• 8 months ago
Google Search or Wikipedia may be the go-to methods for finding out information on the Internet. Perplexity aims to help you go deeper to find concise answers and explain them in an organized way complete with citations and sources. It is a standout on the 2024 Forbes AI 50 list, with $102 million in funding, including investment from Jeff Bezos.

In this video, Forbes staff writer and AI 50 list editor Kenrick Cai sits down with Perplexity cofounder and CEO Aravind Srinivas to discuss Perplexity's rise and goals for Internet search.

Read the full story on Forbes: https://www.forbes.com/lists/ai50/

Subscribe to FORBES: https://www.youtube.com/user/Forbes?sub_confirmation=1

Fuel your success with Forbes. Gain unlimited access to premium journalism, including breaking news, groundbreaking in-depth reported stories, daily digests and more. Plus, members get a front-row seat at members-only events with leading thinkers and doers, access to premium video that can help you get ahead, an ad-light experience, early access to select products including NFT drops and more:

https://account.forbes.com/membership/?utm_source=youtube&utm_medium=display&utm_campaign=growth_non-sub_paid_subscribe_ytdescript

Stay Connected
Forbes newsletters: https://newsletters.editorial.forbes.com
Forbes on Facebook: http://fb.com/forbes
Forbes Video on Twitter: http://www.twitter.com/forbes
Forbes Video on Instagram: http://instagram.com/forbes
More From Forbes: http://forbes.com

Forbes covers the intersection of entrepreneurship, wealth, technology, business and lifestyle with a focus on people and success.
Transcript
00:00 What AI do you use? How do you use Perplexity in your day to day and what other tools do you use?
00:04 Perplexity I use it more as a search tool. I still don't like using AI for writing.
00:08 Makes sense.
00:09 So anything I write, memos or messages, I just write it myself.
00:14 I don't like auto-complete style of writing. It feels very unnatural.
00:20 Hi everyone, I'm Kenrik Kai. I'm a staff writer at Forbes
00:27 and I'm the editor on our AI50 list of the most promising artificial intelligence startups.
00:32 I'm joined by Arvind Srinivas, the co-founder and CEO of Perplexity,
00:38 a newcomer to our list this year. Arvind, thanks for taking the time to chat with me.
00:42 Thank you for having me here.
00:44 So you worked at OpenAI, you worked at Google and then you started this company Perplexity.
00:51 That's right.
00:52 When did you start it? Why did you start it? What's the origin story?
00:56 Yeah, the company was started in August 2022.
01:00 The origin story is that I was always interested in doing a startup ever since 2019
01:08 when I came to Berkeley for a PhD and got a chance to do a lot of internships.
01:14 And AI research was still at a point where it was not ready for products.
01:20 But that started changing in 2021.
01:24 We started hearing rumors that there were startups like Kotb, AI and Jasper
01:29 making a lot of revenue by building on top of OpenAI models.
01:33 And DALI was one of the first things I saw where it was actually a research project,
01:37 but then when you put it out, it seems like a product that people can use.
01:41 OpenAI's image shader.
01:42 Yeah, exactly. And GitHub Copilot was just a model, but then it became a great auto-compete product.
01:49 So it was very clear that until then, the thinking about doing startups was like,
01:56 you got to go build a business and then sprinkle AI after you build a business.
02:01 Now the thinking changed that you can build AI natively as a business.
02:06 The model itself can be a business.
02:08 And so I felt like I could finally have a shot at this game.
02:14 But the thinking was more like, let's try to go build a product,
02:18 not to become another foundation model company.
02:21 And so there were investors like Elad, Gil, and Atfried who were pretty interested in funding such ideas at the time.
02:28 And note that this was also a time when OpenAI was not even interested in building a first-party product.
02:34 It was more an API-shedging team.
02:37 So that was the amount of wisdom at the time.
02:39 And also success from other people outside OpenAI,
02:43 like Mint Journey or Jasper or Copy or CoPilot from GitHub,
02:49 all made it clear that you could use these models and build amazing product experiences.
02:54 So we started off with that ambition to do something like that.
02:57 I was always a big fan of Google and Larry and Sergey.
03:01 They deeply inspired me a lot because as a researcher,
03:05 I always wanted to find examples of entrepreneurs that were similar to my background.
03:09 And all the typical examples like Zuckerberg or Gates or Jobs were all undergraduate dropouts.
03:16 So I couldn't really identify with their story.
03:20 And so there was always a deep desire in me to be part of the journey
03:26 where a research background helps you to build a product in a uniquely differentiated way
03:32 than what a typical consumer product builder would do.
03:35 And it turns out that somehow when you think of all these ideas,
03:39 you converge to search all the time.
03:42 That's how it was.
03:43 We started working on so many LLM-based product ideas, data analysts, AI data analysts, and things like that.
03:48 But then we would converge to just searching over your database,
03:52 searching over your tables, searching over your spreadsheets.
03:55 And at some point, we realized that, hey, why are we restricting ourselves to so many narrow domains?
04:01 That was an investor ask to make us focus.
04:04 What if we try to be a little more ambitious and just say we search over the whole web?
04:08 And that ended up becoming our core product idea.
04:11 So tell me a little bit about the product, Perplexity.
04:14 So it's a sort of search engine without the links or without the links so front and center.
04:20 NVIDIA's CEO, Jensen Huang, Shopify CEO, Toby Lutke have talked about being users of it.
04:28 What is Perplexity? Why is it caught on with business leaders in a way that it has?
04:32 Yeah. So Perplexity is an answer.
04:35 You can go here, directly ask a question and get answers.
04:39 And the answers are supported with citations, references,
04:43 almost like how you, Kendrick, you would do your job, like you say, according to New York Times, so and so,
04:48 according to a few people I've talked to.
04:50 You're having your sources.
04:53 That's why, actually, we have some of the background, like in academia also, when you write a paper,
04:57 every sentence you write in the paper should be backed with a peer review citation to write.
05:03 So the same principles we have.
05:05 Now, how would it be if we baked in that principle into an AI chatbot,
05:09 which otherwise would just say whatever it wants based on what is memorized in its models and its weights?
05:14 This is how ChatGPT launched.
05:16 So we took a different stance.
05:18 We said, look, when humans are interacting with these kind of mediums for their information access, knowledge access,
05:24 it's best if it's accurate, as accurate as possible and as real-time knowledge as possible.
05:31 And the only way to do these two things is to make sure it behaves like a journalist or an academic,
05:36 almost like it always goes and does the background check and then comes back and writes the answer.
05:42 So that's why it's a very fact-based answer engine.
05:47 The difference from Google is you just can converse with it and ask questions and feel like a natural conversation.
05:52 So let's say you're asking me questions right now.
05:55 You're not asking me keywords.
05:57 So there's a fundamentally different mode of interaction.
06:00 And we felt like the search paradigm could finally change.
06:05 When people get used to this new way of asking questions, I remember an episode very viscerally,
06:12 which is I was in this AWS flagship conference called reInvent, and a person from AWS was there next to me.
06:19 And he was like, "I'm pretty sure Jensen invested in Corvue."
06:23 And I corrected him, "Hey, Jensen did not invest in Corvue. NVIDIA invested in Corvue."
06:27 "No, no, no, I'm pretty sure it was Jensen personally invested in Corvue."
06:30 I was like, I took up the Perplexity app and just said, "Who invested in Corvue? Was it Jensen or NVIDIA?"
06:35 And Perplexity gave me the answer in like two seconds.
06:38 How are people using it? Are they using it instead of Google? Are they using it in addition to Google?
06:43 So people use it in all sorts of ways.
06:46 Some people use it as a replacement, but that's not the majority.
06:50 That becomes a narrative, right?
06:52 And then, of course, there are skeptics of that narrative. There are believers in that narrative.
06:57 But our success doesn't rely on Google's failure at all.
07:02 Like people can use Google and people can use Perplexity at the same time.
07:07 All they need to do is pass the Larry Page toothbrush test, which is...
07:12 Larry Page has this famous principle for whether Google should decide to launch a product.
07:16 It's based on whether people can use it at least twice a day.
07:19 I think Perplexity is on the path to getting there.
07:23 It's still not a product that every user of ours uses at least twice a day.
07:27 But there's a path for us to getting there without those users having to stop using Google.
07:32 Literally, all they need to have is two deeper questions to ask.
07:36 Because they can still use Google to quickly get to some subreddit or some quick Amazon site or something like that, right?
07:44 You don't have to use Perplexity for that.
07:46 If you're just navigating, if you're just going to booking.com, if you're just going to airbnb.com, no need to come to us.
07:52 In fact, it'll be slower. Our sources will render slower.
07:56 Instant 200 millisecond load time of Google is never going to be matched by us.
08:00 But if you want to know how do I decide between Airbnb and Vrbo, those are questions that we can answer.
08:07 I see. I see. And what does usage look like right now among the general public?
08:11 We get many millions of queries a day. Single digit millions of queries a day.
08:16 Which is actually a big deal because the first day we launched, we only got 2,000 queries.
08:21 When did you launch?
08:23 December 7, 2022.
08:25 Got it.
08:26 A week after chat.
08:28 A year and a half.
08:29 Yeah, it's been like one year, three months, one year, four months.
08:32 So we've been growing a lot in usage and that's also a North Star metric.
08:37 Yeah, I'll tell you why.
08:38 Tracking the number of users is not a big deal.
08:41 Somebody could have just come checked out your product and never came back,
08:44 but they checked out the product twice in a month or something like that.
08:48 Actual number of queries is what most of you do for you as a company because it tells you usage and it also creates the data fibing for you.
08:55 You can go look at queries that people get wrong and improve on them.
08:59 You can also collect a bigger data set of what pages are important for an AI chatbot to be good at fact-based question answering
09:08 and build your crawlers based on that.
09:12 Prioritize the domains on the web.
09:13 The web is so big, you can never succeed at recreating the whole index like Google.
09:19 It's too late. You're starting two decades later.
09:22 But what you can succeed at is all these great parts of the web that are needed for day-to-day question answering.
09:28 So what does success look like to you then?
09:31 Is it becoming a $20 billion company or are your ambitions different than that?
09:37 I wouldn't say it's based on a market cap.
09:40 It's more that we want to change the paradigm.
09:43 User habits, you change it, that's the hardest thing.
09:47 I think there's a saying that never bet against user habits.
09:51 I think we want to get to the Larry Page benchmark, like toothbrush test for AI answer engines.
09:58 At least twice a day.
10:00 Let's start with once maybe.
10:02 I think it's even less ambitious than that.
10:05 I think if once works a day, then we can go from there.
10:11 We can really go far, build new business models around this.
10:16 Google is not just great for the search innovations.
10:21 It's also the fact that AdWords is one of the greatest business models in human history with such high margins.
10:28 And keeps on scaling as the graphic scales.
10:31 What is the equivalent of that for AI answer bots?
10:34 Nobody knows.
10:35 What AI do you use? How do you use perplexity in your day-to-day and what other tools do you use?
10:40 I use perplexity every day.
10:42 Not because I'm testing it all the time, but I'll do it more than twice a day.
10:46 I at least have 20 queries a day.
10:49 But three or four are probably for testing.
10:52 Ten probably just regular usage.
10:55 Today morning I looked up some stats on companies that are raised more than $250 million in funding.
11:02 And yesterday I looked up history of all the funding rounds of Facebook.
11:07 It's sort of the things that require you to open six or seven different links, take the information, bits and pieces from everywhere, and then collate and summarize.
11:14 But the speed at which perplexity gives me the answer is insane.
11:18 I used it a lot in the past.
11:20 I proposed to my fiancé then.
11:24 And then we had to look for a place that had a diamond ring in an affordable range.
11:30 I hear you say you proposed using perplexity in some kind of way.
11:34 I actually did use it.
11:37 When I was going to Jackson for winter, I wanted to find the right shoes that wouldn't make me slip.
11:48 Because I'd watched tons of YouTube videos where everybody scams in the shoes and you just start slipping.
11:53 And I've had a bad history in the past of slipping in snow.
11:56 And so I can ask very targeted queries of what is good.
12:00 So these are the kind of things perplexity is amazing for.
12:03 I think chat GPT, I use it sometimes for the Dolly Tweet feature.
12:07 If I'm making slides and I want some illustrations, it's easy to just get it done.
12:12 Mid-journey I've used for some design, like, you know, iterating with my designers sometimes on image generator.
12:19 Like brand design.
12:21 I actually don't use much for writing.
12:23 I actually write everything myself.
12:25 Even now, I don't like auto-complete style writing.
12:29 It feels very unnatural.
12:32 The soft perplexity totally changed your work.
12:34 For perplexity, I use it more as a search tool.
12:36 I still don't like using AI for writing.
12:38 Makes sense.
12:39 So anything I write, memos or messages, I just write it myself.
12:44 Got it.
12:45 Well, thank you so much for the time, Garvin.
12:47 Really enjoyed talking.
12:48 Thank you so much.
12:49 Good to see you.
12:50 [end]

Recommended