Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

California governor signs AI transparency bill into law

gov.ca.gov

Post reply on HN
Print view View original

201–210 of 232 posts

Re: California governor signs AI transparency bill into law

#201

Post by Retric »

otterley wrote: post #200
Retric wrote: post #198

Earlier quoted context omitted.

LLMs are being trained on published books a direct equivalent of records. People were able to get one to reproduce over 40% of Harry Potter and the Sorcerer’s stone word for word. https://arstechnica.com/features/2025/06/study-metas-llama-3... There’s zero chance that happened without the book being in their training corpus. Worse, there’s significant effort put into obscuring this.

Yes, they are trained on books, but the courts so far are largely in agreement that AI models are neither copies nor derivative works of the source materials. If you’re searching for legal protections against your works being used for model training, copyright law as written today does not appear to give you cover.

Stuff can take a long time to wind their way through the court system. The worst cases already failed but many are going strong, here’s a 1.5 billion dollar win for authors.

https://www.kron4.com/news/technology-ai/anthropic-copyright...

Re: California governor signs AI transparency bill into law

#202

Post by micromacrofoot »

egorfine wrote: post #182
micromacrofoot wrote: post #90

Earlier quoted context omitted.

well certainly not geoblock a lot of your customers maybe your situation is different, but if we geoblocked all of california we'd go out of business within a year

Yeah I am with you here, but what else is to be done if legislation is passed that is impossible to comply with? Like in UK.

The GDPR and other laws were really hammed up by consultants and lawyers looking to make a lot of money claiming it was really difficult to do right, but in my experience it's incredibly overblown.

Most of who's getting caught up in these laws are very large companies that could comply but consistently don't put forth effort to do so after repeated complaints. Even if you do fall under the eye of regulators (most won't ever) if you show that you're putting forth a good faith effort to comply it's not a big deal.

Re: California governor signs AI transparency bill into law

#203

Post by otterley »

Retric wrote: post #201
otterley wrote: post #200

Earlier quoted context omitted.

Yes, they are trained on books, but the courts so far are largely in agreement that AI models are neither copies nor derivative works of the source materials. If you’re searching for legal protections against your works being used for model training, copyright law as written today does not appear to give you cover.

Stuff can take a long time to wind their way through the court system. The worst cases already failed but many are going strong, here’s a 1.5 billion dollar win for authors. https://www.kron4.com/news/technology-ai/anthropic-copyright...

Anthropic and the authors settled over a portion of the case involving the unauthorized copying of works that were used to train the model. Obtaining the works is a step that happens before the training has begun.

“the authors alleged nearly half a million books had been illegally pirated to train AI chatbots...”

Finally, a settlement isn’t a “win” from a legal perspective. It’s money exchanged for dropping the case. In almost every settlement, there’s no admission of guilt or liability.

Re: California governor signs AI transparency bill into law

#204

Post by Retric »

ndriscoll wrote: post #185
Retric wrote: post #130

Earlier quoted context omitted.

It’s far more recent works that AI companies care about. They can’t copy 50 year old python, JavaScript, etc code because it simply doesn’t exist. There’s some 50 year old C code, but it’s no longer idiomatic and so it goes. Utility of older works drop off as science marches on and culture changes. The real secret of long copyright terms is they just don’t matter much. Steamboat Willy entered the public domain and fo…

Ostensibly copyright is there to increase economic incentives to make things it protects, and like you said, we can massively cut it down without affecting much there. So focusing on economic viability, set it to something like 15 years for code and 20-30 for everything else. Require registration for everything and source escrow for code and digital art to be granted copyright. That would give a wealth of code to tra…

Exclusively training on 15 year source code would make code generation significantly less useful as API’s change.

Economic viability and utility for AI training are closely linked. Exclude all written works including news articles etc from the last 25 years and your model will know nothing about Facebook etc.

It’s not as bad if you can exclude stuff from copyright and then use that, but your proposal would have obvious gaps like excluding works in progress.

Re: California governor signs AI transparency bill into law

#205

Post by Retric »

otterley wrote: post #203
Retric wrote: post #201

Earlier quoted context omitted.

Stuff can take a long time to wind their way through the court system. The worst cases already failed but many are going strong, here’s a 1.5 billion dollar win for authors. https://www.kron4.com/news/technology-ai/anthropic-copyright...

Anthropic and the authors settled over a portion of the case involving the unauthorized copying of works that were used to train the model. Obtaining the works is a step that happens before the training has begun. “the authors alleged nearly half a million books had been illegally pirated to train AI chatbots...” Finally, a settlement isn’t a “win” from a legal perspective. It’s money exchanged for dropping the case.…

The entire case settled, the authors aren’t going to appeal when the company can’t hand out much more than the 1.5 billion in question and the company isn’t allowed to use the works in question going forward.

Re: California governor signs AI transparency bill into law

#206

Post by CamperBob2 »

AlexandrB wrote: post #192
bluefirebrand wrote: post #163

Earlier quoted context omitted.

Then we trust nothing published after 2020 or so AI will have effectively frozen human progress in a snapshot of time because nothing created after LLMs became widely available will be trustworthy

Who is "we"? A lot of people don't care and will trust new AI generated stuff regardless of how wrong it is. What this means for human progress is uncertain.

And if they outcompete you, then what? I guess it'll be a little less uncertain at that point.

Re: California governor signs AI transparency bill into law

#207

Post by CamperBob2 »

drivebyhooting wrote: post #95
ronsor wrote: post #79

Earlier quoted context omitted.

Hackers shilling for copyright is a funny picture.

Information wants to be free, right? Except in this case after robbing humanity’s collective knowledge repository OpenAI and its ilk want to charge for access to it, and completely destroyed the economic incentive for any further human development.

Huh. Funny, my knowledge is still there after supposedly being "robbed," just like Walt Disney's movies and J. K. Rowling's books and Bill Gates's software and Lars Ulrich's music.

And "further human development" is exactly what's happening. We've just found the entrance to the next level. Our brains have gone about as far as they can on their own, just as our muscles did in the pre-industrial era. It's time to craft some new tools.

Re: California governor signs AI transparency bill into law

#208

Post by otterley »

Retric wrote: post #205
otterley wrote: post #203

Earlier quoted context omitted.

Anthropic and the authors settled over a portion of the case involving the unauthorized copying of works that were used to train the model. Obtaining the works is a step that happens before the training has begun. “the authors alleged nearly half a million books had been illegally pirated to train AI chatbots...” Finally, a settlement isn’t a “win” from a legal perspective. It’s money exchanged for dropping the case.…

The entire case settled, the authors aren’t going to appeal when the company can’t hand out much more than the 1.5 billion in question and the company isn’t allowed to use the works in question going forward.

Before the settlement was made, Judge Alsup found as a matter of law that the training stage constituted fair use.

https://storage.courtlistener.com/recap/gov.uscourts.cand.43...

Re: California governor signs AI transparency bill into law

#209

Post by votepaunchy »

array_key_first wrote: post #187
bloodyplonker22 wrote: post #25

Earlier quoted context omitted.

It solves a very real real world problem: putting more money into the hands of government officials.

This is, like, the stupidest and most inefficient way for a government to make money. I know it's fun and all to circle jerk about how greedy those darn bureaucrats are - but we're all aware they control the budget, right? They could just raise taxes. I don't think they're fining companies... sigh... 10,000 dollars as some sort of sneaky "haha gotcha!" scam they're running.

Raising taxes does not solve the problem of getting the money out to the right people.

Re: California governor signs AI transparency bill into law

#210

Post by Retric »

AnthonyMouse wrote: post #176
Retric wrote: post #130

Earlier quoted context omitted.

It’s far more recent works that AI companies care about. They can’t copy 50 year old python, JavaScript, etc code because it simply doesn’t exist. There’s some 50 year old C code, but it’s no longer idiomatic and so it goes. Utility of older works drop off as science marches on and culture changes. The real secret of long copyright terms is they just don’t matter much. Steamboat Willy entered the public domain and fo…

I still don't really get how compensation is supposed to work just based on the math. Models are trained on billions of works and have a lifetime of around a year; AI companies (e.g. Anthropic) have revenue in the low billions of dollars a year. Even if you took all of that -- leave nothing for salaries, hardware, utilities, to say nothing of profit -- and applied it to the works in the training data, it would be app…

I think you’re overestimating the number of authors, and forgetting there’s several AI companies. A revenue sharing agreement with 10% going to creators isn’t unrealistic.

Google’s revenue was 300 billion with 100 billion in profits last year, the AI industry may never reach that size but 1$/person on the planet is only 8 billion dollars, drop that to 70% of people are online so your down to 5.6 billion.

That’s assuming you’re counting books and individual Facebook posts in any language equally. More realistically there’s only 12k professional journalists in the US but they create a disproportionate amount of value for AI companies.

Post reply on HN