Perplexity’s CEO on Its Plan to Displace Google Search With AI Answers — With Aravind Srinivas

Aravind Srinivas is the CEO of Perplexity, an AI-powered search engine that answers queries with a few paragraphs in natural language. Srinivas joins the show to discuss how his company might threaten Google search, and whether conversational AI search is actually similar to traditional search at all. Tune into the second half where we discuss Aravind's perspective on how AI search might help or hurt news publishers. We also discuss Jeff Bezos's investment in Perplexity and where the cutting edge of AI is moving next. Tune in for a candid conversation with the CEO of the hottest AI startup in the world. ---- Enjoying Big Technology Podcast? Please rate us five stars ⭐⭐⭐⭐⭐ in your podcast app of choice. For weekly updates on the show, sign up for the pod newsletter on LinkedIn: https://www.linkedin.com/newsletters/6901970121829801984/ Questions? Feedback? Write to: [email protected] Learn more about your ad choices. Visit megaphone.fm/adchoices

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

• 0:00Alex Kantrowitz?

  The CEO of the hottest AI startup in the world joins us to talk about talent, tech, and his challenge to Google. All that and more, coming up right after this. This episode is brought to you by Genesis. What does it actually take to put Agentic AI to work across an entire enterprise? I recently spoke with Genesis chairman and CEO Tony Bates about why orchestration is emerging as a competitive advantage in customer experience, and what companies need to get right in the AI era. Then, I sat down with Adam Mitchell, head of enterprise business solutions at Voya Financial, to go inside a large-scale Agentic CX transformation, including the business decisions, trade-offs, and lessons along the way. You can watch both conversations now on my YouTube channel today.

• 0:45Alex Kantrowitz?

  It's never been more challenging to get frontier compute. How do you know where to look, who to trust, or how much to pay? Akash is an open cloud that provides easy access, flexibility, and peace of mind. That's how companies like Razer were able to cut their inference cost by 15 times. Visit akash.network/alex for a free consultation. That's akash.network/alex for a free consultation. Welcome to Big Technology Podcast, a show for cool-headed nuanced conversation of the tech world and beyond. Let's go deep today with the CEO of the one startup that's really changing the way that we think about AI, and how it can have a role in challenging incumbents,

• 1:27Alex Kantrowitz?

  and not surprisingly, it's coming in the world of search. We're joined today by Aravind Srinivas. He's the CEO of Perplexity, and he's here to talk to us all about his company and also the state of the industry. Aravind, welcome to the show. Thank you for having me here, Alex. Let me start with this. You have 10 million users, last count. Your company is a search company, so instead of going to Google, you go to Perplexity, you type your question in, you get a paragraph or two in conversational AI with some links at the top telling people where to go, and it's really taking off. When I look around thinking about like, what other generative AI startups there are that have gained traction, it's not a very long list. I mean, character AI is worth a billion dollars, but I don't hear anybody using that.

• 2:09Alex Kantrowitz?

  But your company, people use, you're ranked above Google Calendar on the App Store. I think you were number nine on the top charts last time I checked. Is is this, what what would you attribute this struggle for startups to take off in the generative AI space to? And is my view of this kind of close minded or am I hitting on something that's that's right here?

• 2:29Speaker 01

  It's mostly accurate. There's some nuance. For example, you you don't hear people around you using character AI because it's also not a product that people would be very proud of saying they use it. Uh, it's it's use cases are more uh meant for these lonely uh people who are like, you know, looking for companionship in their life. So it's sort of more like a personal activity and also it's more popular among uh the the younger generation, like fifty to sixty percent of their user base is under the age of twenty. Uh, because they're using it to talk to imaginary anime characters and things like that. So it's a very different uh sector product, more more in the entertainment sector and less on the productivity uh or utility sector. So we are in the productivity sector.

• 3:13Speaker 01

  And uh as for like, you know, in general, for any startup, it's pretty hard to gain uh widespread adoption. First of all, it's hard to launch a good product.

• 3:26Alex Kantrowitz?

  But this technology is supposed to change the world. People are talking about how it's going to be the end of civilization. For something that powerful, you would imagine there would be more.

• 3:33Speaker 01

  Yeah, but it takes time. For for example, it takes a lot of time for people to change their habits, um adopt a new new. I mean, how many times do you adopt a new app? Like, you know, like it it took me personally a very long time to start using Instagram, Twitter, uh even WhatsApp even like like I'm I'm only started using WhatsApp a lot more recently. And I used to stick to like, I I I just want to stick to FB Messenger or like Google Chat. So it takes a while for you to like truly change your habits uh and you know, and we are also facing this, right? It's all relative, okay, but relative to other products and generative AI startups, we have more adoption, but then

• 4:15Speaker 01

  relative to ChatGPT or Google, we are like way lower, right? So uh It takes time and like, you know, I'm not very negative about other startups like. Uh, I also think there are products outside the the sector that you're focusing on that are taking off. Like Levin Labs is a startup that has a lot of adoption. Uh, as a podcaster, you might be aware of it. Uh, there are like a lot of tools there to like, you know, edit stuff. And then, um, many other, many other interesting companies. But for us, the the potential is a lot bigger than than any of these others because the quest for knowledge and information is the ultimate thing ever since like

• 4:57Speaker 01

  uh, humanity has like started evolving as a society, right? We've always sought more information, always sought more knowledge, uh, sought a better understanding of the world. So I think that's why like this product has no upper bound, honestly.

• 5:12Alex Kantrowitz?

  So let's talk a little bit about the product because

• 5:14Speaker 01

  Yeah.

• 5:15Alex Kantrowitz?

  If you're, so you're creating an AI search engine, you've created it. What is the problem with search today that you're trying to solve? Because it's one thing to say, hey, we have this cool technology. Let's just apply it in like a logical way. But you have to see a glaring issue with this multi-trillion dollar business of search. To be like, okay, we're actually gonna do something different.

• 5:39Speaker 01

  So the fundamental problem with search today is that you waste a lot of time, right? Like you're looking for something, you get a bunch of links. Uh, these links are sort of not exactly ordered in the way that should save your time, but ordered in the way that makes Google more money. Uh, you're getting the answer that, uh, ten salespeople are trying to give you at the same time. Not the answer that a friend of yours who has used all these different services or products comes and tells you in a nice summarized way of like what is good about. Let's say you're like going and asking a friend of yours like which car you should buy. They're not gonna tell you what's the best thing about each, uh, you know, car or like each of these cars are trying to bid for against one another to get your attention.

• 6:24Speaker 01

  What they're gonna tell you is like, hey, look, I use all these things. I use all the other, I use two or three cars and this is the best thing about this, this is the best thing about that. And what do you want? Like, what is it that you're interested in? They, they, there, there will be a back and forth interaction with you. Uh, oh, like I actually like mileage more or like I'm looking for electric car because I care about the environment. Like there are that sort of an interaction, that sort of personalization doesn't happen today in search, right? Um, and there is no incentive for Google to save your time in searches, by the way. Because, uh, the whole point for them is that you spend as much time opening like ten tabs on Chrome, uh, the browser they control and give, give the analytics to all of these independent, uh, website publishers and like brands

• 7:05Speaker 01

  of like how much traffic they're driving to each of you, uh, based and how many users are driving to each of you, right? So they're working for the advertiser, not for the, not for the user. So there is an opportunity for a new entrant here, uh, that can directly work for the user, uh, and like build a business model that's more aligned with the user interest than with the advertiser interest. So that's what we are trying to do.

• 7:27Alex Kantrowitz?

  Yeah, but it's interesting because it is almost like a different category of search, right? You've talked about it already, like it's curiosity, right? Like, Google is almost like, I need to get to the facts right away, like hit me with them. Or like, where's this? What is the answer to this question? Well, I guess you guys could do that too. But you actually inspire different queries, right? Like, you, I saw on your Twitter feed, you were asking perplexity, what's the next trillion dollar company? And it said it was the Ozempic maker. And it's like, these type of searches are completely, I would say incremental and new and different than the actual searches we would do with Google.

• 8:02Speaker 01

  Yeah, I mean, I, I generally think that there is like a new segment of searches that these products like perplexity or chat GPT are creating, which is like, you know, pe- making people actually ask well-informed questions. Uh, because we were not asking questions until now, we were just entering keywords. Like in Google, Google kind of spoiled us over two decades to just type in keywords. And we wouldn't even type the full keyword. They would autocomplete it for us. And like we would, we would just like click on that, go and read the links and waste time, right? So the first time you have this unique power to talk to a computer like a human where you can just go and ask questions. Now we all need to get used to that power first. We, we suddenly got it.

• 8:46Speaker 01

  We, we never knew this was coming. It was not incrementally given to us. It just happened all of a sudden. And then um you're like, okay, what do I do with it? Now you're like thinking about what to do with it. But the reality is this should have been the way you interacted with search engines all along. Like you're not coming and talking to me, Aravind, and asking like, you know, Google perplexity, right? You're like asking, what is perplexity doing that Google cannot do? Like what is it? Why is Google not able to do this? Like, these are actual questions. Now you've got the unique ability to ask the same kind of questions in the same conversational way to an AI.

• 9:24Alex Kantrowitz?

  But I'm also wondering how big this can actually be because it's interesting. Like Kevin Roos wrote this story, I'm sure. I mean, you spoke to him for it. Can this AI-powered search engine replace Google? It has for me. But it's not actually a full replace. He even writes later in the article that he's not gotten rid of Google. And so I do wonder like if this is something that can fully replace Google? Or is it just going to be something that you would use for a subset of questions? You might not type into a traditional search engine.

• 9:54Speaker 01

  Yeah, let me, let me give you a more nuanced perspective here. Now, there are two points on a line. The left point is navigational searches. The right point is directly giving you an answer. So the left is traditional search engine. Right is answer engine. We are, we are in the right and Google can be considered on the left. Except Google has done a lot of work already to move more towards the right. By preserving as much as possible of the left. Uh, by the way, there, I don't want you to think about it like left thing, right thing.

• 10:26Alex Kantrowitz?

  Yeah, yeah, no, no, we're not going there.

• 10:28Speaker 01

  Um, it's just like mathematical. Um, now, Google is trying to keep its link UI, link interface, but also try to give you answers whenever possible. Uh, that could include weather, time, um, sports, election results, all sorts of things. So like just getting to a website really quickly, uh, the browser navigation. Now, perplexity is giving you answers all the time, but that's not the only time, uh, only way you want to interact with the web. You sometimes just want the, uh, like you're doing mathematic calcula- calculations in times. You're doing, uh, if you want the time and time in a particular city. Um, and then you want like, uh, NBA scores.

• 11:11Speaker 01

  And like you want, um, you know, if there's an ongoing election, you want to track like, you know, how many seats each part of Canada has won. So all that stuff doesn't need a large language model that can pull sources from different parts of the web and summarize these sources. So we are approaching the problem from a different, uh, end of the line. And Google's approaching a problem from a different end of the line, except we don't have any business model to defend.

• 11:39Alex Kantrowitz?

  But what I'm trying to say is it's not, it might not actually be you versus Google. It might be you and Google.

• 11:45Speaker 01

  Possible, you know, uh, I, I don't want to think about it as, uh, them versus us either. Uh, my sense is that if we are the place for de facto information and accurate information on the internet, um, it will definitely make a dent on them. I'm not, I'm not, you know, clearly like, uh, they, look, their whole image is that they are the only place on the internet for facts.

• 12:08Alex Kantrowitz?

  Is that really their image?

• 12:10Speaker 01

  Well, yeah, like you go, uh, you, you, you go ask anybody, they just say, have you Googled it? Right? Have you Googled it? That, you know, that, that means like-

• 12:20Alex Kantrowitz?

  I buy that.

• 12:20Speaker 01

  Yeah. Yeah. So that, I think that that will definitely change. There's going to be more ways in which you can, you can learn about anything.

• 12:27Alex Kantrowitz?

  I mean, the behavior is also so different. I mean, so let's see, this is a stat that actually perplexity pulled out for me, so fact check it. But in October twenty twenty three, it said you had forty point thirty five million visits. And the average session duration, and this is like the really interesting thing. Twenty one minutes and fifty eight seconds, which you would imagine outside a Google page. It's just might be a minute.

• 12:53Speaker 01

  So that's the thing, right? Like, I I just feel like the idea in our product is to make people more curious and engage and ask more questions. In fact, our the whole idea of a follow up question was something we we innovated on. Uh, where we would suggest to you what follow up question to ask. Uh, you know why we did that? There's number one skill is number one skill that actually is a bottleneck for these products to really take off. It is stability to ask good questions.

• 13:23Alex Kantrowitz?

  Oh, yeah.

• 13:24Speaker 01

  Yeah, your your hu- human mind is not great at articulating a question. We are all very curious. We're all super curious people, but not all of us have the skill to translate that curiosity into a well informed question.

• 13:38Alex Kantrowitz?

  Definitely. It's like with Wikipedia. No one goes to like the Wikipedia homepage and is like, what am I going to explore today? You end up on Wikipedia through search, right? So you have that query in mind.

• 13:49Speaker 01

  But then you you read the whole page. You read the whole page that's written for everybody, but not personalized to you. You care about only some parts of it. And that's why, like, we wanted to create an experience of a dynamic personalized Wikipedia for you. Like, you just you just you just get an article on the fly.

• 14:06Alex Kantrowitz?

  Have you spoken with Jimmy Wales over there? And are they gonna build their own G P T? What's going on with Wikipedia?

• 14:12Speaker 01

  Yeah, they're not gonna build their own G P T. I've spoken to him and he uses our product and he likes our product a lot and, you know. Uh, he used it when he was in Amsterdam and was looking for museums for his kids. Uh, that was like within a few meters from his hotel. And he said like, Google sucked at it and chat G P T hallucinated and like we got the answer right. So he really liked that.

• 14:32Alex Kantrowitz?

  I'm surprised I would think that they would build one.

• 14:34Speaker 01

  He wanted to do some things like, oh, I if I just wanted to create a perplexity like experience, but just on Wikipedia articles alone.

• 14:43Alex Kantrowitz?

  Hmm.

• 14:43Speaker 01

  How can I do that? He he wanted to build something of that nature, which we already had, by the way, like in in our focus searches. Uh, you could just pick one domain and ask questions. I told him about it, but he's like, I want it to be on wikipedia dot com. And I told him, hey dude, like Wikipedia, people go there only through a search engine. They don't go there automatically and start exploring. But he's like, okay, what if I had this on the Wikipedia domain, then maybe people will do that. So that was his idea at that point. I don't know what's the status of that idea today.

• 15:13Alex Kantrowitz?

  There are so many things that you're going to come up against. I mean, you're already coming up against in this battle. First is data, right? You have far less users and we all know that Google, Microsoft Bing even use the data that they have to refine their searches and that feedback to help improve the quality of results. So what's your answer to them on that front?

• 15:33Speaker 01

  Yeah, so look, the whole point, the whole amazing part about this generative AI is that it doesn't actually need as much user data as you'd think.

• 15:43Alex Kantrowitz?

  Yeah, share more about that because that's against the common thought about it.

• 15:46Speaker 01

  That's why OpenAI is a disruption. Why should a research lab like OpenAI be able to create a better chatbot than Google that has like so much more data or like Meta that has so much more chat data? The reason is that like we were able to learn this generic knowledge of like language itself, common sense, understanding just by predicting the next word on the internet. That is the GPT model. All it does is take all the internet, curate it reasonably well, so that you only train on signal instead of noise and try to predict the next word from the previous words. But by doing that, you get such a good understanding of how the world works, how language works, basic common sense reasoning, mathematical reasoning.

• 16:30Speaker 01

  And that enables you for the first time in decades to be able to use little data, learn on the fly and do a lot more things. Right? So somebody who has access to such an amazing intelligence API, which has been exposed to all the developers in the world through OpenAI and Azure and things like that, can now program it in ways they want. Program it for a search engine, program it for a companion bot, program it for a coding assistant. All these things are possible with one generic intelligence layer. And also companies like Meta have open sourced these ways. So you don't even, the flexibility in the programming is even more than what it started off with.

• 17:12Speaker 01

  So the first time, for the first time, we have had intelligence outside Google, like artificial intelligence outside Google, that's of higher quality than what is inside Google.

• 17:25Speaker 01

  Which a new startup like us is able to harness, take advantage and get it out to users before Google can. That's really what has happened. We benefited a lot from OpenAI and Meta, all like making so much progress on large language models and getting them to the user base, developers, other, other small startups, to build amazing products that would never have been done before with way less data, because you can just few-shot prompt it or like fine tune it on a small subset of data that you required. So the advantages that the incumbents had in terms of having a large volume of data has gone away. I'm not saying you don't need user data at all. All I'm saying is you need a fraction of what was needed earlier for the first time.

• 18:10Alex Kantrowitz?

  Arvind Sirnevas is here with us. He's the CEO of Perplexity. We're going to keep up this conversation right after the break. This episode is brought to you by App Point. Everyone's racing to roll out AI right now. Co-pilots, chatbots, agents doing real work. But here's the part nobody loves talking about. All that AI runs on your data and most teams have no single way to see it, secure it and prove it's under control. That's exactly what App Point does. For 25 years, they've been the trusted layer beneath the world's most demanding data, now extended across your entire AI estate, your data, your cloud and the agents acting on your behalf. It's how more than 28,000 organizations deploy AI with confidence.

• 18:54Alex Kantrowitz?

  So innovation scales without scaling risk. App Point, the unifying trust layer for AI. Head to apppoint.com to see how enterprises deploy AI with confidence. Learn more at avpt.co/bigtechnologypodcast. That's avpt.co/bigtechnologypodcast.

• 19:16Alex Kantrowitz?

  Violent threats against executives are growing at an alarming rate. When a company experiences backlash, executives are the first people blamed. When their address and other personal details are sitting online, this backlash can lead right to their front door. The team at Ironwell knows this better than anyone. They've protected some of the most targeted executives and individuals on the planet for almost two decades. As a wide-glove enterprise security service, they protect your people with continuous personal data removal, proactive prevention tools, and human-led emergency support. So when someone goes looking for your executives, Ironwell ensures they hit a dead end. So here's what to do. Go to ironwall.com/bigtechnology, fill in the quick form, and request your free risk assessment. Their team will show you just how exposed your executives are

• 20:01Alex Kantrowitz?

  and how to lock it down before a threat reaches their front door. That's ironwall.com/bigtechnology. Learn how to stop online threats before they become real-world attacks. The link's in the show notes.

• 20:16Alex Kantrowitz?

  And we're back here on Big Technology Podcast with Arvind Srinivas. He's the CEO of Perplexity, which is the hottest AI startup in the world right now. It has 10 million users. It's a search tool, a search engine that you could use, but it's more than search. And it's fascinating. We're just talking before the break about how you don't need as much data for Perplexity as you would for other applications. That is one of the innovations that we're seeing in AI. But what you do need is news, right? And you have a very interesting approach to UI, where you'll put four links, not always news. Sometimes it's a YouTube video. Like when I searched you on Perplexity, I found a couple of your YouTube interviews. And those go at the top.

• 21:01Alex Kantrowitz?

  If you're not training on vast amounts of data, actually urgency and being in the moment is crucial, I would imagine. I mean, you even look at some of the suggested searches you have, and that's there. I'm curious, from your perspective... How do you think news and information creators are going to factor into this world? They're already, the Times is already suing OpenAI. I go back and forth because I run a content business about whether I should let these bots crawl or not. Talk a little bit about what you think there and who it actually benefits.

• 21:35Speaker 01

  So our thinking here has been to rely on our background as academics. First of all, like when we launched and ChatGPT launched, one fundamental difference is we give people information on where the answer is coming from, the sources.

• 21:51Alex Kantrowitz?

  Right at the top, which I love.

• 21:52Speaker 01

  Right at the top, the sources, the citations, right? And that, I believe, is a way to attribute the creator of the content. Okay, you know what? I'm not stealing your content. I'm actually attributing it to you. I'm telling the user exactly where it's coming from. And I'm going to drive traffic to you if they actually want to learn more about other stuff on that website. So in academia, this is a similar concept as writing a citation.

• 22:24Speaker 01

  And if someone's interested, they go and read the cited paper actually on their own. They, you know, we don't, we don't have to like tell them to go read it. They read it on their own because they think it's important. And I think that's very fair use of other people's content. We're not stealing it. We're actually like just being a middleman between them and the, and, and reader. And we're giving them more visibility, right? So it's a different kind of visibility. Until now, the internet economy has been built around like, how much traffic am I getting? Now you're going to be like, how much awareness am I getting? How much awareness am I getting? Like, how many times has some content on my site been viewed by people?

• 23:06Speaker 01

  It may not, it, it can be viewed in a different way too. It can be viewed through somebody else. It can be viewed directly on me, as long as the user knows it's from me. It's not.

• 23:16Alex Kantrowitz?

  Right.

• 23:16Speaker 01

  That's, that's where, that's where we differ from Google or ChatGPT. So not Google, I, I guess Google Bard and ChatGPT. Google itself is going to be pretty transparent about the links. And our sense is also that.

• 23:31Speaker 01

  The value of a link click on perplexity will be more than than than on Google.

• 23:38Alex Kantrowitz?

  You know, your pitch.

• 23:38Speaker 01

  Because like someone has to be way higher intent.

• 23:41Alex Kantrowitz?

  Yeah.

• 23:41Speaker 01

  To still leave the site and go, right?

• 23:44Alex Kantrowitz?

  Your pitch is very similar to like to the one that publishers make to advertisers that like it's not only about clicks. It is about awareness and you know, when you get traffic from a good publisher, it's going to be high quality traffic. That being said, like, if you don't have eyeballs on your site, it's very difficult to make money from it. But we'll see how this plays out. I think it's still an open question.

• 24:05Speaker 01

  It it's an open question. That's that's I I will clearly establish that here that we don't have a complete answers to all this today.

• 24:14Alex Kantrowitz?

  How are you thinking about making money? I mean, you have a pro version that's $20 a month. You don't have ads yet, but you talk very eloquently about the idea that you might eventually have ads and how that could be even better than your experience. You know, we just you talked about Google earnings at the top. I mean, they just added an incremental $6 billion in advertising in the quarter, an incremental in six billion. That would be the 11th biggest advertising business on the globe, period, outside of China, according to Ryan Weiser from Madison and Wall. So people end up spending, what do we say, like 30 something minutes, 21 minutes and 58 seconds on perplexity? And not just, you know, zinging out to websites through Google. How are you going to make money on that?

• 24:58Alex Kantrowitz?

  And how does that change the economics of the internet?

• 25:02Speaker 01

  I think that we just need to rethink what advertising means, right? You go to the first principle of what advertising means. Advertising is a way for the creator of the content or the brand owner to maximize their awareness to the end user. Or or viewership. Now, uh, tend to links is one way to connect the reader and the creator.

• 25:29Speaker 01

  But not the most efficient way.

• 25:33Speaker 01

  Right? The more efficient way is actually answering parts of what exists on your website to the end user directly. What exactly do you want the user to know about you? Which you're putting on your site anyway. You don't have to tell us, you don't have to like pay us for that. In fact, I, I believe it's fair, it's more fair to the advertiser to like not be the middle guy telling, and I'll figure out how to start the order, you, you pay me money for that, but rather you, you write about yourself in the most honest, transparent way that you want the end user to know about you. And let the AI do the job of taking that and reaching the end user more, more efficiently than before.

• 26:15Alex Kantrowitz?

  But advertising interrupts though, like so it's very difficult, very different to be like a blue link on the side or above. Versus like you're having a conversation and then like perplexity ads come in and like, hey, wait a second. You should know about this.

• 26:29Speaker 01

  Yeah, yeah. So, so what I'm trying to say here is advertising has to be rethought. From the perspective, oh, I just want traffic versus I, I want more awareness. And what tools can I use to increase that? Right? How can I write the content on my website even better? To reach the end user for certain kind of questions? And you are in charge of that yourself as a creator. You don't have to be under the mercy of someone creating an auction system and exploiting you. Right? By the way, the internet can be very fair. We, perplexity is not trying to be, uh, creating a trillion dollar economy on top of ads.

• 27:12Speaker 01

  Uh, or, or on top of all the searches that we get. And trying to take all of it and not give away much to others. Like nobody won, only Google won in the past, they were aware, you know, the, the telecom providers did not win. Uh, the, the, the publishers did not win. They took all the moolah out for themselves. We want to change that and create a more fair world. And that's why I think like people will be more willing to work with us.

• 27:37Alex Kantrowitz?

  And I mean, it also doesn't just change like the traditional search model. It can change commerce too. I mean, if you think about it, if you're on Amazon, you could search Amazon and just type a, uh, you know, conversationally into a, a search bar there and they're actually starting to do this. I think it's a product called Rufus.

• 27:54Speaker 01

  Yeah.

• 27:56Alex Kantrowitz?

  That obviously minimizes their amount to make ads, but it's also, how do you think? Yeah, I'm curious. How do you think it's going to change e-commerce?

• 28:03Speaker 01

  Well, it's good. I think it's all like, think, think about it as short term and long term, right?

• 28:08Alex Kantrowitz?

  Yep.

• 28:08Speaker 01

  You, you may think it minimizes your ads, but what if people just purchase more because they're able to make better choices now?

• 28:15Alex Kantrowitz?

  Yeah.

• 28:15Speaker 01

  Right? Uh, most of the time, people don't end up purchasing. They end up shopping a lot. But they don't actually make a purchase because they're not sure what to buy.

• 28:25Speaker 01

  But when you have a great shopping assistant that directly gets to what the user wanted. The conversion to a transaction could be much higher.

• 28:34Alex Kantrowitz?

  Right.

• 28:34Speaker 01

  And the retention could be higher. The number of purchases, cumulative number of purchases could just get higher for user. And uh, you don't have to rely on other people anymore. Uh, you can just, and, and the merchants are happier. So they're willing to pay you a little more for like being displayed on your site. So you can create a different kind of economy, right? So by the way, for Amazon to do this hard, obviously, because they do rely on advertising revenue on amazon.com. To keep amazon.com profitable. So they have like similar problems to Google there. But, but, but look, they can, they, they are such a massive company and have revenue from AWS and like. Can definitely make up for like any short term hits here.

• 29:15Alex Kantrowitz?

  You have Jeff Bezos as an investor?

• 29:18Speaker 01

  Yeah.

• 29:18Alex Kantrowitz?

  Have you made this argument to him?

• 29:21Speaker 01

  No.

• 29:23Speaker 01

  By the way, I want to establish that he doesn't, you know, he, he, he's not like actively involved in Amazon and. His investment in us has no ties. He's chairman, but he, he, his investment in us has nothing to do with uh, us working with Amazon. It's a, it's an independent investment.

• 29:40Alex Kantrowitz?

  How did you recruit Bezos? Or did he just come in the front door? We had, I'm sure it's a great story.

• 29:44Speaker 01

  No, he, he, I, it's, it's not, it's not as crazy as you think. Like we got connected to his um. You know, business expeditions fund and

• 29:53Alex Kantrowitz?

  Hmm.

• 29:54Speaker 01

  Um, you know, obviously in his style, they asked us for like, like, um.

• 29:59Alex Kantrowitz?

  A memo?

• 30:01Speaker 01

  Memo and you know, like that was sort of the process and we had a pitch deck and all that and like demo and all these things. But yeah, they made a decision very fast and it definitely changed our our fortune um in terms of being seen as a widely aware like the awareness that we got through this investment is tremendous.

• 30:25Alex Kantrowitz?

  Interesting. Very quickly, as we come into a landing, uh, you worked at OpenAI. You know, I guess like the fashionable thing would be to ask you about their governance troubles. So if you want to comment on that. Fine, but what I'm hearing about from them is they keep signaling that's huge stuff is coming down. Uh, the pike this year, just like the way that they're talking is that it sounds like some serious advances are going to happen this year. Do you have any sense as to what those might be? And I don't understand like what what could what could inspire such talk? Is it much? I mean, if it's a slightly better GPT model, okay, but it sounds like something more.

• 31:05Speaker 01

  I mean, I don't know what has been said publicly. Maybe you hear something more privately, but uh, I guess I'll just go by what Sam Altman told Bill Gates. I was listening to it like two days ago, which is um. They'll have a lot more reliable models, so uh, right now, sometimes one in a ten or one in a hundred completions of a GPT four model. Uh, is hallucinatory, can can say arbitrary things. It's not reliable. It's undeterministic programs. So I think they'll address that. Uh, they also address the fact of the models being like more multimodal. Like right now it supports images and input, but the most general version will support audio, video

• 31:50Speaker 01

  as inputs and also as outputs, they'll uh interface with text.

• 31:55Speaker 01

  I think they'll make some advances there. I think the third thing they'll make some advancements on is reasoning.

• 32:03Speaker 01

  Whatever reasoning G P T four is able to exhibit today is already amazing, but it's still pretty limited. And I think they'll make some more progress there in terms of multi-step reasoning and doing back and forth thinking. You want models to be able to like think for a while and then come back and give you some things, right? So they they'll probably make more progress on those things. And I guess the fourth thing that Altman mentioned was like just like

• 32:29Speaker 01

  reducing the costs, oh, sorry, increasing the personalization of all these models so that people can program all these things in a more reliable way. Like, for example, why should all the products within G P T four? You know, respond in the same way with the same style of like moralizing the user and like

• 32:46Alex Kantrowitz?

  Right.

• 32:47Speaker 01

  telling them what not to ask and like same kind of like diplomatic positioning of things. You want to have like more variety and diversity in the product experiences created on top of G P T four. Uh, and I think that is something they'll probably address, long context reasoning, things like that. So these are all like, when I say these things, it looks boring to you, right? Long context, uh, better models, better reasoning, more more reliability, multimodal. It's such boring answers, but when these actually happen together, when when these all really happen at the same time, it'll look like the next generation model dropped.

• 33:23Alex Kantrowitz?

  With Anthropic, what do you what do you think about the Claude model? And does it hold a candle to G P Ts?

• 33:30Speaker 01

  So far, not to be very honest, but I'll tell you one thing. They, there are people in the world who actually like Claude a lot more than G P T. In terms of response styles and eloquence, they like Claude a lot more. It's more natural. It feels more like talking to a human than to an A I. If you adopt Claude, so there are, we have, we offer Claude as an option on the perplexity product for pro users. And I I know a lot of people who still use Claude instead of G P T four.

• 34:02Speaker 01

  Even though G P T four is a more capable reasoner, because they like the way Claude responds. So Anthropic certainly has something. Now, can they create a model better than G P T four? I definitely think they can, and if they don't, they are kind of doomed, in fact. This year in two thousand twenty four, if they don't create a model better than G P T four, all the funding in the like, you know, the race is not being put to good use. But I actually think they will end up creating a model better than G P T four this year. Like it it it's sort of almost guaranteed to happen. So I believe it's going to happen with Claude three. Now, does that mean OpenAI is in trouble? I don't think so either. I'm sure there's a G P T four point five or five that will stay ahead.

• 34:44Speaker 01

  So it really is going to be a cat and mouse game there where Anthropic is playing catch up and OpenAI is ahead through multimodal capabilities, reasoning capabilities and things like that.

• 34:56Speaker 01

  Now, the question is like, how long can OpenAI keep on staying in the lead? If the delta between them and their competitors is slowly reducing, uh, what what is the thing that's going to keep them ahead? That's what I'm not clear. And when Llama three comes out and Llama three is as good as Claude three or very close to G P T four, uh, and the weights are just given out to everybody, it threatens the whole economy that these companies are creating with through A P I's. It doesn't demolish it because just because somebody gave you the weights, uh, doesn't mean you can take it and serve inference yourself. You still need someone to serve inference for you in a more efficient way.

• 35:37Speaker 01

  And like all the trickery to like make an inference efficient is still not like easy or open source. But it certainly reduces the monopoly power of like these um close source model providers. And Google is also being going to play in that market through Gemini as an A P I too, right? So it's going to be interesting to see. I think OpenAI's value will lie in the chat G P T product itself, uh, the end to end product. And they're going to face competition there too, like Meta is going to have their own A S and there's Bard. It's all sorts of like Microsoft has Bing Chat, Copilot. So it's it's it's a it's a market, it's a tough market. It's like there are very, there are a few players and they're all going to keep competing. And like, you know who's really winning? Converse.

• 36:22Alex Kantrowitz?

  Yeah, and Nvidia.

• 36:24Speaker 01

  And Nvidia, exactly. So everybody's using Nvidia. I mean, of course, Google doesn't use Nvidia chips. And, you know, I guess Sam Altman is trying to create his own chip company and things like that. So we'll see.

• 36:37Alex Kantrowitz?

  Yeah. All right, Aravind, I know you are pressed for time. Keep shipping. And if people want to check out perplexity, where can they find it?

• 36:46Speaker 01

  perplexity.ai.

• 36:47Alex Kantrowitz?

  Okay, excellent. Thank you for joining. I hope it's not the last time. Really illuminating conversation. Great point to end on. Thanks again for being here.

• 36:54Speaker 01

  Thank you, Aravind.

• 36:55Alex Kantrowitz?

  All right, everybody. Thank you for listening, and we'll see you next time on Big Technology Podcast.

• 37:02Speaker 02

  Now, more than ever, we need someone on our side. But Rob Whitman? Whitman's votes drove up costs, cut Medicaid, and risked 350,000 Virginians' healthcare to pay for tax breaks for himself and his donors. He's not on our side. Shannon Taylor will fight corruption, cut taxes, and bring down costs. This November, vote for Shannon Taylor. She's for us. Paid for by HMP, the HouseMajorityPact.com. Not authorized by any candidate or candidate's committee.