Damien and Nikola brag about their impressive Duolingo streaks... and then discuss the impact of AI on modern journalism - including issues from OpenAI and Perplexity's data-scraping controversies to the ethical implications on the journalistic landscape. They debate (kind of) over the balance between maintaining journalistic integrity and the rise of AI content generators, potential solutions like pre-emptive licensing fees, and the broader effects on information consumption and media bias. Where do you stand on the future of journalism in an AI-dominated world? 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.
NikolaTake your model and you general a lot of data and then you go train that data. You're not really adding entropy or new knowledge into that sharing process. You're just burning cycles. a lot of it. It's kind of like reading, I don't know, junk novels, right? You're not really learning a lot. They're very easy to go through. It can power through a hundred pages in. A few minutes, but if you pick up Dostoyevsky you're not going to power through a hundred pages in a few minutes
DamienHello, folks, and welcome to another episode of the Deep Learning with Polly AI podcast, the only podcast that has a host and guest who have 1000 plus day streaks on Duolingo. I think that's right. Is that right? I think so. Is my researcher correct on that? Because my researcher is me. Are we friends on Duolingo? What's your hand? I don't, I don't think so. It keeps suggesting you. That's how I know you got 1200 days or something like that. You're a beast.
NikolaI'm offended. You must add me. And we, there are, there are all sorts of new streaks. So how do you feel about the LLM features?
DamienOh I the max client, I think most of those are embedded at the max level. At this point, we've talked enough about Duolingo, I think they deserve to send us a sponsorship check, so Duolingo, get on that, please.
NikolaYeah, look, Duo haunting me, in the middle of the night. It's called Nightmare. Gotta keep those streaks alive. I hear and love Duo very much. For those who don't know, Duo is the great green owl, and there are great memes about long term users having phobias about Duo appearing behind them and going, Hey, you're going to miss your streak. Your streak, your streak is in danger.
DamienYou don't want to, you don't want to rile our great god Duo. What do you learn? What do you learn? I'm still stuck on Spanish. I did French for a short stint, but I've been taking Spanish now for, I think, greater part of 15 years and and so I recently got back into it. Now I'm like three years strong on my streak. So same to you, good sir, and may we see each other at the finish line. Absolutely. Aside from Duolingo, which I could spend the entire episode talking about, to be honest, I
Nikolawanted to derailed us
DamienLet's just do a pivot. Let's do a pivot now. I wanted to chat with you about something we've touched on here and there because it's, it seems to pop up in the news going as far back as last December when New York Times sued OpenAI over scraping our content. Recently there was a couple steamy tweets coming across from the editor over at Forbes Tech about how an article that. About Eric Schmidt's combat drones and his fascination with the things got scraped immediately by perplexity and resulted in almost zero clicks to their digital content.
DamienWhereas perplexity got tons of views even though it was lightly cited, still a little bit of just immediate take and rebake. Let's talk about AI and its place in modern journalism some good and bad effects that may come from it and the like. So first and foremost, tech has been scraping content from the web for years. Is what's happening with the complexity and abilities of modern AI, especially in generative with LLMs, is it. Is it really more serious, less serious about the same and it's just in the Zeitgeist now, so we're putting more weight on it.
DamienWhere does generative come into play when it comes to content scraping?
NikolaI'd probably say it's about as serious as it ever was, right? I think, people more than ever now as models multiply. We'll tell you the data is all that matters and that it's the new gold and hence, they're going to go out of their way to protect it and people will go out of their way to use their data to create the best possible product. So yeah, I think it's like the whole perplexity thing. Yeah. I don't know if you want to give a bit more of an intro of exactly what happened. But they feature them one of the new features they have where they basically show an article.
NikolaWhich I guess feels as if it's part of their publication and as if their algorithms came up with it. But really it's like reacting to a publication by, was it Forbes? Yes. Yeah they showed two citations to Forbes and one to, I think, Business Insights or something like that. And another one somewhere else. And I guess, in there they probably didn't do a topological sort of what came first. That, it was probably the Forbes article that caused the other ones and that they deserve, hence, a bit more of the attribution.
NikolaI guess Forbes would have probably liked for them to just Not show it as a new article, but really just
Damienyeah I think the chief complaint coming from the forbes editor was that the article was up for seconds before it was scraped rewritten Yeah, there was a minor attribution, but for all intents and purposes. It looked like it was a primary part of perplexities content and so therefore it absorbed like a ton of clicks and you could see the views it got something like 18 000 by the time the editor had published it. And those were clicks that he felt were deserving of the original piece.
NikolaI think at least I recently listened to, I was on a flight as I often am. And to the perplexity CEOs Lex Friedman podcast. Not a lot of it because Lex goes on for three hours and we don't have that kind of time, but it was interesting. I think that, perplexity as a search engine starts from like a very noble academic oriented approach of let's cite everything diligently and make sure that, I guess their main goal is veracity and like knowing where things come from rather than just, being hallucinated things that kind of look like they might.
NikolaFrom starting from there, genesis, they're still true to that. Nothing there is hallucinated by neural network. I think that technology must evolve in the futures, and I'm sure they'll very quickly. Find a way to maybe have wireframes and say Forbes this is interesting, augmented by us, perhaps, or not. Okay. Yeah. I don't know how they'll deal with it. Really. At the end of the day, it's it should be sorted. It will be sorted. Especially because this is not really so much about training as much as it is about the instantaneous nature there. So they're probably going to have to find a way to share those clicks.
NikolaOr to, we need some way of like sharing that currency, the same way that would page rank, you lend credibility through the backlinks and stuff here, even if they do it may become it depends on how far perplexity goes and how successful it is. Maybe they get to dictate some of these things.
DamienRight.
NikolaMaybe they don't.
DamienAnd maybe to your point, it's, I personally, just as an outside spectator, and as a former journalist, see it as being the training versus commercialization of the content itself. OpenAI, obviously put under the press by New York Times back in December. Who knows? I think that lawsuit is still in action. But essentially the case was being made from open AI is we're trying to train our bots to be sharper, to be better, to give a better output. That's not the case for perplexity and Sam Altman went so far as to say, okay, then I'm going to start, he's currently on a bender of going out and offering licensing fees to all these outlets.
DamienI think who are somebody Associated Press, News Corp, which is the parent company for Fox Financial Times, Fox Media, a bunch of others. He's saying, okay, I'm not going to fight you on this. Here's some licensing fees. Now I'm going to use your content. Do you think that Sam's on to something? Do you think that open AI is basically creating the next commercial model for modern journalism?
NikolaI don't know. Are they, I mean like the, he's really after data and using it to train models, right? Like perplexity is the kind of like. LLM instance of the instantaneous and how people access things. I think, it really depends on, how LLMs find a way. And will they find a way to be the main source of information? Because one school of thought is that with LLMs, things like perplexity will just show you things. But look, what's the, what's the Apple news thing called? Where like it curates the different kind of like things you have subscriptions for, and highlights your attention to certain things.
NikolaI think the news plus feed or whatever that they have the feed, right? Similarly, you open a browser and it gives you some recommendations. I don't really pay much attention to them. If anything, like my YouTube is actually curating some news for me as well. I don't know exactly how we end up consuming things that are purely LLM driven, like that kind of assumes a very inward looking person that is fed things by the media, which I guess most people are, right. Yeah. But I think There is a lot of the there is a great company I'm forgetting now, but it curates basically news articles and tells you which side of the political spectrum they, they come from.
NikolaI'm sure. Yeah. Ground news, ground news, I think is the name. And, they go out of their way to just show you a balanced set of like different viewpoints and things. So I don't really know like where we end up with. With LLMs and like the whole I don't think Sam Altman is after that. I think Sam Altman is after AGI, right? What more powerful models as we approach AGI will do for journalism? I don't know. A lot of journalism, as is really about doing the work, like asking the question, like going out of your way and plugging yourself into things that are not necessarily online.
NikolaI think that LLMs are they're online, right? For now, they're online. And I don't know how that's I don't think he's after journalism. Do you think he is?
DamienI don't know if it's after journalism. I think first of all, open AI is more money than God, right? So they can pretty much pay licensing fees to whomever for whatever they want. I think he's just probably trying to avoid any legal snafus as it comes to, essentially, yes, they're looking to train, but he's not going to go out and, Essentially create a news organization. I don't think that's his intent, but he's, I think it indicates that he's showing recognition that this is livelihoods of reporters, that this is, an entire industry. And it's not a cottage industry. This has been around for decades, centuries, even what I'm concerned about is what the sort of gray area that organizations like perplexity are operating in, which is functionally creating journalism.
DamienBy rewriting other journalist's work instantly. And capturing
Nikolaviews. I mean listen, if you want to be judgy and like strict about it. Oh I do. Okay. I mean I think it's an honest mistake. But you know we can be on different sides of that one. But strictly speaking you're right, it's plagiarism. It's very anti academic of them, but I really think it's an honest mistake. I don't think that, they move fast.
DamienBut it could be, it could be argued as well, that this is what human reporters do as well. So they catch something, they say, as reported by Forbes, blah, blah, blah, blah, blah, blah, blah. They put their own spin on it. They crank out sources. So what we're talking about is a hyper efficient competitive property that goes out and does something. So there are some ethical impacts. The same ethical impacts are not really the same, but the question that could be made now is that with let's look at what OpenAI is doing with licensing content. Now, all of a sudden, it's paid reporting, and reporting is supposed to be unbiased. Reporting is supposed to be pretty organic.
DamienIf you have a benefactor in this You feel that way? In this case, on modern reporting is a totally different beast from back when I was a journalist. But I do think that there is something to be said about what is the overall ethical impact of the what organizations like OpenAI are doing, which is, hey, we want to license the content, yeah, it's to train, but also we're paying you for your reporting. What is that going to do to your reporter's take on stories that involve OpenAI?
NikolaI think it's going to lead to an overall tabloidization. You mentioned use of the world, right? I think that Funniest comment section I've read in a while now is a telegraph article about Rupert Murdoch's sixth, seventh marriage, right? Like it felt like, they were covering a fairytale wedding and without going into it the bulk of the comments were, Hey guys, he doesn't even own you. How afraid of him are you? I think that's what I think about it, which is that. Follow the money, right?
NikolaI think that for the most part, you already have a world in which, unless you I was always surprised, right? Like growing up in like post Milosevic Serbia we were pretty strict for a while around like all media being pretty balanced and biased and sure they had a bit of a twist, but I think I was quite surprised when it moved to the UK and you had The Guardian being very clearly vociferously pro labor and the Telegraph is pro conservatives. And, in America some Fox News, I couldn't believe that, right? And in the end, it really boils down to the same thing, right?
NikolaIt's just the variance that you have in there and the polarization of the country that It's creative for it for what it's worth. It looks like the American model is winning, right? Where that kind of like very strong, clear orientation, at least where the cards are dealt and you can at least create like at least a bipolar thing and you balance it in that way. Now will AI balance it? Hey, I don't know. Ground use for instances where they're like, Hey, go look, take, go hit yourself with the far left and hit yourself with the far right.
NikolaAnd make your own call. I think I struggle to think of truly balanced media anywhere these days. So can AI do better? Will you trust it to do better? I think, you should be
Damiensalient question. Then you took the words right out of my mouth. Could LLMs be the great neutralizer of polarized outlets? I, and to be fair, I don't think that. And I think Fox News even jabs at itself and says we're entertainment. We're not news, because that's that's in the right. So it could be said that this is an actual news, but it is you ask a person what their chief news sources and they're going to point out one of these outlets. That has been accused of, if not directly claims a certain, political bias social bias, et cetera.
DamienSo could LLMs in their probabilistic approach to neutrality of information and could it be the great bias equalizer?
NikolaI think you could set your sights on creating one, right? I think the reasoning capabilities of LLMs are nowhere near the ability to really critically point out the weaknesses. Of all those different things. I think the, the tabloidization of media is a reason for why Fox does what it does, right? They're probably a culprit as well as just a practitioner of this of this form of journalism, if you can call it journalism, but I think the interesting thing is you'll build it. Will people look at it? Will people care for it?
NikolaDo people want the biased version? I think many people do, right? We love our echo chambers and 5 clips on YouTube. That go either way politically and you're stuck in a loop of that part of the echo chamber like no matter where you are actually on the political spectrum. So I think like you just have to follow the money. I think when it comes to these large platforms that monetize based on consumption. They'll do anything to keep you hooked and consuming, right? To think that they would, I think we need strong media exactly because their mission should be to not be just this, but it's not been an easy few decades for the media at large.
NikolaLike when you think about what Alton is doing and just skipping the whole spiel between he goes a bit more gun co uses things, they win a lawsuit, there's an internet. And then they. Subtle they might as well just settle and figure out something that works for all sides. I don't think he's the issue there I think like the media as they are right now, llms Are in my opinion more dangerous for that media because the other big part of course is Everything around twitter slash x and the comment sections, right?
NikolaYes, like most comment sections today's, if you read about russia ukraine, you'll see that for every comment There's a sub comment saying that you're like You're a bot advocating for this site, right? Half of those comments are probably written by bots as well. So I think if anything, technology is adding to the cacophony of the echo chambers.
DamienOkay.
NikolaMaybe there's a way to use them to cut through the noise a little. We're not talking about. So are
Damienyou, are you validating the theory that 90 percent of the content on the internet is now being put out by bots and that it's essentially training the next generation or next iteration of LLMs to think as it's guardrails are
Nikolaintending. No, the truth is the only thing this will do is drive up the share price of Nvidia, right? Take your model and you general a lot of data and then you go train that data. You're not really adding entropy or new knowledge into that sharing process. You're just burning cycles. So go buy some Nvidia, kick back and watch us probably slow down progress for a while until people find a way to just filter out a lot of it. It's like reading, I don't know, junk novels, right? You're not really learning a lot.
NikolaThey're very easy to go through. It can power through a hundred pages in. A few minutes, but if you pick up Dostoyevsky you're not going to power through a hundred pages in a few minutes unless you choose to, skip the classical 20, 30 pages of the description of a step, which so yeah I think it's it's interesting, right? There's a lot of garbage out there and it's mostly done for SEO purposes.
DamienYeah, and unfortunately we're already seeing some, articles that are coming out even in the past couple of days with the hotly contested state's presidential election and campaigning that's going on. There's already accusations about is OpenAI responsible for GPT throwing out misinformation, regurgitating old school misinformation. People are calling in to reinvestigate section 230 of the Communications Decency Act, which basically says a website isn't responsible for content that people put on it. As an outlet, but what if the website is running a bot that is putting content on it?
DamienIs it then responsible for it? And to your previous points where the idea of a modern website is going to be something that's Basically curated by llms in the future anyway, and it's going to go away. So marketing angles are removed, etc i'm curious to see what the next evolution of journalism is going to be
Nikolalook I mean everyone will be augmented by and we'll be using all i'm sure but I think that like Journalism is like a really cool job and a really important one. I don't think it's going anywhere. I think that in its fight to stay independent and, rid itself of these like influences, like that's its own fight, and I think it will always depend on people who are idealists about it. And, want to fight the good fight. I don't think there's anything new here. I think sadly, a lot of what this technology is doing is dumbing down the collective intellect of society because it makes content so easily accessible
Damienand
Nikolapresented and force fed to you, right? That it's For most people, for all of us, really, it's making us use our critical thinking a lot less while doing it. That's a good point. And that's tough. That's really tough. And that's not just an LLM effect. The internet did that to begin with, right? Oh,
Damien100%. But now it's fast tracked as if it wasn't fast tracked before, but it took 20 plus years to get to where we are now. And that's only going to accelerate, I think, because of the instant availability of information on
Nikolathe other hand, it's, created things like the Arab spring and enabled, if you're from a country where, TV outlets are controlled by the government the internet is unless you're in China, the internet is not right, and it's very easy to do it. And even in China, the satirical comments on a lot of their media are like some of the funniest ways that humanity fights censorship today. Like where people are like, the comments are legendary. I don't know if you follow this, it's very common for. An article that is not trusted to get like a hundred comments on the lines of my 120 year old wife just gave birth to twins, right?
NikolaIt's basically an implicit signal from people that that is hogwash. So I don't know. I think that it's just getting more interesting. And I think if you were transported from the 19th century into all this, you'd feel like your head's about to burst from everything you have to figure out, but It was probably always the same, right?
DamienYeah, I think if you're a transformer from the 20th century, your head's gonna burst once you see an iPhone. I think that's that'll be the first thing that sends you overboard.
NikolaYeah, look, I think like you think of journalism back then, I think of like McCarthyism and yeah, all of that Hey, we've always had like authoritarian bands. We've always had like different things that were done to journalists. Fair enough. They figured it out. They'll figure it out again.
DamienThe scale. The scale's a little bit different though. We can agree on that. Yeah,
Nikolait's
Damienalright. I think we're, I think we're wrapping up on time. Appreciate the conversation As always, very thought provoking points. I wish you your 100 and 20-year-old wife and your twins, all the best and we'll talk to you next week. All right. Good,
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...
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...
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...
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...
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...