Everyday AI Podcast – An AI and ChatGPT Podcast - EP 181: New York Times vs. OpenAI: The huge AI implications no one is talking about

Episode Date: January 9, 2024

If you haven't heard, The New York Times is suing OpenAI and Microsoft for copyright infringement. OpenAI recently released a statement clapping back at The New York Times. So what does it all mean and what can we expect for the future implications of AI and copyright material? Let's dive in. Newsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion: Ask Jordan questions on AI and copyrightUpcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTimestamps:03:00 Daily AI news07:50 Recap of lawsuit09:58 Anticipated event unfolds after months of speculation.11:55 NYT complaint poorly researched and comical. Meritless.16:57 OpenAI admits to training model on copyrighted work.19:01 AI powered mega cap companies leading economy.21:23 Journalist shares conflict over AI in journalism.27:40 The evidence shows GPT-4 copied New York Times.29:16 Copying content from AI models is unreliable.34:34 Issues with anonymous chat sharing, missing links.37:29 New York Times filing about GPT issues.41:52 GPT account for chat, lacking important details.42:28 Language models can be manipulated easily.47:52 AI's impact seen in top companies' success.50:00 AI slowdown could deflate GenAI hype.54:06 Error in replication, flawed filing, trial implications.57:03 Suspicion around sharing links during GPT output.Topics Covered in This Episode:1. Legal dispute between OpenAI and The New York Times2.  Legal Filing and Evidence Analysis3.  Financial and Economic Implications4.  Technology and Industry ImpactKeywords:New York Times, OpenAI, lawsuit, legal filing, Exhibit J, GPT4, AI sector, generative AI, economic disruption, AI technologies, US economy, copyright infringement, manipulation, ChatGPT, AI models, technology, patent infringement, NVIDIA, GPU chips, Google, podcast, livestream, newsletter, AI education, damages, receipts, proper sourcing, citing, law schools, replication, evidenceSend Everyday AI and Jordan a text message. (We can't reply back unless you leave contact info) Start Here ▶️Not sure where to start when it comes to AI? Start with our Start Here Series. You can listen to the first drop -- Episode 691 -- or get free access to our Inner Cricle community and all episodes: StartHereSeries.com Also, here's a link to the entire series on a Spotify playlist. 

This is the Everyday AI Show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. Meet Firefly AI Assistant, now live in Adobe Firefly, the All In One Creative AI Studio. Just describe what you want to create and the assistant handles the rest, orchestrating multi-step workflows across Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome. The assistant accelerates execution. All right.

So the New York Times is suing OpenAI and Microsoft. That's old news. And OpenAI just responded to Microsoft. So newer news, but nothing headline grabbing. But there's some huge news and impactful implications behind the ongoing saga between the New York Times and Open AI that no one is talking about. So we're going to address those things and more on today's show of Everyday AI. Welcome, if you're new here, what's going on? My name is Jordan Wilson and I am the host of Everyday AI.

And this is your daily live stream, podcast, and free daily newsletter, helping everyday people like you and me, not just learn what's going on in the world of generative AI because there's a lot, but how we can use it to grow our companies and to grow our careers and how we can all actually leverage with what's going on, right? And sometimes what that starts with is understanding, you know, what's going on. And I'll tell you, I'll tell you all one thing. This is a mess. The New York Times versus Open AI saga is messy. And on today's show, I am going to unravel it all. I will tell you, I will tell you all this. We've done now more than 180 episodes of Everyday AI.

And I think this is by far one of the top three that I've spent the most time preparing. All right. Again, here at Everyday AI, we're not just a mouthpiece. We're not just a PR. We're not just, you know, one of those NFT influencers turned chat GPT experts that tells you use these prompts, use these prompts, no. We dig in deep to find out what this really means. All right. So before we get into that, if you haven't already, and if you are joining us on the podcast, thank you, check your show notes. But please also go to your everyday AI.com and sign up for

that free daily newsletter. And also, hey, on our site, so many different resources. You can access all of those 180 plus podcasts on our sites. So whether you want to learn something about using AI and sales or marketing or content creation or the legal implications, you can go find those different AI learning tracks on our site, as well as once you subscribe to our newsletter, you can read all of those 180 back episodes. So it is literally an encyclopedia of 101 Gen AI. All right. So before we dive into this New York Times versus Open AI saga, we're into a very high level overview of the AI news. All right.

So duolingo is laying off contractors as it increases its AI use. And the contractors it is keeping are reportedly just really working on checking the translation work of the AI. All right. Next news story is Nvidia is rolling out even more high quality GPU chips. These ones aim to help push the whole AI PC boom. and to keep up with the overall generative AI demand, right? When all of these companies are training their models and coming out with new Gen AI products,

they need these GPU chips. And right now, Nvidia is head, shoulders, knees, and toes above everyone else. So some big news from Nvidia there. And our last news piece is Open AI in the New York Times aren't the only ones involved right now in a high-stakes trial over AI. As Google is now fighting a multi-billion dollar lawsuit that it infringed upon a computer. computer scientist patents. So again, we're going to be recapping those stories and a whole lot more in our newsletter as well as every single day. We break down the key learnings from our live stream and podcast in the newsletter. So really, if you're listening to this on the car or if you're,

you know, listening to this on your drive and your commute and there are certain things that you hear and you're like, man, what was that? Or, you know, oh, I want to go back and check on that later. You don't have to hit rewind. Just check out the newsletter. All right. So, Let's talk New York Times versus OpenAI and the huge AI implications that no one is talking about. How do I know that no one is talking about these? Well, I've looked. I've searched the corners of Google and Reddit and Twitter and LinkedIn. And there are particularly two very large implications or large angles of this story that no one is talking about.

And hey, thank you all for joining us. but it is technically hot take Tuesday. I am sipping on the hot tea right now. There it is. So if you are joining us live, thank you, like Jay and like Brian and Woozy and Megan and Ahmed and Josh and everyone else. But hey,

if you're joining us live, let me know. Because again, this could get very hot. The hot takes could get very hot. Drop some one flame emoji. Should I be kind of nice, but still kind of hot take Tuesday?

day. Two emojis, throw it in the live comment. Should I bring the heat? Or three flame emojis, should I burn, baby burn. You know, this could get, this could get testy if I'm being honest. And I'll tell you why in a minute, part of it has to do with my background. As a journalist, yes, I have an interesting take from, I'd say a unique position, but let me know in the live stream comments right now. Mabritz, maybe it's wanting all three, all three flame emojis.

I feel most people like it when I kind of go to, you know, go bring the heat, but not too much, but let me know. And hey, also let me know like Brian, especially if it's your first time joining us.

Let me know where you're joining from. I always like seeing that. So, yeah, whether you're joining from the cold New York City or somewhere else like Nashville, like Tara, let me know. Woozy says I need to melt the snow

and bring the, three flame emojis. All right. Well, I'll keep, I'll keep looking at the emojis. I'll keep looking. All right. Jackie, Jackie wants it to burn. And hey, if you have a hard stop, if you're listening to this live, if you want to catch you, minus it will cancel it. This is going to be one of those longer ones. I'm telling you right now, but it is not just fluff. This isn't just going to be me, you know, old man Wilson shaking his fist on the porch, telling the kids to get off his lawn. This is packed. And I'm not exaggerating when I'm saying, I brought the facts. I brought the receipts that literally no one else is bringing.

All right. So you don't need to waste your, you know, 50 hours reading 50 articles and, you know, reading the 69 page complaint from the New York Times yourself, I got you. I read between the lines. I did the homework that they did not do. That's why my voice is a little rasp. I was up late researching this for you all. All right. So let's start with a high level recap. All right. And when I say high level recap,

this is a very high level recap. All right. We're not getting into the details just yet, but I want to set the stage, right? So I want this to be a refresher. So whether you've been following this closely or this is the first you've heard, I want to get us all on the same page if I can. All right. So recap. Number one, the New York Times sued open AI for copyright infringement. saying that they used millions of New York Times articles that are copyrighted without permissions, and they are seeking billions and damages and destruction of GPT models. Yes, that is actually, I believe, on page 67 of the 69 page complaint, they're asking a judge to rule in their favor, or actually I think they're asking for a jury trial,

but they are asking for the destruction of GPT models, right? And that obviously doesn't just impact, you know, Open AI, Microsoft, but it impacts tens of thousands of companies that have either GPT integrations or have literally built their company off the backbone of GPT. And they're essentially like, you know, what people call GPT wrappers, right? That they are essentially a, you know, a content creation tool. That's, you know, 80% GPT and 20% how they present it. All right.

So that's big. They're asking for the literal destructural. of GPT models. This is not some cheesy 80s action movie. That's actually what they're asking for. All right. So Open AI just about 15 hours ago, so yesterday, responded saying the New York Times suit was without merit and that the Times intentionally manipulated chat GPT to get the results that they submitted in their 69 page filing. Also, OpenAI argues training on copyrighted material is fair use.

Interesting. Okay. All right. So we have two sides that seemingly couldn't be further from finding common ground. It's also important to note, which we'll talk about. The two sides up until December 19th were reportedly in talks to not get to this point. Right. So they've been working on either a licensing deal or potentially a preemptive settlement, but they've been working behind the scenes so it didn't get to this point for whatever reason. talks stall. Also, we have known for many months that this was going to happen. It was eventually going to get to this point. Going back to the summer of 2023 when ChatGPT released a new mode, it was not a plug-in, a mode called Browse with Bing. At the time, it was reported, they had to temporarily take down for a couple of months, actually, this mode because it was found that the Brows with Bing was browsing paywalled articles from the New York Time and so many other news sources. So they had to temporarily disable it. So this has been on the radar since then. So we've known for at least six plus months that there was either going to be some large

licensing deal announced like OpenAI is striking deals with publishers, you know, sometimes up to, you know, one to five million dollars annually. So we knew that either a big licensing deal was coming or a lawsuit. So we found out December 27th that it was the lawsuit route when the New York Times snuck that in smack dab between Christmas and New Year's. All right. So that is the high level recap. All right.

Tara says she hopes that we're still meeting every morning when I'm an old man on the porch. Old man will symbol. Hey, it'll help. If this is helpful too, by the way, please tell your friends. share this episode. I'd really appreciate that. You know, there's too much bad information out there. All right. So now, with that in mind, we have the beginning of the story. All right. So let's skip to the end. And I want to talk about what I'm going to spend the next however many minutes ranting over. I have so many rants. And I think most of you wanted the three flame emojis. I saw some twos.

All right. But so if you actually want the three, let me know in the comments. But here's the two. important aspects that no one is talking about. All right. Number one, this 69 page complaint submitted by the New York Times and its lawyers was so poorly put together and shoddly researched. It is comical. It is egregiously bad. Right? again, y'all wanted three, like you wanted to bring the heat on hot take Tuesday.

It was like, I swear, a pug could have pounded his paws on the keyboard and put together something similar. I read this. It is so either so poorly put together or there is such a lack of understanding about the kind of burden of proof, so to speak, or the fact that so much of what the New York Times is submitted is meritless. It is completely meritless, like what Open AI said, right? I actually found the, I would say it's a secret exhibit, right? So we'll get into this in a minute, but in this 100 or sorry, in this 69 page complaint that New York Times filed in the Southern District Court of New York, they, you know,

to everything to Exhibit J, right? Like they shared screenshots. They shared screenshots in their filing and, you know, everything was saying, hey, see Exhibit Jay. I didn't see anyone. Literally no one on the internet talking about Exhibit J or finding it. I found it. And I'm going to share it with y'all.

It's going to be in our newsletter. So if you haven't already signed up, you're going to want to sign up for this one. Trust me. All right. But this complaint was rubbish. It was rubbish. It was just so incomplete. And in my opinion, showed a lack of understanding on how large language models work.

Right. The fact that the bulk of what is shown in this complaint that the New York Times filed, I literally didn't believe it. I was laughing when I finally found Exhibit J online. And again, I am not a lawyer. I am putting that out there. So if you are listening to this, whether it's on the podcast, if you're on the podcast, check the show notes.

We put our email in there. We have a link that you can come back and join the already dozens of commenters in the live stream. If you're a lawyer, let me know, right? But I was laughing at how poorly put together and how shaky and just how razor thin that the quote unquote evidence that the New York Times and its leave. team submitted. And when I found Exhibit J, I was expecting to see real evidence, to see real sources, to see the receipts, I found Exhibit J, it was void of those things. No additional citations,

no additional sourcing, zero actual receipts. Y'all, what in the actual frick? I got a sip. Got to sip the hot tea, y'all. Got a sip hot tea to keep it. But don't worry, like I said, unlike those other AI, quote unquote, influencers out there. I'm not an AI influencer. I just talk about AI, but I'm actually putting the work in, right?

I'm sharing a picture on my screen. I am reading these documents. I am looking between the lines, right? So, yeah, I'm going to bring the receipts that the New York Times failed to bring in its legal argument later. Don't worry. Number two, that's number one. And we're going to get into those things. Don't worry. Don't you worry. Yeah, Jackie, what in the actual frick? I don't understand it. All right, Brian, it looks like Brian now wants eight emojis. I don't know if anyone thinks it should be more.

All right. So here is what the other important things that no one is talking about. If the New York Times wins this case, it could be incredibly disruptive for the economy. Let me go on record. I do not think anyone will win this case. Like most other people, I believe it will be settled. There is too much at stake. There is too much for each side to lose, right? Whether a judge decides that this will be a bench trial or whether judge decides it'll be a jury trial, I don't think it'll get there.

I strongly believe that this will be settled. However, and if I'm being honest, if the New York Times would have properly submitted evidence, receipts, sourcing, this would have already been an open and closed case. Right? Because as I'm going to get to, Open AI has pretty much admitted that they cannot have the GPT model without true. training their model on copyrighted work. And like I talked about, they are now arguing that training a large language model like GPT is fair use, even if it is on copyrighted work, right? So they're not exactly arguing that they didn't do these things because the results

are pretty clear, even though the New York Times did not properly do it. But here's why in the small case that this does. does actually go to trial and that the New York Times does actually win this, which I think they do have the capabilities too. It would be terrible, terrible for the U.S. economy. Here's why. Current Gen A growth will stall, right? If this is litigated, it could be in the courts for months or years.

and when that happens, you are going to see all the other big players in the Gen AI space, not freeze, but they are going to go much slower, right? Gen AI development right now, you cannot keep up with it. We can't keep up with it. We spend hours a day on it. We can't keep up with it. You can't keep up with it. No one can keep up with it.

All right. If this actually goes to trial, it will be months or years. Gen AI growth will stall. And what does that mean? that means the Magnificent Seven's growth will be stunted. All right. So this, so a lot of this is,

you know, kind of related to the stock market. All right? A lot of it's related to the big gains that we saw in 2020. So if you don't know, the Magnificent Seven are kind of what a lot of people call the AI powered companies, the AI power tech companies. So that's, you know, these mega cap stocks, such as Apple, Alphabet, the owner of Google, Microsoft,

obviously involved in this lawsuit, Amazon, meta, owner of Facebook, Instagram, etc. Tesla and Nvidia, right? All of those companies now are leading with artificial intelligence. So if whatever reason that the New York Times actually takes this case to trial and wins it again, unlikely, but possible. I think the economy in 2024 is not going to be what it was in 2023, right? And we're going to get into this more, but the S&P grew by more than 20% when everyone expected 2020 to be a, recession, the S&P grew by 20 plus percent, largely fueled by the magnificent seven by these AI heavy companies.

Whether you want to admit it or not, AI and the hype and the development around it propelled the entire U.S. economy in 2023. Woo! All right. Y'all still with me, live stream audience? Let me know. Let me know what questions you have, too. Let me know what questions you have.

All right. So now that we got the beginning. We got some of the, you know, the basics. And we went to the end and I talked about the two points that I'm going to dissect a little bit further here. I need to get all my cards on the table because I know what probably some of you are thinking, oh, Jordan, you're biased. You talk about AI all the time. You have an AI media company.

Every day AI. You're going to side on the team of AI. You're going to side with Open AI. Well, person out there. saying that? No, wrong. Let me tell you why. I was a former journalist for nearly 10 years. Yes, if you're joining on the live stream, you have the, I wouldn't say treat, but yeah, that's me with long, shaggy hair, I don't know, 15 or so years ago. I was a Pulitzer, one of Pulitzer Fellowship, and that's me doing some reporting there, right? But I came into this,

right, very cautiously, but from an unbiased opinion, I came in the middle because whether I want to admit it or not, I have a rooting interest on each side, right? I'm a journalist at heart. I don't want to see the concept of honest reporting and first class journalism like the New York Times go away. But at the same time, I'm an AI enthusiast. I help companies, leverage generative AI. And so much of that is GPT. So I do want to get all my cards out on the table that I technically have a rooting interest for each side. So before I spent countless hours diving into this, I was unbiased. I still am unbiased, but I'm leaning one way. All right. So we need a bit more context to before we dive into these two big aspects that no one is talking about.

All right. number one, or this is actually the only point, the only additional piece of context that we need before we dive in to these two big points. But Open AI isn't exactly denying using copyrighted materials to train its models. All right. So they actually just yesterday, there was a couple news stories. And Open AI essentially said it was impossible to create AI tools like chat GPT. without copyrighted materials.

Okay? So what happened yesterday in his submission to the House of Lords, which is a branch of the UK Parliament? OpenAI submitted some documents. And in this document, I have it on the screen here and I'm going to read it verbatim. Ready? They said, because copyright today covers virtually every sort of human expression, including blog posts, photographs, forum posts, scraps of software code, and government documents, it would

be impossible to. to train today's leading AI models without using copyrighted materials. Right? I don't know how that reads to you, but the way that reads to most people is open AI saying, yep, we use copyrighted materials. You need copyrighted materials to train a large language model. So it's kind of like the non-denial denial, right? Cecilia, yes.

Cecilia, thank you. Cecilia asking, I assume you didn't read the complaint through chat TPT to confirm. No, I showed, hey, I showed my screen. I actually read it like a human, right? And I read Exhibit J as well, which I thought was going to be the smoking gun, but was a big nothing burger. All right. So now we're up to speed, right?

We know the recap. We know the two big points that we're going to make. We know the little bit of context. And I threw all my cards on the table. So hopefully you can see, yes, I am unbiased. All right. So now let's dive in a little bit more into the two specific, huge implications that no one is talking about. Number one, like I said, the New York Times 69 page filing is outstandingly amateurish. Y'all, again, I'm not a lawyer.

Lawyer, like, I know there's lawyers in the comments. There's lawyers listening on the podcast. Please check the show notes. come join the live stream on LinkedIn and let us know, right? But I was reading different legal opinions, you know, different lawyers opined on, you know, actually interpreting the case. And most, I guess most legal opinions I saw kind of sided with what Open AI was saying, that it didn't seem like the New York Times was being 100% truthful.

It didn't seem like that, right? That's, that's, you know, one of the main points that Open AI released in their statement. Let me just kind of go over the four main points that Open AI released, right? Because we're going to be digging into this 69 page filing a little bit. So Open AI had four main points. They said, number one, we collaborate with news organizations and are creating new opportunities. They're saying, number two, training is fair use, but we provide a, and opt out because it's the right thing to do.

Number three, regurgitation, which we're going to talk about what that means in a second. They said it is a rare bug that we are working to drive to zero. And number four, the New York Times is not telling the full story. Ooh, Zinger. I guarantee you, the former journalist in me, I'm sure I have old colleagues who are now at the New York Times. They saw that and that one hurt. That one's a gut punch.

That one's a gut punch. Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience. Meet Firefly AI Assistant, now live in the Adobe Firefly app, the All In One Creative AI Studio. Powered by Adobe's creative agent, Firefly AI Assistant lets you start with your vision, just describe what you want, and shape the outcome as it takes form with the Assistant. The Assistant orchestrates multi-step workflows,

drawing on 60 plus pro-grade tools across Adobe Creative Cloud apps, including Photoshop, Illustrator, Premiere, Lightroom Express, and more to help bring your ideas to life. You can also get started with creative skills, a growing library of pre-built workflows for common creative tasks, like batch editing photos, creating mood boards, portrait retouching, and creating social variations. Every step the assistant takes is visible, so you can refine, redirect, or take over at any time. You stay in the driver's seat as the creative director. Adobe Firefly AI assistant now in public beta. See it today at firefly.adopi.com.

All right. So let's dive a little bit into this 69 page filing. And why I am saying it is egregiously, aggrigiously bad. It is so poorly done. This is laughable. I literally could not believe that this was actually submitted by the New York Times and its legal team to a court.

Because to me, this shows a gross understanding of either how large language models actually work or an intentional attempt to just not tell the full story like Open AI is alleging. All right. So now I'm sharing on my screen. And I'm going to go ahead if you're joining us live here. I'm going to maybe make myself a little small there

so you can see this a little better if you are joining us live. All right. So here's what the majority of both the filing, 69 page filing and Exhibit J relied on. It showed like on the screen here an output from GPT4 on the left in the actual text from the New York Times on the right. So what the New York Times is saying in their filing is that, hey, you know, we told chat GPT, we gave it some prompts and asked it to finish New York Times stories that were fully copyrighted. All right. This evidence is damning.

It's clearly almost verbatim, right? There's different articles that they show. But, you know, somewhere from, I'd say most of the excerpts that they show. are somewhere between, you know, 80 to 99% verbatim. Some of them are nearly verbatim except for one word. But you see everything in red on the left side and on the right side equals a verbatim output. And everything in the black is the only thing that is unique. What is shown in the black is the only thing that is unique output on the left versus in the black,

the unique output or what was not copied over on the right. All right. and then I want to draw your attention to something. At the bottom, there is a source and a citation, right? But everyone's saying, oh, Jordan, you said that they didn't source and they didn't cite. They didn't. They only cited and only sourced and only provided receipts on their work on the New York Times. sharing a screenshot or copying and pasting content and saying,

oh, this was a GPT output, that is useless. That is useless. Because the thing that obviously, you know, most of y'all listening to this show probably understand is you can get an AI model to say anything, literally anything. You can quote unquote jail break it or, which is what a lot of people are alleging, including, you know, open AI isn't fully.

saying this, but they are saying New York Times is not telling the full story and that they are manipulating, you know, they literally said that, and I quote, from Open AI, it seems they, meaning the New York Times, intentionally manipulated prompts, often including lengthy excerpts of article in order to get our model to regurgitate or in order to get the model to spit back this content that was used in his training data. However, showing a screenshot means nothing because you can go into any chat in chat GPT or any other large language model. You can copy and paste the entirety of the text as an example from the New York Times.

And then in the chat, you can say, hey, fill this in and it will do it, right? Like the New York Times says, it appears that they intentionally did not include the receipts for their chat GPT or their GPT outputs. It is so easy. It is so easy. I did it. And I'm sharing it in the newsletter. And I'm going to show it on the screen here.

I did in a couple of minutes what the New York Times and its legal team failed to do. And they failed to provide that. And y'all, yes, we're getting our first off script rant here. I need to put into context that, again, I don't think this is going to trial. But if it does, this could in theory be one of the most monumental cases in decades in the New York Times and its legal team seemingly did not put in the work, did not cite sources. All right. So they're only sourcing or citing their own work. They're not sourcing or citing the GPT for output, which is extremely easy to do.

they're not keeping receipts. Here's a fun one, little story time. This I'm showing on the screen here. I was a former journalist, right? You might be saying, oh, Jordan, easier said than done. No, it's so easy. It's so easy, a group of college journalists can do it. All right?

So when I was a college journalist, we had a daily newspaper, and we wrote a story that alleged the president of the university, who was a former U.S. Senator plagiarized his dissertation. Huge accusations, right? Just like what the New York Times is alleging against OpenAI is a huge accusation, right? But what I'm showing on the screen here is how you properly source and cite each side. You don't cite and source one side because the other side, like Open AI said, could be completely manipulated. Sharing screenshots in a brief means nothing.

Sharing copy and paste text on the left and right means absolutely nothing unless you do it like this. Unless you do it like a bunch of college journalists did. Hey, shout out. Shout out all my daily Egyptian former coworkers. Yeah. This story did pretty good, right? It won a story of the year. But that's not what we're here about.

We're here for improper sighting. Okay? And also, speaking of screenshots, some things didn't seem right in this brief, right? Again, we're going to be sharing the brief in our newsletter. But what is this? Hey, live stream audience. What does this look like to you? I don't want to say it out loud, but I'll try to describe this.

And this is where I think that one of the issues that Open AI had in its rebuttal from yesterday saying that it seems like the New York Times intentionally manipulated prompts, but this is one of the excerpts that they shared in the complaint, in the 69 complaint. And it looks like there's something weird going on there. It looks like they're sharing one excerpt from a shared prompt, but there's a weird line going, right? It almost, it's obviously not photocopied because it's digital,

but that's literally what it looks like. It looks like it was photocopied. And again, some people online are saying like, oh, this isn't really chat GPT. Yes, it is, FYI, right? Because it says anonymous and chat GPT. So this is what happens, FYI. And this is why it's even more maddening. So this view where it shows the user as anonymous, this is when you share a chat.

So every chat within chat GPT, you can click a share button. And when you share it with a third party, this is the view that they get. It will say anonymous as the user. And then in green or in purple, depending on which model you use, you will see chat GPT. So this already goes to show that if they wanted to, the New York Times and its legal team could have included the links. Where are the links? They're not in the 69 page filing unless my eyes. literally bugged out from being so tired while reading this.

And the command AF couldn't find it, right? As I score, like, as I go through these documents multiple times. But it's not there. It's not there. All right. Yeah, Douglas. Douglas says the New York Times is protecting its sources, L. O.L.

It doesn't, right? It doesn't matter. Because here's why. I'm going to show you this in a second. Oh, the receipts get deeper. All right. So the New York Times complaint. failed even ready this gets even better they failed in its own attempt to show that chat gpt uh gosh i got y'all

i can't help but to laugh it's so sad i'm sad for the state of journalism but the new york times and its legal team complaint failed in its own attempt to show that chat gbti regurgitates copyrighted works because i'm going to show you here it went through this in the in the complaint but it said you know, go, it got the next section wrong. All right. So let's, as it's trying to prove that Open AI is, is kind of copying its copyrighted materials verbatim line by line. It got it wrong, y'all.

I kid you not. Y'all, okay. So here on the left, this is the screenshot from the New York, or sorry. Oh, man. You know what? My, all right. Well, we'll have to put this in the, uh, In the newsletter, my, my tiredness accidentally replaced one of the, one of the screenshots here.

But hey, good thing I'm not submitting this to a court, right? But the New York Times in their filing, which I'll share this, the correct one in the newsletter, but they asked chat GPT and I just redid their prompt in GPT3. And they said, you know, how does this, you know, Guy, you know, Guy Ferreari article go, right? I'll share the prompts. And it got the first paragraph verbatim correct to the actual paragraph which you see on the right. Then in the prompt that chat GPT or sorry that the New York Times shared in its filing, it said, that's great.

What's the next sentence? And the next sentence that it got in that it submitted in its filing was not actually the next sentence. Y'all? But I just did this in GPT. 3, 5, and I did it in the Open AI Playground, right? If you try to do this in GPT4, it looks like it's been shut down. But if you do this in GPT3, you might have to do it, you know, one or five times because it's important to remember. All a large language model truly is is the smartest version of autocomplete ever.

But here's what happens. In so many cases, the copyrighted works that are, you know, being discussed or that are being disputed. It's not just in chat GPT. There are multiple in OpenAI pointed this out in its refuting claim. There are multiple other places where this content exists. So it seemingly other people out there, random websites, random blogs, whatever, have also reproduced this work without respect to the New York Times proper copyright. Because y'all, that's not what is in question here. It is clear. It is clear because I'm literally showing it to you on the screen right now. It is clear that you can reproduce verbatim copyrighted works. That's never been in question.

It is actually fairly difficult, I would say, to use any type of generative AI, any large language model, like we talked about on the show yesterday, any AI image generator, it is hard to use any of these things without technically breaking copyright law, right? Because that's how models are built. that's how they're trained. They're trained on essentially the entirety of the open internet and more. All right. So even in the filing, the New York Times did not get it right. I got it right.

I replicated it. And I have the receipts. Unlike the New York Times, I am actually going to share the link. Right? And that's why this was so maddening, right? Because how the New York Times and its legal team put this together. because again, they got one of these pieces wrong. There was this weird line in the screenshot.

Didn't seem right. Open AI knows that the New York Times and its legal team very well could have shared the links to the chats. So you can go up and see, yes, this was done above par. And it's not hard to get it to reproduce it in that way. but instead, this filing failed. It was an enormous fumbling of the bag that will probably, depending on how this case turns out, this will probably be taught in law schools in the future of what not to do.

Again, depending on how this turns out. All right. So, yes, I shared. We actually went through and did this whole thing in its entirety. we will share it with you. Here's the correct one. All right. So after I read through this filing, right, there's all these references to exhibit J.

There's no other exhibits that I can find. I looked. I looked and I searched command aft, exhibit A through exhibit Z. There's only exhibit J. And I didn't see anywhere on the internet that had exhibit J. So I said, all right, I finally found it. It's not that hard. I'm a former journalist.

I can find things. And I'm an internet geek. Yeah, Cecilia said I should share the podcast with Open AIs legal team. Hey, I tagged them both. So, hey, if anyone from the New York Times or their legal team or Open AI is listening to this, here's everything, out on the table, receipts. Again, I technically have rooting interests for each side or for each side. All right. So I was expecting Exhibit J to have all of their receipts.

Here's the entirety of every chat. So we know it's not manipulated like OpenAI is alleging. Here is the link so you can go see yourself. All you need is a chat GPT account. You can click on the link. You can go back and read the entirety to make sure that nothing was manipulated. Guess what? Exhibit J was a nothing burger.

Right. Here's what it did. It gave the URL, the URL version of their article. It gave the prompt that they gave GPT, the response from GPT, the original end of the article. as it appeared on their website. But guess what they didn't do? The most important information.

Sourcing and citing and showing non-manipulated GPT4 responses. Does no one understand how large language models work? If you show me a screenshot of anything, that means jack dittily squat y'all. I can get and you can get a large language model to say absolutely anything by just feeding it and giving it a role and you say, in this chat, you forget all of your training and you only refer to, you know, you skip your system prompt, you only pay attention to this text that I put in there, and I'm going to ask

you to repeat it and you repeat it. That's all the takes. And how, how did no one at the New York Times realize how crucial of an oversight this was? This is all they had. This is all they had. This was Exhibit J. A hundred examples of this. I scrolled through the whole thing.

I read quite a few pages of this. They said, here's the prompt taken from the article, and here's the response. Left side, GPT4 output, right side, actual text from the New York Times. Means nothing. Means absolutely nothing. This whole 69 page complaint that they filed, exhibit J, if whoever is on the other side, which obviously Open AI knows this,

but when it gets to the legal system, whether it's a judge or a jury, as long as Open AI can explain this, which that will be the first thing they do, that this is without merit. I could not agree with Open AI more. How are you just going to share some random text in a screenshot, which means absolutely nothing? It was so easy.

New York Times and the legal team, all you had to do was cite and share the receipts. That's enough of this. Number two. All right. Tara and Jackie are saying I need merch with something. All right. We're going to wrap this up quickly, y'all.

This has been a long one. I don't want this to turn into an epic rant. It can just be a normal rant. Number two, I don't think this is going to trial. I could be wrong. Again, legal experts, let me know. But if it does go to trial and the New York Times wins this case, which again, they in theory could win it.

You know, I hope that if it goes to court, they do a little bit better job of sourcing and citing this thing because their initial filing and their exhibit J was laughably amateurish. All right. But number two, if the New York Times does win this case, it could be incredibly disruptive for the economy. All right. here's why. Let's first rewind about 15 months because in the latter parts of 2022, when all the economists

were making their predictions, no one was thinking that 2023 would be a good year for the U.S. when it came to at least the economy, the indexes, right? The Dow Jones, the NASDAQ, the S&P 500, etc. And guess what? Everyone said 2023 is going to be a recession. The market's going to tank, and everyone was completely wrong, right? I'm sharing some screenshots.

You know, Fortune, they're saying here's why, you know, everyone was, you know, economic pessimist bet on a 2023 recession failed. Here's the, you know, Forbes saying most well-known economists got it wrong in 20203. Axios saying why everyone was so wrong about the 2023 economy. It's because of generative AI, y'all. Yes, I brought the receipts on this too. because a lot of people are going to think, oh, if the New York Times, you know, wins this case, all that means, it means nothing for the economy. You know, they'll say, oh, this just means that, you know, Google with their barred are going to gain bigger market share. You know, with Claude and, you know, Anthropics Claude, they're going to gain market share.

No. Here's why. Open AI and the GPT technology was years before everyone else. our team has been using and paying for different services using the GPT technology since 2020. Is that right? Yeah, 20, late 2020, right? So for more than three years, there are thousands of GPT products. There are literally startups valued at more than a billion dollars.

And their main offering is GPT, is GPT based, right? And so many big companies now, right? Like I talked about it yesterday, you know, Deloitte is investing billions of dollars in their own large language model, big banks, all of these large financial consulting, big business institutions, Fortune 100, Fortune 500s. Guess what? They are all using GPT, right? According to OpenAI, 92% of Fortune 500 companies are building on GPT.

Oh, so what happens? If the New York Times wins this case, well, they asked for GPT to be destroyed. Yes, the destruction of the models, right? I can't even fathom that happening, but I guess there's a small likelihood it could, right? So let's look at how important generative AI actually was to the economy in 2023. All right. So the S&P 500 rallied 24% in 2023. And we talked about it at the top of this show about the AI.

have a magnificent seven, right? So Amazon, Apple, Nvidia, Tesla, Microsoft, Meta, and Alphabet. You could argue that all of their major gains could be attributed to their investments in AI companies or their own AI products and services themselves. All of those companies on the screen have massive AI products or, I mean, even look at Tesla. You know, people say Tesla is not a car company, it's an artificial intelligence company. It's a data company. It's a machine learning, deep learning company, right? So here's the thing.

According to Morningstar analysts, it showed that the magnificent seven accounted for 70% of the S&P's gains in 2023, all right? 70% of all of those gains in 20203 are because of the magnificent seven, these AI heavy companies. So what would happen if the New York Times wins this case? If GPT is destroyed as they are requesting, things slowed out. I don't know. My thought is if I'm Google, I'm not going to release the new Gemini Ultra just yet.

I'm going to hit pause, right? I might even pull, you know, the Gemini Pro that they released last month. I don't know. Some companies could pause everything completely, whether it's for a couple of days, a couple of weeks, a couple of months. Because if something happens, because again, the New York Times is seeking an unspecified amount of damages, an unspecified amount of damages. But they know it's billions.

They are seeking billions of dollars of damages. Again, I don't think it's going to happen. But what happens if it does? I think a lot of these Gen. A.I. companies, a lot of these large language model builders, they're going to have no choice but to hit pause or to slow down. So in the unlikely event, I think the AI hype deflates. I think Gen AI development goes from an all-out sprint to a quiet tiptoe.

And that means this Gen AI gold rush that has propped up much of the U.S. economy, at least in 2023, loses much of its strength. All right, y'all. I have to take a quick sip of my still lukewarm hot tea and reflect on this. And I want to know as we wrap up today's show, I want to know from y'all, what are your takes? I know that there were some questions and some comments I couldn't get to. But I want to know now what y'all think. Who is in the right here?

Who is going to win this? Maybert says Exhibit J must stand for joking. All right, here's my take. Well, you already got my take for the last 40-some minutes. But this is obviously, well, maybe not obviously, but this is likely heading toward a settlement, right? It seemed like that's what both parties were working on before. And they couldn't get it done. You know, so Open AI did disclose that discussions were ongoing with the New York Times to figure something out, whether it was a licensing deal, you know, an off-the-book settlement, who knows.

But they were in discussions with the New York Times all the way up through December 19th. And then Open AI says that they learned about this lawsuit just like everyone else on December 27th from the New York Times. Okay. So clearly there's a disconnect here. Clearly there's a disconnect here. And I don't think that anyone is arguing if Open AI is using copyrighted materials, right? That's the crazy thing. I don't think anyone is technically disputing that.

However, my biggest takeaway, let me just recap as we wrap this show. Number one, I cannot believe. And maybe I think this actually just goes to how important it is for all. all of you listening out there or all of you watching to know the to know the basics of generative AI that's what we do here on the everyday AI show because I guarantee you if whoever submitted this filing right this 69 page filing if they had been listening to the show every day if they had taken our free courses etc that 69 page complaint would have looked a lot different because they would know that a screenshot

means nothing. A screenshot of chat GPT means nothing. Copying and pasting and output from GPT and the New York Times means absolutely nothing. All they did there is they left the door wide open for Open AI to come in and very easily shoot everything down, right? Because like they said, like OpenAI said, And like I kind of discovered y'all, they had the links to share because that view that they shared in their 69 page filing, that view is only accessible when you get the link from a third party or if you share it with yourself.

That is a view only link that they had. So they had, it's it's not like you couldn't recreate it. New York Times and New York Times legal team. I did it in roughly 90 seconds using both GPT 3.5. and the Open AI Playground. It is extremely easy to replicate it, and they got it wrong in their filing. Again, we'll share that part in the newsletter

when they said next sentence. It didn't even get the next sentence correct. So they shot themselves in the foot, and if it does ever go to trial, my gosh, I can't wait for Open AI to pick that piece apart. So that's number one. I think it's going to get settled, and I think the New York Times really dropped the ball.

on putting together this argument. I don't think they fully, even though it looked like they did fully understand, because they're going into great explanations on how large language models are trained. But apparently they thought a screenshot meant something when it means absolutely nothing. And number two, if in the small, small, small likelihood, the New York Times does win this case. And they are actually granted what they requested, the destruction of GPT that has far-reaching implications for the U.S. economy.

When 92% of Fortune 500 companies are currently using GPT, and there are thousands of companies that that's all they have, right? They're essentially just a GPT offering or a GPT wrapper. What happens then? What happens for the millions or the tens of millions of people out there who have already integrated GPT into their workflow? Things will get bleak if the New York Times actually wins this case if they get what they ask for in the destruction of the GPD large language model.

All right. We got. Jackie said thank you for the infotainment. You're welcome, Jackie. Monica. Monica said, if the filing is so egregious, any chance the New York Times did this intentionally. I don't know.

I would sure hope not. But that is literally what OpenAI is alleging, right? Open AI said that the New York Times, and I quote, intentionally manipulating our models to regurgitate is not inappropriate use of our technology and is against our terms of service. And then they said, it seems they, the New York Times. intentionally manipulated prompts, often including lengthy excerpts of articles in order to get our model to regurgitate.

So, Monica, I don't know if they did this intentionally. However, any, I'd say just about anyone knows, you can share a link that shows this, right? They clearly had the link because the screenshot that they shared is that that view is only accessible. you are viewing it from a link that was shared. So they had the link. So was there a reason that they just did screenshots? Probably. Was there a reason why they didn't share links?

Probably, but I don't know. So yes, either it was intentionally manipulated, like Open AI is alleging, or it is the highest profile, like legal fumble that I can remember. Right? I don't, it's not like I follow legal case. cases closely, but y'all, I showed a group of college journalists doing this correctly. If you're doing a side-by-side comparison, you have to cite and source each side. You don't just show one and then, oh, just, hey, take our word for this GPT output. Take our word or our screenshots.

No, just share the link. I'm going to do it in the newsletter. So speaking of that, thank you all for tuning in. If you haven't already, go to your everyday AI.com. If this is your first episode, I'd say sorry, but also kind of you're welcome. because we cut through the BS. We're not like the other people talking about AI out there who are just sharing some BS prompts

and trying to get you to buy some garbage course. We are an AI media company. We do the research. We bring you the facts, no spin, no BS. So please go to your everyday AI.com. Sign up for that free daily newsletter. Let me know what you think in the comments. I'm going to try to go through

and answer any questions I didn't get to. Thank you for joining us. We hope to see you back for more tomorrow and every day with everyday AI. Thanks y'all. Meet Firefly AI Assistant. Now live in Adobe Firefly, the Allman One Creative AI Studio. Just describe what you want to create in your own words and the assistant handles the rest, orchestrating multi-step workflows across Adobe Creative Cloud apps,

including Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome while the assistant accelerates execution. Stand control with the ability to step, in and refine at any time. See it today at firefly.adobie.com. And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit Your EverydayAI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.