‘Like Wikipedia and ChatGPT had a kid’: Inside the buzzy AI startup coming for Google’s lunch

Innovation

Perplexity, an AI-powered search engine, is backed by tech VIPs like Jeff Bezos, and counts billionaires like Nvidia CEO Jensen Huang among its frequent users. Now, its early traction puts it on a collision course with the search giant.
Aravind Srinivas, CEO and cofounder of Perplexity CHRISTIE HEMM KLOK FOR FORBES

InAugust 2022, Aravind Srinivas and Denis Yarats waited outside Meta AI chief Yann LeCun’s office in lower Manhattan for five long hours, skipping lunch for the chance to give the NYU professor a demo of their AI program. Once they showed him how their model could search through platforms like LinkedIn, GitHub and Twitter and surface content that GoogleGOOG could not, such as the accounts that most frequently replied to LeCun’s tweets, he was impressed enough with the program’s accuracy to invest.

He was one of several tech VIPs, including Google Chief Scientist Jeff Dean, former GitHub CEO Nat Friedman, OpenAI cofounder Andrej Karpathy and prominent angel investor Elad Gil, who got similar personalized demos — and collectively invested $3.1 million in a September 2022 seed round. “It was very relatable for them to go and search about their own Twitter,” Srinivas told Forbes.

The money gave Srinivas, Yarats and fellow cofounders Andrew Konwinski and Johnny Ho the runway to find an inventive way to bring AI to search. After trying out several ideas, they landed on what is now Perplexity, an AI-based conversational search engine used by about 15 million people to source and summarize information on any topic on the internet — from the best date night restaurants to white elephant gift exchange ideas to the cheapest sneakers for sale online. Featured on Forbes’ 2024 AI 50 list, Perplexity provides succinct answers in four to five sentences along with citations and links to sources by routing millions of questions to a medley of large language models including Anthropic’s Claude, OpenAI’s GPT3.5 and GPT-4 and open source models like Meta’s Llama and Mistral’s Mixtral.

“It’s almost like Wikipedia and ChatGPT had a kid,” said Srinivas, 29, who serves as Perplexity’s CEO.

In less than two years, the buzzy AI startup has raised $102 million in venture capital from some of the most notable names in tech, including AmazonAMZN founder Jeff Bezos, former YouTube CEO Susan Wojcicki, Gmail inventor Paul Buchheit, ShopifySHOP CEO Tobi Lütke and former MicrosoftMSFT President Bob Muglia. The roster of prominent backers has helped the startup, now valued at $1 billion, gain credibility and momentum while attracting top talent, more investors and most importantly, millions of users — among them billionaires like Nvidia CEO Jensen Huang, who uses Perplexity every day, and Dell TechnologiesDELL CEO Michael Dell.

“It creates snowball effects for you,” Srinivas told Forbes. “People take you more seriously.”

“You can never succeed at recreating the whole index like Google. It’s too late… It’s sort of like finding your way in a maze where you’re starting at a huge disadvantage.”

Aravind Srinivas, CEO and cofounder of Perplexity

Since launching in December 2022, Perplexity’s usage has been on a steep upward climb. Some 100,000 users pay a $20 monthly subscription fee to access advanced features on the platform such as searching their own uploaded files and generating images and text from scratch. Perplexity users can also create their own “AI profile” by adding information like their occupation, location, likes and dislikes to get personalized answers and suggestions. It also lets people restrict their searches to specific databases including academic journals, YouTube and Reddit. The startup, which has about $20 million in annual recurring revenue, is considering adding native ads into the product by letting brands influence some of its “related questions” suggestions.

But it still faces the trillion-dollar behemoth that is Google — the first place billions of people go looking for information. There’s still a lot of questions that Google answers better than AI search engines like Perplexity, like suggesting what shows or movies to watch or accurately answering questions about a recent football game. Plus, the tech giant has a two-decade head start on indexing and scraping the web. “You can never succeed at recreating the whole index like Google. It’s too late… It’s sort of like finding your way in a maze where you’re starting at a huge disadvantage,” Srinivas said.

But Srinivas is optimistic because he doesn’t see Perplexity as competing directly with Google. Instead, he hopes that more and more people will turn to Perplexity to find nugget-sized information to make quick decisions rather than being served with 10 rows of blue links. Perplexity especially shines for retrieving information that’s buried deep within different websites, like instructions on tasks like how to renew a passport, or summarizing long passages of text like news articles.

“Our success doesn’t rely on Google’s failure at all. People can use Google and Perplexity at the same time,” Srinivas said.


Quick Shipping

Srinivas and Yarats first met each other over email in 2020 after publishing almost identical research papers on AI training methods two days apart at UC Berkeley and NYU, respectively. The two stayed in touch over the years, messaging each other about the advancements in AI, while Yarats worked as an AI researcher at Facebook and Srinivas went on what he calls a “full world tour” working at incumbents like Google DeepMind and OpenAI.

It was July 2022, and AI was in its pre-ChatGPT era. Srinivas and Yarats decided to team up with former Quora engineer Johnny Ho and Databricks cofounder and fellow UC Berkeley alumnus Andrew Konwinski to build an application that used large language models to create a new search experience. But the technology hadn’t reached the stage where it could generate elaborate and relevant answers on its own. The cofounders hired contractors from countries like India to write templates of questions and answers to train the AI.

Their first product, called Bird SQL, used OpenAI’s code generation tool Codex to turn natural language prompts into code and search databases like Twitter. Publicly launched in December 2022, the tool caught the eyes of luminaries like Twitter cofounder Jack Dorsey and could surface information like “people Sam Altman engages with the most” and “top posts about ChatGPT.”

When Twitter ended free access to its API in February 2023, they were forced to try another idea: an enterprise AI search product that could rummage through a company’s SalesforceCRM and HubSpotHUBS databases and generate handy information for teams. The trio recalled the experience as “painful” and a “nightmare,” because internal databases weren’t well documented and AI models weren’t advanced enough at the time. “You had to spare three or four engineers to just go and fix bugs and we were not interested in that at all,” Srinivas said.

“You’ll mention an idea to them and then a day later, they’ll have something built.”

Elad Gil, angel investor

By that point, ChatGPT had exploded. The cofounders saw an opportunity to create an application that addressed its tendency to create factually inaccurate information by adding in citations. “Citations are a great way to marry search and LLMs,” Srinivas told Forbes. Now, every fact Perplexity spits out is attributed to a source which can provide more context and help verify whether the answer is right or wrong.

The founders were convinced that this new idea would work, and investor Friedman pointed out they had little to lose. Friedman remembers telling Srinivas, “Well, look, your product is sort of irrelevant. The worst thing that happens is you’re still irrelevant.”

The final product, a search engine that scanned the entire web for answers, looked starkly different from the initial Twitter demos given to potential investors. Google’s Jeff Dean even pointed the fact out to Srinivas while congratulating him, Srinivas remembers. “I don’t think Jeff would have invested” if the cofounders had started by tackling AI-powered search and trying to go after Google, he said.

The founders’ ability to rapidly prototype has translated into quickly shipping new Perplexity features, investors told Forbes. “You’ll mention an idea to them and then a day later, they’ll have something built,” said Elad Gil, who wrote the first and biggest check in Perplexity’s pre-seed round in August 2022. Friedman also lauded the company’s clock speed. “They were sending me demos twice a day,” he said about the startup’s early days.

The team, consisting of 45 employees, has also been judicious about spending its capital by making a deliberate decision not to develop a large foundational model in-house, which requires costly compute and infrastructure. Instead, Perplexity relies on AI models from providers like OpenAI and Anthropic, whose applications are likely to become cheaper as they become more efficient, Srinivas said. The company is focusing on distributing Perplexity’s search capabilities through partnerships with consumer hardware startups like Rabbit R1, a handheld AI assistant device, and smartphone company Nothing, which has said it sold 100,000 of its latest phones in the first day. The strategy “won’t get you a billion users,” investor Friedman said, but such partnerships can help the company grow.

As it competes with generative AI search features from established tech titans like Google and Microsoft, Perplexity has another factor working in its favor: novelty, Friedman said.

“I think many people are rooting for Perplexity because they represent the new player, the new paradigm, the new product,” he told Forbes. And if its quick growth and popularity among some of tech’s highest profile people indicates anything, it looks like that novelty has some staying power.

This article was first published on forbes.com and all figures are in USD.

More from Forbes Australia

Avatar of Rashi Shrivastava
Topics: