the innovations we bring to millions of users across India. To kick off this event, it's my absolute honor to welcome our distinguished guests for a fireside chat on the future of AI in everyday life, welcoming Mr. Vijay Shekhar Sharma, Founder and CEO of Paytm, and Aravind Srinivas, CEO and Co-Founder of Perplexity. Over to you, Vijay.
Thank you. Thank you so much. Thank you so much. So, I think we right away dive in. We took some time to do broadcast on two different platforms, and what I'm seeking is that we will be able to discover something new for our Indian customers, Perplexity Indian customers. Let's, without any further delay, hear it from Aravind. Aravind, talk to us about Perplexity India. What are we going to do here?
Yeah, absolutely. Yeah, first, thank you everyone for dialing in this morning. So, I'm very excited to work together with Vijay and Paytm. They're very forward-minded and really trying to bring the power of AI, the most advanced AIs with cutting-edge search capabilities, to crores of Indians without them having to pay anything. I think this second part is very important because knowledge is supposed to be universally accessible, democratized to everyone. And obviously, it's pretty difficult to afford access to the best AIs; they're all behind paywalls. So, if you're on the free version, it's kind of difficult to get the best quality and service all the time.
So, the Paytm app is directly integrating Perplexity as a button on the app. At the bottom, you would see it as an "Ask AI" button, and then you'll be able to ask whatever you want. Now, the immediate question I'm sure in a lot of people's minds is: why is this a payments app? What's the connection between a payments app and an AI button? Why would I do that? I think the whole concept of a well-packaged, well-integrated app that does multiple different things is more popular outside the Western world, actually, the whole concept of super apps and stuff. So, I think people can experience and discover so many different features within one app, and people like that more outside the USA. I hope people in India like that, and they can explore topics in whatever language they want.
Especially when you're about to buy something, you could even go and ask, "Is this worth buying?" Or you're about to pay something to somebody, you could just get the latest financial insights, or in general, you could just ask stuff about the market. All these things that are regularly associated with payments, I think you can just directly ask an AI before making any decision to actually pay somebody. I think this is going to be pretty powerful. The most important thing for us is to democratize access to our technology to so many Indians out there. India is one of the most curious countries. We're curious about so many different topics, and Perplexity is the ultimate solution to your curiosity because your first question is just the start; you can keep asking tons of questions.
And the most important thing is you can fact-check what it says. Even if there's no AI in the world today that makes zero mistakes—if somebody's claiming that, they're definitely lying, right?—at least the only way to give the consumer the power to trust what the AI says is to give sources. That's one of the things we pioneered, and many other AIs out there are following that, but we're still trying to be at the cutting edge of it. I'm very excited to work together with Vijay and Paytm to make it more accessible to all people.
Awesome. I think to call it out, it is definitely the moment in history where to answer and have a conversation like a human... Can we talk about what kind of standard general knowledge, horizontal foundation model answer, and how your reasoning models are different, and what are they giving, and the kind of questions people ask?
Hey, can you hear me now? Okay, so I'll continue here. I think it is also...
Okay, I think we are continuing on Zoom at least, because it seems like on Twitter Spaces there is some confusion, and I'm going to talk about it. So if you can hear me and talk about the kind of things that you believe will be able to help custom, the multi-step reasoning USP and those things.
Yeah, absolutely. So, we're trying to build this auto mode on Perplexity where, depending on your query, we'll automatically decide whether the query needs a small model, or a multi-step reasoning model, or something even more advanced like the Deep Research that we have, in a self-orchestrated way. I think that's pretty difficult to do, but we're working on it, and I think that's the real long-term solution where you got to completely deprecate the model selector or the mode selector and just let the user not have to worry about it, right?
Let's say you're going to ask a question like, "What's the age of the Prime Minister of India?" I don't think you need a DeepSeek answer response to this; it's a pretty easy lookup. At the same time, what if you go and ask some question like, "What are the latest insights from the most recent speech or discussion between Prime Minister Modi and Elon Musk?" It might want to look up multiple different articles and give you the exact summary with detailed takeaways where accuracy might be more important, and so a little more compute needs to be applied to that sort of query. So what we hope is that there's this one mode that is just directly integrated in the Ask AI button and everybody gets access to it. That's still in the works, okay, so that's not what is there in the Paytm app today—I'm not over-promising anything today. But I'm saying that long-term, that's kind of how we plan to democratize access to basically everything to everybody.
Interesting. And I know that you met Prime Minister Modi in Paris, and you also met him in India. Anything that you can share with us, especially with reference to AI in India?
Yeah, it was really a great honor to meet Prime Minister Modi both in India—I got more time to spend with him when I was in India—and I think one thing I was definitely impressed and amazed, I should say, is his level of preparation in terms of how well aware he is with all the different AIs, their flaws, strengths, and weaknesses in all of them. For example, he clearly knew what the differentiation for Perplexity is with the fact that it gives you reliable sources all the time. You don't have to second-guess if it'll give you sources or not and turn on buttons to give you sources or anything; he just knew that it'll always give you the sources.
I think he also was having special insights on, you know, what if we could take all the ancient texts that exist in India, but not necessarily in English or that's not been indexed by the web, and take the scientific knowledge in that and figure out a way for the AIs to draw references to that and what exists in modern scientific literature, and show the world like a lot of the things that we used to follow through our texts kind of had got it right. Similar to how the world looks upon India with deep reverence and respect for being the inventor of zero, famous mathematicians like Aryabhata, I think that's what he was getting at.
And he was also pretty interested in seeing how AIs can figure out smarter ways to do things. We had Vedic math; I think he was pretty interested in seeing when they're asked to solve problems, will they be able to figure out new and interesting ways to solve them? So those were the two main takeaways for me from meeting him. He was also definitely concerned that people might use AIs in a way where they just use it to get work done faster, but not truly curious about it. How do we encourage people to be a lot more curious in a natural way? What he was getting at was if we can make it widely accessible where the barrier to entry and the ramp-up time to learning AIs is very minimal, which he believes the best way to do it is make it work in all Indian languages.
Because the ability to just ask a question comes very naturally to anyone in their mother tongue, but then when you go and have to use modern tools, you kind of have to learn how to prompt in English. That obviously stops multiple people from using these. So if you can just ask in your own mother tongue in just voice, I think the ability to just use this AI comes more naturally to a lot more people at a much younger age, and that could help India to really adopt AI in a way where it's not just meant for cheating at work to get work done faster or something, but is truly meaningful.
That brings me to one of the biggest questions we in India are talking these days: based on local knowledge of different literature which may or may not be on the internet, is there a reason to build a foundation frontier model from India?
I've voiced my opinions about this before, so I'm very consistent with what I've said already, which is: I do think India should build its own foundation models. Not just because it's very important to get the Indian languages right—that's absolutely essential, I would say table stakes—but also to have the skill set to do it and essential to compete on the same benchmarks that everybody else competes on. It doesn't have to literally hill-climb on every benchmark, but at least be able to train really good models that can reason on their own. That's the only way to make sure we have the muscle to train these models ourselves, the infrastructure, the talent. And then the next generation of engineers should have something to feel inspired by, right?
For example, whenever India launches a satellite at such a low cost, everybody gets inspired by those events. I think the same should be true if people are able to train world-class models that not just are good at Indian languages, but are very good at math, coding, reasoning, and are on par with the models trained elsewhere as well.
Aravind, the interesting question that we also have is that there are different open-source models, so India should build it, and the training cost is reduced. Will there be a use for Indian-only applications or will it be an international application, like the India model will be used only in India or will it be a global thing?
Sorry, can I speak on both the X and the Zoom thing? So, will the model be used only in India or global? I hope it's both, right? I think AI is, as Prime Minister Modi already said, a pretty global technology, and that requires models developed by each nation to be widely useful.
Interesting. Speaking of next steps, we want to talk about a little bit of demo, and then we take questions from the public. Is that good?
Okay, hope you can hear me. I'm going to show you now what you can do with this demo. The demo is available on Zoom sort of there, because many of you may be in Zoom. It is, in fact, open it up just like asking a question. What you see here, you can ask, let's say, PM Modi: "Who is Prime Minister Modi and what do you know about him?" So let's see what Perplexity answers. It works in different modes where you can see this is the kind of question-answer you're seeing, and then you can learn more about it and so forth. As simple as that, ask follow-up questions, etc., for a continuous experience. Okay, so we take questions. Is it okay?
Yeah, absolutely. Let's take some questions.
Okay. There is an echo problem. I think Zoom has this because we are speaking in two places, seems like.
I think Indians are already using a lot of AI. I read a statistic actually on Perplexity that knowledge workers in India particularly have adopted AIs a lot, and I think that's pretty exciting. So I hope that it's not just restricted to knowledge workers, but basically everybody. I think that's the next big shift. And the other thing I would say is an AI like Perplexity is not just meant for using at work. I think you can use it for anything: sports, fashion, shopping, travel, all sorts of basic day-to-day activities. Health and fitness is another thing that applies outside work that Perplexity is particularly useful for.
And the fact that the reasoning and stuff works across multiple different languages means you should be able to enjoy it in whatever language you want, and we hope to also bring in all the voice-based experiences to everybody.
Cool. Another question, Aravind. I'll suggest people raise their hand on Twitter because it seems like we are better off here; Zoom has some echo. So instead of giving further demo, I come back to the Twitter Space. Raise your hand, guys, and I'm enabling some of those guys who are asking questions. Okay, Subho Chakrabarti, you can ask questions. You are now allowed to speak.
Hi, I've just enabled some of you to directly ask the question.
Hello, yeah, hi. Thanks for giving me the opportunity. I hope you're able to hear me because there are a lot of audio issues.
We can hear you. I think running on Twitter standalone works good.
Okay, awesome. I have actually two questions. One thing is that it's a very good initiative you're trying to build and integrate Perplexity within Paytm. My first question on this aspect is that Indians, be it in India or abroad, have a very close relationship with money; we are being very careful. Is the data—once you integrate—our financial habits, our expenses, or whatever, would it be safe or would it be shared with Perplexity as well? That's my first question.
Well, the short answer is that obviously the question that you ask to the model goes to the model and then gets answered.
That question part is okay. I'm perfectly fine with that because I'm the one who is asking it, right? So I have to give the relevant information for me to get a pertinent result which can be used. I'm asking about the historical data, because now it is integrated with Paytm.
Oh, no, none of that data goes. This is a perfectly independent shell transaction.
Okay, so none of that data is in sync. Awesome, that's very helpful, thank you. My second question actually is mostly to Aravind. Where do you see Perplexity AI in its near future? What is the goal over here? Apparently, it's like a search and gives good search results along with references, I understand that. But where do you see it going? What is the area that you are trying to exploit in the near future?
Yeah, thanks for asking that. So, I think AI-native search, which is just web sources and LLM-generated answers and summaries, all that is going to be, in my opinion, a commodity. So the frontier right now is reasoning and getting all these models to break down harder questions into multiple parts, use hundreds of sources, doing a lot of research on the web, and make such an AI that essentially is a research agent on whatever topic you want, close to an expert level on any topic, widely accessible. That's the hard part, and that's what we're really trying to push as far as answering questions goes.
And I think the next set of challenges are going to be around making it work on multiple different verticals with not just data from the web, but also proprietary data providers, like merchants for shopping or hotels for travel. All these are different verticals where, even though there are a lot of links on the web, it's very difficult to know which one is really trustworthy for the user, reliable enough for booking, has items in stock, or can deliver reliably within a few days. All that stuff is a lot of domain knowledge that you need to incorporate into building a great reliable agent for everybody.
These are the challenges that we are trying to solve through our shopping or travel experiences, or finance—like we're integrating with data providers like FactSet, making sure people not just get what's the stock price, but reliable charts. All that stuff is something we're working on. And then obviously the buzzword: agents and actions, assistants trying to do stuff for the user, not just give you an answer, but actually do work for you—booking Uber, booking a restaurant reservation, playing some songs or preparing a playlist for you, doing tasks on the browser for you. We announced that we're going to be releasing a browser in a few weeks from now. So all that stuff is where we are putting all energy in for 2025.
Awesome. Thank you. Ayush, your question.
Hello, sir. Hi. Just a question on when it comes to integration of AI on the Paytm app, for example. I just want to get a sense of how can we make sure or how can this evolve to sort of bring in more actionable insights? Like for example, if I have to ask it, "How can I improve my credit score?" How does that evolution happen from search to more in-depth uses?
Local data training, local nuances—this is exactly what we are going to work together and task towards, because credit score for us is different than credit score for other countries, and how does it impact in different local geographies? So in my opinion, just like payment is a very local product and country-specific, so will the financial services product and AI intersection of that be local. So yeah, keep tuned in; that's exactly what we are headed for. Today, we are trying to learn what people are asking so that we can build the product in that direction. Some of the things that you can soon expect us to launch would include AI-first experiences in the app, which wouldn't have been possible if there was no model behind the scene to answer those questions.
Thank you. Pritam, go ahead.
Hi. So right now, when we have to buy a Perplexity membership, we need a US credit card. Is there any plan to do it via Paytm itself, if I could use this Paytm UPI to purchase the Perplexity membership? And also something around purchasing power parity where for Indians it's cheaper than the $20 price right now?
First part: yes, soon. Second part: Aravind, what's your plan for pricing for India, any special...?
Definitely, we're figuring out a way to make the Pro plan widely accessible to everybody without them having to pay the $20 equivalent. Give us a few months or even a few weeks to figure out, because things are moving really fast, so I don't want to give you a poor timeline here. I think the UPI or Paytm payment fundamental bottleneck there is honestly that we work around with the Stripe APIs, and I think it's pretty difficult to accept those kinds of payments right now. I hope we can figure out a solution together pretty soon.
Thanks. Hi, Aravind, big fan of your work, your research, as well as what you did at Perplexity. My question to you is a bit India-specific. There are many use cases in India, things like agriculture and law, where the common man who is not really that fluent in English may have some questions, maybe about agriculture laws or social life and things that happen in India, maybe things like government mandates and policy changes. That could be a great use case where Perplexity could step in, especially considering the fact that you've just integrated with Paytm.
Yeah, absolutely. The point of working together with Vijay and Paytm is that this is just the V0, the beginning of many more things to come in localization and customization for not just the languages, but also the infrastructure that is used there and all the use cases that uniquely exist there. We figure out a way to work together and build solutions together.
