The Real Future of AI with George Hotz

**Amanda Cassatt** (0:00)
This is Endgame, I'm Amanda Cassatt. This episode is with George Hotz, best known in some circles as Kiyohot, hacker of the iPhone and Sony PlayStation, turned AI founder, with the open-source self-driving device Comma, and the elegant and concise machine learning framework TinyGrab. We talk about where AI is going, and of course, the macro, what on earth's going on with the US government and with China. Give it a watch or listen wherever you consume podcasts. Okay, the first question I wanted to ask is, let's just place you in the spectrum of AI people. So we have the Yudkowsky style AI Doomer on one hand, and we have the Dario Amade style machines of love and grace abundance utopian on the other side. You are neither of these things. Let's just lay the foundation. Where are you on the spectrum, and what do you think is going to happen?

**George Hotz** (0:54)
I think you have to separate the what's going to happen from the how do you feel about it.

**Amanda Cassatt** (1:01)
So let's have two separate conversations. What's going to happen and how do you feel about it?

**George Hotz** (1:06)
Machines are going to take over and replace humanity, and I feel good about this.

**Amanda Cassatt** (1:12)
So let's talk about the steps of number one, then we'll return to our feelings.

**George Hotz** (1:17)
How did the machines do it?

**Amanda Cassatt** (1:18)
How does it happen?

**George Hotz** (1:19)
Robin Hansen's Age of M is, despite getting a bunch of the hows wrong, I think the overarching story it tells is right. It's just a question of economic competition. If machines can produce more value for less input cost, they will outcompete humanity. And that is the natural way of things.

**Amanda Cassatt** (1:42)
So what's the timeline? So right now we start with LLMs. Take us on the journey from how we start with these chatbots and how that ends up with human economic value decreasing to zero.

**George Hotz** (1:56)
No, this journey started in 1950 The journey started the minute someone had a computer. This is not a... there's nothing new about LLMs. There's nothing new about AI. AI is just a blanket word for kind of like things machines can't do yet, but now like machines can kind of do everything. So we kind of call it AI. But no, this journey has been happening for a long time. There used to be people whose job it was to sit in a room and add numbers up and then machines came in and did that much more cost-effectively than those people. For a person, you have to give them food, give them housing, for the machine, you don't, and it adds numbers a lot better than people. So it's just a continuation of the same trend, and that trend will continue smoothly into the future. There is no singularity. There is no inflection point. It's just a smooth trend continuing.

**Amanda Cassatt** (2:47)
So the singularity is not coming. The singularity is not nearer. We're just going to slide unconsciously into it and pass it.

**George Hotz** (2:53)
I don't even know what the singularity actually means. If you define the singularity as a point where machines can do most of the productive work, I argue that that probably even happened 20 years ago. If you take the singularity as machines could do all of the productive work, I don't think that will ever happen due to comparative advantage.

**Amanda Cassatt** (3:14)
Explain comparative advantage.

**George Hotz** (3:16)
It's like a trade thing. Even if one country is capable of producing everything for cheaper than another country, it still makes sense for those two countries to trade because, yeah, comparative advantage, it's in the term.

**Amanda Cassatt** (3:30)
Yeah, so it's more proximal for that country to create things to serve their market, even if it's more expensive for them to create things.

**George Hotz** (3:37)
No, okay, comparative advantage. The concrete example is like you have good A and good B, and country one and country two. Country one can produce both good A and B cheaper than country two. But does that mean they shouldn't trade? Well, no, because there could be a comparative advantage. If country one can produce B way cheaper and produce A only slightly cheaper, it actually makes sense for country two to produce A and for them to trade for B. So there will always be this comparative advantage with humans. Even if machines can do everything better with humans, humans will always have a comparative advantage.

**Amanda Cassatt** (4:10)
So human economic value isn't going to zero because of comparative advantage. And so let's talk about where those comparative advantages are. Because so just to plot you on that plot with Yudkowski and Dario Amadei. So Dario Amadei is also of the opinion that machines are going to replace people at every task, except for he's joyful about it. And he's excited to have a life that is free from labor and to devote himself and his essay to mountain biking. And then you have Yudkowski who thinks that, yeah, machines can replace people, but also they might explode the world, and we might be in some doomsday scenario where we all die because of some error made through AI. So let's plot this kind of comparative advantage where humans are going to be. And you also have people like Karpathy saying agency, right? Like the human value prop is having the agency to use all these tools. The locus of agency is still going to be with humans in terms of how they instrumentalize the machines.