Apr 25, 2023 · 39m · no-priors
No Priors Ep. 9 | With Perplexity AI’s Aravind Srinivas and Denis Yarats
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
Perplexity AI co-founders Aravind Srinivas and Denis Yarats join No Priors to discuss the rise of conversational answer engines, their high-velocity engineering culture, and how generative AI is disrupting traditional web search and monetization models.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 18% of the talking time here. How this is scored →
The hosts as informed peer 4.3 Guest teaching 3.8 Guest disagreement 1.6 The hosts pushing back 1.6
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 28:51 Aravind rejects adversarial publisher premise
Aravind directly counters Sarah's suggestion that answer engines destroy publishing incentives, asserting it actually resurrects foundational PageRank citation dynamics.
Hardest push from the hosts ▶ 34:00 Elad refutes academic velocity narrativeElad explicitly pushes back against the guests' claim that academia teaches fast iteration, noting that most academic settings are defined by extensive pre-planning and lack of action bias.
Biggest teaching moment ▶ 16:40 Aravind clarifies citation-first vs chatbot retrofittingAravind educates the hosts on the architectural distinction between cosmetic chatbot citations and Perplexity's strict, non-generative citation-first retrieval paradigm.
The host holds their own ▶ 30:28 Elad details Google's monetization evolutionElad demonstrates seasoned expertise by recounting Google's initial revenue experiments—from selling search syndication queries to on-prem hardware appliances—before discovering ad auctions.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Founding Perplexity and Early Search Explorations | 4 | 2 | 1 | 1 | Elad welcomes the founders and references his early angel investment conversations with them. Aravind reminisces about their initial pitch around visual search and Elad's distribution advice in Noe Valley. The tone is highly collaborative and reflective. | |
| Building a Culture of Rapid Iteration and Engineering Velocity | 4 | 3 | 1 | 1 | Elad highlights Perplexity's rapid execution speed across multiple product pivots and asks about cultivating iteration velocity. Aravind credits academic experimentation habits, operational feedback loops, and recruiting competitive programming talent. | |
| Hiring Philosophies, Trial Work Periods, and Startup Speed | 3 | 3 | 1 | 1 | Denis describes Perplexity's technical trial periods and preference for high-energy generalists over passive specialists. Sarah inquires about false signals, and Denis emphasizes rigorous culture-work ethic alignment. | |
| Debunking AI Pedigree: Software Engineering Versus Academic Specialization | 6 | 4 | 2 | 2 | Elad draws a comparison to the early mobile era where hiring for specialized pedigree proved foolish compared to generalist talent. Aravind and Denis enthusiastically validate this by pointing to OpenAI and Anthropic teams coming from physics and software infrastructure. | |
| Transitioning from Academia to Startups: Radical Focus and Brutal Honesty | 3 | 3 | 1 | 1 | Sarah asks about the steepest learning curve transitioning from research to founding. Aravind and Denis discuss shedding the academic tendency to hedge multiple projects in favor of extreme, singular product focus. | |
| Prioritizing Factual Accuracy, Citations, and Hallucination Reduction | 4 | 5 | 2 | 1 | Elad probes Perplexity's mission around factual accuracy and hallucination mitigation. Aravind explains their citation-first architecture, contrasting it with chatbots that try to retrofit citations after text generation. | |
| Leveraging Reinforcement Learning and Intermediate Ranking Methods | 4 | 4 | 1 | 1 | Elad asks about reinforcement learning integration, and Denis details intermediate techniques like rejection sampling and discriminator ranking on top of LLMs. Sarah asks about chat versus search interfaces. | |
| The Emergence of Answer Engines and Behavioral Shifts in Search | 4 | 5 | 2 | 1 | Sarah asks for a five-year prediction of search architectures. Aravind outlines the rise of answer engines, deliberate link-capping (restricting to 3-4 sources), and changing user query habits. | |
| Proactive Push Models and Disrupting the Legacy Search Ad Paradigm | 4 | 4 | 2 | 2 | Elad asks about pull versus push information delivery timelines. Aravind argues push models are viable immediately, while Denis highlights the innovator's dilemma facing Google's high-click ad revenue model. | |
| Redefining Web Publishing Incentives and Citations in the AI Era | 6 | 4 | 2 | 2 | Sarah raises publisher concerns regarding zero-click AI summaries, which Aravind reframes around citation authority and PageRank origins. Elad demonstrates deep domain knowledge outlining Google's historical progression from syndication and hardware appliances to advertising. | |
| Practical Advice for Academic Researchers and the Future of AI Architectures | 5 | 5 | 2 | 4 | Elad gently challenges Aravind's claim about academia cultivating speed, pointing out that academic culture typically lacks action bias. Aravind explains his advisor's exceptionalism and advises aspiring researchers to seek radical transformer alternatives rather than incremental LLM scaling. |
Statements from this episode (31)
Srinivas: Search is as much a distribution game as a tech game
“No, I think whatever you said still applies. Search is tremendously a distribution game as much as a technology game.”
Perplexity co-founder Johnny Ho was world number one in competitive programming
“He was the world number one at competitive programming. It's like being Magnus Carlsen of programming.”
Yarats: Perplexity prefers raw drive over extensive experience when hiring
“Personally, I would rather like get somebody who has this Burning desire to work on these things rather than somebody who already has a lot of experience and not going to put a lot of, or like as much effort as other people.”
Denis Yarats says sacrificing work-life balance is required to beat competition
“It's kind of like work life balance, maybe not the ideal but you know, but that's the only way to sort of like beat competition and sort of like iterate very fast and do great things.”
Srinivas: Iteration speed is the only path to success without product clarity
“If you don't have a clear idea of like, you know, which product you're going to build or which market you're going after and you still want to be giving, giving yourself a shot at success iteration speed is the only thing that you can hope for.”
Working in AI research or product does not require an AI background
“So it's always been clear to me that in order to work in AI, both research and product, you do not need to already have been in AI.”
Aravind Srinivas argues AI application builders understand LLMs better than trainers
“The people who use the LLMs for building stuff, Understand it better than the people who actually did gradient descent and train these models.”
Yarats: The best AI research breakthroughs came from top software engineers
“Turns out through my experience, the best research scientists, also very good engineers, like very good engineers. And we've noticed like through DeepMind and OpenAI is just like the companies that made the most progress over like last six years or five years …”
Yarats: Perplexity built unreleased Twitter, Salesforce, and HubSpot integrations early on
“Even though we're like six months or like seven months old company, we actually built so many things. We had like, you know, Twitter, we had like a Salesforce integration. We have like HubSpot integration, like many other things that we never like released.”
Srinivas: Trying everything at once destroys team faith in startup leadership
“Also as leader, if you want to lead the company and if you kind of do everything, the people will lose faith in you, right? They think you don't really have any clarity and that's why you're making them do one new thing every week. And so you have more respons…”
Srinivas: Startup founders cannot hedge across multiple projects like academics do
“In academia, you're basically taught to hedge. You have like one first author project and like three or four co-author projects, and like one of them might become a big hit and might change your career. Whereas that's not how you should do startups. Like you r…”
Srinivas: Perplexity grounds even conversational identity queries in web search
“In fact, it's more like a citation first service that it'll never say anything that it cannot say. So if you, if people have tried to play with it as if it's like chat GPT where like, tell me who are you or like things like that. And even for those questions, …”
Standalone LLMs cannot solve hallucination without a platform for user curation
“And I also don't think One LLM will just magically solve this problem. You need to build an end-to-end platform where users can correct the mistakes of an LLM, and that also means you need to design the platform where the incentive is right for the user”
Perplexity to prioritize core RLHF over AI agents for 6-12 months
“We haven't gone beyond that to think of like agents and browsers and things like that. We, we'll probably focus more on the first part for the next at least six months to a year.”
Yarats: Rejection sampling significantly boosts LLM quality before full RLHF
“Full blown, like RLHF is, you know, definitely something we're going to look into that, but there is like several many steps that you can have in between that significantly can increase your quality. So for example, I mean like even using something like a reje…”
Srinivas: Chat UI makes more sense than Google for complex queries
“As the questions get more complex. The chat UI makes a lot more sense than Google's UI.”
Srinivas: Perplexity Was Truly the First Conversational Answer Engine
“Perplexity is the first conversational answer engine like truly the first. I think nobody built it before us.”
Srinivas: Answer Engines Will Become Their Own Separate Market Segment
“I believe answer engines will become its own segment, market segment. Like, just like you have a default search engine, there'll be a default answer engine over time if these things really work.”
Aravind Srinivas predicts users will abandon Google's blue links for answer engines
“They will eventually prefer this experience over the regular like 10 links or 20 links UI that Google has. So that, that's something I'm pretty confident about.”
Srinivas: AI Answer Engines Will Significantly Reduce Traffic to Content Sites
“And I also think the fourth thing is there will be some sort of like Much fewer traffic to the actual content site. Like very few links need to be consumed.”
Aravind Srinivas predicted push-based AI agents would arrive by early 2024
“I feel like we can do these things even better now with language models. So yeah, I think it'll happen in a year, more than five years.”
Denis Yarats argues AI answer engines fundamentally break Google's monetization model
“And it's also kind of like why we believe my, it might be very hard for Google to fully launch this system. Like answer engine, because it just like breaks monetization strategy for them.”
Yarats: AI search monetization will feature fewer but more expensive clicks
“So there is gonna be few, few clicks, but each click is gonna be more expensive.”
Srinivas: Optimizing for LLMs is unlikely to be as easy as keyword SEO
“It's unlikely that it's as easy as SEO with just keywords because LLMs are going to be much smarter in understanding relevance to a query.”
Srinivas: AI search will incentivize publishers to create higher-quality content
“I think it's just going to incentivize people to publish higher quality content. In order to get cited by an LLM powered answer, like Substack or things like that try to do, like you want to own your content and you publish it and make sure it's high quality a…”
Srinivas: Subscription search fails if incumbents offer an 80%-good free tier
“If a bigger behemoth like Google or Bing just put out the same or like even 80% as good as you for free, then you're never going to make it as a subscription product.”
Srinivas: Google and Microsoft will match Perplexity's current search capabilities
“I feel like Google and Microsoft will pretty much do the same thing we have right now as well as us.”
Aravind Srinivas says pursuing an AI PhD is a terrible career choice
“Especially a PhD student is one of the worst Jobs you can go for in a time when AI is so hot and so highly paid and you can do a startup or join a startup.”
Srinivas: Only pursued a US PhD because it was fully funded
“To be very honest, I only came to the United States for doing a PhD because there was no other way for me to come to the United States. Like I could, I couldn't take a loan for a master's or something like that. So a PhD is fully funded and like sponsored. So …”
Srinivas: Academic AI researchers should pursue radical alternatives to transformers
“I think it's best to look for alternatives to the transformer, alternatives to like language models, that sort of radical directions, then Trying to improve them because there's so much incentive for the existing companies to do that.”
Yarats: Aspiring AI researchers should work in industry before a PhD
“I think it's best to not go to PhD right after undergrad and kind of like spend at least like couple of years in industry. I think that goes to my point to me, like the best researchers are those who can, who are also very good engineers.”