this post was submitted on 26 Jul 2023
483 points (96.0% liked)
Technology
59590 readers
5616 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
How can they prove that not some abstract public data has been used to train algorithms, but their particular intellectual property?
Well, if you ask e.g. ChatGPT for the lyrics to a song or page after page of a book, and it spits them out 1:1 correct, you could assume that it must have had access to the original.
Or at least excerpts from it. But even then, it's one thing for a person to put up a quote from their favourite book on their blog, and a completely different thing for a private company to use that data to train a model, and then sell it.
Even more so, if you consider that the LLMs are marketed to replace the authors.
Yeah which I still feel is utterly ridiculous. I love the idea of AI tools to assist with things, but as a complete replacement? No thank you.
I enjoy using things like SynthesizerV and VOCALOID because my own voice is pretty meh and my singing skills aren't there. It's fun to explore the voices, and learn how to use the tools. That doesn't mean I'd like to see all singers replaced with synthesized versions. I view SynthV and the like as instruments, not much more.
I've used LLVMs to proofread stuff, and help me rephrase letters and such, but I'd never hire an editor to do such small tasks for me anyway. The result has always required editing anyway, because the LLVMs have a tendency to make stuff up.
Cases like that I don't see a huge problem with. At my workplace though they're talking about generating entire application layouts and codebases with AI and, being in charge of the AI evaluation project, the tech just isn't there yet. You can in a sense use AI to make entire projects, but it'll generate gnarly unmaintainable rubbish. You need a human hand in there to guide it.
Otherwise you end up with garbage websites with endlessly generated AI content, that can easily be manipulated by third party actors.
Can it recreate anything 1:1? When both my wife and I tried to get them to do that they would refuse, and if pushed they would fail horribly.
This is what I got. Looks pretty 1:1 for me.
Hilarious that it started with just "Buddy", like you'd be happy with only the first word.
Yeah, for some reason it does that a lot when I ask it for copyrighted stuff.
As if it knew it wasn't supposed to output that.
To be fair you'd get the same result easier by just googling "we will rock you lyrics"
How is chatgpt knowing the lyrics to that song different from a website that just tells you the lyrics of the song?
Two points:
Google spitting out the lyrics isn't ok from a copyright standpoint either. The reason why songwriters/singers/music companies don't sue people who publish lyrics (even though they totally could) is because no damages. They sell music, so the lyrics being published for free doesn't hurt their music business and it also doesn't hurt their songwriting business. Other types of copyright infringement that musicians/music companies care about are heavily policed, also on Google.
Content generation AI has a different use case, and it could totally hurt both of these businesses. My test from above that got it to spit out the lyrics verbatim shows, that the AI did indeed use copyrighted works for it's training. Now I can ask GPT to generate lyrics in the style of Queen, and it will basically perform the song texter's job. This can easily be done on a commercial scale, replacing the very human that has written these song texts. Now take this a step further and take a voice-generating AI (of which there are many), which was similarly trained on copyrighted audio samples of Freddie Mercury. Then add to the mix a music-generating AI, also fed with works of Queen, and now you have a machine capable of generating fake Queen songs based directly on Queen's works. You can do the very same with other types of media as well.
And this is where the real conflict comes from.
I don't know if that's true. If Google grabs that book from a pirate site. Then publishes the work as search results. ChatGPT grabs the work from Google results and cobbles it back together as the original.
Who's at fault?
I don't think it's a straight forward ChatGPT can reproduce the work therefore it stole it.
Both are at fault: Google for distributing pirated material and OpenAI for using said material for financial gain.
Copyright doesn't work like that. Say I sell you the rights to Thriller by Michael Jackson. You might not know that I don't have the rights. But even if you bought the rights from me, whoever actually has the rights is totally in their legal right to sue you, because you never actually purchased any rights.
So if ChatGPT ripps it off Google who ripped it off a pirate site, then everyone in that chain who reproduced copyrighted works without permission from the copyright owners is liable for the damages caused by their unpermitted reproduction.
It's literally the same as downloading something from a pirate site doesn't make it legal, just because someone ripped it before you.
That's a terrible example because under copyright law downloading a pirated thing isn't actually illegal. It's the distribution that is illegal (uploading).
Yes, downloading is illegal, and the media is still an illegally obtained copy. It's just never prosecuted, because the damages are miniscule if you just download. They can only fine you for the amount of damages you caused by violating the copyright.
If you upload to 10k people, they can claim that everyone of them would have paid for it, so the damages are (if one copy is worth €30) ~€300k. That's a lot of money and totally worth the lawsuit.
On the other hand, if you just download, the damages are just the value of one copy (in this case €30). That's so miniscule, that even having a lawyer write a letter is more expensive.
But that's totally besides the point. OpenAI didn't just download, they replicate. Which is causing massive damages, especially to the original artists, which in many cases are now not hired any more, since ChatGPT replaces them.
there are a lot of possible ways to audit an AI for copyrighted works, several of which have been proposed in the comments here, but what this could lead to is laws requiring an accounting log of all material that has been used to train an AI as well as all copyrights and compensation, etc.
Personally speaking, I've generated some stupid images like different cities covered in baked beans and have had crude watermarks generate with them where they were decipherable enough that I could find some of the source images used to train the ai. When it comes to photo realistic image generation, if all the ai does is mildly tweak the watermark then it's not too hard to trace back.
All but a very small few generative AI programs use completely destructive methods to create their models. There is no way to recover the training images outside of infantesimally small random chance.
What you are seeing is the AI recognising that images of the sort you are asking for generally include watermarks, and creating one of its own.
Do you have examples? It should only happen in case of overfitting, i.e. too many identical image for the same subject
Here's one I generated and an image from the photographer. Prompt was Charleston SC covered in baked beans lol
Out of curiosity what model did you use?
I think that to protect creators they either need to be transparent about all content used to train the AI (highly unlikely) or have a disclaimer of liability, wherein if original content has been used is training of AI then the Original Content creator who have standing for legal action.
The only other alternative would be to insure that the AI specifically avoid copyright or trademarked content going back to a certain date.
Why a certain date? That feels arbitrary
I'd think that given the nature of the language models and how the whole AI thing tends to work, an author can pluck a unique sentence from one of their works, ask AI to write something about that, and if AI somehow 'magically' writes out an entire paragraph or even chapter of the author's original work, well tada, AI ripped them off.
Not without some seriously invasive warrants! Ones that will never be granted for an intellectual property case.
Intellectual property is an outdated concept. It used to exist so wealthier outfits couldn't copy your work at scale and muscle you out of an industry you were championing.
It simply does not work the way it was intended. As technology spreads, the barrier for entry into most industries wherein intellectual property is important has been all but demolished.
i.e. 50 years ago: your song that your band performed is great. I have a recording studio and am gonna steal it muahahaha.
Today: "anyone have an audio interface I can borrow so my band can record, mix, master, and release this track?"
Intellectual property ignores the fact that, idk, Issac Newton and Gottfried Wilhelm Leibniz both independently invented calculus at the same time on opposite ends of a disconnected globe. That is to say, intellectual property doesn't exist.
Ever opened a post to make a witty comment to find someone else already made the same witty comment? Yeah. It's like that.
Spoken by someone who has never had something you've worked years on, be stolen.
Spoken like someone who is having trouble admitting they're standing on the shoulders of Giants.
I don't expect a nuanced response from you, nor will I waste time with folks who can't be bothered to respond in any form beyond attack, nor do I expect you to watch this
Intellectual property died with the advent of the internet. It's now just a way for the wealthy to remain wealthy.
Here is an alternative Piped link(s): https://piped.video/PJSTFzhs1O4
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I'm open-source, check me out at GitHub.
No dude you don’t understand. A copyrighted book is the same as an internet comment.
I think you said this facetiously... but it literally is.
https://www.howtogeek.com/310158/are-other-people-allowed-to-use-my-tweets/
This may come as a surprise to you but there are internet comments outside of twitter that aren’t protected by twitter rules.
Also you can’t copy the idea of a book and reword it. Which you can do with a tweet.
Copyright isn't Twitter rules...
Then why post a link dealing specifically with twitters terms of use agreement? There is nothing copywritten about this comment. Anyone can copy and paste this without attributing me and there is no recourse for me. If someone copy and pastes a published book, they can sue. How can they possibly be the same thing..?
Okay, seems you need help reading. So let me take DIRECT quotes.
I mean That's literally paragraphs worth of content telling you that YOUR CONTENT IS COVERED BY COPYRIGHT. The whole of twitters Terms of service is you granting twitter a perpetual license to YOUR CONTENT.
But right... Nothing copyrighted about comment content... and yet they mention it over and over that it is covered? Are you okay?
Yes your comment is covered under copyright. Including the ones you just made.
Covered under copyright != how easy it would be to claim damages.
It literally is.
https://law.stackexchange.com/questions/16680/are-comments-posted-on-websites-owned-by-the-website-or-the-commenter
We can even look at this from the opposite direction. Here's a full list of works that cannot be Copyright.
https://www.dmlp.org/legal-guide/works-not-covered-copyright
Notice that "comments on the internet" isn't one of them.
Ok so this is the point I’m trying to understand. If there is no recourse for someone to claim damages, in this case as internet comments because otherwise everyone would be claiming damages any time someone ever copied the text of their original comment, then what would even be the point of listing it as something that “cannot be copyrighted”? It’s like saying, well, the text on this post -it note is copyrighted because “text on post-it notes” isn’t listed specifically as works that cannot be copyrighted.
What was “stolen” from you and how?