The Example Engine: How Exa Is Creating the AI Librarian for the Web with Will Bryk, CEO of Exa

**Erik Torenberg** (0:00)
Turpentine is a network of podcasts, newsletters, and more covering tech, business, and culture, all from the perspective of industry insiders and experts.
We're the network behind the show you're listening to right now.
At Turpentine, we're building the first media outlet for tech people by tech people. We have a slate of hit shows across a range of topics and industries, from AI with Cognitive Revolution, to Econ 102 with Noah Smith. Our other shows drive the conversation in tech with the most interesting thinkers, founders, and investors, like Moment of Zen and my show Upstream. We're looking for industry leading hosts and shows along with sponsors. If you think that might be you or your company, email me at erik.turpentine.co. That's E-R-I-K at turpentine.co.

**Will Bryk** (0:45)
I think there's an interesting distinction between search and research.
Google is a search engine. You know, you kind of know what you're looking for, but when you don't know what you're looking for, you're more doing research. And I think that's where Exa shines. What we're doing at Exa is we're kind of like trying to take all the world's knowledge and putting it into a new type of database, like a neural database. I like this database analogy because it's not really search. It's like you're kind of like with every query, you're filtering the database of all the knowledge into just what you need. There's a problem with, you know, search engines like Google is like you search something and they say, you know, 33 million results at the top. Like, what am I supposed to do with 33 million results? There's no way all these results are actually what I'm asking for. So it's just like, you just feel overwhelmed from the product perspective. Like not every query should require the same amount of compute.
Like Google has kind of made this assumption that like no matter what query you type in, it takes a few hundred milliseconds. But certain queries are extremely complex and might require like scouring the internet for, you know, maybe even seconds or minutes.
Thinking of it as more the optimal trade off is cool. It's like you're always optimizing for exactly the amount of effort to put into the thing.

**Nathan Labenz** (1:50)
Hello and welcome to The Cognitive Revolution, where we interview visionary researchers, entrepreneurs, and builders working on the frontier of artificial intelligence. Each week we'll explore their revolutionary ideas, and together we'll build a picture of how AI technology will transform work, life, and society in the coming years. I'm Nathan Labenz, joined by my co-host, Erik Torenberg. Hello and welcome back to The Cognitive Revolution. Today I'm excited to introduce you to Will Bryk, founder of Exa.ai, a company building a new kind of search engine designed with AI systems and workflows in mind.
Noting that Google has limited people's imaginations about how to search the internet, Exa aims to help people surface things that were previously impossible to find. Now if you've been following the show, you know that I've been exploring a variety of exciting new information and knowledge retrieval tools over the last few months. Today, ChatGPT remains my go-to workhorse for ad hoc coding and other random tasks, though recently Gemini Advanced and now Clog3 are both gaining share. Meanwhile Gemini 1.5 Pro has become my favorite for all write-as-me tasks, including creating the first draft of this introductory essay, which I did edit quite extensively, but nevertheless saved me a lot of time. Perplexity is still my go-to for quick and accurate answers to specific questions, but u.com Research Mode has now taken the prize for deep-dive multi-page research reports. And meanwhile, Elicit is the most useful for structured, systematic academic literature reviews.
So in this increasingly crowded landscape, where does Exa fit in? Well, for one thing, while most of these products are aiming to provide answers to questions, Exa still returns links, like a more traditional search engine would. After playing around with it for a while, I've realized that it's the quality, the controllability, and the depth of results that really sets Exa apart. And I've come to think of it as an example engine.
To understand how valuable this can be, consider that just about every company I talk to would like to scale highly personalized communications, whether for lead generation, recruiting, or something else. Today language models make this dream realistic. Given a list of leads or candidates and a bit of information about them, a language model can personalize your outreach at roughly human quality and far more scalably than was ever possible before. But where does the list itself come from? If the inputs to such a process are low quality, then the whole process will be garbage in, garbage out. And this is really where Exa shines. Given a complicated multi-part query which can be up to a full paragraph in length, something that you wouldn't even bother to try with Google but which language models can very quickly generate, and optionally an example of what you're looking for, Exa returns lists of a certain type of result, whether they be companies, people, how-to articles, you name it. And then you can feed those examples into your broader AI workflows.