Live in the Future

**Naval** (0:00)
Welcome to all the legends out there listening to the Naval podcast, the number one podcast in the universe, and your authoritative source for new knowledge. We've got three founders with us here today. On my right is Garry Tan from Y Combinator. They say it's the most reliable way to become a billionaire is to join the Y Combinator program.
And then-

**Daniel Francis** (0:23)
And start as a trillionaire.

**Garry Tan** (0:26)
That's one way to do it.

**Daniel Francis** (0:27)
We're all drinking, by the way, except one person who's not drinking. You have to figure out the-

**Farbood Nivi** (0:31)
The Russian ruler.

**Daniel Francis** (0:32)
Who's actually sober.

**Farbood Nivi** (0:33)
Yeah.

**Naval** (0:34)
And then on my left, I've got Daniel from Abel Police. They started off by turning body cam footage into police reports. But now they have a whole set of tools for compliant AI chat, translation. What else do you guys do?

**Farbood Nivi** (0:49)
Citizen reporting, all kinds of things.

**Naval** (0:51)
Okay. Then on my right, I've got Farbood from A-List, the health super app. As usual, we don't really care about what these guys are building. We care more about what they're learning about building, what they're figuring out, what they're still trying to figure out, and the principles that they can share with other founders. So does anyone have anything they want to jump in with?

**Daniel Francis** (1:10)
I mean, the topic that everyone's thinking about right now is AI, right?

**Farbood Nivi** (1:14)
AI. Yeah, I think I've heard something about it.

**Daniel Francis** (1:17)
Yeah. I kind of hate talking about it because it's the most perishable thing. Like, you know, the podcast will be obsolete by the time it ships. But it's still also the most interesting because it's the fastest changing thing in the environment that's high impact. So it's the steepest learning curve. And you're always trying to trade notes with people to kind of see where they're at. In fact, that was that a thing with you recently where I bumped into you and the conversation immediately turns to AI, of course.

**Garry Tan** (1:38)
I've got to. Yeah.

**Daniel Francis** (1:39)
So we can avoid it for a little while, but I feel like we'll circle back on that topic.

**Naval** (1:43)
I got a question for you. What are you being floored by at YC in terms of like what you see the company's doing with AI?

**Garry Tan** (1:50)
So I think the awkward thing is like I basically like slept three hours a night for like four months, and then somehow like went from not coding at all to like basically teaching people how to do it and making like one of the top 100 open source packages for how people vibe code called G-Stack. So that is very, very weird. And now like at some point, I met Pedro from Brex. And then after that, I basically converted from like Claude Code, like Colt to OpenClaudeCult within like 24 hours. And then now we use both. But like, it seems like basically if you, you know, the craziest idea that I discovered is that if you're just willing to spend like $100,000 a year on tokens, you can basically live like you are a normal citizen in 2028
Like, it's just pretty clear that token cost is going to come down, compute is going to go way, way up. Like, you know, we think like 90,000X or so, there will be 90,000X the amount of like inference from here to like three years from now.

**Daniel Francis** (2:56)
Or you plot it out in some curve, 90,000 times.

**Garry Tan** (2:59)
I mean, it's like many, many orders of magnitude.

**Daniel Francis** (3:01)
Which is kind of baked in both with the chips and the data centers they're building out.

**Garry Tan** (3:05)
Yeah, I mean, a lot of people say that NVIDIA is way overpriced, but what if it's way, way underpriced by like several orders of magnitude, right?

**Naval** (3:13)
90,000 by when?

**Garry Tan** (3:15)
I don't know, 24 to 36 months.

**Naval** (3:17)
Okay. So five orders of magnitude basically.

**Garry Tan** (3:20)
Yeah. I mean, we might be off by a couple of orders of magnitude.

**Daniel Francis** (3:22)
Yeah, but either way, we should be with friends.

**Garry Tan** (3:24)
I mean, it's actually still a blue ocean right now.

**Daniel Francis** (3:26)
But the thing that's crazy is that if you go up by five orders of magnitude, it's not just what it does in terms of usage, but in terms of capabilities because every order of magnitude, you go up, new capabilities emerge.
I've been pretty behind the curve on AI for a while. I missed it in 2020 to 2022 because I'd grown up my whole life hearing about AI from when I was doing CS and just assumed it was one of those things, like fusion, that was never going to come, and then it came. When it came, it came really fast, and at first, a lot of the stuff people were saying about it sounded breathless and they were talking their own book, but they turned out to be right so far.