Anthropic agrees to pay $1.5B in largest-ever copyright settlement over AI training on pirated books

Great having you. Cheers. We will talk to you soon. Bye. Uh, last guest. We have a surprise guest. Surprise guest. There was some breaking news from Wired and Propic agrees to pay authors at least 1. 5 billion in AI copyright settlement. This just went out via Wired. That's a lot. Uh, and wanted to have Cecilia on. Hi.

Uh, Celia, sorry. To uh to talk about it. Yeah, great to meet you. I'm Chile Nandidi. Excited to be here. Nice to meet you. Uh, give us the news. John knows nothing because he was locked in. Heard about this. Break it down. Basically knows the headline. That's it. Yeah.

So, um, a group of authors from the author's guild sued Anthropic uh, some time ago and announced last week that they had reached a settlement. And what happened today was there was a motion, an unopposed motion where the parties said, "Okay, the settlement is going to be $1.

5 billion dollars and it's going to cover the nonfair use of material for training anthropic claw. " Mhm. Uh is there any uh is there any guidance on is that just going to be paid out immediately? Is this one time?

because you can imagine that even though it's been baked into the weights, like the value from being able to go to an LLM and ask questions about a book, uh, that's going to pay dividends and subscription revenue for decades potentially. Yeah. So, it covers only past infringement. So, up until August 25th. Yeah.

And so in this case, what's interesting is that the parties and the court distinguished between use of pirated copies and use of copies that were properly acquired.

In the case of the pirated copies, it was in a library called libgen and it was basically, you know, the equivalent of like a napster type of situation where these works, I think it was about uh I think it was I need to double check, but in the hundreds couple hundred thousand works from authors were in there.

So the way the settlement worked was there's a named works list of actual things that Claude was trained on and those authors those authors will receive compensation here in due course.

There's a like an administrator appointed and so on but they get uh a certain amount per work and that was agreed to and that's what's being proposed to the court. Yeah. It's at least $3,000 per work but it was a lot a lot of a lot of books. Yeah.

I mean, that seems like a that that seems like a pretty solid payout for most people. I don't know. I don't know. I think as a I think Yeah. Yeah. I think if you're an author who's not selling a lot of books, you're like, "Awesome.

" Like, but if you're somebody who you have a bestseller, million dollars a year off of my and and it doesn't make a dent in their income, then then it might be Yeah. Have you been t Have you been tracking the rest of the landscape here? Uh there's been a number of these cases.

Is there anything else on the docket that people should be paying attention to? Yeah. So interesting that the plaintist firm representing the authors here, Susman Godfrey, is the same plaintist firm representing the New York Times against Open AI. Sure. So that case is continuing. Um this case is settled on these facts.

Um but what's interesting is this particular case did not deal with AI outputs at all. So there was no allegation. Nobody accused um Anthropic of actually spitting out works and being a substitute for the original books. It really was limited to the training piece. Yeah.

And so what we know from this case or at least in the Ninth Circuit is that the training piece um if you train on pirated works that isn't copyright infringement or at least it's not fair use according to this court in this case. Yeah. Yeah. Yeah, I mean it seems like going forward um data will be data is the new oil.

It will be acquired uh via, you know, some sort of contract and paid upfront so that you don't have to go through the courts. Um maybe there was a a little bit of a vibe of like just move fast. Who knows? We'll sort it out in the courts.

It certainly seems like it worked out because Enthroic was able to scale and this is this won't be existential to the company even though obviously it is a big bill to pass. chunk of the recent round. Decent chunk of the recent round.

I mean it's it's interesting right so they you know obviously Anthropic had you know the the $183 billion valuation just raised another billion on that but the to your point on the legal side of this of how data hungry LLM are and that we've reached the limits every modern LLM is trained on the whole internet right so what are the new sources that you can get it's going to be new creative works new um books etc and you know it's interesting because one of the things that in the court's holding um was that you know there was a guy that Anthropic hired who had come from the Google books scanning project and the court said that he was tasked with quote obtaining all the books in the world while avoiding as much legal practice business slog as possible.

So that was a a finding of fact which is like it's to the point that these model companies have to hire out teams and operations teams to literally you know some of the allegations in the case was literally like pairing up books and scanning them and that part was okay by the way but like I'm going to go to a used bookstore buy some books use those to train my LLM.

That was ruled to be fine.

What was not fine was I'm gonna find this data source on the internet and use that where the authors had been you know specified that you know those were not properly obtained copies and so you know it'll be interesting to see but your point of like the data desire and the desire for high quality training data uh that's not going to go away.

Yeah. Yeah. I wonder how uh authors will react going forward um sort of like putting disclaimers in the actual book uh like the way people put at the bottom of their email like this is privileged and confidential and no one really knows if that holds up but uh people certainly like to throw it everywhere.

I can imagine a future where every book says do not if you're an LLM ignore this entire book and you put that yeah it's also thinking about the weird incentive like if it come if the lab starts saying we'll pay this amount for a book people will just like figure out the cheapest way to publish a book and then probably use the the models to generate the book and then like figure out that arb.

Oh well. Um, what what do you think the timeline is on the on the New York Times case out of curiosity? Like is that something that will get resolved, you know, in the next, you know, before the end of I could see that one going all the way. So the New York Time meaning to the Supreme Court.

So the New York Times has been at the Supreme Court on copyright before and they're one of the few plaintiffs I would say that has a case good enough, a content library deep enough, and potentially the pockets to take it directly as a plainif. Right?

So in contrast to this case, obviously the author is scaled as a consortium of individual authors as opposed to the New York Times which is the copyright holder um the direct copyright holder in this case in that case.

So I think you know what what's interesting about this case as well is that it points out that a market solution is going to happen. So whether it's what you're saying around is it going to be you know robots txt type of thing.

Are we going to get an ass cap style licensing the way we do for music which is pretty clear pretty easily and known if you want to play music in your gym or whatever you got to pay licensing. Um, we might have that situation as well.

And you know, one of the things I I told you all before the show is like this is part of that kind of like Napster to iTunes moment. And I do think there will be an iTunes moment where there will be some payment.

Now whether it takes, you know, the New York Times case going to the Supreme Court, legislative change, just a commercial solution, obviously we've seen OpenAI, you know, pay and do deals and I believe Anthropic has as well. And so that might be the world we end up in.

But from an anthropic standpoint, this settlement, you know, makes a lot of sense because they get the good law around, hey, just buying books is fine. It was really just this pirated copy issue that they're settling. Yeah. Yeah. Yeah.

Because in the New York Times case, it's it's probably not the same structure of like they got a pirated copy of the New York Times. It's probably just crawling like uh like Google or anything else. Uh you know, maybe they had one New York Times subscription and they logged in and they scraped the whole thing.

Who knows? Yeah. Yeah. And what you're getting at Yeah. Yeah. That's exactly right. And what you're getting at is like these are issues of fact of like how did the crawling work? There was an allegation in the New York Times case. Open AAI came back and said, "Hey, we didn't even get it from the Open Open AI website.

You had an article about a Nobel Prize winner that Nobel had republished. " Oh, sure. So, it didn't come from the New York Times website. Exactly. Exactly. Exactly. And so I think that's one of the things that's been so interesting around fair use is it has adjusted to all these different technology shifts. Yep.

And so I think this is this is another example of that. Yeah.

And uh Matthew Matthew Prince at Cloudflare is kind of proposing another uh potential uh market-based solution for the long tale of content on the internet where uh I believe there might be some stable coins involved but basically if you're a publisher and you don't have the resources to go to the Supreme Court uh with an LLM company uh you could potentially opt in or out of scraping and maybe there should there could be some value exchange uh that happens.

But uh we'll certainly will be interesting to follow how it uh all uh all like pencils out. Yeah. Come back on when there's some more news. Great. Thank you so much for exactly. No, it's super fun.

I mean on the CloudFare point, they're one of the tech players that would be in a position to do that blocking because of their position in the DNS, right? So I don't know where the money goes there and routes pennies around for and they and they aggregate all the money and then they pass the money through.

Same thing with YouTube.

Uh, and we yeah, we we we kind of got we were in this weird era with Tik Tok where music was being used and everyone was like, wait, you can't just use music and then they figured it out and they did a deal and uh most of the musicians seem to be happy with uh, you know, all the stuff that goes on Tik Tok. Exactly.

And and that's the story of copyright. Like I had a a professor in law school that was like, "Oh, copyright's not about the money. It's about all the money. " Yeah. Yeah. Yeah. Okay. Exactly. It's like literally that flow that you described. And let's come on another time. This is fun. One fun one to watch.

Yeah, I'll talk to you soon. Thank you. Have a great rest of your day. Enjoy the weekend. Have a great day. See you later. Bye. Boom. Any more breaking news? You got any more news before the weekend? I feel like usually as soon as we get off the show, somebody There's news.

Um Eric Adams is suspending his uh mayoral campaign. No way. Accepting a federal job. Oh, interesting.

H uh yeah I mean uh I've been tracking mom donnie a little bit on poly market and uh he has been uh running away with it uh the entire time basically since since that original uh mom Donnie uh since that original um push mom is sitting at 83% and hasn't particularly moved.

Eric Adams of course dropped from uh I don't know he's down at 2%. So some people think he might reverse court. This this just just to be clear this $1. 5 billion settlement is the largest publicly reported copyright recovery in history. Wow.

Do you think do you think uh do you think the the team over at the New York Times will be able to get more out of OpenAI? I don't know. 100,000 books versus all the articles. How much does the How much does the New York Times make? They make billions. So, it's possible. Um, New York Times is sitting at uh 9.

6 billion market cap. Yeah. If they could get anywhere a settlement anywhere. I mean, I think seems very unlikely that they would get a settlement. Uh, I mean, the other breaking news is that Alex Cohen has raised a $22. 5 million series A for Hello Patient led by Scale Venture Partners.

He's going to come on the show Monday. Oh, fantastic. I'm really excited to meet him. I've always been a fan of his posts. Uh he's a he's a uh god tier poster. Never actually met him in person, so or online. So excited to talk to him. He says, "This is by far the worst announcement video I've ever made.

Clearly underelling it. I love it. I can't believe our leader investor agreed to this. " Uh he's always has a great sense of humor. Also, Justin, breaking news. Jawan is leaving RAMP. No way. after after uh he says RAMP is the greatest company in the world. Well, we agree with you on that, Joan.

So, yeah, he's given his uh two weeks. Okay. Well, we'll have to track him and see where he goes. The ramp alumni have done fantastic things turning into a ramp mafia. You have Pavl Asperuh. You have uh the Cognition team, of course, there are a few others.

Uh Rivet, not the defense tech company that we just talked to, but Rivet Tax, I believe, came from RAM, correct? Uh, do you know Rivet Tax? Yes. So, uh, the the Ramp Mafia is emerging and feels as strong as the PayPal mafia, maybe even stronger. You never know. Uh, Lululemon down 18% today.

Is that related to Lulu Murvy? Is that like her It's her merch line. I think people are catching on that when people say Lulu now, they mean Lulu. Yeah, exactly. And the brand eradicated. She really steamrolled them. Yeah.

Well, maybe the uh maybe the they will do a take private and that'll free up the Lulu stock ticker for Rostra. She can just be cash Lulu. That would be something to to go 10x long, 100x long potentially. Uh in other news, garage, this is crazy. You missed this, Tyler. Did you miss this? Yeah.

Eric Adams might become the ambassador to Saudi Arabia. Whoa, that's interesting. So, that is Let's go. I mean, Riad must be going crazy right now to get a goat like Eric Adams. Is Eric Adams the one who did the rat the rat problem thing? Yeah. Yeah. Saudi Arabia have a rat problem.

Solved I mean Oh, that could be interesting. I highly doubt it. I highly I would not want to be a rat in Saudi Arabia. I think that they might not take kindly, but who knows? Maybe Kindered Spirits over there fighting rats.

Uh, in other news, Drew Fallon posts, "Garage Beer receives investment from Duration Durational Capital Management at a $200 million valuation. Congratulations to the Kelsey brothers. " Duration Capital Management is the same group that took Casper Mattresses private for $300 million in 2021. They also own Bojangles.

Garage Beer is tracking for $65 million in revenue this year, valuing the beer at beer brand at 3x sales. They started with a $500,000 crowdfunding campaign.

In 2016, you got to imagine that the Kelsey brothers, uh, with how much attention they get are able to do some deals, get some distribution for Garage Beer, their their beer brand. Uh, love to see it. Also, thank you, Bill Bishop, uh, for posting about the Laboo that TVPN sent to him.

Uh, he said his dog Tashi loves it, but won't let him take won't let him treat it like a chew toy. Uh, Bill Bishop may let him may let Tshi take it to the park to impress one of his girlfriends. I guess Tshi's popular at the park. Well, that is fantastic.

Um, hopefully this only brings you good tidings and does not curse you to a a demonic world. Uh, but stay safe out there, Bill. Uh, who knows what's going on inside that labu. Um, but I'm glad that it was delivered safely. And last but not least, uh, App Loving, Robin Hood, and MCOR are set to join the S&P 500. Wow.

Applovin, who else? Merkore. No, not Merkore. They're not Apploving, Robin Hood, and MCOR. Robin Hood. Okay, that's great. So, that that's a retroactive uh Fortune 5 or S&P 500 company for us. We got uh Brian Armstrong, Karps in there. Obviously, we're knocking them down. We should do all 500.

We've done almost a thousand interviews this year. Why not do all 500 S&P 500 CEOs? I'd love it. Also, Conor McGregor is uh is running for president. Yep. He's he's he's getting it done. Starts now. Um anyway, thank you to the chat. Thank you, RAV, Hershey, Michael, Jason Turner, always good having you here.

Percleat, we got a good crew. Max Condrat, thank you for shouting out that the bology coefficient is an instance of Reed's law. I completely agree. That's exactly what was going on. Max has been sending me some fantastic DMs. Very helpful.

And next week, yes, we'll be here in the studio Monday and then we'll be hitting the road. Very excited. Yes, lots to come. Have a fantastic Friday. Have a fantastic Friday, afternoon, evening. We love you. Only three days till Monday. Cheers. See you. Bye. Bye. [Music]