Episode Summary
Executive Summary: Aravind Srinivas explains how Perplexity AI aims to redefine search by combining LLM reasoning with web indexing, citations, and fast, trustworthy answers. The discussion covers product differentiation versus Google/Bing/OpenAI, data sourcing, prompt orchestration, focused search verticals, monetization through ads and a paid copilot, talent strategy, and the future of chat/voice interfaces.
Main Topics: Perplexity’s search-first AI strategy: Perplexity positions itself as a replacement for traditional search by returning direct answers with citations instead of blue links, emphasizing trust, usefulness, and speed. How the product works under the hood: The company combines search indexes, crawled web links, and LLMs to extract relevant passages and generate concise answers with source citations. Competition with Google, Bing, and OpenAI: Aravind argues Perplexity can win through a superior product experience even while using the same core model providers as competitors. Monetization and advertising in AI search: The conversation explores whether chat-based AI interfaces can support ads, with Aravind arguing relevance and targeting can improve in an LLM-driven world. Focused search verticals and AI profile personalization: Perplexity offers category-specific search modes like academic, YouTube, Reddit, and Wikipedia, and is testing AI profiles to tailor results. Talent, hiring, and building on top of foundation models: The startup’s hiring strategy favors generalist engineers and people new to AI who can move quickly in a product-driven environment. Future interfaces: voice, chat, and wearables: The discussion looks ahead to conversational search through voice and glasses, but notes latency still limits seamless real-time interaction.
Key Arguments: Search is shifting from links to answers; Perplexity is building that future now. LLMs should reason over retrieved web content, not rely on memorized facts. Citations and source grounding are the main trust differentiator versus hallucinating chatbots. Product speed and answer quality matter more than model ownership alone. Even if competitors use the same base models, superior orchestration and UX can win. Advertising is not dead in AI search; it can become more targeted and relevant. Focused vertical search can outperform generic search for certain use cases. Hiring should prioritize adaptable generalists and fast shippers over AI pedigree alone.
Data Points: Launch timing: a week after ChatGPT came out - Perplexity launched almost immediately after ChatGPT, initially as a search bar with direct answers and citations. Early latency: 5 to 6 seconds per query - Initial product response time when Perplexity first shipped in December. Improved latency: almost as fast as Google - Aravind says the product has become much faster since launch. User engagement advantage: 2 minutes more on site than Bing - He cites comparisons showing users spend more time on Perplexity than Bing. Free usage limit: 25 queries a day - Daily limit for free users, including some Copilot usage. Paid usage limit: 300 queries a day - Practical unlimited usage for paid users under the subscription tier. Copilot capability: hundreds of search queries - Copilot can execute many searches to answer complex research tasks. Compliance timing mentioned in ad read: 2 to 4 weeks - Vanta ad claims average SOC 2 completion time with its platform. Compliance timing without Vanta: 3 to 5 months - Used in sponsor read as contrast for SOC 2 timelines. Cost savings from Vanta: up to 85% - Sponsor read describing compliance cost reduction.
Pivotal Quotes: "Realizing that 10 years from now, no one's going to be asking for 10 blue links. You're going to ask for answers." — Aravind Srinivas: Explaining the vision behind Perplexity as a next-generation search product. "We basically ask the LLMs to go read all those links and then pull up the relevant paragraphs from each of those links." — Aravind Srinivas: Describing how citations and answer generation work in Perplexity. "The only way to win against the person who has much more distribution than you, which is a superior product." — Aravind Srinivas: On competing with Microsoft/Google and why product quality is the key moat.
Implications: AI search is moving from novelty to infrastructure. Products that combine retrieval, citations, speed, and personalization may reshape search, research, and ads, while forcing startups to win on orchestration and UX rather than model size alone.
From the Episode
Realizing that 10 years from now, no one's going to be asking for 10 blue links. You're going to ask for answers. So we might as well start it today. And the technology for that was ready. Both ChatGPT and us were basically being powered by GPT-3.5. That was the fundamental breakthrough. And then after that, GPT-4, even better than that. So that's kind of how it started. And we were seen pretty different. It's like, oh, you know, if ChatGPT lies or makes up makes Up things. There's this other side called perplexity. You go there, and like it's gonna be this boring, educated uncle kind of product, but useful, and you can trust it. And that's kind of how we grew. And then BART came out. I believe BART still hasn't solved the fake news problem completely. It does hallucinate and it doesn't actually have real citations sometimes.
How do you get the citations? If you were asking this query I just did about like, hey, what cities should I take my seven years old to, and then what outdoor locations? How do you actually get the citations? Because ChatGPT4 doesn't provide citations, or do they? They have this thing called a browser plugin, which is basically powered by Bing. But people hate that experience in the sense it's really slow and clunky. Yeah, it is slow and clunky. Yeah. And so how do we? Do the citations. We basically pull up the relevant links to your query from a search index. And then we combine that and tell the LLMs to write the answer. We basically ask the LLMs to go read all those links and then pull up the relevant paragraphs from each of those links. And then make an answer out of whatever you thought was relevant, but write down the answer as if an academic or a journalist would write it, where each part of the answer has the corresponding citation, like Wikipedia. You basically say, hey, like, I want you to do the job of what a human does on Wikipedia, where when they're writing something about a new person or a new phenomenon or a new city, they're just basically going and like picking up a lot of web links about that, sifting through them and reading them.
The next big wins are not necessarily going to come from whoever is going to continue to train more. You need some algorithmic efficiencies to make use of compute even better. And you need really good researchers for that. And the best researchers are sort of like NBA players, and they're taken by these organizations who pay them millions of dollars a year. And then, if these guys who are building the tricks for making these models even better are in the close organizations, Then they'll always stay ahead of the open, right? So then, and if and the and if these organizations stop publishing these techniques and these guys to stay in these organizations are paid to stay there forever, it's kind of like closing the walls. So, the only way in which the open source world can catch up is like there are like amazing researchers who kind of like work in organizations that are actively open source models. And I think right now there's only one big org that one to do that which is meta
About This Week in Startups
Jason Calacanis covers startups, tech, markets, media, and all the hottest topics in business and technology. He also interviews the world’s greatest founders, operators, investors, and innovators.