Leopold Aschenbrenner: prediction

Transcript context

…It's a good place to sort of close that part where we know what the crux is and what evidence of that would look like. Let’s talk about AGI to superintelligence. Maybe it's the case that the gains are really easy right now and you can just sort of let loose. Give Alec Radford a compute budget and he’ll comes out the other end with something that is an additive change as part of the code. How many other domains in the world are like this, where you think you could get the equivalent of in one year? You just throw enough intelligence across multiple instances and you come out the other end with something that is remarkably decades, centuries ahead? You start off with no flight, and then you have the Wright brothers. You have a million instances of GPT-6, and you come out the other end with Starlink? Is that your model of how things work? You're exaggerating the timelines a little bit, but I think a decade's worth of progress in a year or something is a reasonable prompt. This is where the automated AI researcher comes in. It gives you this enormous tailwind on all the other stuff. You automate AI research with your automated Alec Radfords. You come out the other end. You've done another five OOMs. You have a thing that is vastly smarter. Not only is it vastly smarter, you've been able to make it good at everything else. You're solving robotics. The robots are important because for a lot of other things, you do actually need to try things in the physical world. Maybe you can do a lot in simulation. Those are the really quick worlds. I don't know if you saw the last Nvidia GTC and it was all about the digital twins having all your manufacturing processes in simulation. Again, if you have these superintelligent cognitive workers, can they just make simulations of everything, off the float style, and make a lot of progress there? You're just going to get the robots. I agree there are a lot of real-world bottlenecks. It's quite possible that we're going to have crazy drone swarms, but also lawyers and doctors still need to be humans because of regulation. You kind of start narrowly, you broaden, and there are worlds in which you let them loose. Again, because of these competitive pressures, we will have to let them loose to some degree on various national security applications. Rapid progress is quite possible. In the explosion after, there are basically two components. The A in the production function, the growth of technology, has massively accelerated. Now you have a billion superintelligent scientists and engineers and technicians, superbly competent at everything. You also just automated labor. Even without the whole technological explosion thing, you have this industrial explosion, at least if you let them loose, because you can cover Nevada and you start with one robot factory producing more robots. It's this cumulative process because you've taken labor out of the equation. Dwarkesh Patel 03:13:21 That's super interesting. Although when you increase the K or the L without increasing the A, you can look at examples like the Soviet Union or China. They rapidly increased inputs, which did have a geopolitically game-changing effect. It is remarkable to see the transformation of cities like Shanghai over just decades. They throw out these crazy cities in like a decade. Dwarkesh Patel 03:13:46 People talk about 30% growth rates from AI. The closest thing— Look at the Asian Tigers at 10%. It's totally possible. But without productivity gains, it's not like the Industrial Revolution. From a perspective of outside the system, your goods become much cheaper and you can manufacture more things. But it's not a sign that the next century is rapidly approaching.…

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