Aravind Srinivas on Perplexity AI vs Google, AI Search & Startup Growth | The AI Rabbit Hole 2025

Thank you so much. And hi Aravind, thank you so much for you know what making it remotely. we know you had to fly to the east coast unfortunately and we couldn't move the conference. We would though for you but you know SFJ was not available later. So first of all congratulations on everything you've built so far so fast. that is the fastest growing company I've ever seen in my 18 years of experience. And the majority of that is just the culture that that you've built in it. I remember in your TED talk you said you see yourself as the carrier of relentless questioning culture. How do you pass that mindset to the fast growing organization? We can you hear me? Okay, there might be some Okay, there we go. Aravind, can you hear us here? Yeah, I can hear you now. Sorry is on my mat.

yeah. So the question was that in your TED talk I remember you said you are the carrier of relentless questioning culture and this is what drives you forward and you know makes the company so successful. How do you pass it on to a company that grows so fast? It's difficult. I don't think you should expect it to be the case for everyone who joins the company and at our current stage. So I guess like lead by example that's one one one approach I believe in which is when somebody comes and says this is how it should be done I tend to ask like why do you think that's the case and I tend to dig deeper into that and if they don't actually know the answers to every follow-up question then that that kind of means they haven't gone pretty deep into the relentless questioning part and I say that okay you know what like you to ask all these questions yourself and then that way you don't even have to like require my help and that's one way to like inculcate that obviously it's not scalable I cannot talk to every individual person in the company so my hope is like people who have direct reports will also start doing that with their people but I don't want to like make everybody lead the same way I do like you know each one has their style and approach but the mentality of questioning assumptions should be there in general.

You recently mentioned that in the last 3 months alone the AI progress has outpaced what you saw in your first two years at Perplexity. So what specific technological shifts are driving this acceleration and how has that changed your long-term view of Perplexity's role in the ecosystem? Sure. So I think the number one change would be reasoning models that can think and before coming back to you with an answer earlier the approach would be summarization and like little bit of tool calls but like a bunch of different models orchestrated together to do oh you go you go do the plan for the query you go like refine the query you you perform the multi-step reasoning you would write some piece of code to do some computations on top of data you retrieved and then you do the sum ization like four or five different models stitched together and we made it all work.

that was the beauty of the system. but then now what happens is you can have one single model that can take your task decompose it into a subset of tasks and then keep all that in context and string together the execution of each of these subtasks. decide what tools are needed for each of these subtasks and like use all of them appropriately and reflect and think and you know reason before actually coming up with a summarized final output. and can do all this over a long context keep the conversation going and not hallucinate or like fail. of course it's not 100% reliable, but the level of reliability is already so amazing that you can do lots of things that you could never do before and you can also maintain the system much more efficiently like fewer lines of code for the product engineers and a lot faster applications.

So that is the main change that happened with the progress and reasoning models. And the number two thing I would say is distillation like bringing down the cost whatever was possible with GPT4 like a year ago is now possible with a very small model that you can host yourself and availability of great inference custom chips for inference that let you like get really good throughput and latency which makes the product feel a lot faster when you're using a bunch of these models together. So keeping up with these two trends has been difficult because like ever since the arrival of Deep Seek and they were the first to do an open source reasoning model and do distillation and made all the frontier labs like cut down the pricing you constantly had to think okay what is the value of your pro plan still like what and then keep building new experiences and while also simultaneously thinking about things to build beyond your current pro plan like stuff that would take many hours of work to do and stuff that you could even potentially charge higher for.

So you need to constantly be pushing the boundaries here while keeping your current product like reliable and like scalable and improving in like the core metrics that's been difficult. that's why I said like the last three months were more challenging than anything else. So with everything changing so fast, which aspects of the product, the tech stack and user experience will you see will remain for a long time as you know something that perplexity stands on and is the moat. Yeah. So this is the Bezos question, right? Like I think he said this in an interview where everybody wants to know what's going to change in the next 5 years. Nobody want nobody wants to ask the question what will not change in the next five years because if you know the answer to the second question you can at least formulate a strategy around it and become really good at it and build a business around it.

I I think we identified this a while ago which is nobody's going to come and say I want a slower product. Nobody's come going to say I want a worse product in terms of accuracy or reliability. nobody wants like things to work for one hour. Like people want instant stuff, even if it's like, you know, super high IQ answers that are like really hard to get. I don't think people are like okay with like being slow. Okay, they're okay to begin slow, but they want everything faster. And nobody wants expensive products. Like everybody wants to save money at the end. Even no matter how much money somebody has, like even the richest people, some of the richest people I've met, they wanted pro codes for using Perplexity Pro for free. It's just the thing that people are just want don't want to like pay and want to get things for free, want to get things fast, want to get things like to work well most of the time.

So this this just means like all the work you keep doing on making your product like hallucinate less, making your product stay stable like not go down, making sure answers are streamed really fast, making the UX feel like amazing and making the product more and more affordable so that most of these experiences are available on the free product. All that will keep paying dividends for you for many years. Thank you. you've built a hundred million ARR business in under two years, a milestone very few have reached this quickly. And this means a lot of conventional wisdom doesn't apply. You're surrounded by very smart people, a lot of really good advisers and investors and obviously employees. But how do you decide when to follow advice and when when to take a contrarian path? when I have pretty deep conviction on on something where I I have my gut feeling that this is the right thing to do, I don't follow anyone's advice.

I just listen to my gut and if I'm wrong, I'm wrong and I live with the consequences of it. when I'm not having my own intuition or gut about something when I'm like clearly like you know neutral I do take into account what other people say and think and then do my own reasoning on top of it and then make a decision. So I have not always made the right decisions. I've made mistakes. but the most important thing is to make a decision like that's the most important thing. the worst leaders are those who are indecisive and they just don't do anything and let multiple projects keep happening in the company. Sometimes even parallel efforts doing the same thing u without even realizing they're doing the same thing. I think those are the kind of things at least I try to be good at where if some project needs to be killed we just decide to do it pretty quickly.

And I think that I should be doing even better. I'm not saying I'm doing really well, but that's actually the most important thing right now is to really decide what to focus on. there are so many projects already that we are working on. Like a lot of people on Twitter and all that consider us to be pretty distracted. but little do they know how many things we've said no to. And I wish we can get the courage to say no to more things because that'll help us do even better on what we choose to do. Well, you you seem to be innovating quite fast on a lot of things and you know a lot of people would argue that it's impossible to build better search experience than Google and well apparently they were wrong and a lot of people would say that like you cannot build a better Siri than Apple and you just did that for congrats on shipping the iOS assistant.

It's really marvelous. how do you enable the company to ship so fast so many different things? I don't think we have a choice and I try to tell people the honest truth that we don't have a choice like this the space is so competitive. everybody wants to win here including Google they're they're you know for whatever fun we make of them they seem to be like you know pretty aggressive in the last 3 4 months shipping AI search on like the AI mode or AI overview or putting sources in search in Gemini trying to get Gemini user base up so they are like going for it finally I don't know what changed there but certainly they do feel the market share probably is not increasing and then other people are growing and so I think there is no choice except to move fast and I just say there's no other way of injecting urgency in people other than saying this is the only way so by committing to work with us on this company on on these set of projects you are committing to this style of work too and you know what the other thing I would say is there is no other way for anybody else either it's not like you go work at Google on AI, you can move slow.

They're also like feeling the urgency, right? So moving fast is even f easier in a smaller startup than a big tech company because there are fewer decision makers and stakeholders and you have way more agency and ownership. So that also helps in like you know moving fast. We just don't have to deal with all this decision by committee. U thank you. Shifting gears in the room today alongside many founders we also have investors that are just entering the space some of them drawn by the fresh wave of opportunity similar to what we saw you know with the internet or the cloud what advice would you give to firsttime investors looking at to like wear invest in AI I generally think investing in products that you kind of use and feel a magical end user experience is always like a good proxy. mainly because very few people manage to build some product that really feels magical.

it takes a lot of good decisions to do that. and and especially if someone has done that with very little f like funding so far, like very little money spent and a very small lean team and actually has a product, no weight list, no recorded demos, you know, not a hype marketing video, but a real product. You can log in and try some things and see for yourself how it works. And you feel something very different from other products that exist. And you know that like they have already planned for the next 3 months of shipping and are like and and and you should feel the trajectory too like okay once they've given you a product and you use it a week later even if it's behind some data like u are you feeling the improvement if all these things are true I think you should go and invest.

because it's very rare to get all these things right. on the other hand there are a lot more companies with like oh like founders with XYZ profiles they're from Stanford they're from Berkeley or OpenAI Google they have like a great LinkedIn and they've raised a bunch of capital from investors you already know. I think that template exists and people are like you know you're going to find plenty of teams like that. on the other hand like this template does not exist that often. So when you find something like that just go for it. That's really good advice. Thanks man. so last question is we spoke last year at the same conference. So a year after everything changed so much. What surprised you the most? I would say that the progress in like reasoning models I I definitely expected progress in models that can think and plan and and do chain of thought but I did not expect it that scaling to be this fast in the sense 01 was released somewhere towards the end of the year I think we are in 04 that's like basically two generations they skipped O2 u and then I expected that it'll go a lot slower And also the other thing I did not expect is for others to catch up this fast.

I thought they'll have the same lead as GPT4 u like where there's like a me period of eight or eight months or something where people are still using open AAI models that that did not end up being the case. The the science behind reasoning diffused pretty quickly including open source models like deepseek and so that dramatically brought down all the costs. So that ensured that like competition is way more intense than ever before and everybody's racing to be the number one model and every 3 weeks or something the number one model changes on the leaderboards. Yeah, that it it is terrifying sometimes to to see how how fast the things move and you know we all expect AI to become smarter. We all expect it to become cheaper but sometimes you know dipick is out there is like oh it can be that much smarter and that much cheaper in such a short period of time.

Thank you. so much for you know making the time for us and this u amazing audience is very appreciative of you showing showing up at least through what what are we on Google meet or zoom are are you going to launch your own video conf calls at some point in the future? I'm working on like already too many things so I want to kill projects not create more so but appreciate the ambition. That's a That's a great mindset. Thank you so much. Thank you.