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

X changes its terms to bar training of AI models using its content

techcrunch.com

Post reply on HN
Print view View original

131–140 of 217 posts

Re: X changes its terms to bar training of AI models using its content

#131

Post by pergadad »

visarga wrote: post #96

Copyright is not going well. The rights of millions of people are trampled by companies, both the content we post on social networks and our private AI chats. Our voice doesn't matter. Copyright was supposed to protect expression and keep ideas freely circulating. But now it protects abstractions (see the Abstraction-Filtration-Comparison test). It is much more difficult to be sure you are not infringing.

Copyright has nothing to do with free expression but was intended to protect the interests of publishers. When the printing press arrived basically any popular book or booklet was quickly copied by others. This meant the original publisher (and sometimes the author, but usually they were paid one-off) saw nothing of the profit.

Re: X changes its terms to bar training of AI models using its content

#132

Post by GuB-42 »

tpxl wrote: post #75
loudmax wrote: post #61

Earlier quoted context omitted.

The current US copyright duration is 70 years after the life of the author. This is absolutely bonkers. 50 years from publication would be a significant improvement. 50 years ago was 1975. If copyright were limited to 50 years, we'd be looking at all of the Beatles works being in the public domain. We'd be midway though Led Zeppelin, and a lot of the best work from Pink Floyd and the Rolling Stones. Also, Superman, B…

> If there were no copyright protection, these works wouldn't have been created. Citation needed. You can freely copy and distribute linux and it still got made.

If you want a point, BSD is probably a better example. Linux is protected by copyright, that's what makes copyleft licenses like GPL possible.

BSD is also protected by copyright, but it matters less for permissive licenses. It still protects attribution (so you can't claim it yours), but it probably would have worked without it, unlike with Linux that is for a big part defined by the "copyleft" protections offered by its licence.

Re: X changes its terms to bar training of AI models using its content

#133

Post by seydor »

mk_stjames wrote: post #99
seydor wrote: post #81

VAT for content should be a thing. Ultimately all users should be getting paid

I wanted to do some quick math on this idea- supposed we trained a vanilla transformer model from scratch, as GPT2/GPT3 was done- the number of seen input tokens is known perfectly, as is the sources of those training tokens (since then, everyone has either kept quiet about the sources post-Books3-fiasco, or have been finetuning on top of previous models making this more difficult of a calculation) GPT-3 was trained…

the payment would probably be based on the usage of that source in generating LLM output for the LLM user. This would probably require training a parallel network that connects LLM network nodes to sources. Then the activation of those nodes could be a surrogate for the contribution of the source

Re: X changes its terms to bar training of AI models using its content

#134

Post by eru »

BoxOfRain wrote: post #115
tetris11 wrote: post #111

Earlier quoted context omitted.

That's a wild reference! https://en.wikipedia.org/wiki/Roko's_basilisk > Roko's basilisk is a thought experiment which states there could be an otherwise benevolent artificial superintelligence (AI) in the future that would punish anyone who knew of its potential existence but did not directly contribute to its advancement or development, in order to incentivize said advancement.

It's Pascal's Wager with a sci-fi reskin, and all the objections that go along with that.

Roko's Basilisk is very, very similar to Pascal's wager, but it has an extra wrinkle:

The Basilisk task you to with bringing the Basilisk into being. Pascal's wager merely asks you to believe (and perhaps do some rituals, like pray or whatever), but not to make the deity more likely.

Re: X changes its terms to bar training of AI models using its content

#135

Post by eru »

thih9 wrote: post #109

I think the rules should be stricter. I’d prefer an explicit opt in from the content author being required for anyone to perform any model training with any given data. Alternatively, require all weights, prompts and chat logs to have the same visibility as the original datasets. None of this is going to happen and current decisions about uncopyrightable ai[1] are already good; but still, it feels like there is room…

Well, you explicitly opt-in to Twitter ToS whenever you post anything there.

Re: X changes its terms to bar training of AI models using its content

#136

Post by ivape »

newsbinator wrote: post #127
ivape wrote: post #124

Earlier quoted context omitted.

X/Twitter has became extremely prohibitive with just about everything since Elon took over. Their API pricing was antagonistic toward even indie developers. Elon is not a generous guy.

> Elon is not a generous guy Why would he be?

It's kind of a "life arc" that gets fulfilled when you've done it all and have all the money in the world, and reach a certain age. It's a very traditional arc for a humane human being.

Re: X changes its terms to bar training of AI models using its content

#137

Post by eru »

gausswho wrote: post #34
Animats wrote: post #15

It would be interesting to have a "classical AI model", trained on the contents of the Harvard libraries before 1926 and now out of copyright.

It does surprise me that we haven't seen nations revise their copyright window back to something sensible in a play to seed their own nascent AI industry. The American founding fathers thought 20 years was enough. I'm sure there'd be repercussions in the banking system, but at some point it might be worth the trade.

What's the connection with the banking system?

Re: X changes its terms to bar training of AI models using its content

#138

Post by eru »

tpxl wrote: post #75
loudmax wrote: post #61

Earlier quoted context omitted.

The current US copyright duration is 70 years after the life of the author. This is absolutely bonkers. 50 years from publication would be a significant improvement. 50 years ago was 1975. If copyright were limited to 50 years, we'd be looking at all of the Beatles works being in the public domain. We'd be midway though Led Zeppelin, and a lot of the best work from Pink Floyd and the Rolling Stones. Also, Superman, B…

> If there were no copyright protection, these works wouldn't have been created. Citation needed. You can freely copy and distribute linux and it still got made.

Linux is under the GPL, which explicitly needs copyright to work.

Something like the BSD licenses approximates 'no copyright' better, perhaps? But also not completely.

Re: X changes its terms to bar training of AI models using its content

#139

Post by eru »

GuB-42 wrote: post #132
tpxl wrote: post #75

Earlier quoted context omitted.

> If there were no copyright protection, these works wouldn't have been created. Citation needed. You can freely copy and distribute linux and it still got made.

If you want a point, BSD is probably a better example. Linux is protected by copyright, that's what makes copyleft licenses like GPL possible. BSD is also protected by copyright, but it matters less for permissive licenses. It still protects attribution (so you can't claim it yours), but it probably would have worked without it, unlike with Linux that is for a big part defined by the "copyleft" protections offered by…

> It still protects attribution (so you can't claim it yours), but it probably would have worked without it, [...]

Well, you could imagine a world that protects the 'moral' rights of authors like attribution, but doesn't otherwise prohibit anyone from duplicating or modifying works.

Re: X changes its terms to bar training of AI models using its content

#140

Post by eru »

simiones wrote: post #121
tpxl wrote: post #75

Earlier quoted context omitted.

> If there were no copyright protection, these works wouldn't have been created. Citation needed. You can freely copy and distribute linux and it still got made.

I think Linus Torvalds has been very explicit that he believes the GPL has been critical to the success of Linux - specifically, the copyright-enforced obligation to contribute back any modifications you make. In a world without copyright, companies would be free to make their own modifications and keep them secret, making it more or less impossible to integrate them into a cohesive whole the way they are more or les…

GPL only forces you to contribute back a modification you make and publish.

> In a world without copyright, companies would be free to make their own modifications and keep them secret, making it more or less impossible to integrate them into a cohesive whole the way they are more or less forced to do today.

Private modifications that are never shared with a third party are fine with the GPL. Eg Google doesn't have to share whatever kernel they are using on their internal servers with you.

Post reply on HN