Set against the backdrop of the tech world, the video delves into a compelling narrative about taking risks and navigating the challenges of startup culture. It is an inspiring tale of ambition, focusing on how viewing eclectic success stories, like those of Elon Musk, can ignite the fire in aspiring entrepreneurs. The storyline is bolstered by personal experiences shared from university days to aspirations compelled by necessity and a relentless pursuit of knowledge.

The protagonist's journey from IIT to working in premier tech companies like Google and OpenAI, and eventually launching their own startup, highlights the paradigm shifts required to undertake audacious goals. Unconventional thinking and an unyielding desire to continuously learn and evolve are depicted as essential traits. The validation of unique and emerging ideas, particularly in AI and tech, is positioned as crucial in driving innovation and competing against established industry giants.

Main takeaways from the video:

Please remember to turn on the CC button to view the subtitles.

Key Vocabularies and Common Phrases:

1. relentless [rɪˈlɛntləs] - (adjective) - Unceasingly intense; showing no signs of abating in intensity, strength, or speed. - Synonyms: (unyielding, persistent, unstoppable)

Relentlessness is the most, like, I would bet on it.

2. conviction [kənˈvɪkʃən] - (noun) - A firmly held belief or opinion. - Synonyms: (certainty, confidence, assurance)

How did that conviction develop?

3. disillusioned [ˌdɪsɪˈluːʒənd] - (adjective) - Disappointed in someone or something that one discovers to be less good than one had believed. - Synonyms: (disenchanted, disappointed, disheartened)

I was kind of disillusioned with the whole thing.

4. philosophy [fɪˈlɒsəfi] - (noun) - A theory or attitude held by a person or organization that acts as a guiding principle for behavior. - Synonyms: (doctrine, belief, creed)

Google has this philosophy of not having role specific interviews.

5. ambiguity [ˌæmbɪˈɡjuːəti] - (noun) - The quality of being open to more than one interpretation; inexactness. - Synonyms: (vagueness, uncertainty, equivocation)

So the kind of people who really thrive here are those who can work with ambiguity.

6. monetization [ˌmɒnɪtaɪˈzeɪʃən] - (noun) - The act of generating revenue from a product or service. - Synonyms: (profit-making, commercialization, revenue generation)

It had a monetization impact because...

7. serendipity [ˌsɛrənˈdɪpəti] - (noun) - The occurrence of events by chance in a happy or beneficial way. - Synonyms: (fortune, coincidence, fluke)

And finally this was completely by accident or serendipity.

8. reluctance [rɪˈlʌktəns] - (noun) - Unwillingness or disinclination to do something. - Synonyms: (hesitation, unwillingness, reticence)

But that kept producing reluctance.

9. paradigm [ˈpærəˌdaɪm] - (noun) - A typical example or pattern of something; a model. - Synonyms: (model, example, pattern)

Highlights the paradigm shifts required to undertake audacious goals.

10. pivotal [ˈpɪvətl] - (adjective) - Of crucial importance in relation to the development or success of something else. - Synonyms: (crucial, essential, critical)

I knew that this was like a pivotal moment for building AI products.

Relentless Journey From Middle Class Roots To AI Search Innovator

Nobody's willing to take the risk of, like, getting deported and, like, trying to start their own thing, living on, like, one subway burger a day. I was not necessarily from a very well-to-do background financially. I couldn't, like, just pay for masters, you know, and come here. I used to just watch videos of Elon Musk. They were like, why should we trust you guys? We don't even know who you are. There's no brand, nothing. There's 10% chance of this working. But if it works, I get hundred x returns.

You can never win there. You can say, oh, like, yeah, IIT. Which IIT? Oh, Kanpur or Madras or, like, Bombay. Oh. Then, like, someone's from Stanford or MIT or Berkeley. It's not the smartest person that makes the most progress after a decade. It's the most relentless person.

Most people are, like, so upset when they use the word rapper. If you like, Salter Kohli, Roy Sharma are people that Donita can bet on and didn't drop them from the teams and played the long game. Arvind, thank you so much for taking out the time. It is amazing to see you in person after using perplexity for so long. And congratulations for perplexity. Turning two this year. This month, in fact. That was like, two days ago, right? Yeah. Amazing.

So I want to start off this podcast with this tweet I just saw by Balaji, which is tech guys don't use Google search anymore. It's not just censored, it just sucks. Perplexity is better, and you can set it as default. And then he shows how to do it. How do you get Banaji to say that? And how did we get here in the first place? Yeah. So related to that, like, biology, first of all, a lot of people replied to him saying, oh, biology, you're just an investor. You're shilling your portfolio company's product or something. But he actually doesn't do that.

He's invested in us one year ago, almost before, and he never wrote any tweets, like, praising perplexity. So he's been using the product for a while. And, like, he would always criticize, "Oh, this thing is not better. You guys should improve the latency here, or you guys should make sure the results use sources from here," and things like that. We've been obviously trying to do our best, taking feedback into account and stuff like that. I think he really liked our integration with polymarket on the election odds. I think he really likes any ideas that are, like, you know, getting to the intersection of two things that are great. Ideas of their own. But the fusion is even more interesting.

So polymarket has its betting market, has the crypto elements to it. Obviously, biology is a big crypto believer, and he also likes perplexity because it can surface so many different sources simultaneously. Doesn't have the political pressure that Google has to succumb to responding in a certain way and things like that, that doesn't need to censor things. And he believes in all of that free access to information and stuff. So that's why I think he promoted us. And the second part of the question is, what did it take to get here? Obviously, it has been two years, long answer, short blood, sweat, tears, but specifically this focus on making the search better and better every day.

A lot of small things, like literally search. The good thing about search is like, everybody knows all the problems you need to solve, but nobody wants to put in years and years of work to solving them because it's a lot of work and boring work. Sometimes you just have to find bad sources on the Internet and ban them. Look at what kind of queries are susceptible to SEO, measure a lot of evals on how accurate your answers are, like all sorts of things. Understood.

But how would you define perplexity today? And how did you get started with the whole idea of it? Because you've worked at OpenAI, you've worked at Google, how did that conviction develop? Because you're going up against the giants, going up against the big tech. So how did you get here? Yeah, first of all, I think if you aim to get to building a search competitor, it's very hard to like even take the first step. I even say this, like, if you go and tell people you want to take on Google, nobody will give you even a penny. But if you say you're going to work on a healthcare search engine, even if that idea is so flawed, people are willing to give you like millions of dollars because they think, oh, that one has no competition. Like, Google's not going to work on like healthcare specifically.

So it's sort of this mentality. And we obviously never tried to compete with them, though. The first idea I did pitch to an investor was, oh, it'd be nice to disrupt Google. What if people could just search through their glasses and ask questions about what they see on the glass and stuff like that? But we were more narrowly focused on just searching over databases, searching over documents and things like that. But each thing as you accomplish it leads you to the next thing. It's almost like very organic. So when we finished searching over tables, and stuff.

We were like, okay, what if we could organize the Internet in the form of tables and search over it? And that became like, okay, what if we scrape Twitter and search over Twitter's dataset? And then that excited a lot of investors. And then that led us like, what if we can crawl the web and then search over it? Why do we restrict to one domain?

And then this unique intersection of a new tool, large language models and an existing tool like traditional search combined together created a very new, unique, differentiated experience. And we thought, okay, it's a good way to make people ask and get answers. No competition with Google or anything, it's just like a tool you can use. But slowly, over time we realize, okay, this can actually be bigger than we thought. There's no reason for people to use a traditional tenduling search engine if this sort of technology works very reliably.

And you always want to work on something where today's state is not perfect, but some years of work put into it will make it so good. You never want to work on something where everything's already ready and like, you just have to go and build a business because that means somebody else can also do it. And the first, and that somebody else, if they have a bigger business team, sales and marketing, they're going to beat you.

You want to have that advantages of like working on something that's on the rise. It is still improving. There's still a lot of grunt work and boring work to be done to make it really work. And while doing that, you build the rest of the stuff as a company, like business and marketing and growth and stuff like that. So that when you eventually land at a place of advantage, it's going to be hard to compete with you.

So this had all those properties, and finally this was completely by accident. We realized that only after serving the product, we realized, okay, why is Google not building this obvious question to ask everybody on the Internet is ready to remind you, oh, yeah, Google is just going to add this as a feature. The one reason they cannot do this is like they've built the whole foundation on driving traffic to others. Like, the more you click on a link, the more Google gets to claim credit for being the referral of the traffic. And the more it can claim that credit, the more it can charge the advertiser a higher fee for bidding on a certain keyword to get that click.

And the more it can charge the advertiser, the more margins it can see on its business. And that is what drives its stock price up. And that's what lets it hire a lot of talented people by giving them a lot of compensation in stock. So when you break the foundation of this, when you start giving rest, referral, the advertising bidding is not as lucrative anymore. Advertisers start paying less prices, your margins go down, and therefore, you're no longer as valuable, even though you have loads of cash and a lot of business otherwise.

So this is the classic your margin is my opportunity situation that, you know, this quote is from Jeff Bezos, and it's historically applied to most innovator dilemma situations. Yeah. So a lot of my audience, which is watching this, are people who are, like, 18 or 20. They're just getting into college. Obviously, what you've just said, there's a lot to unpack over there, but I want to go back to when you were 18.

So you did your electrical engineering degree from dual degree, which was b tech and m tech from IIT Madras. That's like a dream college, which everyone wants to get into in India. Even I participated in the IIT Jee advance, you know, rat race or the competition couldn't get in, but I got into bits. Bilani Goa campus. But take me back to when you were just getting into college.

What was your mindset like, and what all skills did you learn throughout the process of going through the IIT journey and then getting into the PhD that you did over here in UC Berkeley? Look, when I was 18, I wasn't like, you know, and that is ten years ago or eleven. Like, actually a little longer than that. Like, 2012, when I was. That's when I joined IIT. When I was 18, I don't think I had any maturity, right? Like, let me just be honest.

Like, the only reason we did IIT JEe was because I wanted to compete. Like, I was very good at math, physics, chemistry, and there needed to be a yardstick to measure, you know? And, like, you look at people around you and like, are you better than the rest? Look, we. A lot of people in India are driven by the competitive nature, and there are some benefits to that. There are, like, cons of that, too. And ge was a great exam, or all the best students went and studied there in Iit.

So I also want to be one among them. And so I entered with that sort of mentality inside IIT, and that mentality exists inside the IIT, too. The rat race is not just before getting in. It continues even after. It's like, who gets, like, ten out of ten or, like, nine out of 10.8 out of ten in the first semester? Who gets to get good internships? Right? Like, who gets the internship in, like, google, who gets the internship in the end of second year itself? Who gets the best internships? Or for that, in order to get the internships, you got to do all these projects beforehand within the campus. Who gets to be part of these clubs, do all these projects and like, so you're not just competing on the course grades, you're also, like, doing all sorts of other things.

So nobody even actually clearly understands what is it they really want to do. Nobody takes a step back and thinks, okay, why are we even working on this? What is the point? How does the world work? You don't take any time to understand. You're just constantly in a mad rush.

I was also very much like that. But my first year didn't go that great. I got really poor grades because the first semester I wanted to transition from electrical to computer science because that is so mean. Even I got little engineering. Everybody does, by the way. Everybody does.

Yeah. It's not just electrical. I think any department wants to get into computer science. All the cool people study there and you. And then I missed it by like 0.01 or something.

GPA points. Very upset. So the second semester was, didn't go great for me. I was kind of disillusioned with the whole thing. And that's when, like, a God time in the summer, I wasn't doing any internship.

At the end of first year, I actually took time back to, like, just read. I went to YouTube, watched a lot of lectures. I truly understood, like, a lot of fundamentals of electrical engineering. I said, why am I even in this department? Like, okay, I need to stay here for another four years, so I better learn what it is event, right? Why are we working?

I dropped out, but then I studied all these things and I got more and more excited. The third semester, I was truly interested in all the courses I took, and I didn't study for grades, and I just got, like, I just nailed ten out of ten. That got me the internship from end of second year. And from there on, like, I realized I'm not in the rat race anymore. Whatever I'm interested in, I can nail it.

And I'm not going to, like, compete with the computer science friends of mine or something like that. And I got excited about machine learning, uh, much before my. Everybody was like, yeah, whatever, this is not very important. Like, go get an internship at Google. Then you get a pre placement offer.

Then, you know, you, you can make like, you know, after four or five years, you'll have a million dollars. Everybody was in that mentality. And on the other hand, I got genuinely excited about like AI and machine learning and neural networks. Did like Hinton's like, like coursera class and drawings thing. Actually watch the Stanford lectures too. Not just the coursera thing, the CS 231 n. Yeah, exactly.

No, no, no, that's the karpathy thing. His thing is called like 221 or something like that. I forgot the exact course number. Like just. There are some things you just have to really put in the grunt work of watching 1 hour long videos instead of watching a newly released movie.

There's some of these things that you just make your own optimizations. And he chose to watch the lecture and not the movie. Yeah, because. Not. Because like I felt like I'm sacrificing.

I actually preferred this, you see? So that's when you know you like something. Yeah. And that, that got me to where like, you know, doing good research, publishing papers and undergrad, and then admission into Berkeley and things like that. And what made you go for PhD in computer science? Because a lot of people just do bachelor's and they do well. Yeah.

What was it teaching you more that you weren't able to unlock when you were doing the MTEC and B tech? Couple of things. Number one, I really wanted to come here to United States. And why was that? I think like around 2015 or something, when I got a chance to do an internship here through like a program called the Viterbi Indian Viterbi program, like Viterbi USC College of Engineering. They have a partnership there where they send like 30 students from India every year here.

And I interned here. I got a chance to see the labs here, like Los Angeles, how professors operated here, published papers, that pride of seeing your name on an actual peer reviewed publication. I think all that made me feel much better about myself and potential impact I can have. So that one thing, and the other thing is that I was not necessarily from a very well to do background financially. So I couldn't just pay for masters and come here, everybody, even masters, you got to at least be able to pay for one semester or one quarter and get a teaching assistantship after that.

So I couldn't do all that. So PhD seemed like a much better way. But the most important thing out of all is I actually love doing research. I think it was like love thinking about problems like trying experiments, that autonomy to try my own ideas was fantastic. So I thought this was the right career for me.

I would have been happy to finish PhD and just be a research scientist. Somewhere it would have been like life accomplished for me. Yeah. So I feel like even when I was getting into AI and machine learning in first year of my college, I got into the research bed because as you keep doing all the basic stuff, like the Titanic data set, and you keep moving ahead, then you get into the realm of building something of your own. And that's when I realized that I was spending a lot of time on research papers, which was great, but I wasn't focusing on other areas of life, like learning about other skills.

So I prioritized breadth over the depth of the, of the topic. What do you think is, is better? And what was more important at that point? Look, I think I don't have the counterfactual of what would have happened to me if I did more breadth first search than the first search. I got to read a very interesting blog post by this guy, Andre Karpathy, very popular.

Around the same time in undergrad, I was pondering over the exact same question too. Should I try to learn more about compilers, operating systems, you know, like NLP, computer vision, like all these sort of different things, or should I just obsessively focus on that one topic I was working on, reinforcement learning, and he had written a whole blog post of how PhD is all about depth. First search. Like, you have to like show the world that you can go so deep in one thing, and maybe you can try to go deep in like another thing.

But the most important thing is you don't memorize. You go deep into something, that skill transfers to something else you want to do too. And so that's why when you go deep into something new and learn so much about it, usually you are able to translate that success somewhere else.

I can give you another real life example. MrBeast, a youtuber. All of us respect him. He obsessed about minute details. What's the thumbnail? What is the length of content? When do you upload it? Like, you know, all sorts of things, recording quality. And he created a whole career out of it, but he's able to take all that.

And now he's basically running a business where he takes all the money he earns, creates videos that only he can create. But creating a video itself takes so much money. Like I gave everyone like a million dollars. Like, you know, let's see what they did. Or like I made like 100 people compete for a cash prize of like a million dollars. Like, you can only do all these things if you actually have money. But because those videos are so unique, it brings in like hundreds of millions of views, and you turn all the money you spend into more money. And, like, he tried this burgers thing also, right? So.

And that's when I realized, like, PhD is fine. Like, you can go deep into something and it's still fine. Like, it's not like you're unemployable. That skill set is actually worth. And Zuckerberg talks about this, too. Usually when he hired early at Facebook, he really prioritized hiring people who gathered a lot of expertise in one area that was so deep enough that even if that expertise was not valuable to Facebook, he knew that those guys could go learn something else pretty fast.

So having a lot of a players together in one place is what? Something like, they. Something where, like, you know, they were, like, pretty obsessed about it and, like. Like, went so far along than most people have.

You just mentioned that you came from not a well off family. So when you come from a, like, a middle class background, you have that middle class thinking and you have these restrictions in your mind that, okay, I cannot do that or I cannot access that. I cannot become that. Yeah, you have a lot of unknown unknowns. How did you let go of that? Because the whole point of me starting YouTube and making these videos is to help people get awareness of what the opportunities are available to you.

Yeah. And only when you know what is available to you, you can make a better choice in your life. So maybe when you came to United States and SF, what did you notice and how did it open up your mind? Yeah, I mean, I have a very, very, like, cliche answer to this. You might not even think I'm telling the truth, but this is the truth. I used to just watch videos of Elon Musk on YouTube. I didn't know much about Elon Musk before I came to the US.

And the moment I came here, like, people were talking about his rockets and things like that. And, like. So I just, like, you know, went on YouTube and typed Elon Musk. And, like, there's a bunch of videos of him. And I watched those videos of where, like, he put in all his own money and got watched, like, three rocket launches fail and, like, was basically going close to bankruptcy and still went and tried the fourth thing. And, like, his whole story of coming here as an immigrant and trying to find a job and then doing a company because nobody hired him and converting all that winnings into the next company, doing that repeatedly.

And that made me realize, okay, nothing can truly stop you if you don't want to be stopped. Nobody can really defeat you if you don't want to be defeated. You only die if you choose to die. I think that's why he said, like, I don't ever give up. I'll have to be dead or incapacitated or something. Yeah, that's the exact lines. Yeah.

You've really watched it a lot of times then. Yeah, exactly. So I think that gave me a lot of confidence. OK. Like, that has this not even about the well off thing. There's even another layer of problems that Indians go through, which is, oh, like, you know, we. We have all these visa problems, we have all these immigration issues. We're on hedge one b's.

Or, like, it takes a while to become a manager and apply for green card once you're there. Like, you have, like, kids and, like, you know, financial responsibilities. You have to pay mortgages. And then by the time you're, like, you know, late thirties, you lost all the drive and energy to, like, truly create something from scratch. But then, you know what? This is such a great thing. Most people don't even have this, so don't keep aiming for this.

You know, maybe your kid will do a company. This is like how a lot of Indians think, by the way, and nobody's willing to take the risk of, like, getting deported and, like, trying to start their own thing, like, living on, like, you know, one subway burger a day. You've done that? I've done that in an internship where, like, I was an intern in Chicago, like, for a summer beck, I was still in India. Like, this was an internship, and the stipend was so low, and I badly wanted to buy a MacBook. Like, really badly wanted one. And I was using, like, a dell laptop until then. And so in order to save money, I used to just walk from the campus to a subway pretty far, like, for half an hour.

And in Chicago, the prices were lower than, you know, bay area. So you get a nice meal. Like, you get that foot long. Half of it you eat there, half of it you take back and eat for the rest. And you can save a lot of money that way. And then at the end, I got a MacBook pro, and I went back and I started coding a lot on it.

So you can do anything you want. It's just like, a little bit of patience and sacrifice needed. Yeah, yeah. Very interesting. So, you know, one thing I observed when I went to a lot of these friends of mine who are working at big tech companies, they all have ideas, right?

They're all, like, developers. They are PM's, and they all have startup idea in their mind, and they're like, you know, I'm just waiting for the next round of my equity to be vested for me to get that. And then I'll start working on that. And that's how everyone is. Everyone has like a side project or side something that they're working on.

Yeah. And you probably also had that when you were working at these companies, OpenAI and Google. What made you make the switch? That, okay, this is over, I'm gonna quit and I'm going to start something new. How did you develop that, that whole conviction, that confidence that most people in SF don't have? So let me be very honest here that my situation was not exactly the same as people typically doing side projects. So I don't want to, like, claim a lot of credit for, oh, this guy was so brave and did it. I had, like, people were willing to fund me if I left.

I had a certain guarantee of that. Of course I put myself in that position, I'll give myself that credit. But it's not the same as someone who is not known or doesn't have the brand reputation of being farmer, like DeepMind or OpenAI or something. There are some benefits you get from having that. So I thought, okay, like, worst case, I'm gonna have a PhD. Also, I've written papers close to 10,000 citations. I'll get a job.

I'll get a professorship somewhere. Like, I'm fine, you know, sure, it'll suck. My friends would have made way more money. I feel like a loser internally, but I'll be fine. You know, I had that sort of mental cushion.

But in terms of directly answering your question, when do you truly take the bet on your, on yourself, like, on your side project? I think it's just a question of, like, just use the inversion technique. Is the suffering from continuing on your job less than the suffering that. Sorry. More than the suffering that you would endure if you did your own company? Which one is the suffering you're willing to endure? Like, the one where you own your own stuff? That's like a very high chance of failure, but it's still your thing and your baby, and you had at least the chance to die without the regret of not having tried? Or is it the suffering of constantly continuing with the same mentality of, like, working on something you don't enjoy or, like, not something you really want to be doing just because, like, it makes your, like, life less risky and at some point you just make your decision based on, like, hell, yeah.

Like, you know, I'm just gonna go through it doesn't matter. Maybe after, maybe I'm. For me, it was actually an even more direct question. I think I'm good enough to run projects and execute on a vision with people, but I'm not going to get that as that fast in an existing organization by climbing ladders or going through the process. I'm impatient. Maybe there's a miscalibration. Maybe I think I'm more qualified than what the world thinks I am.

The only way to know is to try. And I was ready for the reality check, okay, if my startup goes nowhere, nobody wants to work here, then the reality is I'm not good, and I'll accept that. But what if I was? What if there was a possibility that I might be good at this? I want to know that. I don't want to wait for like five years to know that. I want to know it today.

And I also had this other thing of like, this is my own feeling, by the way. I'm not going to project it on others. I could see in my own biological abilities that, you know, it was some amount of like, decline, which is like, you know, my memory power was like, way better. It's still really good, like, way above, like most people. But at one point was like so good that I felt like almost top of the world.

I can remember anything without making any effort. I can like, go three days and pull off all nighters. I can't do that anymore. I can eat any time of the day before. I can't do that anymore.

All this stuff I could see decline. So I knew that the more years I waited, the decline is going to be even more. And all these are unfair advantages you have, you want to capitalize on, because as a startup founder, you don't have any advantage. The system is rigged against you. 99% of the startups are doomed to fail. So any unfair advantage that you have, you want to capitalize on that.

And I would prioritize starting company less before you turn 30, or at least around that time. And why is that? Because of all the unfair advantages that you have when you're young. Elon also talks about this, like, you don't have, like, mortgages, you don't have like, a family to support yet. Depends on your state, honestly. But assuming all this is true for most people, you don't have all these other distractions. Look, it's like mind is full of distractions most of the time.

Startup is all about focus. So these are not in alignment with each other. So the sooner you get to de risk an idea that is occupying your mind forever, the better, because then at least you know that you gave the idea a fair shot and it failed because the idea was wrong at that time. But you at least have the satisfaction of having tried. So I prioritize all these things, and that's why I push for starting a company as soon as possible.

Yeah, very cool. So you started perplexity in August of 2022. That's also when OpenAI came out with Chad GPT a little later. And what was the state of your mind at that point? It came instantly, GPT 3.5.

Everyone started using. It had that exponential curve. You were launching publicity as well. What was your mindset and what were you pitching to VC's? As you said, betting against Google.

No. Was going to fund you. And so what was, like your pitch, and how did you convince people to believe in your idea? Yeah, so first of all, chat GPT did not exist at the time. Even the product, leave the product, the model 3.5 turbo, that was a model that powered chat GPT that did not exist at the time.

So the pitch was mostly centered around another breakthrough that had happened around the time, which is OpenAI's Codex models that were powering GitHub's copilot, which in my opinion is the most. The first real sign that AI has transitioned from research to delivering real product impact being used by millions of people, and also had a monetization impact because on the day, like, they turned on like the monetization switch for Copilot where they said, okay, people could pay $10 a month. I think they were already, like, having 100,000 people sign up to pay for it on the first day itself. That's huge. Most products take so many months to get there, and then that's now millions of people.

If you looked at the earnings call of Microsoft this quarter, GitHub Gopilot makes like $300 million in revenue. So they've actually 30 x that number since 2022. So that's in two years, and there's a lot more potential. So I knew that this was like a pivotal moment for building AI products. And the idea we pitched was like, imagine we did something like codecs for people analyzing a lot of data over tables, a bunch of relational databases and things like that.

That was how it started. Of course, the search over glass was pitched, but it was not taken seriously. So we tried to say, we will search over spreadsheets and databases and things like that. We wanted to work on other people's internal data. It was a clean idea, but nobody wanted to give us their data.

They were like, why should we trust you guys? We don't even know who you are. There's no brand, nothing. So we were like, okay, if you don't give us your data, we have to build something. So we just use external data.

And I really like Twitter. I'm a big fan of the Twitter social network. So we just took Twitter and organized it into a form of bunch of relational databases, and then we powered text to SQL search over that, and that was a very unique experience. You could, for the first time, ask questions like, what are the tweets that both bezos and Musk liked, or tweets of Bezos that Musk has replied to and would give you that funny silver medal emoji tweet. And we would show these demos, some investors, and they would really like it, and they would search over their own tweets and stuff.

And I think that sort of got us a unique, differentiated experience of search that you can build only with LLMs and you couldn't build before. And then we expanded our horizons of, like, okay, what would it look like if we did it for LinkedIn or, like, GitHub or, like, organize all of them and allow people to search over all of them? And then we're like, okay, hell yeah, let's just do it over the whole web. And that became the perplexity product. It was not really inspired by chat, GBT, or anything like that, but the courage to launch it definitely came in stratub, relaunch.

We're like, okay, this is cool. People are open to the idea of asking questions and getting answers. So we were like, why not? We have a different take on this, where you can ask a question here, get an answer, but you at least have the sources to verify if the answer is right or not. And it's real time.

There's no knowledge cut off, and it's anonymous. There's no need to sign it. You just come here and start typing and ask. I think that people like that idea a lot, and ever since then, we've been growing our usage. It surprised me.

I think what happened is we launched on December 7, 2022. I thought, okay, after a week or two, nobody's going to care, and we just use this demo to get some enterprise contracts. Did you first start with the Chrome extension? Because that's how I discovered it. We just started with direct. That was all inspired by my obsession about Google.

Like I said, the landing page should just be the product. There should be no onboarding flow, nothing. Should just be a search bar, blank page type, get an answer with sources. But I thought, nobody's going to care. And then after two weeks, the vacation started.

So obviously you wouldn't expect anyone to use a two week old product during a vacation, but people were actually using it more and more. The usage kept going up. People were sharing screenshots of, you know, typing their own name and like having this AI summary pop up about them. It was like very novel. And that led to more and more viral growth.

And I thought, okay, that's just like a short term virality. It's going to drop. But it's, it kept sustaining the growth, kept on sustaining over the entire month of January. So organic was like, world of mouth was the biggest organic traction channel. Yeah, I only had like, you know, myself, I had like 4000 followers or something, and the company account had like, you know, 5000 or something like that was pretty minimal, but it kept on growing that we felt like, okay, there's something here, we don't need to pivot, let's keep going.

And then February kept on going. So we said, okay, let's actually raise some capital. And at that point, what was like your North Star metric that you were tracking for number of queries a day? It's always been our north Star metric because any other thing doesn't correlate with like increasing the impact of the product and the value of the product, because the more queries people ask, the more like, like expansion of the index and like the coverage of queries that we can handle, infrastructure and like the quality of the AI models, like more data, we collect flywheels. So that is the one query that matters the most to us.

And that kept on going up. And end of February we were like, okay, this is actually somewhat real, so let's raise some capital. Interestingly, it happened around the same time Bing chat was being launched and Google bar was being announced. So always, ever any fundraising we've done, there's always been so many last minute problems to deal with where convincing people to still invest despite competition. But I've gotten so used to it now that when things go normal, that's when I get even more edgy, because every week or the other, there's some problem. And what have you observed about the VC business? What do they look for?

And when things like this come up, Bing chat is coming up, Googlebot is coming up. What do they think? Why did they choose to invest after all of this competition? Look, a lot of them for like every round we always had more people who didn't want to invest, but that kept producing. What do they look for? Look, I think they're making decisions with limited information. They don't have the vision that you have.

Whatever is in your head, it's your job to communicate that in the best way possible. Some people use Dex for that. I also use Dex, but I. What is Dexd? Like slide decks, like pitch decks.

But I think the better you communicate and pitch, the better your chances are. And I would say that the way we position the a round pitch was like, look, we don't know if this is really going to be a breakout consumer hit company. All we know is that despite us not really trying to grow, it's growing. Look at the trend. The growth rates are clearly there. Yes, the cost per query is very high.

That's why we want to raise money. If it was not high, we wouldn't need that much money. So we are betting on the fact that this technology is on an increasing curve. The models are going to keep improving. The cost per query is going to go down tremendously.

And since the day we said it's gone down by 100 x, and so with the same amount of money, you'll be able to serve a lot more users over time. So you're betting on something that's really going to work. And when the cost per query goes down and the models are going to get more and more capable in smaller sizes, then we are heading towards a world where whatever is like one in the ten queries hallucinate. Today would be one in 100 in a year, one in 1000 in two years, one in 10,003 years, it's going to increase its quality exponentially. So you're like, you cannot understand that world today. It doesn't exist yet.

So you're betting on the potential of getting there. And venture is all about outsized returns. It's not about like, oh, 70% chance of this working, it's about like there's 10% chance of this working. But that if it works, I get like 100 x returns, 90% chance it doesn't work. Okay, but that 10%, I get like 100 x. And in venture, you can only make your name by investing in something that ten x, 100 x, 1000 x.

Anything else doesn't matter because you could get that by putting money into S and P 500, right? You gotta beat the S and P 500. So that's why, like, LP's put money into VC funds. So that is like that. That characteristic was always there in perplexity. This business might not work, but if it works, it's really going to work.

And every time we peeled off the risk layers, when we went to the b round, we were serving at least a million or two daily queries. We had way more traction. We had actually real revenue. We had monetized users through subscription plans. We came forward with a clear growth plan and the way we monetize and things like that, and we raised money for the branch.

And by the time we got to the next round, we had like a lot more users, a lot more revenue. And then similarly. Right. So I think as the risk layers get. Keep getting peeled off.

Think about this like an onion and like different layers. Like, you know, different risk layers. And the later stage you are, the deeper you are into the onion. Yeah. So when you say that generative AI is working, when you were pitching them to them, when you say work, what exactly does that mean?

Because today morning I was checking, like, there was like a number which said that Microsoft in the last one year has lost $5 billion net in this whole thing of like running these genitive models, the Bing chat and other things. Copilot is there. And one thing that we observed in all of these genetic AI company is that they bleed a lot of money and the profitability, the revenue isn't that much right now. Yeah. So what do you think about that?

How does that happen? And what have you thought in terms of the revenue for perplexity in the future? Yeah. First of all, this is not very unique to AI. It's existed in every form of business in the past.

Google never turned in a profit for like, until 2002 or something like that. So since 1997 or eight, it was founded for like five or six years continuously. It was losing money. Amazon is very famous for like, losing more and more money every year. They pay no taxes. Exactly.

Yeah. Until a point where it really started making so much money. So Uber, or in general delivery apps have been famous for being cash burning businesses, and only recently Uber started becoming profitable. So this characteristic, if people were judging based on how you said it, we shouldn't have that many delivery apps. We shouldn't have that.

May like e commerce apps, right? But we did. So just like that. I think in AI too, like, all these businesses will eventually have to figure out how to make money. There's no running away from that.

I think there are two classes of AI companies. Like, one is those that train their own foundation models. That's incredibly expensive. One run could cost you a billion dollars. Something could go wrong.

You spend another billion dollars to get it right and you might want to train even bigger models that might cost you $10 billion. All that hasn't happened yet, but potentially can happen. So any company like OpenAI or Anthropoc, that's like really in the business of like scaling these models by design, has to burn more money than it can make. And that's okay, because whatever model it ends up producing might eventually recover all the costs. By being such an amazing API that you can plug into and build applications.

And for the application layer, companies that are burning money by paying for the API calls, you eventually have to figure out a way to deliver value to the end user that makes more money for you than the amount you pay for the API calls. And that's a bet you're making on the API calls. Costs going down and the value you provide, what do you charge for it is so both have to simultaneously happen, and then that's how you begin to see the margins. So this is generally how things have worked. Uber figured out a way to gain market share by pricing all the rights so low.

And then now just getting from the street to another street costs you so much money. That's how they get back to money or other businesses, other bundles. So you got to be creative in many aspects. Our belief is that as a company we are not, since we're not trading foundation models, been pretty cost efficient, we don't actually burn that much money. And the open source revolution and models getting smaller and more capable, cost of the API going down continuously for the last year or so, and models getting more and more reliable, has made sure that we can continue to serve as many users, spend more on getting the answer as right as possible, and monetize the users also simultaneously through subscription plans.

And see some at least like at a product operations level, we try to be profitable. At a business operations level, like actually paying people and things like that, it's pretty difficult to see profits. So in terms of AI search, which you've built, there are two ways. There's either the subscription plan, which you talked about, or there's the advertising model. How do you decide which is better and in what terms? Yeah, look, I think in terms of margins, advertising has always been a higher margin than subscriptions.

The average revenue per user for Google or Facebook is way higher than chat GPT for a reason. In chat GPT, you charge $20 per month per user, who's paying? And most of the users don't pay like less than 10% of the users pay. So essentially, your average revenue per user is much lower than $20 per month. It's more like single digit dollars per month.

The average revenue per user on Google or meta is like more than $200 a month. So that's how these businesses spend so much money. So you have to figure out something that works. Maybe subscriptions could work. Maybe you could charge up, like, maybe you could imagine a world where like there's a model in GPT five or cloud four, and you could charge like hundred dollars a month instead of $20 a month. But then the access to that technology is also like limited.

It'll become more like a Bloomberg terminal style business where a few people with affordability can pay a lot of money, but most people will not be able to use it. And see, it really depends on what kind of company, what kind of service you want to deliver to the user. It was a really solid research assistant, almost like an employee that works with you all the time, then subscription is the right model. But then not everybody can afford it. But if it's like, oh, high quality answer that just gives you accurate answers to anything you ask and helps you correct grammar and things like that, then you do have to figure out a different model than subscriptions because all that service will be basically available for free for most people.

And you still want to stick to subscriptions and not advertising? No, we are going to do advertisements too. How would that look like in terms of the user workflow? Yeah, it's still not like nailed down yet. It's still like we're in early stages. But I do think there will be one way to do it without compromising on accuracy of the answer, transparency to the user. All those things not go on the same path as Google.

Yeah. So I just got this news that Google just lost the lawsuit and the judges have ruled that they are actually a monopoly in the search space. And now that's something that they've been fighting for a long time, saying that they have so many other businesses and they're not a monopoly in most of them that they operate in. So what do you think about that? And the search space has been dominated by Google, so for you to take up a majority of that space, how do you think about that in the future? I mean, I think in general, I'm pro competition, right? Like, we are a very small player here, and it's important that people get to pick whatever choice works best for them.

And I don't have the specifics of the ruling and things like that to comment deeply there. But in general, I think more competition is only good because that pushes everybody to keep the service as high quality as possible and consumers win. At the end of the day, you guys all winden, there's competition in search. So when you hire people over here at perplexity, what exactly do you look for and what skills mindset do you look for and how did you like learn? Okay, that this is what I want to look for when you're hiring people. We are a startup, so the kind of people who really thrive here are those who can work with ambiguity and the speed velocity that we need to ship at this crazy pace every single week.

And ownership, taking ownership of the work that they do and also ensuring quality is high ability to learn new things really quickly. Also because you got to be multi faceted, you're not just meant to be an expert at one thing. You should be able to multitask and wear many kind of gloves, do different kind of tasks. So we look for some track record of like, you know, these things in the past and we look for young and hungry people, but experienced people are also like working here. So it's just a prediction we make basically.

Okay, would this person like thrive in a fast moving startup and be able to do a lot of things simultaneously and ensure high quality is delivered on record time? Like it's not easy to hire such people, but like we try our best to optimize for that. How is you've been through the interview process of big companies as well. So what have you observed in terms of what they look for and what you're looking for as a startup? It's very different. I think Google has this philosophy of not having role specific interviews and just hiring for the person and then figuring out the role for them later.

That's a philosophy. I don't think that scales well. I think it's a good philosophy in the beginning because it gets you these generalists who can do a lot of things, but I don't think it serves them well. Now, figuring out projects for people is always harder. Sometimes you do want the domain experts to join you and help you build something faster.

I think OpenAI's hiring was more focused on research, at least when I was there. That might attract a bunch of people who are more into really deeply thinking about things and training models and stuff. I think product engineering recruiting is pretty different, so I haven't tried to copy anything from there. One thing I at least try to do is try to interview as many people as possible myself and check for culture fit. Can they really thrive here? Are they really interested in making impact or can they move fast. Are they excited about moving fast?

Is the intensity going to be there? So I check for past track record of doing these things and degrees. I tend not to focus on that much. There you go. People who are, look, you can never win there.

Basically you can say, oh, like, yeah, iit. Which IIT? Oh, Kanpur or Madras or Bombay. Oh, then like someone's from Stanford or MIT or Berkeley, then, okay, like, did they work at Google before? They work at like, Facebook before? You can keep on going and adding more and more, like, criteria for, like, elitism. Like, at the end of the day, what matters is can you get the job done really well and communicate really well here?

And that's the most important thing. If you cannot do that, then it doesn't matter where you study. And if you can do that again, it doesn't matter where you study. So we have like, processes in place to, like, filter candidates, obviously. Look, if you're not from a reputed college, it's harder for you to stand out.

No, I'm not like, you know, denying that because everybody has only limited time to go through all the applications. So you have to figure out a way to stand up somehow get. Get the founders attention or in some, some other manner or do something exceptional that, you know, gets you on the same level visibility. But after you're in the process, it's all the same process. Like, you know, you answer the same questions. Sorry, applied in any different way.

Yeah. You just mentioned about OpenAI. Sam Altman launched search GPT with a tweet and everyone in the comments were just talking about perplexity. Yeah. What do you think about search GPT in general? And OpenAI is now again in this whole competition and recently also raised $250 million at, what, 2.5 to 3 billion valuation. So what do you think of it as a competitor? Yeah, look, I think, like, one thing is good about it, which is until they did this, everybody was like, why do I need perplexity when there's chat GPT? But clearly there is a need for a separate, differentiated UI that always sources content from the web and gives answers grounded in web results.

So the hypothesis of perplexity has been validated by a reputed company like OpenAI. And in addition to that, I think competition is good. Like I said, Google also launched AI overviews, which is similar to search GPT. So I think in general, competition is good. It makes us move faster and us moving faster make them move faster and, like, products will improve at crazier pace.

Of course it's harder on, like, at a personal level, because you got to, like, work intensely for, you know, even longer period of time to fend off competition. But if it. If something was meant to be easy, it's not meant to be great, right? If it was actually easy, yeah. So the other thing I would say is, like, big companies always try to do these things, like, Google launch Google because they thought they could compete with Facebook, or Microsoft tried to do bing against Google.

Or, like, historically, this has always been the case that some other big company has always tried to launch a product that's featuretained or like, separate product that is like full time what another smaller company does. And smaller companies have tremendous advantage in focus. Like, Bruce Lee has his code. Like, I don't fear the man who practices 10,000 kicks once, but I fear the man who practices one kick 10,000 times. So the second is what a startup that this one thing is.

The first is what, like a bigger startup or like a bigger company that does a lot of different things, like ChantGpt, search, APT, Sora, Dali, GPT, five. You know, safety. Do you think is that that is bad for OpenAI to be launching everything, like, into every. It depends on their goals. Like, if their goal is like, you know, I want to be the next Google, then you do have to show your caliber and doing a lot of things, because Google does do a lot of things.

There's like Gmail, maps, Chrome, Android, search, YouTube devices, Pixel G suite. So there's a lot of things Google does, and you do have to show your caliber doing a lot of things. That's an interesting take on competition, because what I have been reading when I first started learning about startups from zero to one by Peter Thiel, is that competition is for losers. You never want to be in that space. That that is a different take.

Yeah, I mean, I think competition will always be there once you've identified something that's valuable. I think Peter Thiel is saying that because it's a provocative way of saying things. But I think competition will begin to exist once you have shown traction. People with more resources than you will try to compete with you. It's happened with Facebook, it's happened with Amazon.

Walmart tried to compete with them. Barnes and Noble try to compete with them. So it's always going to be the case that we'll have competition. When you first came to SF, what do you think? What makes you different from other countries, other places like Bangalore? I would say like this.

It's not particularly about SF in general Bay Area. I would say the people are more interested in hearing a and judging your idea than knowing who you are. Of course, knowing who you are helps. Like, the same thing is said by Mark Andreessen or Peter Thiel. Even if they said such a trivial thing, you're like, oh, wow, that's amazing.

Like, someone else said it, you're like, yeah, yeah, whatever. But still, like, if you can make a point that is interesting and non trivial, very different people are interested in listening to you and, like, hearing you out. And there's always the mentality of, like, what if this was actually correct? What if this actually worked? I think, by the way, I haven't lived in India for years, so I don't know if this is still true. Right. I'm prefacing this because I don't want angry people commenting.

Oh, like, look at this guy. But at least when I was there, it was always like, yeah, that's not gonna work. Yeah. You know, like, like, people are willing to bet on outlandish ideas here. Outlandish ideas.

And also just like, willing to judge. And I like someone's, like, opinion based on the merit of the idea alone and not who they are. I think decoupling the person and the idea is something that exists here more, I would say, yeah, of course, there's a lot more capital here and things like that. The benefits from capital. In the last twelve years, you went from an 18 year old who was just getting into an IIT to this, having a big company over here in GeneTv, in US.

What skills do you think has helped you get here that a 20 year old can take? Relentlessness. I would say that's number one skill. I even wanted to buy that domain, relentless.com. that is. Know who owns it. That is chef.

Yeah. How was, how was raising from him? Did you get to meet him? Yeah. Yeah, he's incredible. Still, like one of the smartest people, you know, like, in terms of strategically thinking about things and, yeah, relentlessness is the most, like, I would bet on it.

Like, if you take a 20 year old, you know, I've seen this repeatedly, both in my own, like, generation as well as, like, next generations. The guys who are like girls or anyone, anybody who's. It's not necessarily the smartest person that makes the most progress after a decade, it's the most relentless person. Of course, you need to have some talent, right? You can't be, like, being relentless in something. You're not having some ability to execute more than other people. But let's assume that's there, then.

I think the relentlessness will truly be the differentiating factor after a decade of work. Yeah. Two last questions. One is about how should a generative AI company think about value? Because you just tweeted that, you know, some of the biggest businesses started as rappers. When you were talking about Starbucks, Pete's coffee, they were taking beans from them.

So a lot of generative AI companies are starting from built on models that other people have built out. Could be open source, could be closed source. But how should they think about building value then? Well, it's a difficult question. I don't even think we have done it entirely yet.

Right. So it's not like I would have the magic formula yet, but I would just say that the more and more orthogonal value of building on top of the LLM, the better. In our case, it's the fact that actually in building, crawling and indexing, it's like a lot of work. We haven't even finished it and it's going to take months or years to truly finish it and it's going to take forever to keep it updated and incrementally improving. So we are working on all of that.

We know all the hard work needed to do that. And we also do a lot of hard work on orchestration, what to do for each query, how to respond to each query, what is the UI, what's the UX going to look like? All the stuff that's so custom and quality improvements and stuff that speed making the cost lower and lower, better models, cheaper models, all this is stuff that you're doing outside of training the better model and things like that.

So I think that's going to add a lot of value. Also marketing how you communicate your product, what the product means to the user, what kind of value, like how their habits change over time and more of their day to day workflows happen through your product. These are ways in which you can build your own value. And I think most people are like so upset and when they use the word rapper, it feels like so derogatory, superficial, like you feel like insulter or something. But look, everyone's a rapper.

Like, you know, OpenAI is Azure Nvidia rapper, or venture capital is a rapper over people who actually have money that are giving the VC's to invest. Or, you know, like, I just think, like at the end when Nvidia, you can joke, it's like a TSMC wrapper because the fabrication doesn't happen inside Nvidia. Yeah. So Netflix runs on AWS, so it's an AWS rapper. So there are like rappers at all levels.

It's just, you don't realize it because they've given you so much value that you don't care. And in the early stages of companies life, the value is still being created. So you tend to focus on, like, places where you think like, oh, the business can potentially fail, but zoom out and think like, okay, this thing really works. And this builds.

Sticky habits and retention and daily workflows are going to be based on this and nobody's going to be discussing which model it's using or which servers it's renting and stuff. Nobody cares, actually. Brilliant. The last question, Arvind, is if you were in shoes of Ishaan Sharma today, what would you ask Arvind Srinivas? Well, that's an interesting question.

Maybe I would just ask, like, what is like, you know, best way to deal with failure? I think. I think that's the thing that I would have. Whether it was you or, like, anybody else, I would say a lot of people overreact to it. They tend to take it seriously, tend to judge a lot, like, make like short term decisions and stuff.

I think the mentality of, like, always looking for the long term is important. Right. You're. I mean, do you follow cricket? I do, yes. So, I mean, like, my favorite example in cricket is Ms. Dhoni. I think he is a truly remarkable captain in the sense. I got to meet him actually in Bangalore.

Oh, that's awesome. Amazing. So he's like, he's, you know, there was these Test series abroad where he used to lose like 4050 and come back and be very humiliating. And he would sit through these horrible press conferences and answer questions. But think about it. Like, all the people that he brought on the team then are the ones who eventually ended up going and winning abroad.

Like, all these guys, Kohli, Roy Sharma, who are like legends now of their own right, are people that Dhoni took a bet on and didn't drop them from the teams and played the long game. And I think, like, that's a lesson to learn there that don't overreact to small failures. Like, always focus on the long term game. And I think if you are a young person and you want to think about building your own thing, one of the best leadership lessons is like, looking at Dhoni's career and seeing what all he did. Yeah.

Incredible. Thank you so much, Arvind. Taking out the time, this was epic. Very insightful. I wish you all the best for perplexity. Thank you.

Artificial Intelligence, Technology, Innovation, Entrepreneurship, Startups, Relentlessness