0:00
[MUSIC] Aravind, welcome to Stanford.
Search this video
0:06
>> Thank you for having me.
0:09
I'm from Berkeley, so hopefully you don't mind >> [LAUGH] >> At least wanted to represent with the blue.
0:15
>> [LAUGH] >> But it's great to be here, thank you.
0:18
>> We're happy to have you here.
0:20
Now many of us in the audience are active perplexity users. >> Thank you.
0:25
>> Especially with free perplexity Pro for all Stanford students.
0:31
>> [APPLAUSE] >> So we couldn't be more excited to have you here. >> Thank you.
0:37
>> Now to get started figured I would turn to perplexity to help me craft my questions.
0:43
So let's take a look at what it said.
0:45
So to get started, I figured I would put this prompt into perplexity, and I asked, I'm interviewing Aravind for a one hour interview at Stanford with an audience of business school students.
0:58
What questions should I ask him? [LAUGH] now.
1:01
>> Hopefully, they're not too difficult.
1:03
>> [LAUGH] >> You probably missed that in the prompt.
1:07
>> You you might have yourself to blame for that one.
1:10
[LAUGH] so perplexity did give me a very detailed response.
1:14
But in summary, it suggested that we talk about, first, your personal backstory second, the early days at perplexity.
1:23
Third, the company today, and fourth, various leadership lessons. How does that sound? >> Sounds good?
1:32
>> [LAUGH] I thought that sounded like a pretty good outline, but I figured we could get just a little more personal.
1:40
So, I thought I would test out my follow-up prompting skills, and I asked perplexity.
1:46
What is something about Aravind that the audience may not know?
1:50
What is the funniest thing about him?
1:55
What are some questions that I can ask him to inject humor into this conversation?
1:59
>> [LAUGH] >> Very direct [LAUGH] what are some rapid-fire questions I can ask you at the end of the interview, so stay tuned for that.
2:07
And finally, is perplexity ever wrong? >> More than you think.
2:15
>> [LAUGH] >> So this generated some more interesting insights than my initial prompt, including, here's the very detailed list, but I'll just hit on a few of them.
2:27
First, I learned that you love cricket.
2:30
Second, I discovered that you actually taught yourself to program after missing getting into computer science at IT address by only 0. 01 points.
2:40
And lastly, I learned about your connection to Sundar Pichai, the CEO of Google, who grew up in the same hometown in Chennai, India.
2:51
Now, perplexity actually suggested that I watch this video of you and Sundar from Chennai.
2:56
So I figured that this might be a very good place for us to start.
3:00
I'll spare you the video.
3:00
But Aravind, is there something in the water in Chennai that's led to so many successful tech entrepreneurs?
3:09
>> Well, I think, it's interesting more than, I'm sure lot of people in India there's so many cities which are producing great people.
3:20
And so one thing that I would say is very common is this sort of culture of really trying to excel and do your best at what you meant to do.
3:33
And the sort of like real emphasis on education is very much present in Chennai and many other cities of course, but in at least in my circles, people valued being scholarly and well-read even more than being rich.
3:52
You kind of got respected a lot for it and I think that translates to like going above and beyond.
3:58
And not just read what's meant for your exam to do well in your exam, but really try to go deep into what you're trying to learn.
4:07
And I think that's sort of common in many people who come from there and excel in Silicon Valley or other parts of the US.
4:15
I feel that it's a very common trait.
4:18
And obviously, this cricket thing is there.
4:21
And one thing China is known for is they call the cricket nerds they really obsess about all the statistics back before google or crick info, or all these sites existed.
4:37
We could still recite all the stats of every player and obsess about things like what's the run rate, what's the average.
4:46
And you learn basic statistics before even it start formally to you that it's not just important to score a lot of runs in one game, but you've got to be pretty consistent.
4:55
And so I think that's kind of common, I would say, for people from Chennai.
5:00
>> So knowledge over wealth and attend cricket games. >> Yeah.
5:03
>> So, it'll take away from that one.
5:04
Now you left India to pursue your PhD in computer science at UC Berkeley, not Stanford, unfortunately.
5:12
>> [LAUGH] >> I didn't get in here that's the truth.
5:14
>> [LAUGH] >> Well, we have you here today.
5:18
We couldn't be back here.
5:19
But how have your academic roots shaped your approach to building Perplexity?
5:24
>> Yeah, it's actually pretty core.
5:28
Perplexity started off with citations right after every answer.
5:33
This is obviously like real.
5:34
I'm not just making it up.
5:37
When I first went to Berkeley and I thought PhD is an amazing thing.
5:40
You watch movies, Stephen Hawking, where you just knock on your advisor's door, this is the idea for my thesis.
5:47
I really romanticize how that's how it's going to be.
5:49
But it's not quite as, like easy as that.
5:54
You have to, kind of earn your way to coming up with your first score original idea.
6:00
So institutions like Berkeley or Stanford, they're kind of like amazing because they give the new student sort of like a framework to get there instead of just leaving them completely unsupervised.
6:16
And so when I first went to the lab, I was asked to kind of help a senior student by work on their idea and try to actually make it really work, and write a paper.
6:26
You learn the process, and then you kind of get from there.
6:30
So after I wrote a paper or two, they told me this concept of citations.
6:38
And so that matters is not necessarily you getting a paper accepted.
6:42
It's that other people have to cite and build on it.
6:45
That's how you build your academic currency in the community.
6:49
And I was okay, that's cool but my paper seems too complicated.
6:54
So there's always this trade-off between writing very complicated like very creative ideas that might get the reviewers to accept your paper, but for someone else to build on it like, if it's too complicated nobody cares.
7:10
So I think like you want that sweet spot of like very simple ideas that will get cited, but also will get accepted in a conference.
7:18
I had to learn the Hard way.
7:20
And once I learned that, I kind of obsessed about citations.
7:22
And, I also learned that was the core inspiration for the Google search engine of how academic citation graph to like marrying that idea to the web hyperlinks.
7:34
So that sort of also came into perplexity because we asked the question, okay, what if an AI always responded like an academic?
7:44
When you write a paper, every sentence you write in an academic paper needs to have a corresponding citation to it, or else it's kind of coming across like an opinion.
7:53
It needs to be something like a source of truth from some other peer-reviewed paper in the past or some experimental To result in your own paper.
8:01
And we thought, okay, what if we do that for AI, where every sentence in the answer, needs to come from some source in the web that has some amount of domain authority or trust score.
8:13
And if we can bake that into the prompt, it's like inbuilt, then that sort of creates a very unique product experience, and that's actually how my academic roots help me in perplexity.
8:26
>> Now I know perplexity is academic focus feature has been a key driver of adoption here at Stanford, and lets you cite only academic journals. >> Yeah.
8:33
>> In your research, so very helpful for all of us who are studying today.
8:37
Now at perplexity, you're very focused that's on improving access to information and you're building the world's first answer engine.
8:46
Why is democratizing access to knowledge so important to you.
8:50
>> Because I love using it myself.
8:52
Like I said right, I came from a culture where being knowledgeable was very valued.
8:59
There is even this quote from Charlie Munger where the best thing you can do for another human being is to help them learn and know more.
9:07
And it's almost like a moral duty for all of us to seek wisdom and become perpetual learning machines, because nothing else can help us keep upgrading ourselves.
9:18
You can probably focus on wealth and your net worth.
9:22
You can try to use that as a metric for your progress in life, but at some point, like it taps out, it stops motivating you if you reach a certain threshold.
9:31
On the other hand, there is no end to knowledge.
9:35
That's why in perplexity, the tagline is where knowledge begins, because there's actually truly no end to knowledge and you can only keep getting better.
9:45
So there's one metric ultimately that we can converge to view ourselves as making progress on that, it's just understanding the world better.
9:54
So if that is so core to the human nature, it's essential all of us have access to the tools that help us get there in the most accessible way.
10:05
And we are trying to do our best to do that and obviously the more premium services are behind paywalls.
10:12
But as these AI models are getting cheaper and smarter, more efficient, more distilled into smaller versions.
10:21
It's going to be possible to create a version that's just widely accessible for all and helps them basically ask any question they wanted and get an instant answer >> Where knowledge begins. >> Yeah, exactly. >> I love that.
10:34
Now I want to go back to the genesis of perplexity.
10:37
To tackle such a big problem, you needed a great team.
10:41
So what qualities did you look for and how did you build your initial founding team?
10:48
>> Yeah, I was lucky enough to know one of my co-founders Dennis during my PhD days.
10:54
This is also where the academic background helps you a lot, you learn people who are very motivated and deep thinkers.
11:04
And so we wrote the same paper, literally the same idea, a day apart and so that's how we got to know each other.
11:10
And he spent some time as a visiting student in my lab.
11:14
And we used to brainstorm ideas on what we could do together, but nothing really came out of it.
11:20
One quality, I would say, you look for your founding team is, obviously, people with complimentary skills.
11:28
You don't want to be as good as them in what they excel at.
11:32
Ideally, they should be a lot better.
11:34
And also, you don't want to step on their toes when they do that.
11:39
And in our case, it's kind of Dennis and Johnny were my core founding team.
11:46
And I would say Johnny was world number one at competitive programming, he represented the United States at the IOI.
11:54
And those who are aware of competitive programming, there used to be this guy called Turis.
12:01
And only one guy has beaten him in the IOI ever, and that was Johnny.
12:07
So that sort of prodigy who can not just write amazing code, but has the problem solving skills to instantly solve hard problems quickly.
12:17
And the sort of AI depth and background and software engineering background that Dennis had combined together allowed me to take bold risks.
12:28
And try to set up the sort of very overarching mission of trying to build a completely new search experience.
12:36
Otherwise it's impossible forget about trying to do this, right?
12:40
So and then over time, you try to hire more and more people who can bring in new skills.
12:47
Obviously, none of the three of us have front end programming skills, so we hired someone really good at the full stack experience.
12:54
And we hired someone really good at writing CUDA kernels, it keeps on becoming an incremental additive as well as multiplicative force.
13:03
And I think that that sort of effect is necessary when there's a lot of people, it's not I think there's a mental model of vector sum of all the people.
13:16
But I actually think if you want to create even truly great company, there needs to be some kind of this Lollapalooza effect, there's factors that become multiplicative in nature.
13:30
And one example I can give is our design team was very well known too, but none of the founding team had that design quality.
13:38
So we went and hired someone specifically for that, where that person actually wanted to build a product like this, but he did not have the AI background or depth.
13:49
He used to work at Cora, where humans came and answered questions.
13:53
So when we gave him the platform to use this AI to be able to answer questions like his imagination skills came in and created something that's completely like multiplicative.
14:03
And that's kind of how I generally try to scale up the team.
14:06
>> Find the people who multiplicate you. >> Yeah, exactly.
14:09
>> Yeah, so less than a year later you're in the middle of raising your series a financing round.
14:14
When you find out that one of your key competitors Open AI has just launched their own search competitor.
14:21
So when you heard the news, how did you respond and what gave you the confidence that there was still room for perplexity?
14:27
>> So you're asking me about my Series A.
14:29
>> The Series A and the news that open AI has a search competitor.
14:32
>> Yeah, actually just this correction there, at that time open AI wasn't launching search.
14:38
It was actually Microsoft was going to launch Bing and this is almost like a Silicon Valley story, the TV show kind of story where we were at the office of NEA.
14:51
And we hand shook on a set of terms there and then I went to Blue Bottle paddle to hear a new app with Dennis and we're just chilling. Okay, finally it's done.
15:04
And then the Verge publishes a story of Bing releasing on Monday, and screenshots were already leaked because of some A/B tests.
15:14
And in venture funding, there's this period called due diligence, like 30 days.
15:21
In fact, another VC would also offer us a term sheet.
15:24
After seeing that, they increased the diligence from 30 days to 45 days.
15:29
And I was okay, this doesn't seem quite right.
15:34
Maybe I was trying to back out here and then the NEA also calls me on Saturday morning.
15:41
Hey, do you have time for a phone call on Saturday morning?
15:44
I was okay, maybe they just going to say, they want to back out, but they actually said, look, we believe in you.
15:52
We saw the Microsoft thing, don't worry about it.
15:56
You figure out a way, So we're not going to back out of the deal, you keep going.
16:03
So that gave us a lot of confidence, and I felt like that was very crucial, because I've heard lots of stories of how you get term sheets and actually don't get the funding, but Dave, we're true believers.
16:17
>> Luckily your investors had your back.
16:19
And as the underdog, you must have needed to get creative several times when fundraising, and you have been very successful at it.
16:25
You have attracted Jeff Bezos from Amazon, Yann LeCun, the godfather of AI, and even Nvidia.
16:32
So how did you assemble such a great group of investors and what war stories can you tell us?
16:37
>> Well, here, this is a funny story.
16:39
Dennis was at NYU, so he kind of already knew Yann, but obviously, Yann is a celebrity, it's so hard to reach him.
16:49
So Yann was on a vacation in France for a long time and we just heard he came back to the NYU Campus.
17:00
So we were already in New York at the time, so we just basically camped in front of his office for multiple hours.
17:08
>> [LAUGH] >> And he went for lunch, and he was like, yeah, you guys are waiting.
17:11
Okay, fine, I'll come back.
17:12
And then we finally got half an hour with him, and we built this search over Twitter demo, where all we had to do is let him search over his own tweets and who's replying to him and how many followers does he have?
17:28
All those kind of interesting questions everybody has for themselves, and he allowed it.
17:32
And he's like, okay, fine, I want to invest, he just made the decision like ten minutes of using the product.
17:38
Same thing happened with other investors, like Karpathy, he's a celebrity here.
17:47
He asked for a deck and I just sent him the link to directly try the product.
17:53
And same thing with Jeff Dean, all these people were just impressed by just using the product.
17:58
So the main takeaway here is, there's this thing called a circle of competence.
18:06
If you're not good at making decks, don't try to do it, right?
18:10
>> [LAUGH] >> And I wasn't good at it, so instead, just make sure there is a link that they can actually use or try instantly and make sure it works.
18:22
Because there are some people who do that and the moment you just click on it, it just crashes or it doesn't work.
18:28
That's not a good experience.
18:30
But if it works, I think it communicates a lot more than having a deck because, number one, most people don't have the time they're on their phones.
18:40
They're not on their computers reading every small part of the deck you optimize for.
18:45
The other thing is, if you're not good at it like me, then don't try to do it.
18:51
And by the way, I haven't really done decks much, even for our Series A, it was very minimal.
18:57
Series B, no decks, C, D, I just write memos or notion documents.
19:03
I actually try not to do it because I'm just not good at it.
19:07
Even a lot of successful decks from the past, like Airbnb, LinkedIn, Facebook, and you see all that and you're just even more confused how to make one.
19:22
because they're all so different, and you don't know which one to copy or like how to be original, it's very confusing.
19:32
So I just never tried to do it.
19:34
>> [LAUGH] So play to your strengths instead of being a copycat. >> Yeah.
19:39
>> It's like a lesson, you have a lot of perspective entrepreneurs in the audience, and so I think this is a good reminder that it takes great determination and hard work, and that that can really pay off >> Or you don't have to be really good at many things that people may, like, if you can be founder, CEO and not know how to make that, it's fine.
19:57
>> [LAUGH] You have a lot of consultants here who all we do is make slide decks.
20:01
>> [LAUGH] >> [LAUGH] So, maybe we have the opposite skill set.
20:04
We're a good team, if you put the two together.
20:07
Now, at Perplexity you're building an answer engine, but you don't own the content and you don't own the models.
20:15
So what is your technical moat, and why is the Perplexity approach better than direct vertical integration?
20:21
>> She's politely asking me why you're just the rapper, so tell us.
20:25
>> [LAUGH] >> Those are your words, not mine.
20:29
[LAUGH] >> But yeah, this is kind of, actually I would be, one year ago, the whole community was pretty divided on which startups to invest in, or which kind of startups to build.
20:43
Should these companies be training their own models, or should they be using APIs?
20:49
And we had a conviction that, number one, models are going to get increasingly commoditized, and if you do want to be one of those players that build it, like our provider of the models.
21:04
You need to have an insane amount of funding and you need to be a company that is losing billions of dollars a year and it's still fine.
21:12
And we were not in a position to be and we didn't want to be either, so we decided to use other people's models and shape them to be really good for a end-to-end consumer experience of searching.
21:23
And we felt like there was a lot to do outside the model there.
21:28
And I think that bet ended up being right in the sense, there are a lot of companies that were trying to build their models who no longer exist.
21:37
And I think that was a clear proof point that you either raise $10 billion or you don't do this thing at all. You do something else.
21:50
And for us, we were working on giving answers to people.
21:55
And if the answer to this question, for giving accurate answers to everybody, do you need to build your own foundation models?
22:03
The answer to that question is an absolute yes, yes, we shouldn't be doing this thing without raising 10 billion.
22:10
But I felt like if open-source makes progress and models keep getting cheaper, the cost of these APIs is going down 2x every four months.
22:20
So assume that trend continues for another year or two, we are at least going to ride the wave of a 10 to 100x reduction in the cost for the same intelligence.
22:29
And the level of intelligence and reasoning is also going up.
22:32
And open source is keeping a check on these closed source models and bringing the price down.
22:37
It's a perfect time to be an application company using these models and post training them to be good at summarization, referencing, formatting.
22:46
Custom UIs for these so many different verticals, finance, sports, reasoning, all these kind of charts.
22:53
There's so many things to do outside the model that we felt like it was just completely worth it to build a differentiated business.
22:59
And at the end, most successful businesses are rappers of some form, right?
23:04
Like, before they existed, something else was the more valuable thing, and then something comes on top.
23:11
There's even a thing of, Coca-Cola wouldn't have really worked if the refrigeration technology did not exist, right?
23:18
But Coca-Cola is extremely valuable direct to consumer product.
23:22
And so, you can always create something that's some magic formula, the right packaging, that works with the foundational technology.
23:30
But in the hands of the consumer provides immense value to them that it's totally worth building.
23:35
And so, that's what we want to be.
23:38
>> What should you build yourself and when can you leverage things that already exist. >> Yeah.
23:42
>> It's a great strategy.
23:43
Now, you've been openly critical of Google's over-reliance on advertising.
23:48
And yet, just this past week Perplexity announced that it was also introducing advertising [LAUGH] for the very first time.
23:55
So what is your monetization strategy, and how big of a role will ads play in the future? Future.
24:00
>> Yeah, so perplexities ads are different from Google's ads.
24:05
Google's ad the problem is the same ad unit is whatever is the answer unit is also the ad unit there in the sense Google gives you a bunch of links as for most queries.
24:18
And that is also the unit an advertiser can influence by paying.
24:24
So that way, when you're looking for relevant answers information, and if the ordering of the links was manipulated with ads, it frustrates you.
24:34
If we can avoid the trap and pick an ad unit that's lower margins, lower Profits, yet allows us to be true to our users and still make money.
24:44
It's a reasonably better, or like you say, I would say a much better sweet spot than what Google went for.
24:50
And we said, okay, there is an answer that should be unbiased and truthful to whatever you ask for.
24:57
But after the answer, there are a bunch of questions that we suggest you to ask next.
25:02
You don't have to literally ask that, but at least it influences you on what you want to ask next.
25:08
And let's say a shot like having a shopping related query of, like I'm looking for running shoes, and this is exactly what I want.
25:15
And these are the brands I like, and it gives you an answer, the follow-up question that we could suggest there is some shoe brand that tries to get your attention there.
25:27
Which could be what makes Adidas better than Nike for tennis or something like that, right?
25:33
That's a question they might have picked because your first question was probably, I'm looking for shoes for playing tennis.
25:40
That's a very high intent question compared to just an ad word of a shoe.
25:46
But we're not making a particular brand appear one in front of the other on the original answer, but we could still get your attention on the brand as a follow up question.
25:55
You can choose to ignore it too.
25:57
So that is an ad unit that we are experimenting with, and we're working with a few brands who are willing to try it out.
26:04
First of all, the major concern is, right now for brands, they're afraid how the answer can come out to be, because they don't really control the answer.
26:13
No brand is influencing the answer, they're only able to pick the question.
26:17
So it first takes courage for some brands come and experiment this style, the ROI is not exactly clear, because it's not necessarily driving a lot of traffic to you.
26:28
So it's still very early days for us.
26:31
But what we are very, very clear on is not trying to influence the accuracy and truthfulness of the answer.
26:37
Because once we do that, then we're going to end up in the same path as Google, where people are frustrated with the answer.
26:44
>> So when I type my initial prompt, for example, for this interview, I will never see an ad response? >> Exactly, yeah.
26:50
>> Okay, reassuring to hear.
26:53
Now, perplexity is obviously innovating very quickly, and yet this pace of innovation has attracted some controversy.
27:00
So News Corp, the parent company of the Wall Street Journal, has sued you for copyright infringement.
27:06
The New York Times has also issued a cease and desist order for inappropriate content use.
27:12
How are you handling these recent challenges, and what is your vision for ethical AI development?
27:16
>> Yeah, so here's what we believe, and we've said this on our blog post to, no one has a copyright or ownership over truth or facts.
27:32
This is true in the world of journalism too.
27:35
If there is an article, for example, right now in our interview, you reference is that the New York Times is zero perplexity.
27:43
Now, that was reported by somebody else, but you're using that in our interview.
27:48
Now, can someone claim ownership over that and disallow you from saying that particular thing, no, right?
27:54
So truth is supposed to be distributed widely, right?
27:58
So the specific expression of truth, the specific way in which something is written that has some copyright angle there, and that's actually the core OpenAI New York Times scenario.
28:11
But what we are doing is, we are referencing truth that already exists in these outlets and summarizing and synthesizing it for the user in the context of a search experience.
28:21
So people need to differentiate the use of AI that trains on proprietary content, versus AI's that just use them as sources and give answers and there's no actual training happening.
28:34
So we made that very clear in our response and we've also made it clear that we can only survive and keep getting better as a product if there is an open and thriving ecosystem of journalism.
28:49
Because we do need real time information to be created every single day, and if there's not the right financial incentive for them to do that, then it's not good.
28:58
So what we did is It's, okay, we're going to make revenue through ads, and we're going to share that ad revenue with publishers.
29:06
And that way, you enter into this publisher program that we came up with, which is not exactly paying you money just to license your data for a certain period of time.
29:14
And then once we've absorbed it, we don't want to create that sort of a short-term model.
29:19
A long-term model, is as we scale and usage as we scale as a business and revenue.
29:24
We want to share that revenue with you on a query level basis.
29:28
So it's very clear, this is more inspired by how Spotify shares revenue and fortune time.
29:36
Der Spiegel have all signed up to be part of it, WordPress signed up part of it.
29:40
And we are also going to announce more partners in the coming weeks.
29:43
So, we are very confident that that program will soon resonate with everybody in the journalism community.
29:50
And we also made grants to Northwestern University to kind of do more research on how tools like ours can help journalists write better, because all journalists do fact checks, and we are an amazing tool for doing fact checks.
30:05
So I'm very confident that this current period of turbulence will go away, and a year or two from now we'll have a system that helps both these different set of people to flourish together Just to follow up on that.
30:21
So earlier today we talked about perplexities, academic roots and the importance of citations. >> Yeah.
30:25
>> So given that, how do you handle these journalist allegations of plagiarism in particular?
30:31
>> Yeah, so exactly, so that when you want to go deeper into the definition of plagiarism, it's if you don't attribute the source.
30:39
That's a core part of it, and when you're always attributing the source, it's very hard to say you're plagiarizing content, and also you're not exactly reproducing things, as sure, AI are unreliable at times.
30:53
And there are times when they're word overlap of more than three or four words.
30:58
And you can argue to what extent that is exact reproduction was just trying to synthesize.
31:04
But what we are trying to say is we are trying our level best to summarize, synthesize from diverse sources, and make sure to give credit to all the original sources.
31:14
And that way, to our best can control these AIs.
31:22
We are doing our best to make sure that the credit attribution part is clear.
31:26
>> I like the Spotify analogy, it's such a rapidly changing.
31:29
>> Yeah exactly, so if we make an ad revenue where you're a source, we're going to share that revenue with you.
31:36
Who would ever share ad revenue?
31:37
Google, because they gave you the traffic, but they made the ad revenue on their platforms.
31:43
And the only way for you to monetize the traffic that you get is to put pop-ups and ads on your site through another product of theirs called AdSense.
31:54
And so that is what frustrates many users who come directly to read on many Journalist sites because there's a lot of ads on the site, and they have to close a lot of pop ups.
32:06
And so this system of just referring traffic and making you monetize with more ads is not sustainable, right?
32:14
You need to create something that the user really wants.
32:17
And we're also offering our APIs so that they can build AI native products and chat bots on their websites.
32:24
So if people want to just come there and ask questions about only articles that they've written.
32:28
We're offering our APIs for free to these people, and we're also offering our tools for free to all the people who work at a particular journalist outlet.
32:37
So that way, we can create a system that is economically pretty lucrative for them.
32:44
>> It's an interesting future ahead.
32:46
So if we take a step back now and we think about the biggest technology companies of all time, in almost every case, these are category creating companies, Uber, Facebook, Airbnb, Salesforce.
33:00
So a decade from now if we look back on this moment in time.
33:04
What is the history defining company that you are building?
33:08
>> I would say if we can help people get answers to all their questions and get help for all their tasks we'll be in that league for sure.
33:22
And we're getting pretty close to being a reliable answer machine.
33:27
I know you asked the question, is Perplexity ever wrong?
33:29
I'm telling you, there are a lot of mistakes we still make on a daily basis.
33:33
But zoom out and think, if models keep getting better and our coverage of the web is getting better.
33:40
The mistakes are going to be whatever is one in a 10 or one in 100 is going to, be reduced to one in 1001 or 10,000 the order of magnitude improvement is going to come.
33:50
So if we are a reliable answer machine to everybody and widely accessible, and not just give you answers, but help you accomplish tasks.
33:57
To make transactions, buy things, book things, book flights, get the best deals, and make your life more productive, give you back all more time.
34:09
I think we are going to be a pretty industry-defined product in a company.
34:15
>> It's exciting, we hope to see you there a few years from now.
34:18
Now, there's so much more we could talk about when it comes to Perplexity, but I wanted to save a few minutes to talk about you and your leadership style.
34:26
So in just two years, you've progressed from a scrappy founder, to CEO, to the leader of a $9 billion AI company.
34:34
What does your leadership journey look like across all of these different stages?
34:39
>> Yeah, I tried my best to keep upgrading.
34:42
And I'm still not the most seasoned, polished CEO.
34:51
But I would say there's an extreme bias for action that I try to bring in and try to encourage everybody else in the company to adopt.
35:01
And I think that's what's helping us continue to be fast, even when you've gotten to about 100 people.
35:08
A founder that I really admire told me, once you get to 100 people, you're guaranteed to move slow.
35:13
And I was very determined to prove him wrong. So, so far, so good.
35:17
But at some point, definitely, we're going to hit the problems of scale and how to move fast.
35:24
So I'm determined to solve that problem and if whatever final solution I come up with, I hope it's helpful for other people too.
35:33
And the other thing I would say is giving people who haven't necessarily become experts at one thing the opportunity to go do something they're not yet proven for.
35:45
Is something I've done a lot, you don't have to hire the former head of growth at Instagram to be the head of growth or head of product Perplexity.
35:57
That's a trap that a lot of people fall into.
36:00
It's, if I want the best person for doing AB tests, I'm going to get the person who did it at the previous best consumer company and hire them here.
36:12
I have not fallen into this trap.
36:14
I've actually tried to hire people with some chips on their shoulders who are very talented, but they have not had their first major hit yet.
36:23
Chips on shoulders, put chips in your pockets.
36:25
So not my original quote, so I don't remember who said this, but it's pretty cool.
36:31
So that's something that I wish more people did the sort of experimentation, putting someone in the waters and letting them figure out how to swim.
36:42
Rather than hiring the most, well known expert at that topic.
36:47
Main reason is that most people are unable to push themselves for the second success in general, I've seen that.
36:55
And I think, it's very hard to be extremely motivated to do grueling hours when you've already had big success in your life.
37:05
So that's one way I've tried to be different and bias for action, trying to do things on my own, to understand what it is.
37:16
And I use the product quite a lot myself, pretty much it's like at least 10 queries a day is my average.
37:23
But there are some users who do it more than me, so I'm very happy, and I think that helps me to make the right decisions.
37:33
If at some point you stop using your own product that the company's building, it's very easy to lose touch with reality.
37:38
And you're just making decisions based on what other people tell you, and it's very essential that you're as close as possible to the source of truth.
37:50
So when people complain on social platforms like Twitter or this thing's not working, that thing's not working, I love doing customer support.
37:58
I think we have people who do customer support too.
38:02
I'm not trying to say they're not needed, but it really helps you to understand what the customer frustration is, the user frustration is.
38:12
And be that sort of a user yourself, complain about your own product to your engineers, your product managers say, this should be fast.
38:20
You don't have to, sort of just be doing whiteboarding and strategy all the time.
38:28
You can actually just sit for hours and hours using your own product, and you can make better decisions.
38:34
>> How we stay scrappy at scale? >> Yeah.
38:36
>> It's a great one to stick with.
38:39
So before we open it up to audience Q and A, I have one final question that we're asking all of our speakers this year at view from the top, our theme is leaving your mark.
38:48
So Aravind, how would you like to be remembered?
38:53
>> I would love for myself and Perplexity, to be known as helping make the world smarter.
39:02
If people who use Perplexity feel smarter after using it because they learn something new, they slept wiser than when they go to bed, then they woke up.
39:12
I would feel really, really glad if we accomplish that because I think that's not easy.
39:19
Most consumer products end up wasting people's times.
39:24
There are obviously not going to, I'm not going to mention which, but there are some products I am addicted to.
39:30
I use it a lot, but I don't feel good at the end, I've wasted so many hours.
39:36
And Perplexity is not that, at least, I don't think it is that sort of a product that may, even the discover feed that we have, people tell me that they learned something when they scroll through it.
39:48
And I want that to continue to be the case.
39:52
And I also want us to help people do things.
39:57
Not a lot of people can afford to have a Assistants, executive assistants, personal assistants.
40:04
And I remember in 2018, when I was an intern at OpenAI, Sam Altman did this fireside chat with Bill Gates.
40:12
And he asked Bill Gates, what do you think the world would look like when there is AGI?
40:19
And the answer Gates gave was very interesting.
40:21
He said it would basically be like living my life.
40:27
>> [LAUGH] >> Live like a billionaire, where if I want to know about a topic, if I want to learn about a topic, I don't have to read any book.
40:38
I can have people read it for me and prepare a report for me and even make a presentation for me.
40:45
If I have to get somewhere, I have a jet, people take care of all the travel planning, meals.
40:50
If I want to work out, everything is done.
40:53
I want the best nutritionist, I know what to eat.
40:56
I don't have to think that life is so easy.
40:57
It's like running life on cheat code.
41:02
Now, I think that sort of a life can be made more and more accessible to most people.
41:07
If AI can do stuff for you, truly understand you, help you plan stuff, help you book stuff, all the mundane work that you have to do on the web.
41:17
If a tool can sort of increasingly get better and better at doing it, I feel like your life will be like a billionaire, right?
41:27
And then the meaning of the word billionaire also sort of loses significance over time.
41:31
So I think if we can be one such tool, I'm not saying we want to be the only tool, but if we can be the one such tool that helps people do that, I feel like I would've made a good mark in the world.
41:42
>> If Perplexity can give me Bill Gates' life, I will be very happy one day.
41:47
[LAUGH] Now, we'd like to open it up to audience Q&A.
41:50
So if you have a question, please raise your hand and one of our view from the top mic runners will come find you.
41:55
If you're selected, stand up, state your question and year, and ask your question.
42:03
>> Aravind, firstly thank you, and then you're one of the young leaders we all look up to given where we are today.
42:10
I'm a fan of the book Atomic Habits.
42:12
So in your journey from PhD student to Perplexity CEO, what is the one habit or daily habit you had to let go of and maybe a new one you learned that helped you become the leader you are today?
42:26
>> So the one habit I let go of, I'm not sure if I had to, but I definitely let go of this and I feel like it helped me, is waking up late was something I stopped doing.
42:37
So- >> [LAUGH] >> Yeah, I think it makes me feel like I get more hours in the day.
42:46
And so that also means going to bed early.
42:48
Early to bed, early to rise.
42:52
I feel like it helped me severally.
42:53
So it's been probably at least maybe three to four years since I've woken up later than 8 AM in the morning.
43:01
And it doesn't matter which city or where, I've always done this.
43:06
What was the other other question?
43:08
>> The one you learned, that you had to pick up I think, yeah, at least trying to get better at getting three days a week workouts.
43:21
>> [LAUGH] >> And I never used to work out much before, so.
43:25
>> [LAUGH] >> Thank you for being here, Aravind. >> Thank you.
43:31
>> My name is Ravin, I'm a second year MBA student, went to IIT Bombay for my dual degree in electrical engineering.
43:37
I think you answered the technical mode question really well, but we'd like to hear from you.
43:41
What do you think is the biggest risk or challenge that you feel the company is facing?
43:45
>> Sorry, could you repeat that?
43:47
>> What do you think is the biggest risk or challenge that you feel the company is facing today?
43:51
>> Yeah, I think it's the same thing I said earlier, which is, I think every startup that's tried to scale somehow ended up moving slower once they got several hundreds of people, a thousand people.
44:08
Somehow it gets difficult to do things.
44:12
You expect progress to be at least linear, and the number of people, you can do more projects.
44:18
But it gets very difficult to simultaneously execute well without some drop in quality.
44:24
And when there is a drop in quality, users notice that, and they think you've regressed, your product has gotten worse.
44:32
There's this whole phrase called enshittification, when you're trying to scale and scale your business, scale your users, product gets worse in quality for the initial loyalists who allowed it for the quality.
44:44
So that's, in my opinion, the biggest challenge for us.
44:51
>> Hey, I've got to follow up to Aislinn's question earlier about the ethical issues that Perplexity has been facing.
44:57
I know that you personally have experienced a lot of criticism surrounding some of those issues, and I imagine that as a leader in those spaces, you spend a lot of time thinking about that.
45:06
I'm curious how you've approached ethical issues, whose opinions you seek out to inform yourself as you step into bigger and bigger roles, and if there's an example where you've seen yourself change your position on something really important?
45:20
I'm sorry, I'm and I'm an MBA 1. >> Thank you.
45:24
So I think we have a good set of people in our company.
45:30
Actually, for the Publisher program itself, it's the brainchild of our chief business officer, Dimitri.
45:40
So clearly, one thing that I've learnt is, just because I'm the CEO doesn't mean I have to be the one who solves every problem.
45:51
If someone's better than me at doing some thing, you should trust instincts there.
45:56
One thing I did believe in is there's a lot you can do by just engaging and trying to educate people on the other side of what you're trying to do.
46:11
I'll give you an instance.
46:13
Forbes, obviously, was pretty unhappy with some of our attempts to do pages and things like that.
46:21
But when I actually met the person who criticized me on Twitter and explained to him what we are doing, he at least shook hands and said, I never understood this is exactly what you're trying to do.
46:35
And so I think there's more work to be done there.
46:38
I haven't met everybody in the community who's talking about us.
46:44
And I think just trying to perceiving something, right?
46:46
When there is even a sign of war or it's foggy around there, feels something my bad might happen, the first thing you got to do is make a phone call and talk.
47:01
So that's what I intend to do. Hey.
47:08
>> Thanks so much for coming here.
47:09
>> I'm Varun, MBA 1 student, computer science from IIT Kanpur.
47:12
You're a big inspiration >> Thank you.
47:14
>> I really resonated with the point where you said that you're trying to help people find the answers that looking for.
47:19
But then also when you talked about introducing ads in suggestions, where you suggest like, why do you think a Adidas is a better shoe for sports than Nike or something?
47:30
Don't you think that affects the kind of information or the view that you're presenting the users?
47:37
That might not be the answer they're looking for.
47:38
So don't you think- >> It's a question, though, it's not an answer.
47:44
It's just a suggested question.
47:46
But if you do decide to engage with that question, the answer to that question is still unbiased.
47:53
It's not something Adidas is influencing us to write relative to the other The brand, and the same way.
48:02
So at some point, we would really understand you quite deeply enough that even those questions are quite personalized to you and not a generic sponsored question that the brand picks.
48:15
And there has been evidence that, for example, a lot of people feel like the Instagram ads are pretty relevant to them, and a lot of purchases happen as a result of that.
48:26
So, I feel like relevance is the true answer to making ads work, and making sure that it's not interfering with the core value of the product.
48:36
Which is, any question you ask if the answer to that is un influenced by ads, and it's always going to be truthful, the product serves its value to you.
48:44
But if we want to bring such a quality product, which is almost never wrong, so you can trust what it says at scale, the company does need to figure out some intelligent sweet spots of monetization too. >> Thank you so much.
49:00
>> I think we have time for just one more question.
49:02
>> Hi, Irwin, I think this might be a good one to end with.
49:05
By the way, I'm Abrar, I'm an MBA 1, so as the co-founder of Perplexity, what is the question that you find the most perplexing? >> [LAUGH] >> This one.
49:18
>> [LAUGH] [APPLAUSE] >> It's a great answer.
49:34
>> [LAUGH] >> I'm glad you stopped there.
49:37
>> We do have one final tradition here at beef from the top.
49:40
Where I ask you a set of rapid fire questions, and you respond with the first thing that comes to mind. >> Okay.
49:45
>> For this one, I use Perplexity to generate all of these questions. Should be easy, right? >> Hopefully.
49:52
>> [LAUGH] If you weren't the CEO of Perplexity, what would you be doing right now?
49:59
>> I'd probably be doing research, that's what I was doing before. So, this AI research.
50:04
>> You're a cricket enthusiast, what is your all time favorite cricket moment?
50:10
>> When India won the World Cup in 2011.
50:13
>> [APPLAUSE] >> If you could have dinner with any tech visionary that are alive, who would it be? >> Larry Page.
50:25
I mean, I'm not saying this, Steve Jobs or Larry Page would be my pick.
50:30
>> Bring them together, dinner for two.
50:32
[LAUGH] And finally, what is the strangest search query you've seen on Perplexity?
50:39
>> [LAUGH] >> So we released Shopping on Monday, and just looking at some of the otters, and someone basically bought this face mask that only has an opening for the eyes and nothing else.
50:53
And so, either they're looking at it in the context of skiing somewhere or like biking somewhere it's really cold, or they're conducting a heist.
51:03
>> [LAUGH] >> A nd we weren't sure of it, so we actually went into the query, and the query was, okay, I'm really trying to bike in this weather in the coming months.
51:12
And I need something that can cover my face, keep me warm, let me breathe, but only as an opening for the eyes.
51:19
That was the intent level of the query, and then we got them some pretty good answer that they just bought a product right from there.
51:27
>> [LAUGH] >> [LAUGH] Well, let's hope that the use case was the former and not the latter.
51:32
>> [LAUGH] >> But that's a great place for us to wrap up.
51:35
So thank you so much, Arvin, for being here. It's been a pleasure.
51:38
>> [APPLAUSE] >> Thank you. >> [APPLAUSE] [MUSIC]