Dwarkesh Patel Explains How AI Could Radically Improve Civilization

If we do get AGI, it would just be so hard to contain it. If you didn't release ChatGPT, you wouldn't have been able to know that this is a feature people really wanted. Just imagine this scaled up like 100x, 1,000x integrated into all our senses, maybe even into our minds. I became more convinced of our glorious transhumanist future.

I think the real bottleneck is just we got to make them smarter. Mhm. I don't think they're already so cheap. It's like 2 cents per million tokens or something ridiculous. I think the real bottleneck for me using them more is not their price, but them being more useful, being able to take over more parts of the economy.

There has been this trend in AI where whenever there's a big breakthrough we think we've automated a large part of what intelligence is and in fact in retrospect it's clear that it was only a beginning. So the big example here is when Deep Blue came out and beat Kasparov at chess. People thought that this was a big breakthrough in intelligence in general because we thought that what chess required was general intelligence and you might have heard this concept of AI-complete problems where if you solve this problem then you've solved intelligence. So people said that about self-driving. The Turing test was supposed to be AI-complete. We've gone through all of these subcomponents of intelligence and afterwards we realized there's actually still more left to it. The thing that's sort of underrated is not even agency per se, although that's a part of it. I think the thing that's underrated is we humans have this global hive mind where the reason we can make iPhones and buildings and whatever is not just intelligence and also not just agency. It's the fact that there's so much specialization. There's so much capital deepening. People are just doing things, trying different ideas. AIs need to be smarter in order to do that. They need to have more agency in order to do that. But once they can, if you have millions of AIs running around trying different things, that's when we get the real acceleration. You'll feel it in your blood.

So to me, I get very excited about it because it's like having any human be able to create beautiful images out of text is fantastic. I became more convinced of our glorious transhumanist future. You're getting a glimpse just from these early images of how cool and beautiful the things AI makes or helps us make will be. Just imagine this scaled up like 100x, a thousandx integrated into all our senses, maybe even into our minds, integrated into the way we relate with the people we care about and so forth. The future could be really beautiful.

Look, we can keep increasing compute that we're putting into these systems for maybe the next 5 to 10 years because compute is just growing at this ridiculous rate where in 3 years we're going to have 10x the amount of global AI compute that we have right now. But at some point, right now we're spending 2% of GDP on compute and data centers and stuff like that. You can't just keep 10xing that forever. If somehow this whole deep learning paradigm is wrong and we just totally miss the boat somehow, then I can see that happening 10 to 20%. Otherwise, if we do get AGI, I'm of the opinion that it would just be so hard to contain it. It's an incredibly powerful technology. Even if there's no intelligence explosion, even if it doesn't help you make an ASI, just AGI alone would make the economy explode. There's a little bit of a force of deceleration like the GDP question, but also I've had this idea that no matter how intelligent you are, you can't break the laws of physics. At a certain point, you need to get the sand out of the ground and turn it into silicon. And at a certain point, just moving the sand around fast enough, even at light speed, you're not 10xing every two years. So it feels like there could be a slowing down even as we're having the robots do basically everything. It's like the robots are still maxed out by physics. I don't know. I was thinking about this this morning actually. The intuition I was thinking about is since the 1750s we've had 2% economic growth in the world. Before that it was a tenth of that. If you were around in the 1500s or 1000 and somebody said there'd be 2% growth, given your reference class you might have said it just takes a long time to learn how to artificially select crops and build new structures and aqueducts. That is a process that takes a while. So why do you think you're just going to be increasing that 2 to 3% a year? It looks like the last 100 years of history, we're discovering all these new things in physics and chemistry. The last 50 years we start with a transistor and now we're talking on this magical screen. And that was just like physics didn't bottleneck that. I think you get another 10x and I don't see any in-principle reason why the next 10x of physics would not allow the robots to move fast enough.

They're not AI-pilled enough. You know, even if you treat it like another feature, it's mysterious why Siri doesn't work on my phone. But it's like more people basically. And if you take that seriously, you're not just going to say, 'Oh, and the 25th department in our complex is about making Siri better at speaking.' No, this is the future.

I think maybe somewhere in between. If you didn't release ChatGPT, you wouldn't have been able to know that this is a feature people really wanted. And I wonder if other features of AGI will be similar where if you don't deploy it to a bunch of engineers on Cursor, you just won't know what actually makes something a good coding bot. The counterargument to this and I think what the SSI people would say is that they actually are deploying but they're deploying towards the one thing they care about which is accelerating AI research and they don't need to do that externally. They can just do that internally. The basic question is can you get this closed loop where you build the AIs which are helping you accelerate AI research, dot dot dot super intelligence. I'm 50/50 on that question. So that other 50% is a big deal.

There's AGI-pilled and then there's transformative AI-pilled where you say look, even if they're just like humans, if they have the advantages that AIs will just intrinsically have because of the fact that they're digital, which is the fact that they can be copied with all of their brain and all of their knowledge. So think of the most skilled engineer in your company like Jeff Dean or Ilya, you can copy that person with all their tasks and knowledge and everything, you can merge different copies, you can scale and distill AIs. Those advantages alone, and that there will be billions of copies as we increase the amount of compute in the world, that alone is enough for transformation in the sense of going from what we were like before the industrial revolution to the industrial revolution pace of growth. So I think somebody can be AGI-pilled in the sense they say yeah I expect human-level intelligence to emerge in the next 10 years but they still don't take that seriously as in okay well what does that imply about what is happening through the economy? Does that just mean you've got a smart personal assistant, or does it mean we're in a very different growth regime? 22-year-olds right now are coming into their careers and they're saying well everybody's going to be paperclipped in a few years. You've got to create value and capture it now so that you're okay in this AGI future. I do feel like that's maybe a common fear throughout history, where there's this sort of pending sense of doom sometimes. But what would you say to somebody that had that sense? I think the way to model out the next few years from a career trajectory is you'll just have 100x of leverage, but you want to be in a position where you can use that leverage. There's a common thing, and I'm sure you experience this now as well, as you advance in your career, before you're like, I've got a bunch of time, but I don't know what to work on. And after you're further along, you're like, I have no time, but there's a thousand different ideas I have for things that would be super valuable or would go really well. So what you should do is just get to a point where in whatever you think is interesting or care about, you're at the frontier and can see what the problem space actually looks like. If you care about AI, I would really recommend moving to San Francisco and then just start working on problems and use the leverage that AI gives you. And if you end up getting paperclipped, look, what's the point of you not doing anything worrying about that in the 80% of worlds or 90% of worlds where we don't get paperclipped? You will get to say you worked on something really cool at a time that was really important in the history of humanity. That's great.