is the developer or the state of California? Do I have to pay for a subscription to access it like a journal or a court document system?

I understand the desire to hold big corporations accountable, but we need societal and cultural changes, including changes to capitalism itself. LLMs pirating works is just one symptom. Wealthy corporations game the system such that they also game the courts and game even the few laws that seek to actually restrain them. And kneecapping open source developers by creating a cost of doing business that the wealthy corporations will definitely be able to afford is just giving the big LLM companies more of a leg up than they already have. They’ve asked in some cases to be regulated so that they can engage in regulatory capture and keep our competition.