Perplexity's Origin Story, Open Source AI, and the Shift from Links to Answers

our mission is to like you know really transition the world from links to answers and build the ultimate knowledge app obviously there's a lot of people that are worried about content creation if I'm writing blog posts or creating podcasts or making YouTube videos and these chat bots in the future will just sort of answer the question does it sort of disincentivize content creators to keep on creating content I'm curious on your thoughts on this this is a technology we never even knew we wanted but now that we have it we just like don't want to go back hey welcome to the next wave podcast my name is Matt wolf I'm here with my co-host Nathan Lans and we are your Chief AI officer it is our goal with this podcast to keep you in the loop with all the latest AI news the coolest AI tools and set you up for success in this next wave that we're entering into with the world of AI and technology today we've got an excellent show for you we've got the founder and CEO of perplexity on the show Arend serenos we had a fascinating conversation with Aran and he told us his story how he went from growing up in Chennai to moving out to California and going to University at Berkeley he's worked at Google deepmind he's worked at open AI he's he's got a pretty impressive resume and now perplexity is one of those companies that all the Venture capitalists in Silicon Valley are chasing after and just throwing a ton of money at because they love the idea and we want to dissect that and break that down in this episode it's probably the best way to do research with AI actually I've started using it to do research for the episodes yeah when you look at the current AI landscape right you've got these large language models like gp4 and anthropics Claud and Gemini and you've got all of these large language models that are sort of put in the form of a chatbot like we see with chat GPT and Claude right perplexity took a little bit different of an angle on it and they wanted to make sure that a they were searching the web and B they were citing all of their sources and they were really sort of the first AI chat query platform that started to site the sources and share where they actually found the information they have a really really great Chrome extension where you install the Chrome extension and it will essentially search anything on the site or domain that you're on I found that to be super super helpful so I love this approach perplexity took and Nathan you and I were talking right before we hit record about how they're totally agnostic to the actual underlying large language model yeah which means they're also like a huge supporter of Open Source which obviously I'm I'm a big fan of because I don't like the idea of just having one company or two companies that rule the future of AI so it it was great to hear from Aron like his his thoughts on open source and and how perplexity he's kind of supporting open source by being agnostic I mean it's it's a cool place to be right because you can go use perplexity and you don't have to worry about all right what is the best model out there right now I mean people like Nathan and I were constantly going all right CLA is marginally better than shat GPT so let's use that now instead right we're keeping our finger on the pulse of that kind of stuff but if you use something like perplexity well it's always just going to use the most beneficial model for what you're trying to achieve yeah so on this episode Arand is going to break down his entire story about what perplexity was before it is what it is now and when he started it was completely different so he's going to break down that whole story arc for you of how it started and how it got to where it is today we talk about the current state of AI we talk about all of these devices that AI is getting rolled out into you're going to learn about the past present and future of AI and how perplexity is firmly placing themselves in the center of all of it so let's go ahead and jump in with aravan serenas Erin surova thank you for joining us today on the next wave thank you for having you Nathan Matt you know I've been a big fan of perplexity since the beginning I think when I first saw you Tweeting about it and tried it out and was blown away so I think it' be it'd be useful to know like how did you get perplexity started like you know from starting out in India to now having like one of the hottest startups in Silicon Valley like what what was that journey I me by the way I think it's better to tell the true story than like something that's retrofitted to you know make it look like a much better story for PR yeah okay yeah yeah yeah yeah the true story is great look I I never intended to start any company but then there was this movie I watched deeply impacted me called the Pirates of silic Valley I don't know if you guys have seen that movie yeah I have yeah it's one of the most authentic portrayals of steam shops and Bill Gates and Microsoft and Apple yeah and I was like okay I really need to be at Silicon Valley it's fantastic and then did not have enough money to go do a masters myself so I thought okay someone else has to pay for you so why why don't try for a PhD and okay the best way to do PhD is to get started in some kind of research and and established like a Tracker card so went to a professor at IID and said hey like can you help me do research and he was like yeah you know that there's this uh paper called Atari games AI like there's this company called Deep Mind that training a had to play Atari games why don't you try to reimplement that whole paper so like he got me excited about all these ideas like transfer learning and hierarchical learning and things like that and like I wrote wrote A Few papers with him and that got me an admission in UC Berkeley for for doing a AI BHD and there I did a little more work and like open AI uh noticed my work particularly this guy called John schillman he's the guy who in basically like the research inventor of chat gbt at that time he was doing research more on our and he invited me to do an internship and until that point I was kind of like on a high I was thinking I was really doing well writing papers like coming all the way from India here and I when I entered open a I was like damn like a whack on your face like so humbling the people here are like Star Wars like Superstars all of them are really amazing like like talented people but it was not a very stable organization at the time I it's probably never been stable for what right so then I got to work on all these unsupervised generative models got an internship at Deep Mind and that's where I think I got the entrepreneurial Ambitions because I always wanted to start a company like that where I knew I would not be successful starting the next like Instagram or Tik Tok anyway even if like luck was on my side cuz I don't have the skill set of like hacking the dopamine of people so my skill set is more okay thinking about like some problem more deeply and trying to see what we can do with some research but quickly ship it to product that was The Sweet Spot I was trying to get at and Google is a great example of that so that was very motivational that doesn't mean I wanted to start a search startup it was just like motivational to try to start a company in that fold and you know one thing led to another I tried you know this TV show slick and Valley right uhhuh you won't believe it I actually thought it was for a com like men for comedy but people told me it's be real I I lived in silon Valley for 13 years it's uh it's definitely uh real I was I was in Berkeley I was in Berkeley right so I was on very well connected silen Valley so I I thought the show was just meant for laughs but people told me dude don't laugh at this I cry watching it because it's too real and it reminds me of my own life and then I was okay fine like this compression generative models all that was like like amazing try to start convince people to work with me nobody wanted to do any company something that you realize as a Founder is like every time you go to your friends and say let's start a company either over drinks or coffee doesn't matter all of them would say hell yeah let's do it and then you just forget about it yeah people don't realize how how hard doing a startup is I think you it's real and you just say yeah I've started it this is the company are you willing to join whether you join or not join I'm going to I'm going to do it and that's when people wait is this real is this serious and then they like spooked and interested right so the reason you're not having co- farers people don't think you're serious enough anyway so all that like one thing led to another and like uh pitched the stupidest idea to one of my First Investors saying hey like we need to disrupt search so it's hard to disrupt Google through the text form factor so so how about we just go through the the vision form factor to the vision pixels so imagine we wore a glass and we all saw this and then we could just ask questions about whatever we seen and he was like okay all this sounds cool but look you're you're like literally one person right now and like you're not going to be able to execute on this yourself start focusing on more narrow things get a team and then try to build up towards this so that was a very good advice given to me by this great investor named elot Gil oh El and then like Nat Freedman and Ela decided to fund me they were like okay look you're from open you have all this like you've done work at Deep Mind research you understand these things but again like you don't have any idea you don't have any product so you're going to give you like one or two million to play around and tinker and we'll see what happens and then I take that money and like they start focusing more on like searching over databases like searching over your own spreadsheets searching over csvs asking questions about like data sets and that was fun like as a data nerd I really loved it and we got like my co-founders Dennis and Johnny joined the you they all excited to experiment too and then like we went to Enterprises and said hey dudes like we have this thing us to show demos what if you gave us your data and we power search over that you just upgrade the functionality for your users like you went to websites like pitchbook and crunch baste and like and all of them would listen to us watch our demos and be like our engineering teams can do this man so thank you and we would feel like really depressing every week where like we would keep do dem was and nobody wants it and then I one day I just realized nobody cares about like a three person startup they think they can do it themselves they don't value you and it's fair it's fair like you've not earned that their value yet I'm sure this will be useful for bigger companies but they are never going to talk to us if these smaller companies don't talk to us like let let’s earn the attention of the bigger guys uh by doing search over public data sets that are really big only then they'll get convinced that we can handle large databases and so we started scraping Twitter cuz I really I mean obviously we all like Twitter we are all using it yeah right yeah X as as it's called today and zck dorsy had Twitter API El Elon also has it but Elon basically charges so high that impossible lose it now but Jacky had the Twitter API and if you're an academic access accounts you can just scrape a lot of tweets every day so we we would just create these academic access accounts and uh non-commercial use and we keep scraping social graphs and tweets and then we would power search over that we power search over like oh like how many followers does Nathan have that Matt is also following or like what are the tweets of Matt that Nathan has liked in the last 10 days or like which of Nathan's tweets as Elon Musk uh replied to or stuff like that right right and and then you can see you can sort of like it it’s fun like these sort of social searches and like or like like tweets about AI or like tweets about like like 3D diffusion models that like like the Mt has tweeted about you can do a lot of these searches that current Twitter just like really sucks at right and once we built this demo we showed it to a few people like like like gun and like Jeff and like they were all like Blown Away by that damn like this is a completely new experience like okay large language models can help you build new search experiences there was never never possible before and they invested in us and then we used that their investment as a credibility to like attract some engineers at least two Engineers joined us after that saying hey okay like look you guys may not be well known but looks like you got funding from some top people so you won't be like randos you know at least I can trust you like work with you guys and uh the demos are actually really impressive so let's work together and then we went to these bigger companies and said hey like look these things are working they want to work with us now they would at least like take our meetings more seriously and say okay you know what these are all our problems like what do you want to do so that was the stage we were in and then one fin day we were like hey like this part of talking to companies and like trying to sell solutions that was not even fun so why why why don’t we just like search over the whole web like make the LM just look at the links take the relevant parts of the links and then let the llm do all the reasoning in terms of whether it has to return a table or a paragraph or citation or whatever and then we built a little more General solution like I think Paul Graham talks a lot about this like how often when you realize a simpler way to do something like it becomes a big unlocked for you and then one weekend like we prototype this idea of just taking the links and summarizing them with citations and then it was working reasonably well and 3 days before chat GPT got released open AI put out this the wici 3 Model yeah and that model just made the summarization so much better better that we were like damn this is truly a big deal it's an inflection point AI everyone’s like this is a technology we never even knew we wanted but now that we have it we just like don't want to go back and they did one thing which is they don't have they had a knowledge cut offly and they don't have citations they don’t have like grounding in facts so there was a space for somebody else to come and put a fact grounded citation powered answer bot and we already had it s like we had to build it in like three days we already had it we just had to put together a web frontend and then we got it really quickly we sent it to a few investors I remember my first feedback was from this guy I really respect Daniel Bross and he said Arvin this is cool you should not have it as as in like it's not a hit button for your query it's a submit button because it's that slow takes like 10 seconds to get an answer so almost like I'm submitting a job so you should call it a submit button and have a queue of queries or something from there on words to now being asked like how is the service so fast that is the progress right we have made not just because of we our own engineering team which is amazing also the fact that chips are getting better faster cheaper models are getting better faster cheaper and we made a bet when we launch there were similar other services in our space that were also launching a mix of search and llms but we are the only ones who had the conviction that it should just be answers the links are only in sources others are like I still want to have the link 10 Blue Links I want to have a site bar with a chat bot I I want to have a summary panel at the top I don’t want to like change it too dramatically and we like dude if you don’t change it dramatically no one’s going to really realize you’re different from Google they’re just going to think you’re Google with some add-ons like that’s not exciting you have to be truly differentiating that okay even if you’re worse than Google even if you’re slower even if you’re suck at navigational queries that people go back to Google they’ll at least register in their minds that you are better than Google on certain things which is like actually asking a question deeper research and they’ll come to you for answer engines they’re not going to come to you for search engines anymore they’re not going to come to you for product comparisons they’re not going to come to you for like ordering San Pino right they’re going to come to you for asking whether San Pino or lacro or should I what should I get it’s going to register in their mind why you’re different and better so that is a position we took we had conviction that like even if we got answers wrong even if people made fun of us for you know like hallucinations over time all these problems will get much better and that ended up being one of the best decisions we made to be called us an answer engine instead of like search engine with lolms on top and so we being proven right our thesis was correct that this is the right format to interact with information on the web and um that ended up being for black City our traffic has been growing exponentially since we started so then we said okay look we we were initially on this treasure hunt uh trying to figure out some product that would resonate with users this is the product that it's growing in terms of traction so that let's commit ourselves to building a company let's not be a seed seed around $2 million project let's try to build a company around it and a business around it and so we went and race Venture funding rounds and use that money to like keep growing even more and and that's our current plan our mission is to like really transition the world from links to answers and build the ultimate knowledge app like if people go to perplexity they should just feel smarter every day that's the Vibes we want people to feel we don’t want the Vibes of dancing girls on Tik Tok or like celebrities posting stuff on Instagram we we just want the Vibes of feeling smarter and I think asking questions is a great great way to feel smarter um discovering new threads your friends sharing like interesting queries with each other these are sort of utility values we trying to add to people’s lives I’m curious on your thoughts on this so obviously there’s a lot of people that are worried about like content creation right if I’m creating if I’m writing blog posts or creating podcasts or making YouTube videos and you know doing my best to like SEO them or whatever so that people will find them and these chat bots in the future will just sort of answer the question without me actually needing to navigate to the site and read the article does it sort of disincentivize content creators to keep on creating content if people aren’t like clicking over to their website anymore I’m just curious your thoughts on on that whole argument around it our model of the citation or attribution is I I would say it’s kind of the right model now you can ask like what about these future AI models that are just training on me like as they like joke all publicly available data so I don’t have a h i just don’t we we don’t do that ourselves like we’re not in the business of training these large Foundation models so we’re not like taking the models and like benefiting from the data you create one thing I think n Freedman has said about this I kind of like this it should be okay to train on someone’s data as long as you’re not like literally we badom reproducing it it’s kind of similar like for example when I watch any of your you guys’s podcasts or YouTube videos is it fair to say I’m training on it because I’m kind of like consuming data right but but if I’m like literally taking that ripping it off and like and and creating value out of it uh without giving you any any kind of attribution is like saying okay according to Matt or according to Nathan without saying that if I’m just literally like reproducing your thing word by word that seems problematic and that that is basically the whole core point that New York Times is beinging against openi and I think there’s some response open I was like they kind of over engineered the prompts to show those cases but the deeper deeper point being made is that like there is a potential to just regurgitate content here so what what what happens like like should should the person be giv credit and I think the current Paradigm of like people fighting for licensing deals and trying to make money out of PE uh the AI companies also doesn’t seem like the right solution it seems like a temporary solution a longer term solution is like whatever value is created per query it should be shared by the person surfacing the answer and the site and the sources that ited which is more of the Spotify model right which works so this is the sort of thing I kind of feel all AI companies should subscribe to not being overly greedy because if people don’t continue to create good content on the web through their blogs or tweets or like journalists writing their good essays or YouTube creators making good videos then there’s really no value in your Bot either your Bot is only as useful because it’s surfacing good content from the web and getting into the hands of people who are asking questions relevant to that and if people stop creating good content your B is also not going to be that useful right you do we need a two-way relationship and so instead of trying to be greedy like and I’m trying to create a company that’s eating all the profits like Google did in the previous era if you’re like less greedy and like more long-term Focus like Spotify I think you can create a much better model here and that’s something we are aspiring to do yeah Google just kept getting more and more greedy over time too right like adding more and more advertising Links at the top where now when you do Google search you’re seeing like five or six or seven or eight or results or or they’re like instantly answering the question or they’re sending you to one of their properties to get their first result yeah a lot of people think mistakenly that Google pay everyone some money for being able to use their content in the 10 L link UI reality is not that reality is they don’t pay anybody anything I’m curious on your thoughts about the you know this the whole open source closed Source debate obviously that’s a very hot topic right now Elon Musk is calling there’s that battle do you think do you think the future of like the large language models do you think it’s going to be more open source close Source a combo of both like what are your thoughts on how this is all going to play out I think it’s a combo of both open source will always lack the best close Source model and that’s probably only one company in the world that has the money and the incentives to keep open sourcing models which is meta everybody hates Zuckerberg but that’s the only guy who’s truly committed to open source rest of the people can are all like kind of like proxy open source or like whatever we does no need to make fun of them because every everyone’s trying to do the best they can right like nobody is able to have a cash car like Zach to be able to like spend so much money and yet open source at all and like give away the benefits because unlike Google he doesn’t even have a cloud business he doesn’t want to have either he’s like I don’t care I just want to like make more ad revenue and and so he has the incentive to just give give it out and own the ecosystem and and and ensure that he profits from the developers who are like building on top so that they engineering can benefit meta and anybody else is not truly committed and I so from so then we should say like when can llama 3 beat gp4 that’s the right question to ask maybe it’s this year maybe you know I I hear they’re trying their best but given that he’s purchased 600,000 h100s it’s inevitable that he beats them right like it’s just a matter of time now then you can say okay by the time he beats them would Sam have a better model definitely like they’ve already had a year for gb4 and they’ve been upgrading gb4 through the course of the year but they’ve had a year to build an even better model so most likely there’ll be a version of closed SOS either it’s open a or anthropic or Gemini that’ll be better than the best llama at the point but that doesn’t mean close sources is getting destroyed like most people just want to use apis and need somebody else to serve these models right but you don’t have to overly depend on one provider I think that’s the future we want you don’t want to overly depend on one provider and you want the ability to take these models and customize them for what you want to build yourself and if you have if there is a lot of friction in being able to train and deploy your own models because literally you have to get a GPU cluster you have to train things you have to deploy you to do evals like people think like oh yeah I’ll just take this model and I create like fake news bots in the world and I’ll destroy the world or something that’s not how internet works actually uh it’s hard to be BS by the way like there are so many layers of security you need to bypass it like there’s so many solutions to like fighting the Bots problem and fake information problem compared to like say Banning the use of Open Source models cuz if the more you block people from having access to powerful technology even more motivated they’ll be to like get access to it you you know the news of how this Chinese engineer was like leaking all the details right from Google and like having somebody else badge him when around view office so this is what’s going to happen if you go too much on the Other Extreme you know a lot of the the concerns that people have about the open vers closed also has to do with you know some of the bias elements right like the people are worried that if Microsoft or Google or one of these companies is in control well now it’s a big Corporation who controls the narrative that is coming out of these Bots right where open source maybe you could steer it and sort of have your own sort of biases preferences whatever inside of the model 100% I think this scary to have like one company that then in the future determines what was human history and they’re like telling you the answer and like and it’s it’s not exactly the truth it’s like some modified version of the truth that fits some agenda that they have close source is going to continue to be way ahead like arvan said but I’m glad open sour is there we definely need Alternatives so it’s not just onean ring everything yeah so what do you think you mentioned uh Zuckerberg real quick I’m just curious on your thoughts on this you mentioned that he’s incentivized to open source it what what is the incentive for meta to be open sourcing it we don’t have to think about anyone as altruistic or like a good or bad person uh just purely capitalistically it’s in this incentive that other engineers build on top of llama than than gpts so that like uh engineering people do in the open source ecosystem meta can learn from that and like use it in their products like if you can see how other people take llama and like make it faster learn how to like fine-tune it get it deployed on the edge devices like like learn how to personalize these llms with like very limited parameter efficient fine-tuning all these are like algorithmic benefits that meta can just look at what people are doing in the open and and put it their products instead of saying oh I’ll hire all the best engineers in my company and then like only rely on their own like brains to do these things because you want the whole ecosystem to benefit faster right and and and and you also benefit from the ecosystem benefiting and you have the cash C you have the user base to like you know go and deploy all this at scale he actually benefits a lot with putting it out and like letting other people build on top now there’s the other argument that I believe he’s making aresty but people can be skeptical of of his true intentions that he’s saying this if you really care about safety uh you you rather want as many eyeballs on it you can’t be the person who comes and says we need to make all this safe this could go really wrong and dangerous so you better trust like these us four or five people in the world who have like all these like billion dollars of funding and like tightly tied to like Microsoft or or you know Google or Amazon and like you know we’ll decide what is good for you like you rather have as many people have access to these things right if it is truly dangerous you would rather have as many people be aware and educated and having access and like trying to be able to have opinions about it right cuz that way even if somebody’s misusing it you at least know how people can misuse things right and and and and that way you’ll be able to build guard RS against it instead of just saying trust us and we know what we’re doing well this has been an amazing conversation everybody needs to check out perplexity doai there is a a free version that you can use of it there’s also a premium version I’m on the premium version I’ve also have a rabbit R1 coming so I’m excited to play around with that with perplexity on board is there anywh that you want people to follow you maybe on Twitter Youtube something like that where do you want to send people after listening to this episode we black City uncore AI That’s our Twitter handle and mine is a AR Shas a r a v s r i n i v as s very cool well thank you so much for uh spending the time with us today and answering all of our questions and and hanging out out with us and uh yeah it’s been a great conversation thank you thank you [Music] Aran.