High Signal Podcasts Evidence ledger
Method
Browse
← Back to evidence

Evidence receipt / belief

Published · transcript-backed

Shawn Wang: belief

22 Mar 2024 Latent Space Why Google failed to make GPT-3 + why Multimodal Agents are the path to AGI — with David Luan of Adept

“I think you have broad thoughts, you know, just about the agent space, but also just in general AI space.”

— Shawn Wang

Source trail

Everything needed to verify it.

Speaker
Shawn Wang
Attribution
Verified speaker
Claim type
belief
Recorded
22 Mar 2024
Publisher
Latent Space

Transcript context

…First off, I think that I'm sure you guys believe this too. The AI space to the extent there's an AI space and the AI agent space are both exactly as she likely said, I think colossal opportunities and people are just going to end up winning in different areas and a lot of companies are going to do well. So I really don't feel that zero something at all. I would say to like change the zero sum framing is why should you be at Adept? I think there's two huge reasons to be at Adept. I think one of them is everything we do is in the service of like useful agents. We're not a research lab. We do a lot of research in service of that goal, but we don't think about ourselves as like a classic research lab at all. And I think the second reason I work at Adept is if you believe that actually having customers and a reward signal from customers lets you build a GI faster, which we really believe, then you should come here. And I think the examples for why that's true is for example, our evaluations, they're not academic evals. They're not simulator evals. They're like, okay, we have a customer that really needs us to do these particular things. We can do some of them. These are the ones they want us to, we can't do them at all. We've turned those into evals, solve it, right? I think that's really cool. Like everybody knows a lot of these evals are like pretty saturated and the new ones that even are not saturated. You look at someone and you're like, is this actually useful? Right? I think that's a degree of practicality that really helps. Like we're equally excited about the same problems around reasoning and planning and generalization and all of this stuff. They're very grounded in actual needs right now, which is really cool. Yeah. This has been a wonderful dive. You know, I wish we had more time, but I would just leave it kind of open to you. I think you have broad thoughts, you know, just about the agent space, but also just in general AI space. Any, any sort of rants or things that are just off of mind for you right now? Any rants?…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.

Search evidence