תעתיק 🔗
00:00:00
Speaker 1: Hey, y'all.
00:00:01
Speaker 2: Brian here. The other day, I had someone slide into my DMs on Instagram and ask how they can best support the work we do here on Question Everything. The original, independent, hopefully surprising, and always thoroughly fact-checked stories and reporting we bring you week in and week out. Well, here's how. Please become a paid subscriber to our show by clicking the link in our show notes. Kicking a few bucks a month or a discounted 55 bucks a year... and you'll get our episodes without ads. And more important, you'll help keep Question Everything growing and thriving in a time when, as all you regular listeners around here know, it is not easy out there for journalism outfits. Direct support from you will ensure that we can keep experimenting, keep reporting, and, you know, keep questioning. If you appreciate the stories you get here, please go to the show notes of this episode right now and click the link to subscribe. Okay, here's today's story. which is about one of the many ways the fundamental business model of journalism is being threatened. Fitting. Thanks to all of you who become subscribers. It means more than you know. A while back, I was on the A train headed to the New York Times offices in Manhattan. I'd been asked to come speak to the staff of the Daily. And just out of somewhat idle curiosity while I was on my way, I wanted to remind myself of the various lawsuits the Times was involved in. So I pulled up ChatGPT and typed, what lawsuits is the Times involved in right now against tech companies in one category and the administration slash government actors in another? And ChatGPT answered the tech part of the question by explaining the Times lawsuit against an AI company called Perplexity. ChatGPT told me that the Times was suing Perplexity for the, quote, large-scale unlawful copying and distribution of of millions of The Times' articles to train and power Perplexity's chatbot. ChatGPT went on to give a bunch of other details about The Times' case against Perplexity, and then near the end said, quote, This lawsuit is part of a broader wave of litigation by media companies against AI firms over copyright trademark violations. The Times, in fact, had previously sued another major AI company. That other major AI company? It's OpenAI. the maker and owner of ChatGPT, the chatbot I was having this very conversation with. I snorted out a little laugh there on the train. I guess ChatGPT thought I wouldn't notice. There have been so many lawsuits against AI companies now that it's hard to even keep track. 143 were filed this year alone. And the most prevalent complaint against these companies is copyright infringement. AI companies like OpenAI and Anthropic and Perplexity require enormous amounts of creative work by humans, writing, video, sound, to train their large language models, their LLMs. The models encode patterns from all this material, and that's what makes them more powerful,
Load full transcript
more effective, and more valuable. So the companies have a voracious appetite for good writing. And in lots of cases, they've ingested vast amounts of it without paying the writers or publishers. As a result, back in July, a group of authors landed the largest copyright settlement in U.S. history, $ 1. 5 billion, against Anthropic for training its chatbot Claude on pirated copies of the authors' books. But the case that might turn out to be the most important, the most precedent-setting in this realm, is still very much not settled. And it's that one from The New York Times, along with four other news organizations, The Daily News, The Intercept, and a couple others, who are all suing OpenAI. as well as Microsoft, which was one of OpenAI's major investors and partners when it was developing ChatGPT. Microsoft also has its own chatbot, Copilot, which is part of the suit. The Times says the two tech companies, OpenAI and Microsoft, working together, scraped the internet, scooped up millions of copyrighted news articles without permission, stole them basically, and used them to train ChatGPT and Copilot. OpenAI and Microsoft argue they were allowed to do this as a matter of law. They're claiming fair use, which allows for the reuse of someone else's copyrighted content under certain circumstances. OpenAI's CEO, Sam Altman, has been asked about this a fair amount because it's so central to how the company has created its signature product, ChachiBT. This is Altman talking to Tucker Carlson.
00:04:38
Speaker 3: And what about copyright? Our stance there is that... Fair use is actually a good law for this. The models should not be plagiarizing. The model should not, you know, if you write something, the model should not get to like replicate that. But the model should be able to learn from and not plagiarize in the same way that people can.
00:04:54
Speaker 2: And New York Times journalist Andrew Ross Sorkin asked Altman specifically about the Times' lawsuit against OpenAI at a New York Times live event in 2024. It was kind of awkward.
00:05:05
Speaker 3: On the New York Times website, Look, I don't believe in showing up in someone else's house as a guest and being rude. But I will say, I think the New York Times is on the wrong side of history in many ways.
00:05:16
Speaker 4: We can discuss and debate that. And we'll do that, I think, in court. Look forward to seeing you.
00:05:27
Speaker 2: Over and over again, in court filings and elsewhere, OpenAI and Microsoft have said that they've done nothing wrong. That their work has been totally misunderstood. They're not copying and plagiarizing news articles, they say. They're using the articles to make ChatGPT's electric brains smarter, to teach it human language and syntax and how to structure an argument. The company's lawyers write, I'm just quoting them here, the purpose of training an LLM on news articles is to endow the model with the ability to generalize so that people can put the tool to myriad
tasks. News content is useful for that, they say. because it is coherent, factually dense, and grammatically sound, while covering a broad range of subject areas. OpenAI and Microsoft also say that the Times in particular has seen no drop in business or revenue due to this new technology. In fact, they point out subscriptions have grown. They claim there's been no harm to the market, which is a big thing you need to prove in order to win on fair use grounds. In short, the tech people say the journalists are not being replaced by their machines. No copyright infringement to see here. That's what they say publicly, anyway. But a couple months ago, this one document in the case was unsealed. And the world learned from it what people inside of Microsoft and OpenAI had been saying privately, behind the scenes, about what their chatbots are doing to the journalism industry. And it was very different than what Altman was saying to Tucker Carlson or Andrew Ross Sorkin and the like. Even hair-raising. So, I asked one of my colleagues, Sean Cole, to read all 92 pages of this pretty explosive legal filing And talk to someone who's closely watching the case. And then report back to tell us, what were the people involved in making ChatGPT actually saying about using all this journalism to train their AI models for free? And do they really think it was legal?
00:07:19
Speaker 5: There was knowledge and awareness and executives inside of both companies who were describing what they were doing as legal. I'm paraphrasing, but destroying the actual supply chain that they depend on.
00:07:36
Speaker 2: From KCRW and Placement Theory, I'm Brian Reed. This is Question Everything. We investigate how the truth gets buried, distorted, denied, and the ways people are fighting to make it matter again. Stick around. And now, gentle audience, here is reporter Sean Cole with our story.
00:07:58
Speaker 1: So this all happened back in September when the lawyers for The New York Times and the other news outlets, New York Times et al., filed a brief asking the judge to just rule in their favor based on the facts in front of him. No trial needed. Just say we won. It's called a summary judgment brief, and it was heavily redacted. Big black blocks blotting out all sorts of quotes and information that OpenAI and Microsoft wanted to keep protected. Lawyers' eyes only. A lot of it came from internal emails and documents that were dug up during discovery, along with depositions, what have you. And just to give you a sense, this is what the first paragraph looked like with all of the crossouts. I'll just read the first half of it. This case is about, as Microsoft's Director of Applied Science put it, redacted. Perhaps the redacted. Defendants repeatedly copied redacted of plaintiffs' copyrighted articles in their entirety without permission... to produce substitutive commercial AI products. OpenAI's head of chat GPT wrote that redacted from those products, which, he said, redacted and
redacted. But then, the news outlet's lawyers petitioned to have a lot of the document unredacted, saying, this is an important case, it affects the public, it should be litigated in the public, not in the shadows. And the judge agreed, and let them refile it with a bunch of the secret stuff unsealed.
00:09:22
Speaker 5: The first paragraph, you know, the lead, if you want to use a journalism term, was probably the most eye-popping of anything.
00:09:31
Speaker 1: This is Jason Kint, CEO of a nonprofit called Digital Content Next. It's a trade organization for media companies who have their content up online. The New York Times is a member, as is Fox News and about four dozen others. A lot of DCN's work is trying to figure out how news and entertainment in the future is going to be funded. So they have a dog in this fight. They even filed an amicus brief on the side of the Times and the other plaintiffs. Anyway, Jason's been following this case like a hawk. And I know this is kind of a dramatic question. to ask about a legal brief. But I asked him if he remembers where he was and what he was doing when he saw the unredacted version. Oh, I do.
00:10:09
Speaker 4: I was at a event in New York. I think it's called rebooting. It was an industry event.
00:10:14
Speaker 5: I think A lot of the topics being discussed were AI, and it was just about to be lunch break, and I saw it pop onto the docket. And so I opened it up and stayed in my seat where everybody went off and had lunch. I didn't want to leave my seat because lots of the evidence that I hadn't been able to see until that moment was from Microsoft and OpenAI executives.
00:10:40
Speaker 1: So this is what the same passage really said under all the black boxes. This case is about, as Microsoft's director of applied science put it, quote, an astonishing theft of unprecedented proportions. Perhaps the, quote, largest theft of labor in human history. Defendants, that's OpenAI and Microsoft, defendants repeatedly copied millions of plaintiffs' copyrighted articles in their entirety without permission to produce substitutive commercial AI products. OpenAI's head of chat GPT wrote that, quote, Publishers face an existential threat from those products. These are direct quotes from executives at the same companies, OpenAI and Microsoft, that are claiming they didn't do anything wrong. There and then, Jason started posting sections of the document up on Blue Sky, highlighting in pink anything that had been previously redacted, and including his own running commentary, using words like eye-popping, blood-boiling, touche.
00:11:39
Speaker 4: You had.
00:11:40
Speaker 5: The acknowledgement of the theft of labor. You had that it was an existential threat. You had that it was largely substitutive.
00:11:49
Speaker 1: And, quote, we'll get more and more substitutive as they get better. Meaning chatbots like ChatGPT and specifically Microsoft Copilot were becoming substitutions for the New York Times and the
other organizations on the plaintiff's side. Again, this is a guy at Microsoft saying all of this.
00:12:07
Speaker 4: And then.
00:12:08
Speaker 5: Maybe most importantly to the actual legal case is that it makes a complete mockery of the idea of fair use.
00:12:14
Speaker 1: This is the kicker. The same guy at Microsoft said that if Microsoft and OpenAI win, if the judge decides it's OK for them to train their AI models on massive amounts of news articles without paying, it would, quote, make a complete mockery of the idea of fair use. Fair use is the defense that OpenAI and Microsoft are using for commandeering all of this content to train the large language models or LLMs. which again are what power the chatbots. And you may have heard of this little corner of copyright law before. The fair use exception is built into the Copyright Act of 1976, Section 107. Basically, there are four factors that judges consider when ruling on a fair use claim. And it's not cut and dry, per se. The analysis can be kind of subjective.
00:12:58
Speaker 5: You look at things like, is it commercial? Is it to make money?
00:13:02
Speaker 1: That's one of the four factors. Another, what's the nature of the copyrighted work being used? Is it a news article, which can be harder to protect than an imaginative work, like a song or a poem? Another factor, how much of the copyrighted material was used? Was it used in full? Just an excerpt? Was it the most recognizable part of the work? That could play into the assessment.
00:13:24
Speaker 5: And also, is it a different purpose than the original work in such a way that it's transformative?
00:13:32
Speaker 1: This is the fourth factor used to decide if something is fair use. and it's central to this Times OpenAI case, is the reuse of the copyrighted work transformative. Has the way you reused it transformed that work somehow?
00:13:45
Speaker 5: Often in journalism, you'll see people use that factor to say, well, I've transformed it in a different way that it's being used for an entirely different purpose.
00:13:55
Speaker 4: Right.
00:13:55
Speaker 1: Instead of to entertain, I'm using it to explain something.
00:13:59
Speaker 4: That's a very good description, yeah.
00:14:01
Speaker 1: I've done this a lot in my work over the years. I used a portion of a song or a movie in a radio story about that song or movie. One story I did, I had to summarize the entirety of A Clockwork Orange in one minute and 37 seconds, complete with clips. Here, I'll just play you part of it. This is from a piece I did for This American Life. About a third of the way through, Alex accidentally murders someone during a break-in and goes to prison. And after serving a couple years, the government chooses him for a new experimental type of aversion therapy. They gave him this drug. which makes him basically allergic to violence. Anytime he's so much as pictures hitting someone, he's overwhelmed by nausea and dread. Also, whenever he hears the music
of Beethoven, but that's another thing. And so on. The point is, I used tiny bits of the movie as part of a completely different work that had a different purpose. That's transformative. And now I'm using a bit of the This American Life story in this story, so fair use all the way down. Conversely, there was this famous case that's cited in this lawsuit in which a photographer sued the Andy Warhol Foundation, saying that Warhol co-opted a portrait she had taken of Prince, the musician, and did his Warhol number on it. Saturated the colors, whatever, whatever. But he didn't license it and apparently didn't transform it enough. The case went all the way to the Supreme Court and the photographer won the day against the Andy Warhol Foundation. So that's the law under which OpenAI and Microsoft say they're allowed to scrape the web for scads and scads of articles and feed them into their large language models, their LLMs, in order to train them. Fair use. They're like, this couldn't be more transformative. And you can see their reasoning. If I teach my parrot every line of dialogue from the West Wing to improve its vocabulary, that shouldn't be a problem for Aaron Sorkin. But the Times side says that training AI models on their news content can't be transformative. because of what happens after the LLMs in question are sufficiently trained. When users go to ChatGPT or Microsoft Copilot and ask it questions about the news. Here's Jason Kent again.
00:16:11
Speaker 5: What they're spitting out will cause the user to not have to go to the actual source content, including not buy a subscription or not have ads that they have up on their screen.
00:16:25
Speaker 1: I should say, Both sides seem to agree that it's bad when LLMs spit out long, verbatim passages of news articles or, as Sam Altman said earlier, plagiarizing. But they differ widely on how much that's happening. The OpenAI side says it almost never happens and that they take pains to prevent it. The Times says it's happened millions of times. This is important as it goes to just how competitive, how substitutive the LLMs are. And if you go strictly by the opposing filings in this case, It's impossible to get to the bottom of.
00:16:57
Speaker 4: But it's also substituted.
00:16:58
Speaker 5: This is kind of interesting to think about, too, from a business perspective, because then they don't have to go license the content. So it harms the licensing marketplace for all the copyright holders, too.
00:17:10
Speaker 4: I'm sorry.
00:17:11
Speaker 1: Say more about that. I don't understand.
00:17:13
Speaker 4: So there's actually a market that has developed.
00:17:16
Speaker 5: There's already a licensing market in publishing. There's been for more than a century. But Now there's actually a licensing market where publishers are able to license their content to Microsoft, to Perplexity, to Google, to OpenAI to train their LLMs. And so why would those companies pay to license the content if they're allowed to use it for free?
00:17:42
Speaker 1: Oh, you mean that other companies are actually going through, in the, I guess, mind of the plaintiffs, the appropriate channels saying, hey, can we use your stuff? We'll give you money. And then we'll train our LLMs like they're going through the front door as opposed to through the back door or the side door.
00:17:58
Speaker 4: That's right.
00:17:59
Speaker 5: And OpenAI has done deals with publishers.
00:18:04
Speaker 1: Why didn't they do it in this case?
00:18:06
Speaker 5: Because the challenge is if they have to, for them, is if they have to license everything that's held under copyright across the web, it's going to cost them a lot of money and they're probably not going to have the same margins and probably not be quite as interesting of a company.
00:18:24
Speaker 1: Interesting to investors. That said, for a company that's valued at $ 1. 25 trillion, even if the licensing deal was for a billion, they could probably handle it. Maybe they just thought it was easier to beg forgiveness than to ask permission. Or maybe they genuinely thought that the wholesale trawling of the web for however many millions of articles truly is in the realm of fair use. But the behind-the-scenes quotes from Microsoft and OpenAI in this summary judgment document make it sound like they knew what they were doing was wrong, even destructive.
00:18:55
Speaker 5: You know, there was knowledge and awareness and executives inside of both companies who were describing what they were doing as destroying, I'm paraphrasing, but destroying the actual supply chain that they depend on, which they characterized as, you know, the doom loop.
00:19:16
Speaker 4: One of their executives.
00:19:18
Speaker 1: That's a quote, doom loop.
00:19:20
Speaker 4: They use the term doom loop.
00:19:22
Speaker 6: Yeah.
00:19:23
Speaker 1: Let me just pull out the exact quote. It's from an internal Microsoft document, and it says, our AI strategy has started a doom loop that will hurt the performance of our models and the entire web at the same time. It is highly unusual that an end product threatens the economic foundations of its essential suppliers, but that is the situation we have created for our LLM business, with respect to its content supply chain. So someone at Microsoft itself is saying, it's like killing the goose that lays the golden eggs. If you make newspapers and magazines and other media companies unviable in the future, well then ChatGPT and Copilot won't have any news and information to retrieve and spit out to its users. In the same part of the summary judgment brief, It says that Microsoft's CEO, Satya Nadella, agreed under oath that, for many people, chatting with chatbots, quote, has substituted, giving you the information right there on the website on the AI platform versus needing to go to the underlying source. And later in the filing, a source that isn't named, either at OpenAI or Microsoft, says straight up, quote,
LLMs are a product that destroys its supply chain. It's funny to think of content, specifically news articles and the organizations that generate them, as being part of a supply chain, as though they were ever intended to be grist for some larger mill or pinions that make a car engine run. Also, is it a supply chain if you don't pay the supplier of the raw material? Although to hear Sam Altman tell it in public, he thinks that people who create valuable data, that's how he puts it, valuable data, should be compensated for the use of it. He even described a little of what that might look like at that New York Times event with Andrew Ross Sorkin.
00:21:18
Speaker 3: You know, a particular passion of mine has always been, can we figure out how to do micropayments where if you generate a story in the style of Andrew Ross Sorkin, you can opt into that for your name and likeness and style to be used and get paid for it.
00:21:31
Speaker 4: There's many other ideas, too.
00:21:33
Speaker 1: But of course, at least looking at it from the outside... None of those ideas seem to have made it to the conference room table at OpenAI or Microsoft before the harvesting of written works from the Times and its co-complainants. Coming up, what Microsoft has to say about their own people sounding the alarm about OpenAI's, quote, theft of copyrighted material. How can they possibly come back from this? And further revelations from inside OpenAI. The person who called the hoovering up of Times articles and other articles an astonishing theft of unprecedented proportions is Brent Hecht. Again, he's Microsoft's director of applied sciences. Hoovering up were his words also. And I was trying to imagine how Microsoft in particular could possibly try to get around having one of their own dead to rights, describing their corporate practices as the largest theft of labor in human history. But I didn't have to imagine it. Because the defendant's legal team filed two written responses to the brief, one from OpenAI and one from Microsoft. And in Microsoft's response, they sort of gently disown Brent Hecht. First, they say that all of his quotes and those of Satya Nadella are deployed in a, quote, context to be damned fashion by the Times lawyers. And then they say this. News plaintiffs call Hecht an executive. In reality, Hecht is a researcher and a self-styled provocateur. They say, quote, Hecht himself says he does not speak for Microsoft. And, quote, Hecht is not a labor historian qualified to opine on historical theft of labor or a lawyer who can speak to fair use doctrine. Although, slightly more generously, they also say, quote, Microsoft values his input, even when divergent. Also, to claim fair use, it helps if you proved you haven't harmed the market for the copyrighted material. And on that score... Microsoft says training their LLMs with articles by The Times and the other news plaintiffs hasn't hurt the news plaintiffs' business at all. They even quote a speech that
Times publisher A.G. Salzberger gave at Michigan State University back in September, where he said that The Times is thriving. I guess two can play at the using your own words against you game. Now, there's data that disputes Microsoft's disputation. For instance, The click-through rate, the rate at which people read an AI summary and then click through to the source where it came from, the click-through rate for the Times and the Daily News has dropped by anywhere from 83 to 93 percent. Also, when surveyed, 36 percent of Times subscribers who've also used ChatGPT for news say using ChatGPT, quote, means I no longer need news from the New York Times at all. Funny thing. I read both of those data points in that same summary judgment brief we've been talking about. And they are sourced to Microsoft itself. It's Microsoft's research that generated those statistics. But all of the excerpts I've pointed to thus far, while juicy on their face, are kind of philosophical. A Microsoft insider sounding a warning bell to his colleagues... Even the impact of the reduction in click-through rates and people reading AI summaries instead of sources of record. Like, you can't really conjure a mental picture of a specific person at either company in the actual act of stealing copyrighted content. Except in this one part, which Jason Kent brought up in our conversation.
00:25:16
Speaker 4: You know, one of the more interesting things that was unsealed, one of the OpenAI co-founders, Greg Brockman.
00:25:22
Speaker 1: One of the co-founders with Sam Altman and others. Brockman got a message from one of his employees at OpenAI who had.
00:25:29
Speaker 5: Actually hacked, I would call it hacked, but bypassed the subscription wall at the New York Times and was actually able to get the content behind the subscription wall. And he was kind of seemed to be celebrating that fact.
00:25:42
Speaker 4: Right.
00:25:42
Speaker 1: The quote that they have Greg Brockman saying this is in response to it looks like someone named Nick Ryder saying, oh, we figured out a way around the paywall. And Greg Brockman says, ah, nice.
00:25:54
Speaker 5: Yeah.
00:25:55
Speaker 1: I don't know if that was his tone of voice. Oh, nice. The words were, ah, nice.
00:25:59
Speaker 4: Yeah. So I take that as a positive.
00:26:03
Speaker 5: If I got that back from the co-founder, that's a good thing that I did that.
00:26:07
Speaker 4: And so you could make a lot of arguments around.
00:26:10
Speaker 5: How illegal that would be to be bypassing that subscription wall in order to steal the content.
00:26:17
Speaker 1: Is that one of the things that boiled your blood?
00:26:20
Speaker 4: It could have been, yeah, yeah.
00:26:22
Speaker 5: Because at that point, you're putting through considerable effort to get around the common sense rules of, hey, you can't use this content.
00:26:31
Speaker 4: It's copyrighted.
00:26:32
Speaker 5: If you just take it on its face, what's incredible here is that humans aren't allowed to bypass and get subscription content in order to, you know, as individuals to inform their thought, they have to pay for a subscription. It's not like Fair Use says, hey, Jason, you can go ahead and Break through that subscription wall because it's fair because you're informing yourself and making yourself smarter. No, you still have to pay for the subscription, Jason. That's the way the market works. But because it's a machine that can actually take millions of pieces of journalism to make a company that spends tens of billions of dollars, but it's suddenly okay.
00:27:16
Speaker 4: It's fair in that case. So, I mean, it just– it doesn't make any sense.
00:27:19
Speaker 5: Why does the machine have– rights that I don't have as an individual human, right, to go steal people's content and train itself.
00:27:26
Speaker 1: Microsoft CEO Satya Nadella would seem to agree. According to the brief, Nadella stated that, quote, anything that is paywalled should be licensed by anyone who wants to use it. And then it says that if Nadella, quote, had been made aware that OpenAI had scraped and trained on information that was behind a paywall, he would have, quote, invoked Microsoft's right to require OpenAI to retrain its models. OpenAI's response to the hacking anecdote, which is in their response to the summary judgment brief, lots of briefs and responses here, is frankly too technical for me to understand. They basically said they did it in a way that was above board. Quote, the fact that content may under certain conditions appear behind a paywall in a standard browser does not establish that OpenAI bypassed a paywall to obtain it. And just yesterday... or the day before we put out this episode, Sam Altman and Greg Brockman appeared once again with Times journalist Andrew Ross Sorkin on stage. And in what appears to be an annual cringe ritual with these guys, Sorkin asked Brockman directly about this very moment in the summary judgment document.
00:28:38
Speaker 6: Greg, there was some email exchanges that came out where you had a colleague who who was scraping effectively, got past the paper of the New York Times to go scrape the Times. And they sent you this email and you wrote back, nice, like, this is great. This is a great thing that you were able to do this. And I would like to give you the opportunity to comment on that.
00:29:03
Speaker 3: Well, first of all, I guess I would just say active litigation, not going to comment on it.
00:29:10
Speaker 4: I've actually, I've talked about this one in my deposition, so it'll be public. We're out of time, guys.
00:29:18
Speaker 6: But I will say this was fun and it was very insightful.
00:29:33
Speaker 1: While the defendants in the case claim that they're not harming the news market or trying to compete with it, Greg Brockman wrote that large-language models are, quote, particularly good at predicting text of news articles, quote, excellent at news, and, quote, very good at any
news task. Which, when you think about it for a second, how can they possibly be good at news? All of the material that was fed into the machines in order to train them is, by definition, about stuff that happened in the past. That's not news. It's history. News is new. And the answer is this other process that's talked about in the summary judgment document, a process called grounding. Put simply, the LLMs aren't just trained on old data. That's their memory. They're also taught to go and look up the answers to questions on the Internet if the information isn't already in their memory banks. They're literally grounded in present-day reality. People sometimes compare it to a student taking an open book test instead of having to cram.
00:30:34
Speaker 5: One of the big realizations when ChatGPT came out was, oh, it's amazing, but it actually only knows about everything up until some date in 2021.
00:30:45
Speaker 4: And it doesn't know.
00:30:46
Speaker 5: Anything after that. It's basically been trained on everything up until this date. So imagine a world today where you go into Clod or you go into ChatGPT, And it's super amazing, answering every one of your questions, but it doesn't know about.
00:31:00
Speaker 4: Anything after January 1st, 2025, right? Right. It becomes really stale very quickly and a problem.
00:31:06
Speaker 2: Yeah.
00:31:06
Speaker 4: And so news is what grounds it.
00:31:09
Speaker 1: It'd be like, what do you think of the new Nathan Fielder movie about Elizabeth Holmes? It'd be like, I don't know what you're talking about. It's like, well, you're the only one who doesn't know what.
00:31:17
Speaker 4: I'm talking about. I have no idea what you're talking about.
00:31:19
Speaker 6: Right. Like, I mean, or yeah.
00:31:20
Speaker 4: What do you what do you think of of how President Trump is doing in office? Right. It doesn't know anything after January 1st, 2025. You'd be like, I don't know. I didn't know. You know, maybe I'd say I don't know he's president.
00:31:31
Speaker 3: Right.
00:31:31
Speaker 5: So so like news is what everything that's new every day in a prompt that somebody asks when it's a completely new query. News is what is the critical, vital input and trusted news is. is even more important, trustworthy news, I should say, that the person on the other end is going to get reliable information.
00:31:52
Speaker 4: And so news is really important.
00:31:55
Speaker 1: But on the defense side, the lawyers for OpenAI and Microsoft are essentially arguing that news can't be owned. I haven't mentioned this yet, but the defense team also submitted a summary judgment brief, asking the judge to just rule in their favor. And in it, they state, without equivocation, that nobody owns facts. News plaintiffs, they write, have no right to prevent OpenAI from telling its users that Gustavo Dudamel has been appointed music director of the New York Philharmonic, or that the Dow hit a 52-week high of 54,744 on August 5, 2026, or that the battery in the new iPhone lasts for 24 hours and 31 minutes. Skipping ahead a bit here, those who expend sweat or shoe leather to report the facts do not gain a monopoly on the truths they report. And this isn't just their opinion. They're actually citing case law here, citing precedent. Specifically, a 1991 case called Feist Publications Inc. v. Rural Telserv Co., in which one phone book publisher sued another for cribbing name and number listings without permission. I'm not kidding. And the quote that Microsoft and OpenAI's lawyers pull out of that case raises a much larger question that I didn't expect would be at the heart of this case.
00:33:14
Speaker 4: But it is.
00:33:16
Speaker 1: And that is, what is the purpose of copyright? The quote from the Supreme Court decision in the phone book case that, again, two state-of-the-art tech companies, OpenAI and Microsoft, are leaning on, is, the primary objective of copyright is not to reward the labor of authors, but to promote the progress of science and useful arts. On the Times et al. side, well, they go back to that Warhol case, which I looked it up. has been cited a lot in copyright disputes like this one. Anyway, what they said was at risk in that case was, quoting that case, disincentivizing artists from producing new work by decreasing its value. The precise evil against which copyright law is designed to guard. That's the word the judge used in the Warhol case.
00:34:06
Speaker 4: Evil.
00:34:08
Speaker 1: And I've just never thought of copyright that way before. It's having an existential quality. People want to make stuff, and they want it to be theirs, and they also want to be rewarded for it in some way. And if they think that it might just be taken from them willy-nilly and stripped of their fingerprint, they might be less likely to make it in the first place.
00:34:29
Speaker 5: And that's important also to the work of journalism. It's not just about the facts themselves, but it's the expression of the facts. If somebody else then is able just to take that and use it for their own profits.
00:34:44
Speaker 4: It doesn't have to pay for that original work. Then what's the incentive to do the original work?
00:34:48
Speaker 1: Right. And that's something that we haven't talked about that actually exists in this document is that it's like they're describing journalism, not just as a collection of facts, but as an expressive pursuit that like I, you know, I learned all this stuff. I, I, analyzed it, sifted it through in my brain, and I came up with a way of expressing it that is more than the sum of the facts.
00:35:18
Speaker 4: That's right.
00:35:19
Speaker 1: The facts plus the person saying them. That's what journalism is.
00:35:24
Speaker 5: And the context in which they're said. So you and I are not just the words that are going back and forth between us.
00:35:33
Speaker 3: Right.
00:35:33
Speaker 5: Or, you know, the AI companies would even argue they use the term tokens, right? They're not even words. They're letters that make up words, right? And so the sequencing of those words, but that we are humans that are actually relaying these back and forth in different sequences and expression back and forth. And that entire context matters. And if it, by the way, if it didn't matter, then they wouldn't need to hoover up the entirety of the web.
00:35:59
Speaker 4: They would just suck in the facts that they want, right? Yeah.
00:36:03
Speaker 5: So I think that's the basis of copyright and why this case is so important. But you also, just as a pure capitalist, have to ask that question of why is premium content the only critical input to AI? Unlike data centers and cloud and engineers and all that stuff, why is premium content the only critical AI input that should be treated as free?
00:36:32
Speaker 4: Why just this one? Well, everything else has massive cost to it.
00:36:38
Speaker 1: It's true that the Times et al. does not own the facts. But there's clearly a reason why OpenAI and the like would want to train and ground their models on a paper like that and its cohort. Facts don't just lie around like pebbles waiting to be picked up. They have to be sorted, prioritized, verified, clarified. And the Times, for example, has a track record of trying pretty hard to do those things pretty well, which OpenAI and Microsoft clearly value. Whether there will ultimately be a dollar amount put on that value, that's going to be for the judge to decide, which could happen early next year or mid to late next year. And in terms of legal compensation for the journalism companies, if there is any, that could be awarded per edition, like per issue of a given paper, or even awarded per article, anywhere from $ 750 to $ 30, 000 apiece, which, if I'm doing my math right, On a per-article basis? On the low end, that judgment could be more than $ 7 billion.
00:37:55
Speaker 2: Sean Cole is a contributing reporter for our show. Thanks for listening. As always, you can follow us at QuestionEverything.com.. podcast on Instagram. That's questioneverything.podcast. Today's show was produced by Sean Cole and edited by Neil Drumming. Robin Simeon and I are the executive producers of Question Everything. Our team also includes executive editor Kevin Sullivan, producers Sam Megan, Zach St. Louis, and Tracy Hunt, contributing editor Jen Kinney, and associate producer Kevin Shepard. This episode was fact-checked by Annika Robbins, mixing and sound design by Brendan Baker. Our music is by Matt McGinley. Evie Kwong runs our social media. Our partners at KCRW include Arnie Seipel, Tejal Ajmera, Natalie Hill, and Jennifer Farrow. See you next week.