Nov 1, 2023 · 52m · mad
Perplexity AI CEO on Dethroning Google & Redefining Search
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this Data Driven NYC fireside chat hosted by Matt Turck, Perplexity AI Co-Founder and CEO Aravind Srinivas discusses how his team is building an AI-powered answer engine to redefine web search. He shares insights on product innovation, open-source AI strategy, web indexing infrastructure, and factual integrity.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 13.6% of the talking time here. How this is scored →
Matt as informed peer 2.4 Guest teaching 3.2 Guest disagreement 1.1 Matt pushing back 0.6
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 50:54 Rejection of OpenAI's plugin strategy
The guest forcefully rejects OpenAI's approach to third-party plugins, arguing that current LLM reliability requires first-party product orchestration rather than developer plugins.
Hardest push from Matt ▶ 33:02 Host confronts guest on index dependencyThe host directly probes the company's technical core by pointing out their reliance on third-party crawlers like Bing and questioning their ability to build an independent web index.
Biggest teaching moment ▶ 33:32 Explaining the real cost and difficulty of web indexingThe guest educates the host on search infrastructure economics, refuting the idea that crawling is expensive and explaining that index ranking quality and user feedback loops are the true bottlenecks.
Matt holds his own ▶ 12:48 Host articulates Text-to-SQL concept clearlyThe host demonstrates deep domain fluency by concisely summarizing the core purpose and technical promise of Text-to-SQL for enterprise data democratisation.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Welcome and High-Growth Valuation | 2 | 2 | 1 | 0 | The host opens with a warm welcome and cites recent press reports regarding Perplexity's $500M valuation rumor. The guest shares his career trajectory from Berkeley PhD to OpenAI and DeepMind internships before starting the company. | |
| Working as a Research Scientist at OpenAI | 1 | 3 | 0 | 0 | The host asks open-ended questions about the reality of working as a research scientist at OpenAI. The guest explains the internal GPU allocation dynamics and contrasts the patience required for research with the impatience needed for a founder. | |
| The Pivot from Text-to-SQL to Perplexity Answer Engine | 5 | 4 | 1 | 1 | The host demonstrates strong knowledge of data tools by explaining the mechanics and value proposition of Text-to-SQL. The guest details why Text-to-SQL failed to find product-market fit and recounts Nat Friedman's advice that spurred them to launch Perplexity Search. | |
| Deep Dive into Product Features: Copilot, Labs, and API | 3 | 3 | 1 | 1 | The host uses Perplexity to generate prompt ideas for the interview and asks for a tour of core features. The guest outlines Copilot, Labs, and the API, directly addressing critics who label the product a simple ChatGPT wrapper. | |
| Go-to-Market Strategy and Research User Base | 3 | 3 | 2 | 1 | The host asks about go-to-market strategies for consumer search versus developer APIs. The guest explains how they captured a niche research audience first and argues that Google is making a mistake by competing purely on model size rather than search orchestration. | |
| Building an Independent Web Index | 5 | 4 | 1 | 3 | The host challenges the guest by bringing up Perplexity's reliance on third-party search indexes like Bing. The guest clarifies that web crawling is cheap and straightforward, whereas index ranking quality and user feedback loops represent the true barrier to entry. | |
| Open Source AI and the Future of Synthetic Data | 5 | 3 | 0 | 0 | The host quotes one of the guest's recent tweets regarding compute-data feedback loops and synthetic data. The guest elaborates on open-source model ecosystems and recursive self-improvement as the path toward AGI. | |
| Audience Q&A: Misinformation Safeguards and Publisher Crawling | 0 | 3 | 1 | 0 | An audience member from LinkedIn News asks about misinformation safeguards and publisher crawler blocking. The guest explains algorithmic domain trust scores and defends citations as fair use. | |
| Audience Q&A: User Experience Evolution and Model Inference Strategy | 0 | 4 | 2 | 0 | An audience member explicitly challenges the guest, calling their open-source model inference API an 'odd tangent'. The guest acknowledges the critique, admitting it looks like a distraction, but justifies it as necessary infrastructure testing for long-term margin control. | |
| Audience Q&A: Infrastructure Choices and User Data Customization | 0 | 3 | 2 | 0 | Audience members inquire about cloud GPU infrastructure choices and third-party data customization. The guest explicitly rejects OpenAI's plugin architecture, advocating instead for native internal orchestration. |
Statements from this episode (21)
Srinivas: Jasper and Copy.ai initially generated more revenue than OpenAI
“News started spreading that there were companies like Jasper and copy.ai that started making more money than even OpenAI at the time.”
Srinivas: GitHub Copilot gained hundreds of thousands of paying users day one
“GitHub Copilot when they moved away from the waitlist to the paid version, they just had like hundreds of thousands of people paying from the first day.”
Srinivas: Stable Diffusion and Midjourney likely generated more revenue than DALL-E 2
“Similarly, like, Stable Diffusion and Midjourney were all, like, pretty competitive with Dolly too, and actually probably making more money than Dolly too.”
Incumbents will dominate text-to-SQL, making it a bad startup problem
“The bigger companies, like Databricks or Snowflake might build these tools and offer it on their sequel editor or, like, some kind of, like, you know, assistant And that could probably solve the problem. It's not a great problem for a startup to work on.”
Perplexity AI reaches millions of MAUs and millions of daily queries
“We have a lot of millions of monthly active users and millions of daily queries.”
Srinivas: Conversational AI will replace 10 blue links in 5-10 years
“If you have like deeper, more complex questions with five or 10 years from now, like you're not going to consume 10 blue links. It's going to be a chat bot or a system that you just ask questions like how you would ask another person.”
Srinivas: Google will lose its single search monopoly in the conversational era
“And in that era, I don't think Google is going to be the one single monopoly because they're already like far behind in this interface.”
Srinivas: Consumer AI products shouldn't expect users to be prompt engineers
“If you wanna take AI and like Make it like be widely accessible and like regularly used by every person in the world. You shouldn't expect people to be great prompt engineers. You should work on the user's behalf and try to understand their underlying intent a…”
Srinivas: AI search will not be dominated by the largest model
“This is not a problem that will be dominated by the company with the largest language model.”
Srinivas: Google is making a mistake chasing OpenAI with Gemini
“Hence why, like, you know, I actually think Google is making a big mistake by trying to do whatever OpenAI is doing, right, like, going after them with the large, oh, I'm not, they're having GPT-IV, I'm going to turn Gemini.”
Srinivas: Building a web search index is harder than competing with GPT-4
“We believe that building your own index is even harder than trying to compete with GPT-IV because it's not a problem that's just solved with money.”
Srinivas: Building a web search index requires active user traffic
“Without people using a product, there's no way to build an index.”
Perplexity CEO: Perplexity won't train open-source models, will serve them
“We are not the ones building these open source models because that requires a lot more capital. But we want to, like, make sure anybody can access them easily. So we are democratizing the access to them through our APIs, through our labs, and, like, you can co…”
Srinivas: Next major AI jump will come from self-teaching models
“At some point, like we lost saturated on data that, that exists on the internet. That's interesting enough for models to keep improving on that. The next big capability jump is probably going to come from the models teaching themselves.”
Srinivas: Perplexity will not crawl websites that block web crawlers
“Yeah, I mean, like, if they don't want to be crawled, we shouldn't crawl them. You know, that's just being how the internet has worked. Like, you know, people have the robots, the text file, and like, you know, if they don't let you do it, then you should just…”
Srinivas: Citing web sources in AI search constitutes legal fair use
“And it's actually fair use. The fact that you're attributing credit is fair use.”
Srinivas: Open-source AI models do not match closed-source model capabilities
“That needs the open source models or fine tuned versions of them to match the closed source, the closed ones. And That's not the case today.”
Srinivas: Self-hosting open-source LLMs is more expensive than closed-source APIs
“Even if there exists a checkpoint today, that's as good as the close models. It's still not an incentive to use them and serve them yourself because you're going to burn more money than paying for the close ones because the close ones are being served in such …”
Srinivas: Perplexity rejected CoreWeave due to high pricing
“Yeah, we don't use Corby because we decoded a very high pricing to us and like we decided not to use them.”
Srinivas: Perplexity does not believe in OpenAI's plugin vision
“We don't believe in the OpenAI plugins vision.”
Srinivas: Reliable AI user experiences require full internal orchestration
“Our vision is like at this point, like today, or like maybe the next year, The reliability that you want for a great consumer experience requires you to like do the whole orchestration yourself and not rely on like developers to do it for you.”