Dwarkesh Patel: observation

Transcript context

…Well, they add basically a France or UK on average every year. Extrapolate that forward. So far we've been focused especially on the training of AI models. When these models start becoming super economically productive—basically by the point at which they are at human level—the more important question will be how many can you deploy? Why is China such a powerful country or America such a powerful country? A big part of the reason is we just have more people. That's why China could take over Taiwan. China has like a billion people and Taiwan has like 20 million. Now you just have more AIs, more AI people if you have more power, which is ultimately what's upstream. That was all just a preface for asking, in 2035 what is the story of how China doesn't absolutely dominate AI? It's a good question and we're verging a little bit beyond my realm of competence here because I don't pretend to understand AI very well. I'm still very unclear as to economically where all the benefit accrues. Let’s abstract away from AI and just think about this. This is a situation where you've got a new technology, it's entering society. What is the innovation cycle that's going to drive it? I'm pretty partial to theories of innovation like the one that was propounded by a guy named Amar Bhidé who wrote a book a number of years ago called The Venturesome Economy. What he stresses there is the feedback loop between consumer demand and what producers are doing. His view is that we focus too much on what the producers are doing and not enough on what the consumers are driving. I think that's a pretty good insight generally. It helps explain a lot of why the US has maintained its position with such a dynamic economy, even though we've abandoned a lot of the physical processes that underlie it. So my question is, “Okay economically, when we say AI, what's the most important part of it? Is it the large language models?” I think the consensus is no. That's just the substrate. What really matters are the specific applications that get built on top of that. That's where most of the economic benefit is going to accrue. If that idea is correct, then the question you want to ask yourself is, “What economic system offers the most fertile ground for people figuring out how to use AI in specific zones?” There I would give the US a fairly strong edge over China because China has become more self-enclosed. They have become more obsessed with self-sufficiency. They've made it very, very difficult for other people to operate in China. It's creating some difficulties in people operating outside of China. They have this walled garden of Chinese people and whatever they do that creates applications. But whoever the big US or Western AI operators are, they have the entire rest of the world to play with. That's a much bigger universe. I would rate the likelihood that you get interesting applications higher outside the Chinese walled garden. So if you want the case for why China does not dominate, I think something like that would be the reason why.…

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