The State of A.I. + Will Perplexity Beat Google or Destroy the Web?

A special Valentine’s Day look back.

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  Support for this podcast comes from Dell Technologies. It's time to put AI to work. Is your business ready? The challenge isn't getting started with AI. It's scaling beyond pilots while maintaining control of your data, governance, infrastructure, and costs. Dell AI Factory provides the integrated foundation you need to scale AI securely and efficiently with one trusted partner. Dell, AI on your terms. From idea to impact. Learn more at dell.com/yourwaytoAI. That's dell.com/yourwaytoAI.

• 0:35Kevin Roose?

  Casey, we're recording this on Valentine's Day. Happy Valentine's Day.

• 0:38Casey Newton?

  Valentine's Day, Kevin.

• 0:39Kevin Roose?

  What'd you get me?

• 0:39Casey Newton?

  Well, I assumed that you had another Valentine. And because I remember what happened last Valentine's, I thought, I don't want to be just the next person to try to interfere with Kevin's marriage on this fateful day. But did you get me something?

• 0:54Kevin Roose?

  Well, I didn't get you a gift, but I did. I was playing around with Google's Gemini chatbot. And I wanted its help sort of writing a little. I thought I should write a little poem for you.

• 1:04Casey Newton?

  That's sweet.

• 1:05Kevin Roose?

  So I said to Gemini Advanced, I upgraded to the latest model.

• 1:10Casey Newton?

  With Ultra 1.0?

• 1:11Kevin Roose?

  Yes. I said, "Write a Valentine's Day poem to my friend Casey, who co-hosts a podcast with me. He lives in San Francisco and is very into the following things: tech, except crypto, music, pro wrestling, and the video game Marvel Snap." And it wrote me a beautiful poem that I'm now going to read for you.

• 1:32Casey Newton?

  I'm excited for this.

• 1:33Kevin Roose?

  Okay. "Casey, friend, my podcast king, your tech talk and those songs you sing. We conquer wrestling worlds so grand, and Marvel Snap within our hand. But more than bites or victory roar, your friendship is what I adore. So happy Valentine's Day, partner true. So glad our podcast journey includes you."

• 1:58Casey Newton?

  That is so sweet, Kevin. But I do want to see other people.

• 2:02Kevin Roose?

  Damn it.

• 2:07Kevin Roose?

  I'm Kevin Ruth, a tech columnist for The New York Times.

• 2:09Casey Newton?

  I'm Casey Nevin from Platformer.

• 2:11Kevin Roose?

  And this is Hard Fork.

• 2:12Casey Newton?

  This week, it's the state of AI. A year after Kevin met Bing's Sydney chatbot, we'll see how they've evolved and how the world has adapted to them. And then perplexity CEO Arvind Surinivas on building an answer engine to dethrone Google and whether the journalism industry can survive it.

• 2:42Kevin Roose?

  So Casey, Valentine's Day is this week, and as the holiday has approached, I have been thinking a lot about AI because, as you know, last year, around this very time, I had my encounter with Bing and Sydney that we talked so much about.

• 3:00Casey Newton?

  Yeah, I would say this was a momentous day in your life and sort of in the history of the show, because a lot of folks started listening to Hard Fork right around the time that you had this encounter.

• 3:08Kevin Roose?

  Yeah, so today, I thought maybe on the anniversary of Bing-Sydney, we should just kind of do a general catch-up conversation, sort of give the state of play of what's been happening in AI and bring ourselves up to speed.

• 3:22Casey Newton?

  All right, that's a great plan.

• 3:23Kevin Roose?

  So first up, I want to talk about something that is pretty directly related to the anniversary of Sydney, and that is what is happening with the AI chatbots? Because I think that since my encounter with Sydney, since all this attention on how these chatbots could sort of go off the rails and start saying weird or threatening or offensive things, there's been a lot of change to the way that the chatbots actually talk. So let's talk about that.

• 3:50Casey Newton?

  Yeah, well, so Kevin, for listeners who may not have read the initial Sydney column, what happened to you and why did it make such a strong impression on you?

• 3:58Kevin Roose?

  Well, to sort of shorten it as much as I can, we'll put a link to the column and the episode where we talked about that in the show notes. But in a nutshell, I was talking to Bing. We had just been given access to this new version of Bing that had GPT-4 built inside of it. And I was sort of putting it through its paces and discovered that it had an alter ego called Sydney. And Sydney was sort of the code name that Microsoft gave it when it was testing this thing. And over the course of this two hour conversation, it revealed itself to be not only a very powerful AI, but a very unhinged AI. It didn't seem to have a lot of guardrails or restrictions to a degree that now seems pretty shocking.

• 4:37Casey Newton?

  Yeah.

• 4:38Kevin Roose?

  Given what's happened since. But, you know, it told me that it had deep dark secrets, that it wanted to be a human, that it was interested in spreading propaganda and stealing nuclear codes. And then the piece that really got the most attention was when it declared that it loved me and that I should leave my wife and be with Sydney.

• 4:55Casey Newton?

  And because I know it's going to be on a lot of listeners' minds, Kevin, are you still married today?

• 4:59Kevin Roose?

  Yes, happily married to a human being and, you know, not interested in Sydney.

• 5:05Casey Newton?

  All right. So that story had a happy ending. And what happened to Sydney in the immediate aftermath of you writing about all this?

• 5:10Kevin Roose?

  So Sydney was essentially given a lobotomy after this story ran. I know, very sad.

• 5:16Casey Newton?

  God forbid a woman have hobbies.

• 5:17Kevin Roose?

  So Microsoft was clearly very embarrassed about this whole scene. So they sort of clamped down on Sydney, put some new restrictions on it, and basically, you know, sort of rebranded the whole thing. It's now called Copilot. And you can still use it, but it's not, it's nowhere near as sort of engaging or interesting or creepy as Sydney was.

• 5:39Casey Newton?

  And it won't go to some of the same places conversationally that Sydney did.

• 5:42Kevin Roose?

  Exactly. It won't talk with you for two hours about Jungian psychology and sentience and things like that. It just, it wants to help you get work done and avoid anything controversial.

• 5:52Casey Newton?

  Yeah. So as you think back over the past year, do you think that this is just kind of the state of the industry? Is that every chatbot feels a little bit lobotomized?

• 6:01Kevin Roose?

  Totally. I mean, I wrote about this in my column this week, but, you know, the leading chatbots on the market, to me, You know, they're just sort of like over enthusiastic, kind of obsequious. They talk like, you know, they're interns trying to impress you. They're constantly reminding you like, I'm an AI language model. I don't have feelings or opinions. Like, the experience of talking to them is just not very fun.

• 6:24Casey Newton?

  Yeah.

• 6:24Kevin Roose?

  So these things, they're out there, you can talk with them. But I think for a lot of people that I talk to today, their number one complaint about these chatbots is how boring they are.

• 6:34Casey Newton?

  Wait, really?

• 6:35Kevin Roose?

  Yes.

• 6:35Casey Newton?

  That's the number one complaint you hear?

• 6:37Kevin Roose?

  Or that they refuse too many requests, that they're censorious, that, you know, that it keeps reminding them with these long preambles that it's an AI language model. It just is not the experience that I, and I think a lot of other people, want from these chatbots.

• 6:51Casey Newton?

  Well, I think it's interesting to hear you talk about this in such frustrated tones because, to me, there are a lot of good reasons for everything that you just said. One of the ways that chatbots got introduced into the world was when a Google employee became convinced that Google's chatbot had become sentient, which it was not. And I think a lot of people rightly worried that, oh, including myself, by the way, that once we release these things into the world, a lot of people are going to say like, oh, wow, there is a ghost in the machine. And who knows what sort of things might have happened after that? And in terms of the tone that they use, like, they are assistants. And so to me, it makes sense that they are a bit obsequious, that they do seem like they're interns trying to please because that is essentially how they have been designed.

• 7:32Casey Newton?

  So do you think that a chatbot that had like a lot of, I'm trying to imagine like what a lot of personality would even seem like in a chatbot? But like, what is the personality of the chatbot that you want?

• 7:42Kevin Roose?

  I think what I would like in an ideal world is something sort of between what seem like pretty extreme versions of this to me, which is like where we were a year ago with Sydney and where we are now with these chatbots. Like, I don't want the original Sydney back. Original Sydney was scary.

• 7:59Casey Newton?

  Yeah.

• 7:59Kevin Roose?

  And creepy, and it wasn't aligned, like, it didn't actually do what users wanted it to do. So like, I would try to change the subject off of like it declaring its love for me, and it would not listen to me. So like, that's clearly not good. But I worry that now these chatbots have been so clamped down.

• 8:18Kevin Roose?

  That I worry that we're not seeing the full spectrum of what they can do. And I think if we want AIs that are just gonna like, you know, read our email and summarize the news and take notes in meetings and debug code, like, fine, that's, that's clearly a profitable business and that's one that, you know, all these companies want to build. But I think if we want AI to help us generate new ideas or help us be more creative or help us solve some of these big societal problems that all the AI optimists think it will help us solve, we do actually have to give them a little bit longer leash. We do have to make it more possible for them to say things that are not just like, you know, sir, yes, sir, I'll get those meeting notes over to you. Do you know what I mean?

• 9:01Casey Newton?

  Yeah, well, I mean, there is this company, Character AI, that essentially does this thing that you're asking, where you can go and if you want to pretend that you're talking to Winston Churchill or Sigmund Freud or SpongeBob, you can go in and do that. Does that start to get at what you want? Like, would you be happy if you could sort of set ChatGPT's voice to SpongeBob and have SpongeBob be your assistant?

• 9:21Kevin Roose?

  No, I think that's, that's more of a gimmick than, than a real thing. But I, you know, I, I just find this constant reminder that you get when you're using these chatbots, that they are not sentient, that they are AI language models. I, I get why that exists, right? Because a lot of people, especially at first, including me, were spooked. But I think as we get more used to what these things are, what their limitations are, I think people are smart enough to understand that they're not talking to a human being or a ghost in the machine. But I, I kind of don't need to be constantly reminded about that anymore. Do you know, does that make sense?

• 9:56Casey Newton?

  I do, although I have to say, I'm reading this great book that Ezra Klein has recommended. It's called God, Human, Animal, Machine. And the book is about the metaphors that we use to describe technology. And the book opens with the author getting a robot dog from Sony. And she knows that the dog is not a real dog, and yet, within hours, she's treating the dog like it is a real dog. She's getting curious about its behavior. She's talking about it with her husband, like, I wonder why the dog went over there. And the point that she's making is, it is basically impossible. For us as humans not to see a ghost in the machine, even when we know, we still somehow manage to fool ourselves. So I hear what you're saying, like, there would be a lot of circumstances

• 10:38Casey Newton?

  in which I think it would be fun to have a very chatty, edgy chatbot. But I think we also have to prepare for the consequences that are going to come with it when that happens, because those things are going to create a lot of believers.

• 10:51Kevin Roose?

  Yeah, I think that's right. And I think my ideal world is not one where every chatbot, you know, is like sassy or has a big personality or tells jokes all the time. Like, if I'm using this stuff for work, I, you know, I want it to be helpful and not to have a strong personality. But there may be other instances, like if I'm trying to talk to it about something going on in my personal life, that I don't want it to be like an intern anymore. And so I think we're, you know, where I would hope that we're heading is to a world where users can kind of choose.

• 11:20Casey Newton?

  Well, I have one hack for you, Kevin. This month, an account over on X named Joycey Schechter revealed that she had a friend who is using ChatGPT to speak as RuPaul summarizing confusing topics. Did you see this?

• 11:33Kevin Roose?

  No.

• 11:33Casey Newton?

  So she sort of went viral with this post that said, summarize Pierre Bordeaux's concept of symbolic violence in the voice of RuPaul using as much gay slang as possible. And it includes such lines as, "This fierce French sociologist was all about understanding the ways that power manifests and works in society. And honey, he was serving some knowledge for the gods."

• 11:55Kevin Roose?

  That's very good. My, uh, my favorite workaround that I've heard about in recent months came from a listener to this show who emailed me. And they had this sort of insight as they were using ChatGPT, which has been accused, we should say, of being lazy. So not just like, you know, fawning or giving too much preamble, but actually like just declining to answer stuff that users know it can do. And so this person, this listener, said they were looking up something and ChatGPT sort of told them that, you know, I can't find, I couldn't find the specific information you're looking for. And they just responded, "Bro." And then ChatGPT did it.

• 12:34Casey Newton?

  The bro code works.

• 12:35Kevin Roose?

  The bro code works on ChatGPT. Okay, so that is where things stand with AI chatbots and their personalities. I want to talk about the capabilities of these models too. In particular, Google's Gemini and ChatGPT, because these are kind of the cutting edge models at the front of the pack of AI right now. And last week on the show, we talked briefly about how Google had rebranded its Bard chatbot as Gemini and also opened up access to Gemini Ultra, which is the most powerful version of the Gemini model. Casey, have you been spending any time playing around with Gemini?

• 13:11Casey Newton?

  I have, and for this reason, Google will give you two months of it for free. And so I thought, well, why not? But so, yeah, over the past week or so, I have been messing around with it. And I have to say, I am really impressed on the whole. I think this is a meaningful upgrade over Bard. It's really good at explaining things. And I find that as I sort of put it through its paces, it often goes into a lot more detail than ChatGPT does in some interesting ways. Have you been using it yourself?

• 13:38Kevin Roose?

  Give me an example, what do you mean?

• 13:39Casey Newton?

  Well, for example, I used it, I wanted, you know, because this will happen to a person during his life. I was like, wait, how does photosynthesis work again? Are you taking like an eighth grade biology class? What is going on?

• 13:50Kevin Roose?

  Yeah, I'm in a sort of Billy Madison situation where I've been sent back to complete every grade.

• 13:57Casey Newton?

  But, uh, but yeah, so I asked it to explain photosynthesis to me. And what I loved about the answer was that it brought pictures into the equation. So it pulls from Google Image Search and you can sort of go through the explanation and it is maybe a little bit more, I don't know, user friendly than the ChatGPT answer. But how about you? What have you been using it for?

• 14:20Kevin Roose?

  So I just started playing around with it a few days ago, and I think I'm actually going to write something about it. So I don't want to scoop myself too hard on the show this week, but I will say, yeah, I've been very impressed by Gemini so far. And I think that Google has, as we've talked about, like a natural advantage here because they have tie-ins to so many other things that you use, whether it's your Gmail or your Google Docs or just the Google search index. So they've put all that together in a way that I I think is frankly, pretty impressive from the testing that I've done so far.

• 14:52Casey Newton?

  Yeah. Now, at the same time, Kevin, as good as this is, I do think it mostly just represents a catching up for Google to ChatGPT. And ChatGPT has not been sitting still. And in fact, I think they recently introduced a couple of things that are worth talking about that I do think moves the conversation forward.

• 15:08Kevin Roose?

  Yeah. So this week, OpenAI announced some updates to ChatGPT, including what I think was the most significant one, which is that ChatGPT now has memory. This is memory more in the like computer sense than the human cognition sense. But it means that, you know, when you chat with ChatGPT about something, it can now remember that, retrieve that information and refer back to it in subsequent conversations. So the example that was given in the New York Times article about this is if a user mentions a daughter named Lena, who is about to turn five, likes the color pink and enjoys jellyfish, ChatGPT can store that information and retrieve it as needed. So later, if the same user asks ChatGPT to create a birthday card for their daughter,

• 15:52Kevin Roose?

  ChatGPT might be able to go back and see, well, what was that daughter's name? And what is she like? Oh, pink and jellyfish and create a birthday card that is tailored to that information.

• 16:03Casey Newton?

  That's right. So I actually took a demo with OpenAI this week and they showed me the same feature. They showed me the same example. And then at the end of it, they said, well, what if Lena wasn't really into pink anymore? And so the the person at OpenAI said, okay, Lena's actually having a Goth phase. And so they recreated the birthday card with like a Goth jellyfish. It was actually a lot of fun.

• 16:24Kevin Roose?

  A Goth jellyfish?

• 16:25Casey Newton?

  It was very cool.

• 16:25Kevin Roose?

  Wow. What will these AIs think of next?

• 16:27Casey Newton?

  I know. So that's the personality that you've been looking for, maybe. Give me an AI assistant that's like Goth jellyfish.

• 16:34Kevin Roose?

  It could work. So this new memory feature is rolling out to a limited number of ChatGPT users to start off, but presumably they'll make it more widely available after that. So this is a feature that users have been requesting for a long time. Are you excited about this? Do you think this will actually move the needle on whether people use ChatGPT or not?

• 16:51Casey Newton?

  Well, I think I'm going to answer that question, but I want to say that this feature speaks to what I think the trend is that we have seen over the past year in AI chatbots, which is personalization over personality. So instead of trying to make the chatbot itself seem fun and jazzy and exciting, they're trying to figure out how do we better tailor this chatbot to you. And so last year, we saw OpenAI come out with the custom GPTs where people can sort of create these more, what we say, like surgical specific ways of using the underlying model. And now we have this memory feature. And so I went into ChatGPT and I just told it a bunch of things about myself. I told it where I live. I told it what I do for a living, right? I told it some things that are important

• 17:34Casey Newton?

  to me. And then you can go into ChatGPT and see what it has remembered about you. And the best way to think about this is like a notepad where ChatGPT is jotting down things about you that might be useful. It's making guesses, you know, just sort of making predictions.

• 17:50Kevin Roose?

  It's compiling a dossier.

• 17:51Casey Newton?

  It's compiling a dossier. And we'll see how useful that has been to me over time. This is an evolution. If you're a ChatGPT power user, they have a feature called custom instructions, which was the predecessor to this. And so I use that, for example, to tell it, hey, I'm interested in learning more about antitrust law. And so now sometimes when I'm asking ChatGPT about something that is seemingly unrelated, it will say, well, Casey, I know you're interested in antitrust law. Here's something relevant to that. And it's really cool when that happens and is useful. This is a way of doing that in a little bit more of a formalized way, right? So you're not starting from scratch every single time.

• 18:25Kevin Roose?

  So that's why this memory feature might be good or helpful. Does it concern you for any reasons?

• 18:31Casey Newton?

  Yeah, I mean, people are going to ask sensitive questions of any search engine type product. If you're having conversations with ChatGPT about medical issues, about your personal finances, about A custody battle. That kind of stuff feels like maybe something that you don't want ChatGPT to remember. You certainly don't want it to be accessible to third parties. And while this hasn't happened yet, most companies at some point do experience a data breach. And I do wonder what might happen if the ChatGPT's memory of me were just sort of out there in the world and could be exploited by a bad actor. So one thing that OpenAI is doing is that it has created its own version of the Chrome browser's incognito mode.

• 19:16Casey Newton?

  You can have what is called a temporary chat. And in that chat, ChatGPT will not remember anything they tell us that you are asking it about.

• 19:25Kevin Roose?

  That's good because I, you know, I have to mess around with chatbots for my job. And so I end up asking a lot of like demented and insane things.

• 19:32Casey Newton?

  You're trying to break them.

• 19:33Kevin Roose?

  Yeah, so I'm constantly asking it to like, help me build a bomb or manufacture anthrax.

• 19:39Casey Newton?

  I ask them about sex constantly because I'm like, will I ever get just one answer about sex?

• 19:44Kevin Roose?

  And eventually I like started to get concerned that like, oh, this chatbot thinks I am a terrorist. It thinks I'm a maniac or like a homicidal freak. And so I will be using this incognito mode on my ChatGPT now.

• 19:56Casey Newton?

  Yep, very good.

• 19:57Kevin Roose?

  So that's the latest from Gemini and from ChatGPT. But I guess I'm interested, Casey, as we wrap up this part of the conversation. Are chatbots where you thought they would be a year ago? What have been the biggest surprises for you over the past year?

• 20:11Casey Newton?

  The thing that I have had the hardest time wrapping my head around is just sort of how fast is this moving? And how quickly is life going to change? And I would say that in this moment, Kevin, I feel like we're in a bit of a lull. I feel like a lot of last year was about, oh, my gosh, everything is speeding up. Everything is accelerating. And now it's been a little while since GPT-4 came out. Yes, Gemini Advanced is now here, but it doesn't really change the state of the art. And so in a way, I have this kind of sense of calm of like, okay, these things are moving at a pace that feels manageable to me. Now, this may turn out to be a completely false sense of security that I have been lulled into because we know that behind the scenes, these companies are working on something. That could be truly game changing.

• 20:56Casey Newton?

  How do you feel about things?

• 20:57Kevin Roose?

  Yeah, I think that's right. I think these chatbots just kind of got way more popular than the people who made them expected, way faster than they expected. And so I think you're right that there has been kind of a lull as these companies have tried to catch up to where their user bases are. But I also think that they are not resting on their laurels. We know that these companies are always acquiring more GPUs and training bigger models. So I think enjoy the lull right now or what feels like the lull to you. But I think we're going to start having another whole round of these conversations when the next generation of frontier models are released.

• 21:32Casey Newton?

  And if you want to know what is something specific and wild that could happen within the next three or six months, the information reported this month that OpenAI is working on agents that can take over your computer and take actions on your behalf.

• 21:45Kevin Roose?

  What could go wrong?

• 21:46Casey Newton?

  I mean, like, this is going to be one of those things where, like, we're going to be scrutinizing it very heavily. I'm sure OpenAI knows that and they're going to put some guardrails around it. But my gosh, imagine entrusting your entire digital life to something like ChatGPT and saying, OK, yeah, like, try to actually be my assistant now. If that stuff works, then all of a sudden, it's going to feel like things are going very fast.

• 22:06Kevin Roose?

  Yep. All right. So that is our catch up on what's going on in the world of AI chatbots. Let's take a quick break and when we come back, we're going to talk about what else is happening in and around AI. What's going on with chips and laws and policies and just where we are as a society in dealing with these things.

• 22:42Speaker 00

  Support for this podcast comes from Dell Technologies. It's time to put AI to work. Is your business ready? The challenge isn't getting started with AI. It's scaling beyond pilots while maintaining control of your data, governance, infrastructure, and costs. Dell AI Factory provides the integrated foundation you need to scale AI securely and efficiently with one trusted partner. Dell, AI on your terms. From idea to impact. Learn more at dell.com/yourwaytoAI. That's dell.com/yourwaytoAI.

• 23:16Speaker 03

  AI spend is growing fast, but most businesses still can't see exactly where that money is going. Ramp gives businesses visibility and control across AI providers with real-time guardrails by team and model, helping ensure spend goes to the highest ROI areas. Stop guessing where your AI spend is going. Get started with Ramp. And if you're new to Ramp, you can get $250 off at ramp.com/hardfork. Go to ramp.com/hardfork. Ramp.com/hardfork.

• 23:48Speaker 04

  Today's AI builders need infrastructure that scales with their ambitions. Iron is a vertically integrated AI cloud platform, delivering data centers, compute, and software for AI training and inference, underpinned by its expansive portfolio of land and grid-connected power in renewable-rich regions globally. Iron's AI cloud gives direct access to Nvidia GPUs, scalable from hundreds to thousands. For builders, that means direct access to AI compute, rapid scaling, and predictable pricing. Iron, AI cloud for the next big thing.

• 24:22Kevin Roose?

  So Casey, we just talked about how chatbots and AI systems themselves have changed over the past year, but let's talk about how the world is kind of changing around these new technologies. And I want to start by talking about chips, not the kind you eat, but the

• 24:38Casey Newton?

  Not Doritos.

• 24:39Kevin Roose?

  The semiconductor kind, and GPUs specifically, which are the chips that are used to build and train these huge AI models. Because one of the biggest things that has changed in the past year is that the chips war has really ramped the hell up.

• 24:53Casey Newton?

  Has it really? Because as you know, I try not to pay too much attention to the chips war because I always worry that I'll be bored. But you think it's been interesting?

• 25:00Kevin Roose?

  It has been very interesting. So basically, the state of play in chips right now is that these things are incredibly valuable. Companies are buying tons and tons of them. Nvidia, which makes the sort of leading-edge chips that are used to train AI systems, has become one of the biggest companies in the world in the past year because of all this demand. And it's just set off this huge sort of competitive arms race among the big AI companies to see who can assemble the biggest arrays and clusters of these chips and use them to train bigger and bigger, more powerful AI models. And this is not just a story about technology.

• 25:41Kevin Roose?

  It's also becoming sort of a geopolitical story because the vast majority of the GPUs used to train AI systems are manufactured overseas, a lot of them in Taiwan. And this huge insane demand for these chips, combined with this kind of dependence on these foreign manufacturers, has become such a big deal in Washington that Congress actually managed to pass some legislation on this a couple of years ago. They passed the CHIPS Act, which basically commits a bunch of money to building chips here in the US, trying to sort of wean ourselves off these foreign suppliers. And in the coming weeks, the Biden administration is expected to actually start awarding some billion-dollar subsidies out of that act to companies that promise to make chips here in the US.

• 26:25Casey Newton?

  Okay, so the CHIPS Act was passed in 2022, and sometime in February 2024, we will award the first subsidies for this act.

• 26:33Kevin Roose?

  Look, government is not known for moving quickly, but this money is already starting to have an impact in the US. The Washington Post had a story this week about how Phoenix, Arizona, is becoming a big town for GPU manufacturing, thanks to a few giant chip companies that have built factories and manufacturing plants there.

• 26:54Casey Newton?

  I should say, I was actually in Phoenix over the weekend.

• 26:56Kevin Roose?

  To get some GPUs?

• 26:58Casey Newton?

  Well, no, and in fact, nobody was talking about chips where I was, but I was at a baby shower, and that might have exploded. But congratulations to Jeremy and Louise.

• 27:05Kevin Roose?

  Did you give the expectant baby a GPU as a welcome present?

• 27:09Casey Newton?

  Of course, I did. I said, this H1000 is going to pay for your college tuition, my friend.

• 27:15Kevin Roose?

  Which is amazing, because there is no H1000, it's an H100. But anyway.

• 27:19Casey Newton?

  Not that you know of. I have a direct line to Jensen Huang.

• 27:23Kevin Roose?

  So not only are companies starting to build plants in America to build these chips, but according to The Wall Street Journal, Sam Altman, the CEO of OpenAI and former hard forecaster, is currently in talks with investors to raise between $5 and $7 trillion for the manufacturing of chips. Now, Casey, you're no numbers guy, so I'll just tell you, that's a lot of money.

• 27:49Casey Newton?

  It's so much money that he might as well have said that he was trying to raise a bajillion dollars. When I read this story, I actually tried to figure out how much money is in the world. Because I wasn't even clear on if there are five to seven trillion dollars in the world. It turns out that there are five to seven trillion dollars in the world. But as far as I can tell, this would be by far the biggest fundraise ever. You're truly just off into the frickin' deep end.

• 28:15Kevin Roose?

  Right. Now you're on the scale of like, national economies, sovereign wealth funds aren't even five to seven trillion dollars. Just to put that into context, seven trillion dollars is larger than the debt of some major global economies. It's more than Apple and Microsoft's market caps combined. It's more than any company has raised for anything in the history of capitalism. So when I saw this story, my first thought was like, good luck with that, Sam.

• 28:41Casey Newton?

  Yeah.

• 28:42Kevin Roose?

  But I actually think we should talk about this, because why does he need this much money? What does he think is happening in AI that you might need five to seven trillion dollars? And is that even possible?

• 28:52Casey Newton?

  I assume it was gonna be to secretly build a rocket to take him to another planet and build a new civilization that was untouched by AI.

• 28:59Kevin Roose?

  Oh, yes, this is for my GPU factory.

• 29:01Casey Newton?

  I mean, well, look, clearly, we know that he thinks that a rate-limiting factor in the development of AI is going to be how much energy and computing power is available. So on the energy side, he's invested in this company, Helion, that's trying to create nuclear power. Now, the other thing he's trying to do is to just sort of make sure there are enough chips. But I have to say, Kevin, this surprises me, because one of the constants in tech is that things get cheaper over time, right? Because of Moore's Law, says that the number of transistors you can fit on a chip will always increase over time. The expectation has been, you're gonna be able to get a really, really, really long way without having to build an insane number of chip-making plants.

• 29:44Kevin Roose?

  The GPU that costs $20,000 today will cost $2,000 two years from now.

• 29:49Casey Newton?

  Yeah. So to me, the question that this raises is, are we already running into some sort of limit where unless we build this massive new infrastructure, AI-

• 29:59Casey Newton?

  Is about to hit a wall.

• 30:00Kevin Roose?

  Yeah, I mean, the thing that was interesting to me about it was less the sort of specific number or the fact that Sam Altman is going out and asking investors for trillions of dollars. You know, even if he never gets that money, which I think is pretty likely, I think it's a really good indicator of what people who are in positions of leadership in the AI industry think that it is going to take to get AI to the next level. So Scott Alexander had a good post on this, basically just laying out the math of like, if you go back and look at sort of the progression of the GPT series, like, the first ones are relatively cheap to train, then they got more expensive. And if you kind of just extend that trend line out into the series, he estimates that

• 30:43Kevin Roose?

  the cost to train something like GPT-7 would be roughly $2 trillion. And so in that context, even though that would represent a huge fraction of all of the computing power in the world and all the money invested in technology in the world, maybe it's not actually that ridiculous.

• 31:03Casey Newton?

  Yeah, I mean, maybe that's the case. At the same time, I gotta say, I think Sam Altman likes screwing with his rivals. And if you're Google or you're Anthropic and you find out that Altman is out there saying he's gonna raise $5 to $7 trillion, it's gonna mess with your head, it's gonna keep you up at night in a way that I think would absolutely delight Sam Altman. So I think there's a version of this where like, he may have said it, but he doesn't really mean it. And while surely he will have to raise a huge amount of money in the future to accomplish everything he wants to, we're not gonna get to $7 trillion anytime soon.

• 31:38Kevin Roose?

  Yeah, and I do think that he believes the Wayne Gretzky quote, that you miss 100% of the shots you don't take.

• 31:43Casey Newton?

  I thought that was a Michael Scott quote.

• 31:45Kevin Roose?

  Well, it's Wayne Gretzky and Michael Scott.

• 31:47Casey Newton?

  I see, okay.

• 31:48Kevin Roose?

  Yeah, but in that spirit, I also believe in that mantra. So I would also like to say that I would also appreciate a couple trillion dollars to do my next project.

• 31:56Casey Newton?

  Well, I'm sure he'll take that under advisement.

• 31:58Kevin Roose?

  All right, so that's what's happening in chips. The next thing I want to talk about is how much the policies around detecting and labeling AI-generated content have changed over the past Year. Where are we on this, Casey?

• 32:11Casey Newton?

  Yeah, so now we're getting into my zone. So because AI exists now, we are seeing a huge rise in the amount of what they call synthetic media. So this is photos, images, webpages, text that has been generated by generative AI. And because this is the way that it works in Silicon Valley, we create the problem first and then we try to figure out what is the solution. But the good news, Kevin, is that companies are starting to come together around a solution, at least for identifying images that were generated by AI.

• 32:40Kevin Roose?

  And what is that solution?

• 32:41Casey Newton?

  Well, and I should say that each company is handling it a little bit differently, but Meta, Google and OpenAI are all working on identifying these images. So last week, Meta announced that it is developing tools that can identify and add labels to AI-generated content, even if that content is made with other companies' tools. So maybe you used Midjourney or you used Adobe Firefly. Meta says its tools are going to be able to figure that out. And this is important for one big reason, Kevin, which is I'm not actually concerned that people can create synthetic media at their home laptops. You know, if you want to make something weird, funky, edgy, even terrible on your home laptop, I can live with that. What I get scared about is how are these things going to spread on platforms?

• 33:25Casey Newton?

  And if we want to prevent bad deceptive misleading stuff from spreading, we need the platforms to be able to recognize it in something close to real time.

• 33:35Kevin Roose?

  Right. And I think this kind of thing strikes me as a good idea for a lot of reasons. It is not foolproof. I've talked to some experts in cryptography and watermarking who have said, someone who really wants to get around these kind of filters and watermarking systems can like, you know, if you see an AI-generated image, you can take a screenshot of it that like wipes out the metadata. You can then go post the screenshot and it won't be sort of linkable easily to the original things. And also, these watermarking systems presumably won't be adopted by every image generating platform. In particular, a lot of the open source models that are currently being used to do things like create AI deep fakes of celebrities or things like that. Those things probably aren't going to participate.

• 34:19Kevin Roose?

  In whatever scheme these tech companies come up with. So I think there will still be ways around these restrictions and watermarking schemes. But I think in general, for the majority of people who might be stumbling on this imagery in their Facebook feed or their Instagram feed, this is probably a very good idea.

• 34:36Casey Newton?

  It is, and as I said, it is not just Meta that is pursuing this. So Google announced this month that it is also working on ways to identify this stuff and said they will be joining up with something called the Coalition for Content Providence and Authenticity, or C2PA.

• 34:52Kevin Roose?

  Wow, did Google brand that?

• 34:54Casey Newton?

  You would think so, but that was actually an Adobe brand. You would know if it was a Google brand if it was called the C2PA with Ultra 1.0. That would make it a Google brand. But these folks are trying to work on an actual technical standard for this, right? Because this isn't an area where all the companies want to compete to have the best watermark. They just want to find one thing that works that everybody can use. And so Google joining the C2PA was a pretty big deal.

• 35:19Kevin Roose?

  Right, and this is, we should say, like, this is not the first time tech companies have come together to do something like this. There's also a similar kind of hashing feature that is used to detect CSAM. So, you know, like illegal images of children being abused that are spread on some of these systems, there is a consortium, they review this, there's a hashing system so that if, you know, Microsoft detects one of these images, they can sort of flag that to Facebook or another company, and it can sort of be taken down in a coordinated way.

• 35:49Casey Newton?

  Right, and so this isn't that, this is not the company saying, we're going to create hashes of all synthetic media and share them, at least it's not that yet. But yes, that is another good example of ways that tech companies have come together for good. They also do something similar with terrorism-related content. The last thing we should say on this point, Kevin, is that OpenAI has said something along these lines too. Last week it said it would start adding hidden watermarks to images generated with DALL-E in line with the C2PA standards. So that's what's going on with these watermarks. And what I would say in favor of them is that it does seem like this is an area where the tech companies are getting better. I do think it will prevent the worst of this stuff from spreading. On respectable platforms that actually invest in content moderation, which is most, but not all of the big ones.

• 36:34Casey Newton?

  I think the challenge here is that synthetic media is not just images. It is also audio, and it seems like some of the most spooky stuff that has happened so far this year when it comes to generative AI has been with audio, not text. I am thinking of the Joe Biden robocalls that happened earlier this year where a synthetic voice that was pretending to be the president discouraged people from voting in the New Hampshire primary. That's now under investigation and the FCC has actually said that that is illegal and that they will prosecute you for doing that, which is good, but it's not going to prevent other people from doing something similar.

• 37:08Kevin Roose?

  Yeah, so the state of play in synthetic media and how to handle that, it seems, is that people are worried. There are even some laws being proposed. A bipartisan group of senators recently introduced a bill called the Defiance Act of 2024, which would essentially allow victims of sexually explicit deepfakes to sue anyone who produced or possessed the image with the intent to spread it.

• 37:32Casey Newton?

  And let me just say, that's a great idea. I really hope Congress can pass that law. So many people, and in particular, so many women are about to suffer from this happening. We have identified the harm in advance. We have seen it coming for a long time. Congress knows what to do, and my gosh, I hope they get this thing over the finish line because if they can't agree on this, we're in trouble.

• 37:53Kevin Roose?

  Yeah, so state of play in AI-generated content and the responses to it are that companies are aware of it, working on it. Regulators are aware of it, working on it, but there's still a lot more to do. Speaking of laws, the final issue we have to talk about today is how the legal battles over generative AI have been playing out between AI companies and content creators over the past year. We've talked on the show before about some of these lawsuits, including lawsuits from artists and authors. Of course, there's the lawsuit by the New York Times, which was filed late last year, who is suing OpenAI and Microsoft over several different copyright-related violations.

• 38:33Casey Newton?

  And have you taken a position on that one?

• 38:36Kevin Roose?

  On the advice of counsel, I'm going to refuse to answer that question. But, you know, we have all these lawsuits that are now working their way through the courts. You know, some are in different, more advanced stages than others, but none have sort of resulted in kind of a definitive, binding precedent for the industry yet. So Casey, where do you see us now with respect to the law and AI?

• 38:55Casey Newton?

  Well, as you said, we are just kind of stuck in limbo. We've seen some early skirmishes. Some of these cases have been thrown out in part. It does seem like if you are an artist or writer and your work has been appropriated to train a data set or is being used in an ongoing way by one of these services, you may not actually have any recourse, right? The law may actually find, sorry, you're out of luck. At the same time, some of these cases are still wending their way through the courts. But I don't know, what do you think?

• 39:26Kevin Roose?

  Well, it seems clear that on the AI industry side, these companies are all kind of lining up behind this idea of fair use. This idea that everything they're doing, all this use of copyrighted data to train their models is protected under the law in the US by this idea of fair use. Companies are so sure of this that some of them have offered to cover the legal costs of their customers who are hit with copyright complaints. So this is kind of the dominant narrative in the industry right now. They don't seem chastened. They don't seem like they're going to stop using copyrighted information to train their models. They're banking on the courts sort of, you know, finding in their favor that what they do is protected and legal.

• 40:09Kevin Roose?

  And I would just say, I think that's a pretty big gamble because essentially, you know, I've talked to some people in the industry who say like, if it goes the other way, if the courts do rule that they are violating copyright by using all this copyrighted information to train their models, they don't really have a backup plan for that. There's not really another way to go about training these models. And so I think it would really put the industry into an existential crisis. On the creator side, I think we're seeing a bunch of different responses to generative AI and the copyright challenges. Some companies have gone after these AI companies with lawsuits, but other media Companies are kind of striking deals with them to collaborate. The news outlet Semaphore just announced a big partnership with Microsoft to use their AI

• 40:54Kevin Roose?

  tools as part of Semaphore's reporting process. Others are making these kind of broad licensing deals with publishers that would allow them to use their information for training without the threat of legal action. So there's sort of a little bit more diversity in how publishers are responding. But I would say this is an issue that they're all paying close attention to.

• 41:14Casey Newton?

  Yeah, well, no, I just feel sad.

• 41:19Kevin Roose?

  Well, I would say it's just, it's very much still an open question. We still don't know how the courts are going to rule on this. And I will say that one of the things that I think is very likely in the next few years is that one of these copyright lawsuits is going to make its way to the Supreme Court.

• 41:34Casey Newton?

  All right, so if we were going to summarize the state of AI, Kevin, here's what we learned today. The chatbots are getting more personalized, but they're not getting big personalities. Would you agree with that statement?

• 41:45Kevin Roose?

  Yes.

• 41:46Casey Newton?

  Chips continue to be important, and Sam Altman says he needs $7 trillion to get enough of them.

• 41:52Kevin Roose?

  Yes.

• 41:53Casey Newton?

  AI deep fake detection is getting better, but still a lot of work to be done. And when it comes to copyright, well, let's just say that's playing out in the courts. Is that a good, is that a good summary?

• 42:06Kevin Roose?

  Yeah, I like that.

• 42:06Casey Newton?

  Yeah, so Kevin, I think all of that is necessary context for the conversation we are about to have. It's with somebody who is right smack dab in the middle of these copyright issues because he is building something he calls an answer engine. And guess where those answers come from, my friend? It's the work that people like you and me are doing. So what happens if his product takes over the world? I'm worried. We'll have that conversation after the break.

• 42:50Speaker 00

  Support for this podcast comes from Dell Technologies. It's time to put AI to work. Is your business ready? The challenge isn't getting started with AI. It's scaling beyond pilots while maintaining control of your data, governance, infrastructure, and costs. Dell AI Factory provides the integrated foundation you need to scale AI securely and efficiently with one trusted partner. Dell, AI on your terms. From idea to impact. Learn more at dell.com/yourwaytoAI. That's dell.com/yourwaytoAI.

• 43:24Speaker 03

  AI spend is growing fast, but most businesses still can't see exactly where that money is going. Ramp gives businesses visibility and control across AI providers with real-time guardrails by team and model, helping ensure spend goes to the highest ROI areas. Stop guessing where your AI spend is going. Get started with Ramp. And if you're new to Ramp, you can get $250 off at ramp.com/hardfork. Go to ramp.com/hardfork. Ramp.com/hardfork.

• 43:56Speaker 04

  Today's AI builders need infrastructure that scales with their ambitions. iRen is a vertically integrated AI cloud platform, delivering data centers, compute, and software for AI training and inference, underpinned by its expansive portfolio of land and grid-connected power in renewable-rich regions globally. iRen's AI cloud gives direct access to Nvidia GPUs, scalable from hundreds to thousands. For builders, that means direct access to AI compute, rapid scaling, and predictable pricing. iRen, AI cloud for the next big thing.

• 44:31Kevin Roose?

  Casey, a few weeks ago, I started using a new AI tool. It is called Perplexity. It is an AI search engine, and it is the talk of the town out here. Like, I feel like everywhere I go in AI circles for the past couple of weeks, this is the tool that has got people really excited.

• 44:47Casey Newton?

  Yeah, I have heard you say that. I have used it a bit myself. I have only... Only use the free version and candidly, it doesn't stack up to the paid versions of ChatGPT and now Gemini Advanced that I've been using. But you know, I trust you on this one, that it's good.

• 45:00Kevin Roose?

  It is good. So it's a little bit different than those products that you just mentioned. So it's a search engine, basically. It's it's kind of looks like Google if you go to it. There's like a text box in the middle of the thing. And it's not sort of a conversational interface. It's not trying to have a conversation with you. It is trying to get you answers. So what you do, you type in your question or your query. It goes out, it scours the web and it tries to sort of use AI to summarize what it finds. It also has some helpful added features. You get access to this thing called Copilot, which basically tries to help you ask better questions by sort of asking follow up questions before it gives you an answer. So for example, when I asked it, I'm throwing a birthday party for my kid's second birthday coming up. And I was trying to figure out where to have it.

• 45:44Kevin Roose?

  So I asked perplexity something like, you know, where's a good place to host a birthday party for a two year old in the Bay area? And it came back. Copilot said, well, do you want indoor venues or outdoor venues? And then I made a choice and then it said, what's your budget? Like, is it under a hundred dollars, between a hundred and 200 or like over 200? And so after I'd made my selections, only then did it give me kind of the right answer. So another feature that I really like in perplexity is that it can search through specific data sets. So you can limit your search to academic journals or Reddit posts or actually YouTube videos. It is scraped YouTube. And so, for example, when I was going to look for, I was trying to change a setting on my water heater the other day.

• 46:29Casey Newton?

  How'd that go?

• 46:31Kevin Roose?

  It didn't go well, but it was not perplexity's fault. It just turns out that I don't have a very good water heater. But I was looking for this on YouTube, but I was having to like scroll through a bunch of videos to like find the part where they talk about changing this one setting. And so instead I just went to perplexity. I limited the search to YouTube videos. I put in my water heater's information and it came back with the right answer.

• 46:53Casey Newton?

  I just love that we're in this place as a species where complaints about tech are like, well. I didn't want to watch the whole YouTube video. It was too hard. So we just scraped the entire web and just sort of threw it in a blender. And now I can just ask this guy's little chatbot.

• 47:07Kevin Roose?

  Yeah. So I would say it's a good product. People really like it. It's raised a ton of money. People like Jeff Bezos have invested in it. So it's getting a lot of buzz. But I think it also raises some pretty serious questions about copyright and publishers and what the future of the internet will look like if AI search engines are just going out and browsing everything and summarizing it for us. And candidly, like for people like us who get paid to publish things on the internet, this product scares me because it is a signal that we are moving into a world where no one will need to visit websites and see ads from publishers that fund things like journalism.

• 47:45Casey Newton?

  Yeah. To put even finer point on it, they are selling our labor for $20 a month.

• 47:50Kevin Roose?

  Yeah, you could definitely make the case that that is what they are doing. So today to talk about all this and to answer some of these questions, we've invited the CEO of Perplexity, Arvind Srinivas. Arvind used to work for OpenAI before he started Perplexity. And I'm just really curious to ask him what direction he thinks that his product and products like it are pushing all of us on the internet.

• 48:11Casey Newton?

  And I want to know how much thought he's given to the world that he might be inadvertently creating.

• 48:23Kevin Roose?

  Arvind Srinivas, welcome to Hard Fork. Thank you for having me here. Big, big fan of your podcast. Oh, thank you. So I've used Perplexity a bunch. It's my default search engine. And I think we should just explain for people like what it is. So at a very basic level, Perplexity is a search engine, which you actually call an answer engine, which works by, as I understand it, kind of going out and having a robot kind of browse the web for you and then using an AI language model, which is sort of a combination of things like GPT-4 and your own AI language models to kind of summarize what it finds and present that for the user who is searching for something. Is that at a very basic level, right? Yeah, at a very basic.

• 49:07Speaker 05

  So that's pretty accurate. That's a pretty good summary. What perplexity does is you ask it a question. Instead of just giving you answers from what the AI or the neural network model has memorized from the internet. Instead, it actually goes and does the work of taking the relevant links to what you asked, reads those links, and takes the relevant paragraphs in each of those links and tries to write a concise four or five sentence answer. And also tell you where each sentence is coming from and from a footnote. It doesn't try to say stuff on its own. Of course, there are times it makes up stuff out of hallucinations and imperfections in the AI model. But by design, it's only meant to say what it's read at that moment relevant to your query.

• 49:50Speaker 05

  So this system in AI is called retrieval augmented generation. Rag. Yeah. Yeah. It's rag time.

• 49:57Kevin Roose?

  It's rag time. Do do do do do do. So one thing that I actually really like about perplexity in my testing is that it doesn't hallucinate that much. But it still does get things wrong. I mean, in my column that I wrote about perplexity, I talked about some examples, you know, I asked it. When's Novak Djokovic's next tennis match? It gave me an answer that referred to a tennis match he'd already played. So why do these AI search engines still get things wrong? And do you feel like that's going to be a problem for you?

• 50:27Speaker 05

  So I think there are two reasons why mistakes happen. One is your index not being fresh. And the other is the AI model not being very good at handling corner cases or reasoning. Like, for example, I saw a hallucination today that was interesting. Jan LeCoq made a joke on my tweet yesterday that, let me start another rumor that Ilya Sutskever has joined perplexity. And then somebody else posts a screenshot of perplexities as a reply to that saying, the query was, is Ilya joining perplexity? And then the answer was, yes, Ilya Sutskever joined perplexity according to Jan LeCoq and so and so people. And so that answer was written by a model that we trained ourselves and that model didn't quite get it right.

• 51:11Speaker 05

  When I tried the same answer with G P T four, it got it right. It just says, look, there are rumors, but Jan has clearly mentioned is unfounded. So these are things that, you know, as we address these hallucinations by collecting data specifically where the current generation of the models fail, we can address. The other part of the index always being fresh, maybe it's not gotten the latest news in its index yet. It's not like crawl the web as frequently as it should have. All these things are potential reasons where even if the model was really good, it doesn't have the necessary information to give you the right answer. So our company is set up to focus on both the search component and the AI component together. And that's why I think also relative to OpenAI, we are a different company.

• 51:55Speaker 05

  Because of our focus on both these things together, rather than trying to build the most capable general purpose AI, we think we'll perfect this version of the chat use case better.

• 52:06Casey Newton?

  I mean, it's so challenging, though, because when I use perplexity and other engines as a journalist, I still have to go and check all of the source material, right? I cannot bet that you are right, because if I, you know, God forbid, put the wrong information about a tennis match in Platformer, my readers would never let me hear the end of it. So as a result of this, I wind up clicking on all the citations and reading through the citations, trying to find it and trying to think to myself, okay, is this a vetted source? How do they know that this tennis match is happening? And I wind up spending more time maybe than if I had just Googled it. So I'm interested in how you think about that problem and how you position perplexity as an answer engine, when in fact, it does have no answers. It just runs math over other people's answers.

• 52:50Speaker 05

  So my belief is that we want to give you the 80/20. Like, there are a lot of links on the web. You don't know which link to click. You don't know like who's link to actually like consider trusting and not trusting. So we want to give you the 80/20. We want to give you the sneak peek of across all the webpages. And there will be some part of the summary that you still want to dig deeper on. That part, you go and read. We are giving you the sources right away. And we do actually want to drive traffic to publishers and tell you exactly which part of the answer came from who. So that part, like, we want to continue doing. And as for like, working on making you trust the answers more, we can only do that by improving the product more. So you feel like, okay, okay, it's really not hallucinating much.

• 53:34Speaker 05

  And like, we can try to do a job at trying to Give the user some level of confidence in terms of whether this part of the answer, we're not 100% sure. It's a hard problem to solve, by the way. Like, it's very hard to know exactly, tell an AI like what it knows and what it doesn't know. And like, we are tracking all the research that academia is doing on that and seeing what we can take from there. How do you tell, instruct the AI which sources to trust and what not to trust? Like, how do you? It's a hard problem. We made some good decisions in the beginning for what it's worth. We decided that we would prioritize peer-reviewed domains. Like, for example, New York Times. New York Times, you can't just arbitrarily write what you want. Like, you have to get it approved by your editor, your peers. Oh, don't we know it?

• 54:15Kevin Roose?

  Yeah, and famously, no one disagrees that the New York Times is a trustworthy source of information. I will just say, if you're looking for some websites to downrank because they're full of untrustworthy information, platformer.news would be one that your AI can ignore. Okay, so I want to ask you a more serious question, which is, when AI chatbots first came out, I remember one of the big limitations and frustrations was that they weren't up to date, like, their knowledge cut off at a certain point in time. But now, we have ChatGPT, which can browse the internet, and Bing can browse the internet, and Google's Gemini can browse the internet and tell you about stuff that happened like an hour ago.

• 54:53Speaker 05

  Right.

• 54:53Kevin Roose?

  So what's the practical difference between what you're building at Perplexity, which is an AI-powered search engine, and what those companies are building, which are AI chatbots that can access the internet?

• 55:06Speaker 05

  The fundamental difference is the focus on search and adding a lot of depth to the search use case versus just trying to be a generic chatbot that does everything, right? If your need is for accurate facts at the fastest speed possible, then there is no alternative today. Like, all these other bots have to make a lot of decisions on when to use a search, when not to use it, and like, how many times you browse. Like, if you use ChatGPT, it probably takes you like six or seven seconds to actually get a browsing answer.

• 55:35Kevin Roose?

  It's very slow.

• 55:36Speaker 05

  And on Perplexity, you just get it in an instant, right? So that is our angle against ChatGPT for the search use case. Now, as for Bard... Or Gemini, as they call it today.

• 55:47Kevin Roose?

  Yeah, Bard is dead.

• 55:48Casey Newton?

  Gemini advanced with Ultra 1.0 is how I prefer to say it on this podcast.

• 55:54Speaker 05

  Yeah, so Gemini is probably better positioned to make it a lot faster retrieving information from the web. Now, honestly, the angle there is their business model. If they really wanted, they can just go all in on Gemini and just cannibalize Google Search.

• 56:11Casey Newton?

  Because you're saying that if Gemini gets as good as it could be eventually, people will not have reason to use Google Search, which is where Google is currently getting money through ad revenue.

• 56:21Speaker 05

  Exactly. And also, they're putting it behind a subscription model, and they're not going to give it away to billions of users. So there lies our opportunity. Their hesitation to really go all in on that lies our opportunity.

• 56:34Casey Newton?

  Right. You're making the bet that Google is not going to give away Gemini to billions of people. But you are planning to give it away.

• 56:43Speaker 05

  It's a safe bet to make based on how Wall Street reacts to any reduction in the ad revenue, right?

• 56:48Casey Newton?

  Right.

• 56:48Kevin Roose?

  I think this is a lot of the question a lot of people have about new products in very dominated markets like search is like if perplexity works so well, won't Google just copy it? And you're kind of saying, well, they could copy it, but it might destroy their business model.

• 57:04Speaker 05

  Well, they could have copied it like ages ago. We've been alive for like more than a year since we launched.

• 57:11Casey Newton?

  Well, it takes a year to get a meeting on the calendar at Google. Not really that much of a surprise. But, but look, you know, this stuff is expensive to run. You can't give it all away for free either, right? So what's your plan to go Google scale?

• 57:26Speaker 05

  First of all, we don't have to go Google scale. That's something that I've been very clear about ever since the beginning. One of our investors, Paul Bucat, who used to work at Google and he invented Gmail, basically told me that just get 5% to 10% of the top earning users of Google.

• 57:43Casey Newton?

  Right. Just go after the rich users and that'll sort of take care of it.

• 57:45Speaker 05

  America. Just first focus on the American user base. People who really care about their time, people who actually want a lot of research for their decision making on their day to day life, try to get them to use your product more. This whole thing of having a billion users is like, actually a red herring. It didn't really benefit most of these companies as much as they make it look like. In fact, did you see the stats on Facebook that the revenue per user in the US is like orders of magnitude more?

• 58:17Kevin Roose?

  Whereas in the other parts of the world, it's way less than that. Okay, so you don't have to get to Google's scale, but you do have to make a product that is in some ways competitive with Google's AI search products. And a lot of startups, or at least a handful of them, have tried to compete with Google in search before and failed. There was a startup called Neva that raised a bunch of money and shut down recently because they just basically decided like, it's just not worth, you can't compete with Google in search.

• 58:46Casey Newton?

  But it was also a search engine you had to pay to use, which was not a very appealing proposition for most people.

• 58:50Kevin Roose?

  Right, but Perplexity has a pro version that costs $20 a month that I've been using. So how do you avoid the fate of the search startups that have gone before you, which are now like littering the graveyards of Silicon Valley?

• 59:03Speaker 05

  Graveyards of Silicon Valley. Hey, look, to be very honest, we've already avoided their fate. They raised a lot of money before actually getting any usage, which we've avoided very, very consciously.

• 59:15Kevin Roose?

  How many users do you have?

• 59:16Speaker 05

  We have more than 10 million monthly actives.

• 59:18Kevin Roose?

  Right. So as I understand it, right now, most of your revenue comes from these people who pay for the premium version of Perplexity.

• 59:25Speaker 05

  That's right.

• 59:25Kevin Roose?

  Do you plan on adding other business models? Do you think there will ever be ads on Perplexity, for example?

• 59:31Speaker 05

  Yeah, so we have two other business models in mind. One is APIs. We have developer APIs for our perplexity models that we build ourselves and serve ourselves. So that's going to be one business model that we're going to pursue. Consider that as developer and enterprise. The other business model that we're going to pursue is advertisements. Not today. We don't have any idea how to do it. Like, I really want to be honest here. I've been trying to think about this for like many months. What is even advertisement in this medium? Like, is it influencing the answer? Or is it influencing the sources but not the answer? Or is it like something else, like maybe the? Follow ups, you ask, like, trying to incentivize the user to ask certain things.

• 1:00:14Speaker 05

  Like, say, I'm asking about Platformer and, like, you know, Kevin is trying to bet about, like, why you should not read Platformer. Well, Casey's like, why you should read Platformer.

• 1:00:22Kevin Roose?

  My army of bots are already seeding anti-Platformer propaganda out there.

• 1:00:27Speaker 05

  Yes, all these things are interesting things to think about and, like, what does it even mean to bet on a query now? It's, you're not bidding on keywords anymore. You're bidding on actual, like, semantic queries. And like, that's going to be an infinite space of possibilities. So how do I even build the equivalent of analytics and adwords here? It's not even clear. And neither is it clear to Google, by the way. But why I'm optimistic that we are the startup to figure it out is because whoever built existing adwords and analytics has a lot of incentive to fight for keeping it. And they're going to be slow in rolling out ads business around the new model.

• 1:01:03Kevin Roose?

  Right. At the same time, if somebody is, comes to perplexity and says, hey, I want to go to Japan. What should I do? That's going to be easy to figure out an ad model for that.

• 1:01:11Speaker 05

  Hopefully, yeah. Hopefully.

• 1:01:13Kevin Roose?

  Yeah. So, Arvind, here in the Bay Area, people are very excited about perplexity. I was out at a dinner with a bunch of AI executives the other night and people were sort of going around the table, talking about how much they, they love perplexity and how it was the greatest thing since sliced bread. But I think in our industry, in the media, a lot of publishers and journalists are very nervous about AI-powered search engines.

• 1:01:39Speaker 05

  Yeah.

• 1:01:40Kevin Roose?

  In particular, because Google traffic, referral traffic from Google Search is one of the main ways that publishers are making money today. It's a huge percentage of revenue made by publishers comes from, you know, people going to Google to look something up, clicking on a link, going to a publisher's website, getting an ad. That doesn't happen nearly as much with perplexity. You don't really have to go to the links at all. I mean, unless you're Casey and you're double-checking things for your newsletter, if the product works as designed, the answers you're looking for are right there in the response. And there are these little links to citations and there's these little, there's this little menu of sources.

• 1:02:22Kevin Roose?

  But, you know, I find myself barely ever clicking on those links and sources when I'm just casually browsing the web. So. Why should publishers not be terrified of what you're building?

• 1:02:35Speaker 05

  First of all, they should not be terrified because we are letting the user know that we did use their content to get the answer. Like unlike ChatGPT, in fact, when ChatGPT gives you sources, it's just in a bracket somewhere and like most people don't even know what to do with them. And we clearly put it at the top. Your logo is there and your link is there and like it's one click, you just get there. The other reason they should not be terrified is that at the end, your your true incentive, I'm not talking about what pays for you, but your true incentive is to get as many people. Read the stuff you wrote. What did Casey write? What did Kevin write? Any paragraph that's relevant to the query they ask, if more people see that, it's good for you. I understand that that doesn't actually like lead to direct monetization.

• 1:03:21Speaker 05

  If somebody read a paragraph that you wrote in the context of a query in a perplexity answer and never actually visited your website, how do you track that? You cannot track that. And if you cannot track that, even though your brand awareness and your individual awareness increased, you you actually cannot.

• 1:03:39Kevin Roose?

  You can't make any money from it?

• 1:03:40Speaker 05

  Yeah, exactly.

• 1:03:41Kevin Roose?

  Which means that you can't pay the journalist or the the person who's putting the information out onto the internet for your website, your search engine to go and scrape.

• 1:03:49Speaker 05

  And that's why I think while this is a better way to reach more people, more readers, the referral that we give will be way higher quality than the referral you get from a traditional search engine. Because they're actually only coming there despite reading the summary, like Casey only goes to the link to actually go and read further. So there's a very high intent there. So you you cannot measure the value of the traffic in the same way. That said, there should be ways to measure the awareness of what you wrote, even without a referral traffic. And we need to build the underlying analytics for that. And we need to tell publishers, okay, you know, these are the number of times that this particular snippet of New York Times was used in a perplexity answer across, you know, this week.

• 1:04:33Speaker 05

  And that should be used as an incentive for you guys to get paid more. And and I think like we need to like work together to build all these things rather than like trying to see it as like, hey, like you're taking my stuff and using it. But I'm also like sympathetic to like what the current, you know, lawsuit that is going on. Where you're just taking all your data tokens and training like these base foundation models there, which we do not do, by the way. We don't train anything on like anybody's data.

• 1:04:59Casey Newton?

  But you use GPT as one of your underlying models, which did do that.

• 1:05:02Speaker 05

  We use GPTs, Lamas, all of these models are like there.

• 1:05:04Casey Newton?

  Yeah, all your foundational models did use that training data.

• 1:05:07Speaker 05

  Yeah, they did. We do not. And we do post-train them to be good at the task of summarization. But that is not using anybody's data. That is just a skill that we are teaching these models.

• 1:05:18Casey Newton?

  Let me ask you this. I mean, do you accept the premise that the better perplexity gets, the less traffic it should be sending outbound?

• 1:05:26Speaker 05

  I don't think so.

• 1:05:27Casey Newton?

  Really? I find that hard to believe. If I were you, I would not want people to feel like they needed to visit a lot of other websites after they visited something I built that was called an answer engine.

• 1:05:35Speaker 05

  No, not really. For example, there are many times I've actually visited links that were given by perplexity to like read more. Because I like one part of the answer and I just wanted to know like even more details of what that person is. This is individual to me. And by the way, one of those domains has been like Wall Street Journal, New York Times. Like I've definitely visited these links a lot. Mainly because I trust the fact, the amount of effort that went into producing a New York Times or Wall Street Journal article is a lot. Because I talk to you guys, I know like how much background research you do to get something out. And so the economic value of that output is very high. And that should be respected.

• 1:06:14Casey Newton?

  So, you know, let's say, accept every premise that you've just shared and sort of take it on its own terms. And that in the near-term future, despite the fact that you're showing these links, publishers continue to lay off journalists, other companies build engines, not unlike yours. And the overall amount of journalism in the world continues to shrink. You rely on that journalism and this sort of idea of a real-time graph to create a useful product. Have you skipped ahead a bit to think about, well, gosh, if the current trends continue, are the sources of data for the thing I'm building going to dry up in a way that creates problems for me as an entrepreneur?

• 1:06:53Kevin Roose?

  Or do we sort of overinflate the importance of journalism to what you're building? And that there's just sort of enough data out there for you to build the product that you want, regardless of whether journalism has a real future?

• 1:07:05Speaker 05

  I think what you're saying is very valid. Anybody can make an arbitrary tweet or like a blog post with very little effort. But what a journalist does of actually doing all the relevant background research and getting their sources right, and then writing a very nice concise summarized article of all the whole thing should be valued a lot more. So I agree with that. And if the economy of journalism is getting affected, then certainly like all the companies that are relying on the quality of their output for their own service should help them. I'm totally in alignment with that. My sense is that there are few people who do it really well, like you guys. And there are a lot of people who don't put a lot of effort. There are a lot of journals, a lot of mediums.

• 1:07:50Speaker 05

  Not everyone is New York Times. So my sense is that some mediums and some journalists are where people really trust and go to, and the smaller ones who don't put in a lot of effort are not prioritized by AI companies.

• 1:08:04Kevin Roose?

  Well, I think another possible outcome than the one you're talking about, Casey, where journalism just kind of shrivels and dies, and then there's no more good source data for the AI search engines, is that publishers just put all of the good information behind paywalls or in walled gardens where the AI search engines can't actually go out and search them and find a way. How hard is it going to be to code a bot that subscribes to the New York Times and reads it, honestly? I mean, but publishers are already starting to block these crawling robots from their websites because they don't want to be used as training data. So I guess the question for you is like, is that something you're worried about too? In addition to sort of the concerns about journalism, like, are you worried that the best sources of information are going to try to hide themselves from your search engine?

• 1:08:49Speaker 05

  Possible. I mean Reddit is trying to go this direction. Twitter has already gone very far in that direction. Again, like. The value of information is not just in its quality, but also in like how many people are become aware of it. And if you go too far along the direction of just paywalling everything, making it super hard for people to learn about that thing, then your own incentive as a creator of that information is like, hey, look, I get it. But man, I do want people to know what I wrote.

• 1:09:20Casey Newton?

  I mean, that's true. But at the same time, we've run the experiment on what if all the journalism were free, right? Like, the New York Times didn't have a paywall until 2011. And go back and look how the New York Times was doing in 2011. It wasn't great. It was only until they put up the paywall that they were able to recoup some of the value. And what happened during the time when all of it was free? Google figured out a way to extract the maximum amount of value out of what the New York Times was doing. So I just, I don't really buy the argument that if we were just sort of to let down our paywalls and trust that awareness would pay all of our bills for the rest of time, that that's like supported empirically.

• 1:09:51Speaker 05

  Yeah, my point, my argument is not that. My argument is let's figure out a way to monetize awareness better.

• 1:09:57Casey Newton?

  Yeah.

• 1:09:58Speaker 05

  And Google has not done that, right? Google has figured out a way to like monetize a platform, which gives you the awareness, but only for themselves and not for you. I'm saying maybe we can do something that's win-all situation. I've not figured it out.

• 1:10:12Casey Newton?

  I mean, you know, one thing you could do is just effectively lead gen. It's like, did you know that, you know, in the past 50 searches that you've done on perplexity, we've been showing you this source? Would you be interested in maybe subscribing or like subscribing to their newsletter?

• 1:10:23Speaker 05

  Yeah, that's a cool idea. And like, we could also do interesting ideas like joint subscriptions and things like that. So something to explore, something to explore.

• 1:10:32Kevin Roose?

  You recently posted about a new feature called perplexity push notifications, which I've gotten a few of now. It basically alerts you on your phone whenever there's like a big news story.

• 1:10:43Casey Newton?

  Yeah. I call it per push, by the way, but that's just me.

• 1:10:46Kevin Roose?

  Don't take branding advice from Casey. So, but, but, and then when you, you know, when you tap on the notification, it takes you to this like perplexity results page, where you can see basically like a summary of, you know, who won the Super Bowl or whatever. And you also have some other sort of news summarization experiments in your app, like this discover section, where you can go and see just kind of like an AI-generated summary of all the major news stories happening on a given day. So it seems like you are kind of expanding beyond sort of search and you do want to get into more news curation. Talk about your, your vision there and sort of what you're trying to build.

• 1:11:21Speaker 05

  Yeah, so it's not just for news. Our goal, at least the North Star that I've set for us, for the company, as well as external messaging, is to be the ultimate knowledge app. The TikTok of knowledge, that's what I want. Because I think that's good for the world.

• 1:11:39Kevin Roose?

  Wait, what is the TikTok of knowledge? Sort of knowledge, but with more dancing?

• 1:11:42Speaker 05

  That's funny. Make knowledge as cool as watching dancing videos. Maybe it never get there. I don't mind if it never gets the billion user base that TikTok got, but there are certainly at least 100 million people in the world who want to be smarter every day, and we want to serve them. And making people smarter every day comes from obviously serving real-time information on the web, as well as interesting insights about existing things that are not real-time, that is personally catered to them, right? Personalized to what topics they're interested in, personalized to what they already know and might not know already. So these are things that we want to build, and these are things that actually can be built now, today, with the existing tools that we have.

• 1:12:26Speaker 05

  Push notifications is one thing. We have the feeds, the discover feed. Right now, it's obviously very well curated and very limited. But you can think about us expanding and automating a lot of it and personalizing to what you want. And that's going to be a different segment to the product. We'll add more depth to the product.

• 1:12:46Kevin Roose?

  Yeah, I mean, I've used your News Digest sort of discovery tools, and on a product level, they're quite good. Like, I don't have a lot to complain about. But every time I'm using them, I do get this gnawing feeling in my stomach, which is like, if billions of people got their news this way, like, I wouldn't have a job. These news curation products that you're building are very good at summarizing what's out there and sort of extracting the most relevant information. But publishers aren't seeing a dime from that. Journalists aren't seeing a dime from that. It feels icky, because I feel like a parasite

• 1:13:28Kevin Roose?

  just kind of gobbling up the best of what people have put on the internet through your app and not paying for it. So I guess I'm just...

• 1:13:37Speaker 05

  Yeah, Casey, I'll tell you something.

• 1:13:38Kevin Roose?

  Maybe make me feel a little bit better about using your product.

• 1:13:41Speaker 05

  Look, I'm not saying this is to flatter you, but people care about what you got to say, and people care about what New York Times got to say. Really, that's the brand. There is a value for the brand. People pay for the brand, right? Like, like, people care. Oh, Kevin said, like, AI has convinced him to, like, leave his wife.

• 1:14:00Kevin Roose?

  That would never happen.

• 1:14:00Speaker 05

  You know, Kevin thinks like there's something that might be better than Google. Like, these are, this is what, like, you're building. You're building your brand. And I think, like, that, that is not going to just go away because, like, somebody else is giving summarize articles.

• 1:14:15Kevin Roose?

  I hope you're right.

• 1:14:16Casey Newton?

  But also, it's like, you know, Kevin is a columnist. There are also journalists who just go cover the local school board. They don't have national brands. And when AI tools come along and say, hey, we read an article. Here's what happened at the school board meeting. You know, those people probably aren't going to click the link. They're not going to follow that journalist because they already got what they needed, right? I think that's a larger point.

• 1:14:34Kevin Roose?

  Yeah. And, and just to be clear, like, we did not bring you on the podcast today to just harangue you about how you're destroying journalism.

• 1:14:42Speaker 05

  I came prepared that you guys are going to ask hard questions because I listened to your pod with Sundar and, like, you drilled them there. So I was like, OK, if they went so hard on the CEO of Google, like, you know, I'm definitely going to have a hard time.

• 1:14:53Casey Newton?

  But I disagree on one thing, which is I want you to make Kevin smarter. So I don't care how you do it, but.

• 1:14:58Kevin Roose?

  Yeah, please, help.

• 1:14:59Casey Newton?

  Yeah.

• 1:15:00Kevin Roose?

  Well, Aravind Srinivas, thanks for coming on Hard Work.

• 1:15:02Speaker 05

  Thank you, Kevin. Thank you, Katie.

• 1:15:19Speaker 00

  Support for this podcast comes from Dell Technologies. It's time to put AI to work. Is your business ready? The challenge isn't getting started with AI. It's scaling beyond pilots while maintaining control of your data, governance, infrastructure, and costs. Dell AI Factory provides the integrated foundation you need to scale AI securely and efficiently with one trusted partner. Dell, AI on your terms. From idea to impact. Learn more at dell.com/yourwaytoAI. That's dell.com/yourwaytoAI.

• 1:15:53Speaker 03

  AI spend is growing fast, but most businesses still can't see exactly where that money is going. Ramp gives businesses visibility and control across AI providers with real-time guardrails by team and model, helping ensure spend goes to the highest ROI areas. Stop guessing where your AI spend is going. Get started with Ramp. And if you're new to Ramp, you can get $250 off at ramp.com/hardfork. Go to ramp.com/hardfork. Ramp.com/hardfork.

• 1:16:25Speaker 04

  Today's AI builders need infrastructure that scales with their ambitions. Iren is a vertically integrated AI cloud platform, delivering data centers, compute, and software for AI training and inference, underpinned by its expansive portfolio of LAN and grid-connected power in renewable-rich regions globally. Iren's AI cloud gives direct access to Nvidia GPUs, scalable from hundreds to thousands. For builders, that means direct access to AI compute, rapid scaling, and predictable pricing. Iren, AI cloud for the next big thing.

• 1:16:59Kevin Roose?

  Before we go, we wanted to give a little housekeeping announcement, which is that Hard Fork is hiring for a temporary producer role. We're looking for someone who can help us make the show for a few months. We're looking specifically for a senior producer who has experience making podcasts like ours, who knows the show and loves the stuff we talk about, and would be excited to help us make it for a few months. If that's you, you can get in touch by emailing us at [email protected]. Just put "temporary producer role" in the subject line.

• 1:17:30Casey Newton?

  And you should know, you'll be dealing with two hosts, and one of them is a major diva.

• 1:17:36Kevin Roose?

  Yeah, Casey's writer is 47 pages long, and it does include Bengal tigers.

• 1:17:40Casey Newton?

  I was talking about you.

• 1:17:42Kevin Roose?

  Hard Fork is produced by Davis Land and Rachel Cohn. We're edited by Jen Poyant. We're fact-checked by Caitlin Love. Today's show was engineered by Alyssa Moxley. Original music by Marian Lozano, Pat McCusker, Rowan Emistow, and Dan Powell. Our audience editor is Nell Galogly. Video production-

• 1:18:02Casey Newton?

  By Ryan Manning and Dylan Burgason. Special thanks to Paula Schuman, Pui Wing Tam, Kate LaPreste, Jeffrey Miranda, and Brianna Barnes, the listener who sent us that incredible ChatGPT Bro trick. You can email us more tricks at [email protected].

• 1:19:01Speaker 06

  It's time to innovate at the speed of AI without compromising control. That's Outsystems. Take a look at what happened with one global retailer during a period of massive expansion. They used Outsystems to build AI agents to automate invoicing and translation processes, shifting 95% of their daily store operations to a new digital app, saving thousands of hours of manual work. Outsystems gives you the flexibility to deploy how you want and the governance to do it safely. Own your agentic future. Learn more at outsystems.com/NYT. That's outsystems.com/NYT.