Comment by 101008
1 day ago
The problem is not AI per se (which are only a mix of algorithms). The problem is that this new wave of AI is trained in propietary content, and the owners/creators didn't allow it in the first place.
If this AI worked without training, no one would say anything.
> If this AI worked without training, no one would say anything.
I don’t believe that for one second.
People are rightfully scared of professional and economic disruption. OMG training is just a convenient bit of rhetoric to establish the moral high ground. If and when AIs appear that are entirely trained on public domain and synthetic data, there will be some other moral argument.
Yeah I'm not interested in "art" created by a computer. A watercolor by a first-grader is more interesting.
Same goes for music. If you need AI and autotune, find another way to earn a living.
So you need to know the provenance of art before you can decide if it’s interesting?
Do you think the Lord of the Rings movies were bad art?
Yea, it definitely is just a convenient argument to people that feel threatened. I in no way feel as though the same internet that has so consistently disregarded copyright laws with such reckless abandon is now clutching their pearls about this.
Seriously. What percentage of these pearl-clutchers were mocking the MPAA and Lars Ulrich and supporting Napster as proof that art should be free?
It’s a high percentage. But time, increased personal wealth, and “OMG this might affect me” all have a lot of power.
People would still be griping about how it devalues the hard work artists have put in, "isn't real art" and all the other things. The only difference is the public at large would be telling them to put a sock in it, rather than having some sympathy because of deceptive articles about how big tech is stealing from hardworking artists.
Yes they're two different issues from AI:
- LLMs were trained on copy protected content and devaluing the input a worker puts into creating original work
- LLMs are a tool for generating statistical variations and refinements of work, this doesn't devalue the input but makes generating output easier
Form vs Function issues. So it would be preferable to give people a legal pathway to continue making money and own their work instead of allowing their work to be vacuumed up by the people at corporations looking to automate them away. The functional issue still exists but doesn't put your personal work at risk of theft/abuse outside of it's economic intent. Then the social stigma doesn't really matter because "an LLM is just a tool" is now a solid argument not causing abuse or deterioration of existing legal protections.
their consent was not required. https://en.wikipedia.org/wiki/Transformative_use
petabytes of training data are transformed into mere gigabytes of model weights. no existing copyright laws are violated. until new laws declare that permission is required, this is a non-argument.
>If this AI worked without training, no one would say anything.
adobe firefly was trained on licensed content, and rest assured, the anti-AI zealots don't give it a pass.
the copyright is just one of the many angles they use to decry the thing that threatens their jobs.
There is no final word on the matter yet and there are counterpoints to the "Transformative use" argument.
https://www.reuters.com/legal/litigation/judge-meta-case-wei...
> "You have companies using copyright-protected material to create a product that is capable of producing an infinite number of competing products," Chhabria told Meta's attorneys. "You are dramatically changing, you might even say obliterating, the market for that person's work, and you're saying that you don't even have to pay a license to that person."
> "I just don't understand how that can be fair use," Chhabria said.
https://ipwatchdog.com/2025/05/12/copyright-office-weighs-ai...
> Stylistic imitation even without substantial similarity would likely be implicated under such a [market-dilution] theory, which could be considered as a market effect under factor four that diminishes the value of the original work used to train the model.
that's one sympathetic judge's opinion vs written law.
I'm not American, but it's clear to me that it will be the supreme court that ultimately decides whether licensing is necessary or not. the parties involved - the megacorps with infinite money and the litigious publishers - won't settle for less.
and given that the ruling in favor of the publishers would all but kill the American AI efforts (good luck licensing millions of works needed to train a coherent model) and greatly benefit China (who doesn't give a fuck about IP), I find it highly likely that it will not happen.
I don't know how they verify it, but the article claims the model mentioned ("Moonvalley") trained an entirely clean/licensed data model.