Can Perplexity AI Disrupt Search? Google Created A Terrible Internet, Says CEO Aravind Srinivas

hello and welcome to this money control interview we are in conversation uh with Arvin shrinivas who joins us all the way from the US he is the co-founder and uh CEO of perplexity AI uh an AI startup that has been making waves in uh Silicon Valley now this is a startup it's a search engine that's focused on generative AI uh they recently raised $73 million valuing them at half a billion and they capable includes the likes of Amazon founder Jeff Bezos Nvidia among others Arin thank you very much for joining us on money control and uh you know I've been using perplexity playing around with it seeing how it Compares with Google and uh you guys are really redefining search As We Know It uh I'll come to that but I first want to start about your own story uh it's been quite a journey for you from uh Chennai I Madras btech mtech Berkeley uh to now co-founding uh perplexity take us through your own story your origins in India what got you started what made you interested in AI I mean I started around 2014 I believe uh to like think about Ai and machine learning it was actually called machine learning at the time I did an internship in India in one startup in Bangalore um they they took me because I got a 10 out of 10 GP in like my second year in I Madras and you know like that was pretty rare so they just took me out of the promise and um the problem they gave me was hey like we have a lot of e-commerce data and just you know build a recommender system using it and I had no idea what any of these terms meant so all I knew was it's a huge Matrix with users and items and it's heavily spared like most users users haven't bought or clicked on most items so it's going to be a lot of zeros and a few ones and I have to figure out how to recommend new items to users especially new users who haven't like browsed much because that's the coldstar problem and the most uh interesting part was me finishing the whole internship in like a month uh using by reading a paper and reimplementing it and like uh deploying it to production and they were like already making uh like money with the contractor like consultant they were working with and I had one month of time just to like study machine learning because I got so interested in it and I watched Andrew in and Jeffrey hinton's lectures on Cor era and like you know like got really excited about all these topics went back to it Madras I was in electrical engineering so I machine learning was a computer science topic so i s the professor in itm that who was also one of the the only professors in India at the time doing ml research Professor balaraman ravindran he he's an i madress and he uh told me you know take my class and I'm I don't think e guys would handle it uh but you know if you if you can handle it maybe you can you can um um continue and I almost to this class like basically the top one or two I believe don't forget exactly and he really like like like wanted to work with the top students for doing research with them so I continued did a lot of research with them uh worked on deep reinforcement learning at a time and it was not very popular uh got to do an internship with one of the touring Award winners of deep learning yosua Benjo in Montreal all these like things got me an admission in UC Berkley for PhD um and then like you know there I got noticed by open AI nice uh researcher John schillman who was also like a senior student in my own lab and um that got me into a journey of like working at top research labs in the US and learning from the best and um over time that translated all into like starting company perplexity you know I uh I noticed you interned at Google and deep mind and interned at openi and then worked as opena as a research scientist for a year before you decided to start up perplexity sometime in August 2022 how did you come up with the idea of a perplexity uh so my there was no clear one moment where I had an idea uh in my head and like translated it to a company uh those are usually how like movies are made but reality doesn't work that way uh my co-founder like like we were always like you know trying to translate many different ideas of how to use large language models and build commercial products using them uh we had early demos and prototypes of searching over like databases like Twitter or like you know I always wanted to know which VC or investors investing what startup so we would scrape all the funding news and search over them we just realized how we were approaching the problem of like scraping a lot of data organizing it into real like structured tables and then powering search over that was a round about way and a more efficient solution is just to scrape the links and let the large language model the AI model do the reasoning on the fly when the user is asking for a question and and give the answer um that was an Insight that my co-founder had uh and he prototyped it in a weekend and then we decided as a company that we would just do the more efficient longer term thing right um Arin over the years you know there have been many many attempts to disrupt Google search as we know it from um you know being searched to du du go to neva research which shut down recently why do you but you know none of them have really taken off in a significant fashion despite showing a lot of Promise initially what's different about perplexity why do you think you will have staying par and you will be successful I mean there is something to be said about the right place right time right like um you you everybody else who tried to disrupt Google uh through a focus on private see or like like add free models and those were not something that a large section of the population wanted or was not truly disruptive to Google either um but um on the other hand um when when it came to actually like offering a answer based experience where links were no longer The prominent part of the user experience but you could just get answers directly uh that was a true disruption because that whole basically threaten their core user interface right the user interface of Google is tied to their business model like people are advertising for how many times their links get viewed on the 10 Blue Link UI or how many times their link gets clicked on the 10 BL link UI now when that gets disrupted when you do not need that interface anymore when you can just Converse and ask more questions and engage in a conversation and get sources but directly get the answer that disrupts the whole incumbent in a way that even if they wanted to do the exact same thing it's going to take them a lot of time to work around the existing business model existing users and Wall Street and expectations and that's the perfect uh mode in which a startup can operate because you can take all the risks that the big company cannot now the timing to do this uh was 2022 when there existed sufficient progress in generative AI right especially in in this one tool called the large language model when it got sufficiently good not great but sufficiently good to roll out a product that was getting the 802 on most questions you could ask that was the right moment for a startup to start and like write the next wave right uh and Jeff basil says this you always want to start companies where you're not like writing the existing wave but you're writing the next wave and Nea had got this wrong they tried many like like versions of pivots like you know from privacy to adree to crypto crypto and then by the time they realize it was an AI and llms they R ran out of funding it was already the four-year period for them right right Al I also want to understand you know so essentially what you're saying is Google in a way is very constrained by its existing business model which as you said is driven by clicks and ads and you know how many people click on these links in pages but they are also investing a lot in improving their generative search experience I think Gemini can you know throw up results um in nearly real time U it can also give you a summary of information like what perplexity is doing it will not throw up links uh using you know the search engine results pages but it will give you a summary uh it's something that Microsoft Bing is also trying to do and there are reports that openai is also working on a you know browsing kind of product so again what's your Mo going to be here our Moe is going to be our user experience our speed at which we deliver the answer and our accuracy of the answers right um if we are not the best in these three categories then we will be displaced by whoever is better whether it's open AI or or or Microsoft or Google it doesn't matter right like our goal should be the best you spoke about these three aspects Arin but the other aspect is again um how do you ensure that you know you're serving up the most accurate information that you get around the edge cases where you know there is scope for perplexity to get something wrong or um hallucinate or whatever so how you solving for those edge cases I mean how do you sort of ensure that um you know it gets better and better and better because there are no reference links here that you immediately click on I have to go by the summary that you give me on a particular topic well we are giving you sources at the top that is fun very tiny tabs Arin I'm sure hardly I mean I'm sure for people who want quick answers I don't know how many people are actually clicking those links I I know you know like that's the thing right I I do know how many how many queries actually have a user clicking on a link and it's pretty significant so uh to be very honest with you like from the beginning we believed in a model where we are transparent about where we get the information from rather than pulling all your information and not attributing it to you and whe if the information is from money control we do tell the user where it's coming from we give give the citation money control and like if people actually want to go and read the article they can go and read it and uh this is actually a big difference from what Gemini is doing they don't even give the sources and uh you know why because they don't want you to like even use this mode of search they want you to go back to the Tim Blue Links for search so at least open AI gives you the sources in in Brackets but that's not as good experiences like the user interface that we offer so and like more than 10% of our queries actually have people clicking on the links I know you're smiling but like really that's the truth so people I grant I'll be honestly I'll be happy if that happens because you know redirected traffic is good for us so I'll be happy if that happens it it really doesn't like a lot of people praising Us online on Twitter or X as they call it now of how they're get seeing an increase in referral traffic from perplexity understood in terms of your revenue or business model Arin I mean I think you have over 10 million active users now um and some of them are you know paying users where they pay you what $20 a month can you tell us how many paying users uh you currently have yeah so we haven't like made that statistic public but what I would say is like you know we that that fraction is increasing by more than 50% month over month so at some point we'll make that number public but um our belief is that the most important metric is usage um not the number of paid users um and uh that said there is a big correlation between the number of paid users and number of queries they produce per day to if you pay for a product you tend to use it more and um you might think $20 a month is expensive but $20 a month is like four or five lattes in in the US you know per month people buy a latte every day in Starbucks it doesn't always taste that good it's sometimes like not hot enough sometimes it's burned coffee yet you don't care you just pay for it we want search like information and knowledge to be that useful like the amount of time it saves you amount of value it provides your life is way is worth way more than the $20 per month you pay it's like it's like buying four or five coffees are you seeing a lot of usage from India as well I don't know if you track geography by geography but considering we have the largest number of Internet users are you seeing increasing activity and the second part of that question will you have a different pricing strategy for India $20 going by us standards you know maybe three or four cups of coffee but for India will will you look at a different pricing yeah so we have more than a million users in India right and we want that number to expand to like 10 to 100x more because Indian population is so big and we need to do little more work on supporting Indian languages and you know like uh support like uh sometimes the queries answer in English even though your query was in your like original Source language we need to address these problems and like make the app better for like internet nationalization but once we do that I believe that we can certainly uh command a bigger user base in India and India is like the Next Generation economy um I know you're thinking $20 a month is expensive for India today and I probably think you're right but India's GDP is growing so much that I believe uh the optimal pricing is something we'll arrive at maybe in some time to come um and I I don't think people are going to be uh in the the previous mode of like trying to optimize for prices like how they thought about Netflix or Spotify because people are going to think about it as that this tool makes me more productive at work let me use it um so I don't know exactly if the $20 per month Works in India uh but I certainly don't think it'll be very cheap either like there are going to be people who would be willing to pay up for the value this sort of product has to them right Arin since you mentioned you know how you you to sort of work towards supporting more Indian languages now if I look at the foundational models that you use you use GPT and Claw um will you also look at tying up with someone for instance a suram which is building you know an there and um I like one of my friends is one of founding engineers in saram and you know I I have a good regard for okay and uh um I think like I I I certainly hope that the model model you train specifically for India ends up being a better model to use than the generic model like GPD or CLA or the models that we build uh so that it makes a lot of sense to just use that as a default experience for Indian apps and I'm excited to be a customer of llms built by saram or other Indian language model companies uh you already use their llm not yet they haven't produced not yet they haven't produced it yet when they are ready for go to market Market uh we would love to collaborate with them and work on like you know because we are a big platform for them too yeah if we use their models uh we have a user base that they can already bootstrap from so we're excited to collaborate with them Arin currently you rely on uh sub I mean paying users what are the other models you will look at I I think I saw an interview of yours recently where you said you know you will be open to advertising um at some point take us through what other you know what other scope do you see for an engine like perplexity where else can you make money yeah uh advertisements is something we shouldn't rule out and we should just do it in the right way uh if we go the way of like um you know Google where we corrupt the quality of the answer in order to make more revenue and satisfy our shareholders then we're also going to end up in the same fade and somebody else who wants who does it the right way will disrupt us so we should just figure out what is the right real estate for ads and what is the best way to do it in this new experience where the answer quality is not corrupted but just the source will Partnerships also be a focus I mean I think you've already tied up with rabbit R1 and yesterday I saw your exchange with nothing scar so yeah yeah we we we're going to partner with other devices which is actually you know one way to distribute PX City API and we're also going to try to tap into getting our apps and like like like U bundling our uh subscription plan with their devices uh like these are all creative distribution opportunities for us to grow our usch more and like build more awareness to perplexity and um there are a lot more people coming up there's a glass company called brilliant which is making smart glasses uh we are going to work with nothing and then we're also going to collaborate in more ways with future devices of rabbit so there's going to be a lot of opportunities for us here and um you know like like the form factor is going to be a disruption too when people start interacting with voice they're going to be able to truly feel the difference from the 10 Blue Link interface which was not conducive for voice F Factor at all to uh directly getting answers to your questions right right AR I also want to understand you know as someone who's worked um I mean uh in Google openi now running perplexity we all Wonder every day you know how is the future of the web going to be how is it going to be monetized more specifically because for digital Publishers and I'll give you money controls example I'm sure this is the case with many uh search engine optimization you know is a vital part of our function we spend significant amount of time every day trying to figure out the keywords the URL the tags that you know will make Google scrollers capture our articles and rank them higher but with business models like perplexity and I know it's taking you back to the source website it's still going to be a challenge how do you see this evolving will search as we know it today change or you know will people search use URLs for different things search for maybe summaries on perplexity how is it going to evolve I think I think the way it's going to evolve is uh you're not going to value a generic visit to your site right like everybody thinks oh I got like x million visits a month therefore I'm doing well but you it's not just about the visit but how much time they spend on your site too right uh if somebody reads an answer on perplexity and the content for that answer was pulled up from money control despite that if they still chose to click on the money control link and go read the article on money control that's way higher intent referral traffic than uh some keyword search on Google where money control popped up because of SEO and somebody clicked on it and realized it wasn't relevant and just went off right so the amount of time a user spends on money control and like uh the quality of the referral is more important than the quantity in terms of number of visits and the time like you know lack of engagements uh time so that's one obvious uh realization from how the llm answer Bots are going to change uh the internet like what people value and what people are incentivized by is going to change a lot and analytics products will also change instead of measuring like referal traffic you're going to measure how many times your your domain got cited in an answer bot like ours and we are forced to also give you the statistic because we rely on you creating good content for our products to be good too and additionally there'll be lot more opportunities to work together the kind of Internet that Google created might seem like good to you because that's what they want you to believe but honestly they've created a terrible internet where they scrape all your content without giving you any money or creating any incentive for you to benefit from their ad business uh in the in the pretext of telling you that you get all your refer traffic because of them but reality is if you stop stop sharing sh in content with them uh their service or platform becomes way worse and that is going to disrupt their ad Revenue so if we can establish a better relationship where uh clearly like where your content is being used in an answer bot is shown to the user and you get lot of advertisement and brand value through your citations being used for high quality answers and uh a lot of like co-marketing opportunities where like we help you create more engagement on your side by suggesting interesting question to ask and uh ways to share revenues through subscriptions I think that will be like a pretty good outcome for all of us and that's the future that we we we want to create in perplexity it's pretty ambitious but I think that's a better internet than what Google created right which brings me to my next question Arvin the nyt lawsuit against openi and Ms how does it you know impact you what do you think about striking data licensing deals with content providers will that be the way forward I don't believe in data licensing deals I what I believe in is like uh a win-win situation that we can establish by saying hey look we scraping our content to answer our questions but because we do that you're getting this and uh in we can help you in these so and so other ways and so let's keep working together uh and I think we can establish a better win-win situation than what Google has done with the current publishing Publishers on the internet I I also wanted to you know get your thoughts on where you think India is in the AI race I mean there are many Indian uh U you know AI Experts of Indian origin striking it big in the US but where is India in the AI race I I I think that India is certainly not like very behind uh if like there are you know like two couple of startups trying to train large language models in India uh there are a lot of early St like like like deployments of Chad Bots uh in um it's just the awareness of Indian public to generative AI is actually pretty good uh a lot of people are aware of chat GPT and a lot of people are aware of like large language models and people are using text speech services I've seen seen an ad where somebody prototyped an like a Shah ruk Khan advertisement in like so many different uh Brands like so people are like using these Technologies already in like very creative ways so uh I think like we're going to see a lot more faster movement in the AI revolution in India compared to the previous web and mobile revolutions SP like we were a little behind the rest of the world there but what do you think it needs to do to play catchup in terms of compute in terms of talent what what else needs to be done I there are also deeper pools of capital you know for AI startups in the US yeah yeah yeah so wec need to take bets on Indian Founders who are trying to disrupt the existing incumbents right like um and um there are two categories of companies you can invest in one is like you can call them the rapper like you know people start off with existing apis but build great product experiences for end users and don't need to like worry too much about like you know having your modes or things like that just get the user base in India and create like economy at scale for for that user base and then figure out how to like optimize costs and build your own infrastructure and or partner with other Indian infrastructure companies um and then the infrastructure Builders like people build their own llms their own Hardware inference chips training chips data centers so that's going to be a longer term commitment and a longer term project to invest in and people should take bu bets on both of them and the government should also be involved if possible and um that's how you can you will see um a lot more activities right when Sam Alman came to India last year you know he said people I mean Founders here or companies here shouldn't waste time building a foundational model do you agree and you know do you have plans to build an llm I mean he's saying that because he doesn't want competition right but but then like um I I think there are some good thing advice too like the reason he's probably saying that is because um building a great Foundation model isn't just about raising a lot of capital you also need really good people like top talent to work on training these models utilizing the hardware effectively like making sure all the data mixes are done carefully well and that requires a lot of muscle into like running scaling law analysis at a smaller scale and making sure that your prediction of what is the the right data mix and right parameters to scale are all like done accurately so that you can just launch the large run once but pre-training is like very hard it's very challenging and you can raise a lot of money and still end up as a failure I believe at least one or two companies in India should aim to do that and I think they are like the Ola guy is and and S yeah yeah exactly I find it hard to tell the name that's why the guy but um yeah maybe they one of them will succeed or both of them will succeed and I I hope one of them at least succeeds and um that's good for India in general right um Arin does perplexity use AI uh you know to write codes what's your assessment of productivity Improvement that this is leading to uh I think people do uh like a lot of times I see people are are using it for basic questions about doc pages and stack Overflow like queries and you know quick quick sneak peek into docks without wasting time you know parsing and sifting through the uh you know hundreds of links so it's very useful very good productivity ad for day-to-day work and I see a lot of people using CH PT for coding as well right um final part of my interview Arin you know when I see the AI optimism the funding rush in the US I mean what is this a refle ction of is it the overall optimism around AI that you know this is a generational shift that we are seeing or are startups raising money because we could see a down cycle again in a so-called funding winter what is your own assessment of how long this is going to last I think it's going to last for much longer than we think because AI is the final frontier um so we AI is the final frontier and we we not going to need much more after AI is done in terms of software uh most of the things that we built in software is built through like a lot of hardcoded rules and once AI is able to learn all of that and like able to do most of it on its own um building software just gets like exponentially easier so the value AI is creating is exponential so this is going to be the ultimate thing and and so even if the amount of progress the speed at which the progress is being made today doesn't last which is fine like you know uh the the first derivative or the second derivative of progress is not always going to remain at this high value as it exists today what's the one or two things that we can expect from perplexity in 2024 2025 um 25 I don't know like you know first one year one year one one year at a time one week seems like a long long time in AI yeah like we usually don't even do three Monon Plans by the way we are just like going month by month because the world changes so fast that in fact that's a good reason to also be a startup right now because it rewards you for being Dynamic and flexible and um what you can expect is we'll always keep making the answers more accurate we we keep making the product faster and we keep making the answers more readable and this is our manthra too like three these three things if we just keep doing it every week every year uh after five years if you look back like it's just going to be incredible like it's basically your value as a companies how how hard it is to recreate you how much would it cost to recreate you right and if we make constant improvements every week it's going to be hard to recreate the whole process what would your advice be Alvin to an 18yearold in India or a 21y old who's just graduated who wants to start up start up in Ai and maybe do something disruptive like what you've done follow your heart don't don't look at markets and decide like you know don't say oh the market wants a sales analyst product I'm going to build it if you don't even care about sales like why do you build an AI for sales like build what you care about and what you would use uh because the first user of your product has to be yourself nobody cares about your product that you're building in the beginning other than you and your co-founders so and and it's hard to care about something if you don't actually like it so follow your heart and get some users and then the market will automatically guide you to building something that has some commercial value so uh that's that's the right order rather than following the market and then trying to like it on that note Arin thank you very much uh for talking to us and taking us through the perplexity journey and the future of search and finding information as we know it thank you very much for talking to money [Music] [Music] control

Channel: moneycontrol

Views: 36,262

Keywords: moneycontrol, moneycontrol live, ipo wrap, ipo this week, new ipo, moneycontrol ipo, ipo news, stock market, new ipo news, ipo market, new ipo 2024, ipo watch, stock market live, ipo live, juniper hotel ipo review, juniper hotel ipo, juniper ipo

Id: ev9OS1eooiM

Length: 32min 22sec (1942 seconds)

Published: Wed Feb 21 2024