In this interview I speak to Aravind Srinivas,the CEO of Perplexity, an AI powered answer engine that is transforming the world of search by answering questions conversationally.
Perplexity extracts data from lakes including the web, images and video, before applying LLM’s to produce incredibly natural answers to questions, with references and citations for accuracy. The company was founded in 2022, and has already raised over $100 million from investors including Jeff Bezos, Nat Friedman, Elad Gill and Susan Wojciki. Perplexity is already revenue positive, and has attracted tens of millions of users each month who want to know everything from the origin story of apple and whether ageing can be reversed, to the best techniques to bake sourdough at home. You might even want to use Perplexity to research this very site!
Q: What was the hypothesis that led to Perplexity?
[Aravind Srinivas]: I’ll delineate the core principle of Perplexity separately from the existing flaws. Historically, the concept of providing 10 blue links to deliver the right information was essentially a makeshift solution. Larry and Sergey (Google’s founders) were aware that Google’s ideal evolution would be transforming into an answer engine — directly furnishing users with the answers they sought. This engine would autonomously navigate through links, sift through webpage content, identify relevant information, and present a concise response. That encapsulates the fundamental ethos behind it. The question then arises: why was this not feasible before, and what makes it achievable now? This breakthrough was propelled by advancements in large language models, particularly post-GPT 3.5, marking a pivotal moment that enabled the creation of practical technologies. However, these advancements come with their limitations, notably in managing the accuracy of generated content.
Another cornerstone of Perplexity stems from my scholarly background. The cardinal rule in academic research is to base your assertions on citable evidence rather than conjecture. This principle sets Perplexity apart from ChatGPT, which has the freedom to generate content without such constraints. Perplexity, by design, is restricted to sourcing information directly from the web, eschewing any reliance on pre-existing knowledge within the model. While this may limit its ability to entertain or offer opinions, it significantly enhances its utility for those seeking reliable information. This approach, rooted in academic discipline, underscores our vision of the ultimate search engine — one that not only answers queries but does so with the integrity of a scholar. Thus, Perplexity embodies the fusion of delivering precise answers and adhering to scholarly principles, laying the groundwork for a new paradigm in information retrieval.
Q: How do you approach the revenue model of Perplexity?
[Aravind Srinivas]: …looking back on the launch of our answer engine, we admittedly hadn’t fully considered its potential to disrupt the AdWords business model. Had we realised this aspect sooner, our enthusiasm for the project would have been even greater. It’s quite rare to enhance the product experience and simultaneously challenge an existing business model. Typically, you manage to achieve either one or the other. For instance, envisioning a search experience without ads, offering just the 10 blue links, doesn’t significantly alter the product experience nor does it disrupt the existing model. Even if Google satisfies the majority of user needs (the 80/20 rule), the absence of ads for specific searches won’t significantly impact user satisfaction. Eventually, a similar monetisation strategy becomes inevitable.
However, Perplexity introduces a superior product experience by saving users time and eliminating the need for them to click through links, among other benefits. This approach not only enhances user experience but also disrupts the traditional pay-per-click model anchored to the standard 10 blue links interface. This disruption is nothing short of revolutionary.
Addressing your direct query about our next steps without the traditional UI and AdWords model, we’re pivoting towards revenue sharing and distribution strategies. We’ve introduced the Pro Plan, a premium subscription offering access to advanced models and in-depth searches, along with unlimited usage of features such as image and file uploads. This Pro Plan has attracted considerable interest, and we’re leveraging creative partnerships to extend its reach. For instance, we offer the Pro Plan for free for a few months through bundle deals with brands like Nothing Phone, Rabbit, and SK Telecom. These partnerships significantly expand our user base.
By showcasing the superior experience our product offers, we’re confident many users will see the value in continuing their subscriptions. Plus, we remain committed to offering a free service tier, staying true to our belief in the freemium model. That’s our strategy moving forward.
Q: Will answer engines change the fundamental nature of the web?
[Aravind Srinivas]: The idea of transforming the frontend into something more intuitive is fascinating. Imagine using Perplexity to ask any question without relying on outbound links. Instead, links could open directly within the app or site, making it more like navigating through natural language rather than the traditional method of entering URLs. This concept is intriguing, though, to be frank, it may not yet rival the browsing experience of directly typing a query and visiting a website. Plus, users are accustomed to starting their search from a familiar browser search bar, like quickly jumping to R/WallStreetBets.
Currently, our product isn’t optimised for this seamless experience because we need to enhance our latency. The goal is to display sources in less than a second, making users wonder why they would need any other frontend. Achieving this would be groundbreaking, especially as we anticipate this browser replacement effect to be more pronounced on mobile devices. On phones, the need to open a separate browser could become obsolete. Imagine having various apps for different needs, with the Perplexity app serving as your central hub for inquiries. Eventually, this could evolve into a super app, controlling a suite of applications—though this ambition faces the challenge of requiring some control over the operating system.
The notion that the Google or Chrome app could become redundant on mobile devices isn’t far-fetched. In a future where direct URL visits are less common, the convenience of an all-encompassing app could significantly alter how we interact with our devices. While it’s uncertain who will lead or succeed in creating such a super app, the potential for changing our browsing habits is an exciting prospect we’re eager to explore and work towards.”
Q: Will answer engines lead to you interacting with knowledge differently?
[Aravind Srinivas]: Let me share an interesting story. I was having a conversation with someone from Amazon, and we touched on a variety of topics, including Coreweave and AWS. During our chat, he mentioned that Jensen Huang, the CEO of Nvidia, had invested in Coreweave. I corrected him, pointing out it was actually Nvidia that had made the investment in Coreweave, not Jensen personally. But he insisted, quite confidently, that it was indeed Jensen who directly invested.
To settle the debate, we both reached for our phones to verify the facts. I quickly queried Perplexity, asking whether Jensen or Nvidia invested in Coreweave. The answer came promptly. Meanwhile, he typed his query in a rather awkward manner: ‘Coreweave investment Jensen.’ He was presented with a slew of links and seemed unsure of which one to trust. By the time he was still scrolling through his options, I simply showed him the answer on my phone. He was taken aback, admitting that this was a far superior approach.
This experience exemplifies the revolutionary nature of directly conversing with your search engine as if it were a person. Imagine the entire internet personified, available to chat with you. That’s the magic Perplexity offers. Granted, this ‘person’ doesn’t possess emotions or opinions; it can only relay what’s available online. Yet, the ability to access information in such a natural, conversational manner is immensely valuable.
Q: Will perplexity also enable multi-modal search across types of data?
[Aravind Srinivas]: As it stands, you can already upload images and pose questions about them, which is just the beginning. Imagine pasting a URL of a YouTube video into our service. We’d be capable of parsing the transcript, comprehending the video’s content, and allowing you to inquire about specific moments or details within that video. This functionality is on the horizon, promising to be a reality very soon.
Eventually, the type of file you upload won’t matter. Be it a hefty text document, a folder, or a series of videos, the flexibility of these advanced models will handle it all. This is the strength of versatile general-purpose models: a single backend engineered to accommodate a wide array of tasks.
Contrast this with the conventional search interface, which is considerably restricted. Introducing new functionalities requires navigating around these limitations, a task that proves much more challenging for established players like Google. They are constrained by the need to maintain the traditional structure of ten blue links on google.com, making the integration of generative AI without altering the interface a complex endeavor. The result is a cluttered and confusing user interface, leaving users puzzled about what Google offers them. This is the kind of innovation barrier we aim to transcend, offering a seamless and intuitive experience.
Q: What have been some of your learnings leading such a fast growing company?
[Aravind Srinivas]: Being a CEO is truly an honour, and frankly, it’s been an unexpected journey. It might surprise some, but we’ve only been at this for a year and a half. Yet, we’ve achieved remarkable progress in such a short time. The essence of start-up culture is all about maintaining momentum, and currently, our momentum is soaring. The key for us is to keep this momentum alive.
The recognition we’re receiving is encouraging, but it’s crucial not to let it cloud our vision. The focal point remains our commitment to serving the user – it’s a privilege to work for and with them.
Many argue there’s no predefined path to becoming a founder. Before embarking on this journey, advice often pointed towards gaining experience at another start-up or leading a product launch elsewhere as the best preparation. However, the true enlightenment comes from diving into the deep end. Much like teaching a baby to swim by gently introducing them to water, start-ups teach you to navigate through experience. Unlike babies, though, the buoyancy isn’t guaranteed; it’s a heavier challenge, and while support from investors might be there, the onus of figuring things out rests squarely on your shoulders.
The internet, used wisely, is humanity’s most significant repository of knowledge. Platforms like Perplexity, Google, and YouTube offer an infinite wellspring of information. From my PhD days, I’ve delved deep into studying the trajectories of tech giants and their founders – Google’s Larry Page and Sergey Brin, Amazon’s Jeff Bezos, Apple’s Steve Jobs, and Facebook’s Mark Zuckerberg. I’ve absorbed their methodologies, read their biographies, analysed their management styles, and discerned the good from the bad. All these insights are gleaned with the understanding that relevance is key, and I’ve applied what’s pertinent to our journey.
This approach has been foundational in our rapid growth and will continue to guide us as we navigate the future.
Q: What do you hope your legacy will be?
[Aravind Srinivas]: I aspire for my legacy to be intertwined with Perplexity’s potential success as the definitive source for knowledge and information on the internet. That’s the dream. But even if we don’t reach that pinnacle, I’ve already seen us make remarkable strides. I want to be remembered as the leader who nudged the giants — be it Google, OpenAI, or any other entity — towards a paradigm shift in how we access information online, moving from mere links to direct answers. This shift, I believe, has been catalysed by our efforts, and it’s a mission I’m passionate about continuing.
My hope is to be remembered as someone who significantly impacted the way people enhance their knowledge every day. This personal value, deeply rooted in my culture and upbringing, where knowledge was held in the highest regard, drives my vision. If our work can democratise access to information for those who lack the resources — the expert guidance, the access to prestigious universities — then I believe we’ve achieved something truly transformative. Imagine anyone, regardless of their financial standing, being able to inquire about complex topics like nuclear fusion and receiving an answer so comprehensive it’s as if an MIT professor personally explained it to them. That level of accessibility and clarity is what I aim for our service to provide. Ultimately, I want my legacy to reflect this ambition — to have played a part in making such profound knowledge readily accessible to everyone, everywhere.