Evidence receipt / evaluation
Published · transcript-backedWill Bryk: evaluation
10 Jan 2025 Latent Space Beating Google at Search with Neural PageRank and $5M of H200s — with Will Bryk of Exa.ai
“People think of searches as like, oh, it takes 500 milliseconds because we've been conditioned, uh, to have search that takes 500 milliseconds.”
Source trail
Everything needed to verify it.
- Speaker
- Will Bryk
- Attribution
- Verified speaker
- Claim type
- evaluation
- Recorded
- 10 Jan 2025
- Publisher
- Latent Space
Transcript context
…Does that mean you have like some kind of compute budget that I can specify? Yes. Yes. Okay. And like, what are the upper and lower bounds? Yeah, there's something we're still figuring out. I think like, like everyone is a new paradigm of like variable compute products. Yeah. How do you specify the amount of compute? Like what happens when you. Run out? Do you just like, ah, do you, can you like keep going with it? Like, do you just put in more credits to get more, um, for some, like this can get complex at like the really large compute queries. And like, one thing we do is we give you a preview of what you're going to get, and then you could then spin up like a much larger job, uh, to get like way more results. But yes, there is some compute limit, um, at, at least right now. Yeah. People think of searches as like, oh, it takes 500 milliseconds because we've been conditioned, uh, to have search that takes 500 milliseconds. But like search engines like Google, right. No matter how complex your query to Google, it will take like, you know, roughly 400 milliseconds. But what if searches can take like a minute or 10 minutes or a whole day, what can you then do? And you can do very powerful things. Um, you know, you can imagine, you know, writing a search, going and get a cup of coffee, coming back and you have a perfect list. Like that's okay for a lot of use cases. Yeah. Yeah. I mean, the use case closest to me is venture capital, right? So, uh, no, I mean, eight years ago, I built one of the first like data driven sourcing platforms. So we were. You look at GitHub, Twitter, Product Hunt, all these things, look at interesting things, evaluate them. If you think about some jobs that people have, it's like literally just make a list. If you're like an analyst at a venture firm, your job is to make a list of interesting companies. And then you reach out to them. How do you think about being infrastructure versus like a product you could say, Hey, this is like a product to find companies. This is a product to find things versus like offering more as a blank canvas that people can build on top of. Oh, right. Right.…
Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.