California bill (AB 412) would effectively ban open-source generative AI

[Jfdlsjfd]

And well, if someone does train models using only public domain, and donated material for creating open access models, I'm sure people will gladly help. Not a real problem unless they use petabytes of pirated material.

Well, the one that would be strengthened by this bill is probably Adobe, since its Firefly system was trained only on licensed and copyright-free material, before profitting open-source initiative. OS initiatives might even be the last to profit compared to company located in other countries with more AI-friendly legislations (I am thinking of Europe, where the producers of the best free-to-use image model, Flux, is headquartered, but especially China, which is probably the leading country in AI so far).

The F-lite model, a prototype base model open sourced and trained entirely with copyright-free image, is probably too young to verify your prediction about people gladly helping (it needs more training, but since its totally non profit, it will require people donating to improve it), but I didn't see a lot of support for the concept here in discussion on AI art so far.

 

[Necropolitan]

If enforcing existing copyright laws will put AI companies out of business that's the fault of the AI companies that decided to base their businesses on theft.

 

[Mistwell]

How exactly can you ban AI at a state level?

If you regulate it in California, where most of the AI companies are based, 1/7th of the nations GDP, and 1/8th of the population, that generally is enough to force the issue. They're not going to run one version for California and another for the rest of the nation, and other states and often nations follow once California passes such a law.

 

[trappedslider]

If you regulate it in California, where most of the AI companies are based, 1/7th of the nations GDP, and 1/8th of the population, that generally is enough to force the issue. They're not going to run one version for California and another for the rest of the nation, and other states and often nations follow once California passes such a law.

Google is incorporated in Delaware. So is Meta. DeepMind (Google's lab) is from the UK. Almost 0 of their data centers are in California. The training happens in Texas, Oklahoma, Tennessee, etc.

The only thing in California are offices.

 

[Scribe]

Many of the e-books there are from very old books (the reason why they are free), written in a language style and using words that nobody would have used for decades. On top free old e-books didn't mean they are written in a very good language quality anyway. Language has evolved a lot and using this data without processing/filtering it as training material does not make the AI better, but only worse, because language from the last 100+ years is wildly mixed together and the AI naturally cannot separate which language comes from which time, it mix it all together. LLMs have problems understanding the aspect of time anyway.

So what you are saying is, if you cannot rip the creative work of modern media which you/AI programs should not have access to, to have it regurgitated back to you in a way that makes you believe that a chatbot is actually being 'creative' despite knowing it is NOT creative, as it in your scenario would only be able to 'create' in prose that is not modern...that its worthless?

 

[Jfdlsjfd]

Google is incorporated in Delaware. So is Meta. DeepMind (Google's lab) is from the UK. Almost 0 of their data centers are in California. The training happens in Texas, Oklahoma, Tennessee, etc.

The only thing in California are offices.

And that's for language models. For image models...

Seedream is Chinese.
Stability AI is UK.
NoobAI is is Chinese
Black Forest is German
Alibaba is Chinese
Wan, Lumina, Hunyuan, Kling are Chinese.
Illustrious is Korean.
AuraFlow is Korean...

I am hard-pressed to identify a single California-made model outside of OpenAI's. Edit: and Firefly, which is already compliant.

 

[Kaodi]

If enforcing existing copyright laws will put AI companies out of business that's the fault of the AI companies that decided to base their businesses on theft.

Will be fun when we get to the "copyright law is too stifling" stage of this debate.

 

[trappedslider]

Will be fun when we get to the "copyright law is too stifling" stage of this debate.

I remember the outcry when Disney and others had copyright laws extended, I guess copy right law is okay when it's not a big company.

and now we can go right to the avenue of "big companies are evil and not your friend (unless you're a shareholder)" then right into the "capitalism is bad" dead end of the street.

 

[Mistwell]

Google is incorporated in Delaware. So is Meta. DeepMind (Google's lab) is from the UK. Almost 0 of their data centers are in California. The training happens in Texas, Oklahoma, Tennessee, etc.

The only thing in California are offices.

Where a company incorporated is irrelevant for this topic. That's purely a tax issue, not a regulation issue. Same with data centers (which are copied across the globe including in California for load balancing) or training, all irrelevant. Where the humans doing the work are located is all that matters, along with where the users are located.

There isn't any question a law passed in California about this will change it all. I'm not going to debate it with you. They're not shutting off AI for just users in California, nor making two versions. They will just change it to conform to California, like all major companies for all major products do.

 

[CleverNickName]

The issue isn't that smaller companies will be more affected than larger ones. That's a(nother) misdirection.

The issue is, and always has been, that AI was trained from stolen material, and the artists and writers are entitled to compensation for their stolen works.