@EUCommission Licensing and payment in full at best (i.e. for every time the model is accessed, as the corpus can't be separated from the model). Otherwise public domain or bust.

@EUCommission These AI fuckbois are not special. Stop giving them special treatment.

@EUCommission ARE YOU INSANE? They’ve stolen all the internet and instead of doing something you are asking people for “how do much do I let them keep fucking you, but this time legally?”

Wow. Just went through the process. What a byzantine mess.

@EUCommission when there are guarantees that said AI systems are chiefly committed to furthering human understanding and serve the public, instead of a for-profit hype bubble. control over the training process, and what to do with the resulting models should be put under democratic control (including whether to even start this project at all).

in other words: just about the opposite direction AI companies have been sprinting in for the past few years.

in practice, i am very burnt on genAI as a concept, and i really don't see why we would need to train more models. what would this hypothetically even add that, when in doubt, couldn't be served by existing open-weight models? if the answer is "nothing", then shouldn't we conserve resources and reuse those instead of throwing more and more precious resources into this bottomless pit? just take that money and fund FOSS and research instead, that will be cheaper and will give a much better return on investment in the long run.

I will happily sue for copyright infringement given the fact that the companies in question simply disregard robots.txt and licenses in public repositories. The sammelklagen are coming.

@EUCommission you’re too late, all copyrighted
content has already been used to train AI

@EUCommission

> Under what conditions should copyright-protected content be used for AI training?

Implies the only option is to accept machine learning bullshit training on others people work so technofacsists can "legally" ripoff people's work to make profit… 😒

But I guess, this shouldn't comme as a surprise, since the EU commission is sabotaging the GDPR for "AI training" on PII…

@EUCommission hey uhhhhh, if anyone asks all my downloads are purely for the purpose of training AI models. Also, I may have accidentally stored exact copies of every single frame of every single movie I’ve ever played on my computer in the frame buffer of my video card. Whoops, I hope I don’t get in trouble for piracy!

@EUCommission Like every other copyrighted work?

Only after explicit permission of the creators (not the right holders because they "merely bought a license to the art style" to phrase it how the software companies always do) AND with revenue sharing because otherwise artists can't survive and acquiring and using copyrighted materials without a license is piracy, no?

@EUCommission never without the permission of the copyright holder. We should consider this a specific right that can be granted by license. Also, ignoring the contents of a robots.txt file on a server should be a punishable offence.

@EUCommission This is such a bad form that I gave up after the first few questions. Does not seemlike genuine consultation.

All I want is:
1. people/companies pay for my books
or
2. similar scheme to where authors get paid for library use.

Anything more than that and a company pays for the rights in the same way as any other publisher (eg in negotiation with my publisher)

In short,
Input to LLM same as library use and I get paid.
ANY output: LLM company buys the rights.

@EUCommission Licensed specifically for AI training (I do not agree with the idea of letting works licensed under MIT, Creative Commons, *GPL, etc. as also being de-facto able to be used for AI training), and public domain works. Everything else should just fall under regular copyright violations if it's used for AI training.

AI training should be explicitly opt-in, not opt-out.

@EUCommission

We declare Anna's Archive, Library Genesis, and Sci-Hub public domain, and we reverse all of the anti-circumvention laws.