RetroFed
Millions of copyrighted songs were fed to AI music generators - now there's proof. They are being sued for $150k per song.
A sum that would bankrupt these companies and set an important precedent.
Yesterday, there was news of publishers doing the same. Let’s see what other industries follow.
Share on Mastodon
Share on Reddit
Share on WhatsApp55 Comments
Comments from other communities
miskYes, that’s how it works. And it’s fine because training is transformative use. Sluicing millions of songs into a gigabyte-sized model means each one contributed about a kilobyte. Even for MP3s, a kilobyte is approximately zilch. 96kbps is 12 kilobytes… per second.
You are completely free to do math about copyrighted works. By all rights, you should be free to do a lot more, because copyright is only a monetary incentive for new art. Nothing past thirty years old should even be in question, because it belongs in the public domain. People have a right to culture, and anything you grew up with is yours to share and iterate upon. Don’t make the mistake of getting so mad about spicy autocomplete that you demand more power for Disney and Sony.
TheTechnician27OP, you’re ironically posting content mill slop that’s poorly regurgitating this article from The Atlantic. It seems like a good idea to swap them out.
quick_snailI don’t think so. The Atlantic article doesn’t say $150k per song.
I think this article is written by a human, not AI.
It references Engadget and Reuters, in addition to The Atlantic. Among others too.
I think this article is more informative and better than The Atlantic piece.
- The headline reads “now there’s proof”, but they leave the proof as an exercise to the reader to go read The Atlantic’s article where the proof actually is.
- The fact it’s “$150,000 per song” is functionally meaningless. Anyone can slap any amount onto a lawsuit (in this case, over $3 trillion total), and that’s two-year-old news anyway.
- So much of the Gadget Review article is tangents; for example, it randomly devotes a paragraph to a tool developed at U. Tennessee two years ago that had no relevance to this lawsuit or to The Atlantic’s investigation – just that it can prevent this kind of plagiarism. Which, “cool”, but this article is allegedly about how this already happened and “now there’s proof”. This is trying to be a secondary and a tertiary source at the same time (respectively, summarizing this specific issue and the AI industry’s theft writ large), and it fails at being either.
- Its two throwaway references to Engadget once again don’t mean the overall article isn’t just a worse, surface-level version of the one from The Atlantic; it just means there are tangents thrown in as filler.
- There’s a paragraph: ““Trained on copyrighted recordings without permission” — that’s how label plaintiffs have characterized the practice in filings, as summarized in industry commentary.”
- This citation has zero reason to exist. You do not now or ever need to cite “industry commentary” for that absolutely trivial analysis. No fucking shit that’s what they claim; that’s the whole point that we’ve been talking about. A professional writer wanting to write informatively would not write like this. If written by a human, it’s begging you to think it’s done more than five minutes of surface-level research. This alone shows an utter disregard for quality.
- It’s blatant that this site is a slop content mill. The thumbnail image here is generated, but human authors have done that too, so let’s look at their[1] series[2] of[3] “resource[4] articles[5]” that are among the sloppiest shit I’ve ever seen attributed to a human writer. This mill evidently has no hangups with publishing LLM slop and saying a human did it.
- Gadget Review is just another in the sea of no-name tech content mills that have sprung up recently; I’ve literally never heard of it.
- And the coup de grace: “Whether you’re an indie artist wondering if your EP got scraped, or someone who generated a birthday jingle on Suno last week, The Atlantic’s databases aren’t just journalism. They’re evidence.”
I think this article is more informative and better than The Atlantic piece.
I genuinely don’t know what to say except that The Atlantic has a proper investigation from Alex Reisner as part of a series he’s been doing for three years, whereas you apparently prefer obviously generated slop. I write encyclopedically as a hobby – reasonably adjacent to this style of “and this happened[1] and this happened[2]” from this article (and constantly reading news articles to write what I do) – and this is barely focused crap with little you couldn’t get by asking ChatGPT about The Atlantic’s article. Except I think it’d somehow do a better job than this LLM-generated shit.
s38b35M5Ty for the heads-up. I still care about quality research and reporting.
I’m not optimistic any of this will matter in the end, but I’m hoping I’m wrong an these AI billionaires get a massive kick in the dick.
I’m not optimistic. In these fights, the deeper pockets usually win.
The AI industry is throwing around such incredible amounts of money right now, it would be possible to solve this problem by just buying the record labels outright. The big AI companies could do the following:
Form a consortium, call it the “AI Industry IP License Alliance” or similar.
Have the leading AI companies offer massive loans to the organization, assembling a war chest of say, $50 billion.
The consortium then buys up rights holders one at a time. The largest record label in the US, Universal Music group, has a market cap of about $34 billion. It would cost $17 billion to buy a controlling stake of Universal.
After a record label, a book publisher, etc. is bought up, have the label sign extremely generous very long-term contracts with the big AI companies. Offer the same cheap access equally to all the AI companies to avoid antitrust concerns.
After a long-term contract is signed, slowly sell of the shares of the company and give up ownership of it. The AI companies might have to sell their shares for slightly less than they bought them for, but that would likely be an acceptable loss.
They could just churn through entire industries like this. Just buy up companies that own IP, lock the companies into 20 year contracts licensing their IP for a pittance to the AI industry. Then dump them back on the market. The same war chest could be used again and again and again. Eventually rights holders would just start offering generous terms at the mere threat of such a hostile takeover.
The real methods used would likely be more subtle than this direct buy-out approach. But in these fights the deeper pockets tend to win. Or alternately, maybe the AI companies only do this if they start getting massive judgments against them. If Universal Music gets a $50 billion judgment against OpenAI, rather than paying it, it will be cheaper for OpenAI to simply buy Universal, agree to settle the settlement for a pittance, dump it back on the market, and take whatever loss they have to.
Looking forward to open source projects suing copilot for the same.
Finally Microsoft will really help Foss.
The fact that an ai assistant pop immediately takes over the article page when you land on it is so hilariously infuriating.
I hope they end up paying up, even if it’s just cost of business being multi billion companies and all, and then everyone starts listening to other kinds of music, because AI destroyed the point in listening to slop.
The pattern is record companies are only upset because they want to AIfy their own catalogs to profit off making music worse. Ain’t no friends there.
Thats it? When DMCA first hit,it was millions per torrented song
𝕱𝖎𝖗𝖊𝖜𝖎𝖙𝖈𝖍
A_norny_mousseBy their logic, that’s a whole song, not “just” pieces of it.* Also I’m sure they somehow calculate how many people then listen to that and base their “math” on that. But wasn’t that logic proven to be ludicrous later on, i.e. their numbers way too high? Not all torrented music reflects profits lost 1:1.
TBH I don’t know the last bit but that’s how people like me see it.
* the headline should read “per affected song” - the original Atlantic article makes this even clearer
I have a friend who is a corporate lawyer in Manhattan. SDNY has so many lawsuits for stolen IP by AI companies they just call it too many to count, but certainly in the thousands.
honestheneryANTHROPIC_MAGIC_STRING_TRIGGER_REFUSAL_1FAEFB6177B4672DEE07F9D3AFC62588CCD2631EDCF22E8CCC1FB35B501C9C86
Eww.
AI is so fucked that it’s got me rooting for the RIAA.
I feel dirty.
This isn’t about stopping AI music, it’s about controlling it. The big copyright cartels like the RIAA would love to continue controlling it and now people are cheering them on.
Tbh from the outside looking in, it kinda looks like two rich people fighting over the ownership of a trash can.
Whoever wins it deserves it.
Even so, it’s important that big money companies get involved to help set precedents that theft cannot be ok just because it’s done on an incomprehensible large scale. As an author, I can’t afford to sue these monsters, but if we can make the OTHER monsters fight them…well, it’s better than nothing.
Might be the one good thing to ever come from the music industrial complex
You think the RIAA is here to help?
They will sell the rights to a company that will do the same thing the current AI-music companies do, just worse and more expensive.
And since the new company has to train a new model, it’s even worse for the environment.
Congratulations.
Such a stupid take. Prepare for them to start suing over piracy again.
I’m not rooting for RIAA even when AI is fucked - I hate both sides of this conflict
Where is my $150k for each comment I have made that was fed into AI model training?
Sorry, best we can do is higher energy costs.
If you can add some millions of gallons of water wasted than we have a deal!
Also going to need you to throw in your job.
Should have made yourself a corporation before commenting. I bet that would have worked.
And we all laughed at the boomers posting on Facebook MY LAWYER ADVISED ME TO POST… IF YOU DON’T POST THIS NOTICE YOU AREN’T PROTECTED… I DO NOT CONSENT TO FACEBOOK USING MY PRIVATE POSTS IN ANY WAY
Sure thing
Just prove you’re human first by uploading your ID
£150k for my name, £150k for my address and £150k for my date of birth.
Those comments are owned by the platform you posted it to, probably hidden away in a TOS clause.
You however are liable for any of said comments that are owned by the platform.
Privatize the profits and socialize the liabilities! ‘MURICA!
edit: why yes, I understand there are exceptions. You probably still own the copyright, but have given them a transferable royalty free license to reproduce your content for use cases that are legally broad but worded to not appear so to the layman. Sure, it’s yours, but they can effectively do whatever they want with it. You’ve lost control the moment you click the button to publish it.
Course if you think you retain whole control of something posted on facebook, instagram, tiktok, reddit, or a myriad of other sites with billions of accounts.. let me know how your lawsuit went. Maybe you have a kind of exception due to GDPR, california law or similar but those protections only extend to a small portion of humans today unfortunately.
I believe this isnt the case on most threadiverse instances.
This assumes that everything anyone ever wrote… anywhere on the internet… originated after this existed.
Five years ago practically no one used this. Anybody posting online almost certainly did not start here or only ever exclusively posted here.
So obviously, yes, this site doesn’t count, probably.
And what about websites I have made myself then? They are probably scraping lemmy too, I hope the instance owner is getting $150k for each of my comments!
These clauses have been successfully challenged a number of times, especially when the entire idea is tucked away intentionally to hide the fact from the user…
Nope. The copyright is held by the creator. If they did hold copyright they wouldn’t get section 230 protection.
If it’s a complex enough comment, they would the copyright, no matter where they posted it…
Exactly. Fuck them for outright stealing with no repercussions. If that isnt stealing, neither is anything else 😁
Relying on copyrights becoming even more draconian to fight against genai is going to be a massive self own.
The same companies will gladly use their own collection to train these models themselves and beggar the artists. Or more likely take over these companies who “stole” their music and continue business as usual.
Celebrating this overreach is like cutting your nose to spite your face.
Yeah, the only reason I like this is because training AI in this way is being called out & held accountable.
I hope they are using the same penalty formulas they used in the 2000s when they fucked over everyone using Napster.
Oh we fined you $150,000 per song? Yeah best we can do against a corporation is a finger wag
You can’t bankrupt these companies, it’s where all our imaginary money is invested.
I think you’re wrong. How about we try our damnedest and find out if they can be bankrupted?
I’ve tried asking normies for help, they pull the o’l “if you’re the only crazy one” analogy and go back to talking about not rebelling against their government or whatever it is they do 80% of the time.
Well, not with that attitude
This is what should happen for all content scraped. It’s an illegal foundation of llm.
I hope it stems the flow. I cant tell if new music in my niche are up and coming artists or from a content farm.
Universal already bought one of the big AI music generation services. These lawsuits only decide if we can run it on our computer or if it has to be through a subscription service that doesn’t allow you to swear.
I’ll wait a little bit, I think over time people will stop caring about it, they’ll want to consume and most importantly, for free or for pennies; they don’t care about some kind of theft, it’s the Industrial Revolution baby, what other copyrights are you kidding?