Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Deep Learning with PolyAI

Deep Learning with PolyAI artwork

014: Why the New York Times is suing OpenAI - Copyright, Lawsuits & Fair Use

Dec 28, 202323 minAudio
ShareXLinkedInFacebookWhatsAppEmail

Join hosts Michelle and Nikola as they dive into a potentially industry-shaping lawsuit: the New York Times vs OpenAI. They explore various aspects of this case, including implications for copyright, fair use, and the value of journalism in today's AI-driven world. The duo also touches upon potential threats to traditional journalism from AI, the struggle for content attribution, and the market's response to this unprecedented lawsuit. All this, and The Artist FKA Prince gets a shout. Follow PolyAI on LinkedIn Watch this and other episodes of the Deep Learning pod on YouTube

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

Michellebut the really interesting thing there is this idea that isn't really copyright infringement to train a model, to train a student, to train a person, like you think about Prince, like one of the greatest Like artists ever, right? He didn't become Prince by just learning to play piano, in a silo and like playing guitar. He like studied Sly Stone. He obsessed over James Brown videos to figure out how to dance Hi, everyone, and welcome back to deep learning with poly AI.

MichelleI'm Michelle. I lead marketing here. And I'm joined by Nicola. Hey, everyone. How's it going? Awesome. So we hope you've been enjoying the holiday break. And if you haven't had time to be, but just perusing the news headlines surrounding AI, we've done it for you. I think the biggest 1 of this week is the New York Times suing open AI. And it's a potentially monumental precedent setting case. Around the whole notion of copyrights, fair use what's the value of journalism in a day like, today yeah, where do we, and where do we go from here?

MichelleLike, why is this historic? So I guess we'll start there, like really, what are they suing for? Yeah. And what does that, what does it mean, for the AI world and for anybody who's using open

Nikoladata? Yeah. I think that in the whole thing, to not quote it verbatim, but I think they are right. Billions of damages from like opening using their data to train and deliver a service that is valuable and they're making money off of it. I think Altman said that open AI revenue is 1. 3 billion as of last October, right? So I'm finding it hard to see how it's billions of damages so far, but Hey, I don't think, I think you people would think that New York times does not have a, an inflated sense of self worth.

NikolaSo, gDPR run amok, right? Where, for example, with a machine learning model, anyone whose data you've ever used can go back and say, Hey, remove my data. And then if you did, you'd have to go and retrain the whole thing. So one thing they're asking for is for the models trained with their data to be destroyed and recreated. So, like they're probably asking for at the very least, a rerunning of a training sequence. It probably costs a few million, if not more, right? And To me, it sounds a bit like NIMBYism, right?

NikolaIs your local neighborhood newspaper going to ask for it as well? Yeah, it might be fair to say, hey, don't use my data. Right. But

Michellewell, I think, in order to think about that whole thing, like our old journalistic, publication is going to link arms and join a lawsuit like this or join the bandwagon. Think about GDPR, how many people actually requested to have their data removed and why, and like what groups of people. Feel infringed upon when when companies are using their data, usually to market to them. I think the way to get at the answer to that, or to try to predict the future would be to say really, what is the threat to these. To these companies to people who look like the New York Times.

MichelleAnd so if you go to them and it's it's billions of dollars and, actual and statutory damages. What does that mean? What are those? Where are those damages coming from? I don't know. I saw 1. The site, the, one of the things in the complaint was this article about a cab, about the New York city cab come, business, it's a huge business versus Uber, I believe. And they had, it was almost word for word, the training model, which was exposed by chat GPT. We could ask it to say Hey, how did you get this information? Tell us a little bit about how you were trained on this and it'll expose stuff that should have been behind a paywall.

MichelleSo is the threat their profit, like it's the threat that they're concerned that the day when GPT catches up to, the time that we live in now, and it's not based on information from 2021 that they're going to be threatened, their intellectual property is going to be out there for the world to see, and will that have a monetary impact? iS it?

NikolaYeah, I am 100 percent confident that, way more about journalism. Then I do so, please tell me if I'm wrong, but there are really two value streams for the consumer for publication like New York Times, right? One is reporting real time reporting, right? Like you are my trusted source of information. You know what the latest happening in Ukraine or Gaza is you will provide that information or trustworthy way. Another one won't challenge that the best they can do is.

NikolaNot be retrained to provide that information because that would be monumentally expensive and also just impractical because that information has to come from somewhere physical, real world, right? So, yeah, it can, then you basically move into a mode where you're using the LLM to do API calls off of Twitter, off of anything else. And that will bring the old and familiar misinformation and everything where, the same question is, does a New York Times journalist sit on Twitter? And YouTube and watch for, geolocation reports from Ukraine to establish where the battle lines are.

NikolaOr do they actually have people in the field who are speaking with commanders there and everyone and, providing that information and good real time form to their readers. I think they are doing that and that's clearly not gonna happen. So people who are subscribed to New York Times because they want this real time stream of information. I don't see how that's under threat, right?

MichelleYeah. It's a lot of the processing. So like the thing, the thought of, or the promise of journalism, right? Is that you have these people with the best intentions, performing a public service of taking information and putting it in a format that people understand. And that helps them understand the world around them and understand, what's likely to happen next. Or why is this politician seeing things this way or saying these types of things, or why are these problems happening? And then they are uniquely, in fact, I think they used that, that word. They said they had uniquely valuable content.

MichelleAnd the way that it's unique is by the process is by the lens through which they're filtering truth. Right. And an LLM has its own lens too but just like something that you said before, it just made me think of this. There was a time when I was in college, you could get a discounted New York times subscription because it's expensive, even, especially for a poor college kids, but but yeah, you could get a discounted kind of subscription. And when I think about that, just the fact that their public service, this sort of thing that they want to uphold the integrity of journalism, that they want to give you the truth in the most unfiltered form, but in a way that's tempered and easy to understand, it's a similar, maybe a similar kind of philosophy that would be behind wanting to train a model well, wanting to be.

MichelleKind of train it with integrity, with information. That's not, just regurgitated garbage from, like the blogs, the internet blogs or like Twitter. People are going to

Nikoladeal with this. Hey guys, loosen up. So what's your data help the world. Right. It's you'd expect that from the way that they see the world. Right. I was going to say okay, so the first part is like the real time piece, right? If you trust New York Times above other newspapers, get to be that source of information and you're willing to pay, well, nothing changes, right? The second stream is like the opinion pieces, right? Like you told them to provide a unique, valuable perspective based on not necessarily, information from the last six hours.

Nikola12 hours, maybe we get to the point where we're tuning LLM so fast that they can be, knowledge up to yesterday, knowledge up to six hours ago. Look, we may get there very soon, right? At which point you're like, okay. What if this LLM writes something so thought provoking that makes you completely question your perspective on something and then you go and you're like, wow, like I don't need to read columns from New York Times or Guardian, but instead, I have a stream and you know The thing is if an LLM can produce this thing, it's not only true that it can produce one Then it can produce a thousand, right?

NikolaAnd you could then, if you are, the CFO of New York Times, and you look at like your future revenue streams, I could see it, as like a meaningful split between the two services they provide, right? And then yeah, this is under threat, but it would be pretty arrogant and self absorbed of them to think that their content alone will create that and that they have any shot whatsoever. Of that future coming to be, I'm not saying it is coming, right?

NikolaI think that there will be, we will increasingly have different autopilots and we will use different things to do all our work like you and I, but also a writer at New York times. And, maybe the best opinion pieces of, five years from now, we'll be heavily co created with AI. Why not? And I think New York times should stop. Doing NIMBYism and like looking at its own data as this, single thing. Unless they're really preparing a revolution where every single newspaper jumps on a Union train and gets their data.

NikolaLike they'll slow down OpenAI, but at the end of the day the discussion goes to there will be someone in Beijing or in Moscow who will go buy one subscription to The website crawl, the whole data set, and then train that model. Sure. So the model will be there and in that model, one way or another, we'll find its way there. So I think if it's just a play at really finding a way to. Get royalties out of their data, they're probably doing the right thing.

NikolaRight? They're not non-business savvy people. Yeah. So that's really what it's about. I think it's fair, right? Is it billions of damages? Well, American law works, right? Ask for hundreds of billions in the end. We cut a deal somewhere and the whole thing goes away.

MichelleI wonder if the New York Times is fighting something else entirely. And it's a battle that I think journalism has been like they've been under siege from social media and this concept of bespoke realities, which this great sociologist, I can't remember her name, came up with. I actually read about it on the New York Times, ironically, but this idea that you're being served the reality In media that you want to see. So, the algorithm serves you stuff that you're interested in and truth gets distorted, or into that wormhole you go down.

MichelleanD so the New York times is fighting the same way the Washington post might've been before they were, acquired by Amazon, but they're fighting to maintain their own reality and be able to write from their own perspective, take truth, turn it into opinion and serve that out as its own intellectual property without it getting distorted and diluted and transformed by someone else's prompt, I think that's really, if I had to guess what the battle is actually about but the really interesting thing there is this idea that isn't really copyright infringement to train a model, to train a student, to train a person, like you think about Prince, like one of the greatest Like artists ever, right?

MichelleHe didn't become Prince by just learning to play piano, in a silo and like playing guitar. He like studied Sly Stone. He obsessed over James Brown videos to figure out how to dance, how to move, how to sing, how to put all the pieces together. He obsessed over other artists and did hours and hours of copying, not study, copying. And then because he's his own unique artist with his own perspective out came. What we know of as Prince or what we know of the artists formerly known as Prince.

MichelleBut that's not copyright infringement the output, the product that we see and we all celebrate, is not you can't tie it back. James Brown can't sue him. It posts it's just that's just crazy. So I think that at the core of this lawsuit is the scariest thing. For anyone who's an AI creator, anyone who's, a creator at all. I don't know. Very

Nikolainteresting. Yeah, I don't, I mean the whole thing about protecting their right to be that and to not have it distorted, well, the one problem we have with LLM's is that like the whole like, accountability and saying I'm telling you this because I read it in the New York Times. It might hallucinate if you ask it where it's from, but probably not. And I think the mode of using these models is going to become such that you question it all and you go to the source to look for it. And I think there will be a lot of good tooling and methods that will come out for this in the very near future.

NikolaSo I don't think that the threat to them of hey, New York Times said this and it's fake news. I don't think that's really going to happen or that there's a high risk of that being a thing. But yeah, I think they are a major powerful publication, but they're not that big. Right. So I think like the evolution of AI honestly, they're inconsequential and okay, the easiest thing I would do if I were. At opening, I dealing with this would be like, Hey, cool guys, right? Okay, here's some money disappear from my life, and I will never use your data again, because your data is 0.

Nikola000000. Yeah. This for 30 seconds, 1 percent of what I've been trained on. Right. And I've been trained on far more quality opinion pieces, which are, like classical novels and, works of great philosophers that are better than yours, honestly, like your stuff. Well, it is, but come on, get a grip, right? What you think about, what the fed should do relative to this or culture wars or whatever is bust some scribbles on a piece of paper in the grand scheme of history, anything that humanity ever produced.

NikolaSo get a grip influence this model at all. Fine. Don't. Right. Exactly. The

Michelleproblem. Maybe that's exactly the problem is that they are mounting to scribbles on a piece of paper that they are staring down their own fate by the LLM like capture of what they do. It's like we talk about corporate capture in the form of regulation. Maybe at the heart of this, New York, the New York Times does want to slow down. They do want to slow down the progress. But to say that their percentage of content that they're, contributing to the universe of training content is significant, but maybe that is what they're trying to do.

MichelleMaybe they are trying to inspire others like them. To slow down the pace of progress because it is absolutely a threat. I think to their,

NikolaI don't think, that would be like, shots fired and you just, I'm teetered another round of bullets into them, but I don't think they think of it that way. Right. But like the real threat of, Hey, is this coming for us? Am I but some scribbling on a piece of paper? What are our lives at the end of the day? A small imprint on, the general flow of human history, right? And what can they do, right? Their piece there is, how are they different from The Guardian? Or, pick your favorite, left wing or right wing publication in any country, right?

NikolaThey produce thoughts and they're all filled with intelligent people with slightly different worldviews. The business model versus the idealistic view of journalism. Yeah they're they need to be separated here. And I think what you're getting at is something really important, which is is what I do important? If I scream and do something now, can I like, cling on to this thought that I do something meaningful, that it's really important, that it's worth a lawsuit against something that is a historic development. And honestly, like just turn on Serbian skeptic mode, could it maybe just be a cry for attention?

MichelleYeah, what is that attention by them? I think that the audience is

Nikolasitting there in a room and it's let's say you and I are like, you're the editor in chief and I'm like your deputy. I'm like, Hey, you know what? Someone should sue these guys, down the seventh drink. You know what? Why shouldn't we sue these guys? Like you totally got on the New York Times to be the ones that actually pulls the trigger on it. Right?

MichelleI think, though, let's turn the cynicism all the way down. Let's say what do they actually want? Do they want billions of dollars? No, I'm guessing no. Or maybe they do, but I'm guessing no. Yeah, royalties. But really what is the solution to this? What is the version of AI meets content that exists out on in the world? What's the best kind of most harmonious version of that? Is it citations and attribution of like where this thing is learning things from? Is it not accessing things behind a paywall?

MichelleIs it? I don't know,

Nikolayou you lead marketing here. You, we all know how hard attribution is, right? They just don't come as an inbound, outbound, both partners, sourced influence, et cetera. The real issue we have with LLMs and deep learning overall is that we can't really attribute, did the model, right? This genius opinion piece, that guy in New York times from year 2018 to year 2020 was the catalyst for it. Well, you know what? Maybe, right. And then the thing wrote the pivotal piece that, prevented World War III.

NikolaYeah. Yeah, but how are we going to even attribute that? It's stuff, right? I think collective action by, like, all media, et cetera. Well, I don't know. This is what I mean look at

Michellea company like Wikipedia, right? Which built their foundational sort of tenets on the fact that they were going to be able to cite The internet, that they were going to be able to be the source of truth, this neutral kind of, repository of information that cites everything. And if anybody like, leading marketing, if anybody's ever tried to create a Wikipedia page for a company or a person and realize how difficult it is and how stringent, yes, there's workarounds, but like how stringent their rules are for for like content approval. Maybe.

MichellePart of, or behind a lot of the impetus for this lawsuit, which is can we get to a place where, you know, where you have a Wikipedia esque, I don't know, attribution model would not work in our

Nikolabusiness, but it's so much data though, right. And I, if you were to audit anyone who's built like a data lake for something like this. Like I'm betting that it wouldn't be trivial to actually go through and like fully identify exactly where everything came from, unless, far more disciplined than I ever did when I did anything. it's a lot of data. And then just even attribution wise, it's these words model learns from, they look very similar to a lot of other words, right?

NikolaI don't know. I think like the simplest, the path of least resistance for open AI here is just a sick list. Like you're up, like your data is cool. I can guarantee that's proper GDPR at the end of the day. Right. Although not, I think fully applicable here in that sense, but AI. Nothing

Michellewill change for them, except that opened a very interesting door where anybody who's competitive now with them can pay people or, do things that are not right in order to get other people to join and pull their information. And there are big repository repositories of data. Wikipedia is one. There, there are many other things if they did all start to pull out would make a meaningful difference.

NikolaYeah. And I guess, if if you're training on something that references the New York times, then did you indirectly still infringe on their content? I don't think it'll be that easy to enforce. Now, of course, you could maybe have a self serve model for all the people going after you, where you reveal your data and you go, tell me what your data is, we'll review it, like Wikipedia, and remove it if so. And I think that works for Wikipedia because it's fully public. You and I can go and edit any article now. Well, we can try. Right. And it'll be reviewed and we'll see what happens.

NikolaThe one major thing here is rather than, there are training procedures and the know how of how they train their GPT models and the data pre processing and post processing steps between the crawling that they do, but the data mode is really the thing they have that is the most important one. And they wouldn't be able to just share it with everyone like that without losing that competitive differentiation. So it's not really in their interest to serve it in Wikipedia or that.

NikolaSo I don't really know. I it's interesting. It is interesting. Maybe New York Times will write something interesting about why they're suing and then we can use your subscription to, to read the thing and comment on it in the next one. I don't know. Yeah.

MichelleAnd then talk about it here and infringe on their copyright because we want to get Maybe we'll get sued as well, right? Yeah. It's, yeah. It's super interesting. There are probably a lot of, who do you think sues them next? Yeah. Yeah.

NikolaI'd wager that, the irony is not lost on if New York times, the similar left leaning ones all off that, that, we get a Breitbart fueled 55 and I'm not sure how that's in the public interest to be honest.

MichelleI think right now it's in the public interest because hopefully they cancel each other out and get something neutral in the middle. That's not, I don't know if that's what's happened either. It's almost like this bespoke reality thing where you ask what you get, the way that you prompt it, influences what you expect. Yeah. It's super interesting. I can see Incredible impacts on the academic world as well. For these types of questions that are being asked and brought up by this lawsuit. But yeah, I think we're going to need to see what happens. What does the judge say? Did they get a jury? Are they granted like a real these things get resolved?

MichelleTypically? Oh, goodness. Like we could be talking about this two years from now. But yeah yeah, we'll see. We'll definitely see the New York. I think so. Yeah. I'm just,

Nikolayeah. Federal district. Then we'll get a Brussels one and a London one.

MichelleYeah. Those will do better than the New York ones. I'm sure. I should have filed it in California or Germany based on the GDPR precedent. What we know happened there. Anyway it's

Nikolasuper, super interesting. Right. There's definitely opening up a Pandora's box of these things. Someone will just ignore it. Right. That's the real. It's like downer here is they might slow them down. They can't really stop the whole field. No. Best case they make money on it, which I guess is a decent enough expected outcome for them.

MichelleI, that just seems so, that seems to me somehow that seems like the most cynical take is that they're doing it for money. who knows. Awesome. How's your Christmas going? It's been going great. It's been going great. Yeah no, no drama. Lots and lots of it was speaking with our, the producer of this podcast earlier and he was saying that our shards of Christmas wrapping, wrapping paper everywhere. That's the state of fortunately my parents house where we had Christmas and not here. What about yours?

NikolaNot bad. My daughter just figured out how to stand up, so I think that's just the fascination, so

Michelleyeah. Well, what goes up, unfortunately, hopefully for Christmas she got some padding, some yeah, Yeah. Nice. Well, I hope you enjoy New Year's. You too. Speak

Nikolasoon. See you guys next year. Bye.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. What if your contact center became a product team?

    Nov 17, 202635 minAudio

    Three years ago, Simplyhealth's PolyAI deployment was only resolving simple FAQs over the phone. Today, they run voice, in-app, email, chat, and WhatsApp conversations on the platform. Nikola Mrkšić talks with Paul...

    Credit the people on this episode

  2. What happens if your AI agents get 1% better every day?

    Sep 10, 202639 minAudio

    An agent that gets better every day compounds into a very different agent within months. That's the idea behind Wren, PolyAI's new release: an agent whose job is to make your other agents better. Some of what it finds...

    Credit the people on this episode

  3. Who's coordinating your army of AI agents?

    Sep 3, 202644 minAudio

    Companies are pouring money into AI, but not everyone is getting the returns. Ian Jacobs, VP and Lead Analyst at Opus Research, thinks he knows why. In this episode, your host Nikola Mrkšić sits down with Ian to work...

    Credit the people on this episode

  4. Why should CX leaders care about MCP?

    Aug 20, 202636 minAudio

    When PolyAI built its Builder MCP, one client helped shape it: Vixxo Facility Solutions, one of the largest facilities providers in the US. In this episode, guest host Damian Sasso talks with Vixxo's CTO Derek Neighbors...

    Credit the people on this episode

  5. Can AI really hear a call the way a person does?

    Aug 6, 202622 minAudio

    PolyAI just launched Dialog-RSN-1, its first audio-native model, and your host Nikola Mrkšić sat down with its builder, Matt Henderson, to unpack why it’s a game-changer for building voice agents. Most voice AI either...

    Credit the people on this episode