How Perplexity is revolutionizing the way people find information

Not a lot of startups aim directly at Google’s core business. The most obvious reason is that it’s usually not a good business strategy to go head-to-head with one of the biggest and most successful companies of all time. But most people also consider Internet search a solved problem. 

Aravind Srinivas didn’t see it that way. As a PhD student, he drew inspiration from an old interview where Larry Page described search as one that would require powerful artificial intelligence to truly get it right. Obsessed with learning and knowledge, he and his co-founders decided they were up for the challenge. 

Today, just two years after it was founded, Perplexity, a conversational search engine that uses LLMs to understand queries and summarize answers, has 10 million monthly active users. 

Ann Bordetsky, NEA’s lead investor in Perplexity, has been working with the company for most of that time. She spoke to Aravind about how he became interested in search, the challenges of running a startup, and how his reading habits have changed. 

Ann: When we first met it was clear you were on a mission to make knowledge more accessible. Where did that spark come from? 

Aravind: I come from a background where being smarter was more valued than being rich. When I was in second grade, I would be able to tell you the capital of any country on the planet. We didn't have the internet until I was in 6th or 7th grade, so I actually had the physical encyclopedia, all the books of Britannica. And when Wikipedia came about, I'd spend a lot of time rabbit-holing. So that was my upbringing. 

Ann: What inspired you to start the company?

Aravind: I've never built products before, but AI has changed what it means to build a product because most of the magic is happening through the AI layer. That is, if AI gets better, the product of the company should also get better. And as the product of the company gets better, more users should use it. And by more users using it, the underlying AI gets even better and it becomes a flywheel that keeps fueling itself. 

One interview of Larry Page that's very old inspired me a lot during my PhD days. He says that search is an AI complete problem, meaning you can only truly solve the mission of search, which is giving people information in the most efficient way, if you fully solve AI. Sometimes you even have to give people what they want without them even having to ask it precisely. And so that inspired me to identify a company mission that is of similar nature to Google, where that company mission is solved only if AI is solved.

Ann: Historically, we just hadn't seen many founders have the courage to question whether the Google experience was the best possible experience that a user could have. 

Aravind: Perplexity is a marriage of Wikipedia and ChatGPT. You get answers, you converse. It's almost like you're talking to a person. So that's a ChatGPT part. The Wikipedia part comes from always having to source what it says from the web, and citing what it says with appropriate footnotes. 

Ann: When I think about what makes Perplexity different, just to add to what you were saying, there’s a commitment to simplicity in the product, almost in an Apple-like way. 

Arvind: We’re really obsessed about the correctness of an answer, how fast the product is, and how readable an answer looks. We’re constantly thinking of the user, trying to earn the user's trust, and working hard to retain them. The whole point we exist is because another company went [stopped focusing on those issues] to the extent that people got tired of the quality and now there's a new chance for somebody else. 

Ann: Speed is really the weapon that the startup has against an incumbent. And you probably ship features and product faster than most companies and you're maintaining that velocity as you continue to grow. 

Aravind: I think the challenge has always been keeping up the pace and relentlessness. We have 60 people or something right now, it's very different from when you have four people. When we raised money from NEA we were like five people. Back then a bug report would be fixed in an hour. Now it goes to days. When there are a lot of people working on one code base, you can't just change it so frequently. So trying to preserve that fast-moving nature, along with having the stability of a reasonably mid-sized company is challenging. 

Ann: Getting perfection and speed at the same time is really, really hard. 

Aravind: The only thing that tires the human brain is constant contradictions. Because the brain just likes to converge to some version of truth. But a startup is all about contradictions. You want to move fast, but you also want to stabilize and grow. You want to build a business, but you also want to have fundraising and enough runway to keep trying to build up a great business. And you want to exploit what's working, but you always want to have the hedge when a big company does the same thing as you. And in AI especially the world is changing so fast it's very hard to build a business, build a product, build moats all at the same time.

Ann: What do you not want to change about Perplexity as you go on to conquer the world, and scale the company and the team? 

Aravind : I mean, there's a whole phrase for it. It's called N-stratification, when things get bigger the quality degrades. That happens to every company, every app that we've loved using in the beginning. Often almost degrades in quality and service because they've captured that market and outspent all their rivals, and users don't have a choice anymore. So the thing I want is to make sure that never happens.

Ann: What is your company culture?

Aravind: Culture is what people implicitly do every day. You’ve got to be authentic. Our product has three evergreen goals — accuracy,  speed, and readability — and therefore our culture is the same way.

If you want your product to be accurate, truthful, fact-fully accurate all the time, you must be yourself. If you as a person are trying to be truth-seeking, then you would obsess about hallucinations and you would go fix it. So we're truth seeking. And we're fast obviously, that's when you really care about every minute, millisecond latency that you can improve if you generally are impatient, and try to get things done faster. 

And then the last thing is the readability. The whole design aesthetic. It's not enough to have a chatbot and a bunch of texts being streamed. You have to really care about how the user is consuming that information that you don't want to just put an answer, put some panels, put some links, put them ads and everywhere just clutter the whole UI. You want to be thoughtful and opinionated on what makes for a good reading experience. And that works at the level of what is the color shade of white for the light mode or shade of dark for the dark mode? Every single detail matters, pixel matters.