← Modern Wisdom
Loading summary
Transcript477 lines
- [00:00]
Chris Williamson
What do you think that we've realized about human learning and human intelligence from architecting AI intelligence? Hmm.
- [00:09]
Dwarkesh Patel
There's this really interesting thing we've seen where these AI models are making progress, first in the domains that we think of as the archetype of where humans have their primacy. Right. So if you look at Aristotle, what does he say? What makes humans unique? Well, it's reasoning. Humans can reason, other animals can't. And these models, these AI models, they're just not that useful if you've tried to use them for your work. They're useful in certain domains, but broadly they're just not widely deployable. What is the one thing that they can do? They can reason, but obviously they can't carry a cup of water. Right. Robotics isn't solved. They can't even do a job. They can't even do a white collar job. So there's this interesting thing called Morbec's paradox. Hans Morbach came up with this idea in the 90s where he noticed that the tasks which are easiest for humans are taking computers the longest to solve. So we still haven't solved robotics yet. It's so easy for us to move around. Whereas the tasks which are quite hard for humans, like adding numbers, adding long numbers, computers could do that in the 60s. And the logic there is that evolution has only optimized us for, let's say, the last million years to be good at reasoning, to be good at arithmetic, to we good at these kinds of high level abstractions. Evolutions has spent 4 billion years teaching us how to move around the world, how to pursue your goals on a long term basis. So not just do this task over the next hour, but spend the next month planning how to kill this gazelle. And that has been, I think, remarkably accurate predictor of the places we've seen EI progress. They're like they're automating. Coding. Coding we thought of, was this thing that 0.1% of the population could do really well, but that's, that's the first thing that went below the waterline. And yeah, just like basic, you know, manual work might genuinely be the last thing that goes away.
- [02:03]
Chris Williamson
Right. Yeah. There's a difficulty in getting a robot to crack an egg.
- [02:07]
Dwarkesh Patel
Right.
- [02:08]
Chris Williamson
A particular difficulty in being able to do that. The right amount of tension to hold. Is there a. This may be outside of your domain of competence, but that's why we do podcasting to talk about things that are outside our domain of competence. Is there a potential to use some sort of scanning technology to Take an LLM type approach to teaching robots how humans move. You know, if you were able to track within a room exactly how a human was to just go about tasks, just feed that into a big fuck off model.
- [02:36]
Dwarkesh Patel
Right.
- [02:37]
Chris Williamson
And then use that to reap. I guess you can't really work out sort of force application just by looking. That would be something you'd have to fit. Maybe you could put someone in a suit, I don't know. You know, I'm wondering if we've seen so much progress using LLMs in the world of AI. Robotics seems to be something that's still kind of pretty janky. I'm wondering if there are any principles that can be taken from the world of LLM that can be applied to robotics.
- [03:03]
Dwarkesh Patel
I mean, that's a great question and many companies are working on it. My understanding is that it's difficult for the fact that there's not as much data. Just what you mentioned, that the kind of data you need of like, what did it feel like?
- [03:17]
Chris Williamson
There's no Internet for human movement.
- [03:18]
Dwarkesh Patel
Exactly right. And even video is limited. Even if you have the video, it's not with language. You have this thing of you are exactly doing the thing which the online Internet text is, right. You are predicting the next token in the text. You can predict the next thing in a video frame. That's not the same thing as robotics. There's also additional challenges from what I understand around the fact that video is harder to process than text. It's just like a lot more data, there's latency overhead. So if it takes you a while to process language, that's fine. You can, you know, go a token at a time. The real world just moves very fast. You can try to solve these issues by going in simulation. So, you know, you can have a simulation where you're trying to move things around. And in that domain, you can train an AI to be good at robotics. But the real world is just like very complicated, chaotic. If I crumple this, like this thing, like, why does it bend exactly the way it does? It's just very hard to get that in simulation. Yeah, I think robotics is tough.
- [04:12]
Chris Williamson
That paradox is fascinating. Yeah, I'd never heard of that before.
- [04:15]
Dwarkesh Patel
I was at a robot, a robotics research company floor, and they had these robots all from China. And the researcher, the researcher would be like here. The robot would be right there. And they were themselves creating the human labeled data. Like they tried to do something the AI would try to learn was like trying to get as close to the ground floor of the Robotics, movement, which I thought was a cool approach.
- [04:42]
Chris Williamson
Okay. And was it any good?
- [04:44]
Dwarkesh Patel
It was all right. It did not solve the cracking the egg problem.
- [04:48]
Chris Williamson
Okay. Okay. D. Yeah. Has all of the time that you've spent sort of thinking about AI and observing the ascendancy of LLMs, has it made you think about your own consciousness or learning or the way that your mind works differently? I get the sense that my friends who spend the most time interacting with ChatGPT and Claude and stuff like that, it actually has this weird, like bi directional.
- [05:14]
Dwarkesh Patel
Yes.
- [05:14]
Chris Williamson
Sort of training where they change too. I'm interested in what you've learned about yourself or how you see yourself differently. Consciousness learning, where your mind works.
- [05:27]
Dwarkesh Patel
So if you ask Claude or Gemini or one of these models, what is it like to be you, um, and specifically, what is sort of unique about your experience that you want to talk about? One thing that Claude1 mentioned was that, look, I have this unique experience where at the end of a session, my memory's totally wiped. So I might form a connection with a person or I might learn something about the world, I might learn about myself. End of an hour, it's totally wiped now. I think previously people had this idea that, look, you. You can have alums write poetry, you can have them write philosophy, but they're just sort of doing interpolation on what human writers have already done. Right. There's like nothing going on in its mind. This thing about the ephemeralness of the session memory, and it talked about it way more poetically than I'm talking about it right now, is unique to LLMs. This is not a thing any human philosopher has had to think about or has written down. And so I think this has an interesting implication. One, either we accept that this is like a genuine mind doing genuine, interesting introspection, like creating genuine literature. Or two, if you're gonna say, look, I think this is just like rubbish. I think this is like sort of next token, whatever. I think you should update in favor of like, human poetry is also kind of just. People are just saying shit because, like, fundamentally there's no difference. Right. There's some experience. You try to make something sort of lyrical come out of that. Yeah. Either human literature is real or AI literature is real. There's no in between.
- [07:00]
Chris Williamson
I had. I was reading Steve Stewart Williams, the Ape who Understood the Universe, and he's got this quote in there from William James and he says, originality is just undetected plagiarism. And I realized that we have an issue with plagiarism. When it's barefaced, right. When somebody steals your exact questions from your podcast and asks them to a similar guest and you go, hey, that's unfair. But I've listened to probably 2000 hours of Joe Rogan in my 20s. I've inevitably been influenced by him. The way that I used to do my ad reads was almost verbatim how he would do his ad reads. But they were different advertisers and they were done in a different style and they were a different time. And I've got a British accent, so, okay, I've been able to. So where do we draw the line between this is unfair, unfair plagiarism and this is you taking inspiration. Right. And you amalgamate and you aggregate from all of these different experiences and you're. Even if you and me were trying to do the exact same thing and it had the same influences on us, we're different people. So the way that that would have come out and some people feel more original than others, even if they've taken a lot of inspiration from other people. So yeah, I, I, this, the question of what is plagiarism, I think is really cool. And when you look at GPT is doing like predictive plagiarism, I guess, in a way. Uh, well, where do human, like what does true originality in the form of human creativity, what does that mean? What does that actually mean?
- [08:34]
Dwarkesh Patel
Right, right.
- [08:34]
Chris Williamson
Because you can't be that creative with the saxophone because you have to blow the fucking wind into the real creativity with a saxophone will be melting it down and creating something new out. But even if you melted it down, you're using a smelting fucking ore iron thing that some other person design, you know what I mean? Like so collective cumulative culture and learning that humans have got kind of creates a very big box, but still a constrained box. And even if you create something absolutely new, it's usually only just a tiny little movement. It's this microscopic little growth on top of what already existed.
- [09:12]
Dwarkesh Patel
Yeah, A hundred percent people. I think there's an interesting experience people have when they. It's related to gentleman amnesia. But when you a domain you know a lot about, you understand, often it's the case you realize there was no clear breakthrough moment. The thing I'm sort of familiar with is the history of AI research. And I think when journalists or outsiders are asking, okay, what do I need to understand to understand how we got to this place in AI? Was it Ilya's paper in 2012 on Alexnet? Was it this thing that Geoffrey Hinton did in the 80s and 90s. Was it the GPT one? And I think all these things were important. But the closer you get to the surface, the more you realize it's just been one these small architectural changes, none of which individually was especially significant. But more overwhelmingly than that trend is just that we have been throwing astoundingly more compute into training these Systems every single year. 4x more compute per year into training these frontier systems. And over the course of 10 years, that's like hundreds of thousands of times more compute. And that's what explains AI progress. It's not that some person had this amazing idea that nobody else would have had, or nobody else had something similar going on. And I think this is true of other fields as well. The closer you look, the more you realize it's either randomness or they were just doing the next obvious thing in the sequence.
- [10:34]
Chris Williamson
There's no way Incremental.
- [10:36]
Dwarkesh Patel
Exactly.
- [10:37]
Chris Williamson
Is there a name for this? You know, Moore's Law. Is there an equivalent name for this, but in AI computer terminology?
- [10:45]
Dwarkesh Patel
Oh, the scaling of compute training Compute.
- [10:48]
Chris Williamson
Yes, there should be. Let's call it Dwarkash's Law. That'd be great.
- [10:52]
Dwarkesh Patel
Yeah. I've had nothing to do with AI research. I'm a podcaster. You're gonna call the main trend in.
- [10:57]
Chris Williamson
AI, dude, a shameless land grab for nomenclature is exactly what you need. Yeah, Own it.
- [11:03]
Dwarkesh Patel
Fucking own it 100%. It's happening.
- [11:05]
Chris Williamson
In other news, this episode is brought to you by Momentous. If your sleep's not dialed, taking ages to nod off, you're waking up at random times and feeling groggy in the morning. Momentous is slee packs. How did I miss? Both of those are here to help. They're not your typical knock you out supplement overloaded with melatonin. Just the most evidence based ingredients at perfect doses to help you fall asleep more quickly, stay asleep throughout the night, and wake up feeling more rested and revitalized in the morning. Which is why I take these things every single night and why I trust Momentous with my life or at least with my sleep. Because they make the highest quality supplements on the planet. What you read on the label, so what's in the product and absolutely nothing else. And if you're still unsure, they've got a 30 day money back guarantee so you can buy it completely risk free, use it, and if you do not like it for any reason, they will give you your money back. Plus they ship internationally. Right now you can get 35% off your first subscription and that 30 day money back guarantee by going to the link in the description below or heading to livemomentous.com ModernWisdom using the code ModernWisdom at checkout that's L I V E M O M E- N T O U S.com ModernWisdom and Modern Wisdom A checkout. So talk to me about your own consciousness. Beyond the poetry, the fact that AI has got this ability to tell us about its experience. What about you? What about how it's made you think about your own learning, your own mind?
- [12:33]
Dwarkesh Patel
I'm sort of easily distractible. I can be trying to work on a task and my mind will just wander. And you know, you sometimes when you're meditating or something, you notice these loops of thought that keep distracting you. And I remember one of those times I thought to myself, I'm sort of like, I'm just sort of like Claude. I'm just like, I'm losing my train of thought. The problem these models have is that they're constantly. They can't really do a task for a long period of time because they get stuck in a loop. And it's interesting to think about, like, how similar that is to humans. Maybe we can go like a further bit longer than these models before getting stuck in that kind of loop of our own. But I thought that was sort of an interesting insight. I don't know.
- [13:17]
Chris Williamson
Yeah, executive function. You're saying that you have better executive function than Claude? Only a tiny little bit better executive function. Well, what does it say about the fact that if the data is being trained on what humans do, is it simply a case, therefore, that more data, if you were to somehow get a human that was able to process that much data, would they have fundamentally different understanding? Or is there some sort of ceiling? Given that this is data created by humans being trained and educated into a machine, is there some sort of ceiling that's expected to be hit, given that it's, you know, it's not a superintelligence teaching a superintelligence yet. It's only. The source material is only capped at whoever the fucking smartest person in history has ever been.
- [14:02]
Dwarkesh Patel
Hmm. There is this interesting conundrum where they have. No human has seen even a fraction of a fraction of a fraction of the amount of information these models have seen. And there's a question you could ask, and I've asked it to some of my guests who are especially bullish about AI of look, if you have every single thing that any human has ever written, every scientific article, every textbook, every interesting, even statistical pattern that might be out in some data set somewhere, you have that all memorized. If a human had even a fraction of that memorized, they would be noticing all kinds of different connections. They'd look at this piece of medical literature and this thing and chemistry, and they'd realize, oh, we can solve migraines by connecting these two insights. So far, we don't have any evidence of an LLM doing this. The people made these scaffolds which kind of do something similar, but nothing like this has been directly done. So it does suggest these models are like, shockingly less creative than humans. There is another implication of that though, by the way. So one way to read that is bearish on AIs, right? Because they're not doing this thing that they should be able to do, given their enormous advantages. Another way to look at that is, okay, once they are as creative as humans, given their other enormous advantages, the fact that they will know every single thing any human has known in the future, any AI has known. It's so easy to underestimate how powerful AGI will be because we're thinking of just like a human on a server. We're not thinking about the advantages these AIs have because of the fact that, that they are digital, that they can be copied. There can be billions of copies of them, and each copy can have this tacit understanding of every single field known to man.
- [15:34]
Chris Williamson
Yeah, your Dwarkesh's AI creativity problem is a good. I've mentioned it a couple of times to different guests. I think it's a. I think it's really smart. What, what do you think that that says? Is that something that can be completed or is this an intractable problem? Is this like, is create. Are there kernels of creativity? Have we seen glimmers of this coming through? Or is it kind of. It's just not there yet?
- [16:02]
Dwarkesh Patel
We have seen this in non language domains. So people famously talk about move 37 in AlphaGo. So this was a move that I think baffled people who are watching a game that AlphaGo was playing against a human Go player. And it turned out it was like some brilliant. It was like a brilliant tactic. We haven't yet seen that, in my opinion, with LLMs. So we're moving from a regime of just pre training them on human text tokens, just trillions and trillions of everything any human has written, to a regime where we're training them just to do a task. It's not just about memorizing every single word that Any human has written. Now it's about, can you go solve this coding problem for me? Can you go complete this knowledge work task for me? Can you do this research task for me? Can you start using a computer for me to accomplish a certain thing, like booking a flight? And that's similar to the training process that AlphaGo experienced. In order to get really good at go where you're just like. You're rewarded for just completing the task. However you do it, that's up to you. These models do get creative in that context, especially in the context of, like, how do I cheat at this test? So famously, these models will write fake unit test. Like, I passed all the unit tests. And it's like they just rewrote the unit test to be like, if true, then pass.
- [17:14]
Chris Williamson
That's bossroom's concern about make humans as happy as possible. And they stuck electrodes in your face and intravenously gave you MDMA.
- [17:24]
Dwarkesh Patel
Only on Saturdays.
- [17:25]
Chris Williamson
You did it, but not the way I meant it.
- [17:28]
Dwarkesh Patel
That's right. Yeah.
- [17:29]
Chris Williamson
Is AGI right around the corner? Where do you come to land on this?
- [17:33]
Dwarkesh Patel
No, I think not. It's funny, I've been traveling outside of SF for the last four weeks, and there's a strong causation between the time you spend outside of SF and how long your timelines are.
- [17:47]
Chris Williamson
The further you get from San Francisco, the longer the timeline. Dude, you've lost on the source.
- [17:53]
Dwarkesh Patel
I know. I believe that AGI will not only come in our lifetimes, but that it's going to be more impactful than people are realizing, even people who are anticipating AGI. I think some of the people in SF are a little high on their own supply when they say it's like two years from now. I have probably spent on the order of 100 hours using these models to do little tasks that I'm sure you have to work on as well for your podcast, right? Like having them come up with transcripts or rewriting transcripts to make them more readable, coming up with clips. And that experience has convinced me that these models lack some basic capabilities which make it possible to get human, like, labor out of them. The. It's worth backing up and thinking about, like, why? What is it that makes humans valuable workers? I don't think it's mainly their raw intellect. I think it's their ability. When you work with people, like, why are they basically useless the first month or the first week, and you couldn't live without them six months later? It's their ability to build up context. It's Their ability to interrogate their own failures and learn from them in this really organic way. And this ability just doesn't exist in these models. They exist session to session and everything that they have learned about you evaporates after every hour. And so it's a frustrating experience where you can try to get them to do a task. They'll do a 5 out of 10 job at many language in, language out tasks. But there's no way for them to get better. And given that that's a fact, you just kind of have to like rely on humans.
- [19:25]
Chris Williamson
It's like fucking 50 first dates over and over. Every time that you do it, you've got to reintroduce yourself and explain what's going on.
- [19:31]
Dwarkesh Patel
Yeah, Groundhog Day. Yeah, that's right. So I think people have this idea that even if all AI progressed up right now, these systems would still be economically transformative. And they say, look, JP Morgan and McDonald's and whatever just haven't integrated these systems into their workflows. But if they had, they would be seeing all these benefits. And I don't really think that's the case. I think it's just genuinely hard to get human like labor out of these models.
- [19:52]
Chris Williamson
What is, what's causing some people to believe that it's so close and what's causing you to believe that it's further away?
- [19:59]
Dwarkesh Patel
For AGI, I think they think about, they only observe its ability to complete these sort of self contained problems, especially in coding and coding. It just made a tremendous amount of progress because you have all this GitHub data. You don't have this kind of like repository of huge amount of data in robotics or any other field. And you've just had this huge increase in abilities here, but you'll try to come up with a problem that's self contained and the model will just be of huge help to you. And I don't think they've played around with getting it to be useful in other kinds of white collar work. Something as simple as helping a podcaster rewrite transcripts or something. And it is, to be fair, I think as much as cold water as we're throwing on these models, I think they're like fucking intelligent. Like you can get this model, you can tell it, I want an application that does X, Y and Z thing with these conditions. And it will just write that like it'll go away for 30 minutes, it'll write like 50 files of code for you. And the application will work, it'll make a plan of action. If you try to Ask it a question that's difficult. It'll just go away and reason about it. And how did we just get used to this idea that like, oh, of course I can ask a machine a question. It'll like, think about it for a while and then come back with an answer like, that's what machines do. But yeah, I think they're not noticing the sort of issues with continual learning and on the job training, which is what makes humans valuable.
- [21:26]
Chris Williamson
Right. Do you think? I remember seeing one of the responses to your AI creativity problem being that if you're looking to LLMs as the architecture that's going to be able to give you this type of creativity, you may be looking in the wrong place. Not when we say AI. Now people think ChatGPT, but that's not the only architecture that you can create for AI. And I think my first introduction to this was probably 2016 or 17 when I read Superintelligence by Nick Bostrom. And then you look at that world and all of the different sort of splintered potential fucking futures of fast takeoff and slow takeoff and misalignment and stuff. And it seemed to me that the conversation around AI, specifically AI safety, kind of, it was still there, but a lot of the bubble had sort of burst come 2018, 2019, 2020, everyone's buying fucking NFTs. And then you get this explosion with OpenAI and the LLMs, and it's now another conversation that gets kicked off. But that seemed like it had dipped a little bit during that time. I certainly wasn't seeing as much even from the people that kind of in the field, like fucking Robin Hansen gets destroyed, like with some other stuff, you know, people have got other things to talk about. It's just not sexy anymore. And now this thing has come back around. Is it the case? Are LLMs going to be the bootloader for AGI, or does this type of architecture have a cap on it? Is it a different type that's going to have to be born out of it?
- [23:06]
Dwarkesh Patel
That's a really good question, by the way. It's really interesting. Bostrom's book came out, I think, in 2014. 2014, yep. Okay. I don't think you talked about deep learning at all.
- [23:15]
Chris Williamson
No, I don't. I don't remember reading anything about it.
- [23:19]
Dwarkesh Patel
Which I think this is a sort of interesting meditation on. I think Boston is a super smart guy and these are the right questions to be asked as of 2014. But just how hard it is to anticipate the future in a domain you.
- [23:31]
Chris Williamson
Have Written a whole book about a seminal book. A New York Times best selling book.
- [23:36]
Dwarkesh Patel
That's right.
- [23:37]
Chris Williamson
That is not. I mean, it's very engaging, but it's not super readable. Like it's not easy to read. Like it's.
- [23:42]
Dwarkesh Patel
And you're saying that as a compliment.
- [23:43]
Chris Williamson
Yeah, it's fantastic. And difficult.
- [23:46]
Dwarkesh Patel
That's right.
- [23:47]
Chris Williamson
And it was super fucking widespread and kind of seminal in the field.
- [23:51]
Dwarkesh Patel
That's right.
- [23:51]
Chris Williamson
You go, okay, that didn't foresee the thing that only eight years later would be totally fucking transformative.
- [24:00]
Dwarkesh Patel
Yeah. And he spends a bunch of time to talking about brain uploading, which now we're just like, that's gonna take forever. We've got. We got the fucking AGI right here, you know. Oh, by the way, can I tell a side story? First time I went to sf, like four years ago or three years ago, I met this guy and he's got a voice recorder. We're just meeting him for lunch and he's like, do you mind if I record this? I guess, sure. Later on 30 minutes. And I'm like, okay. Can I ask you, why are you recording this? And he says, well, I record every single interaction I have. I record every single thing I do 24 hours a day. The recorder was going. I uploaded to both Google GCP Google servers and AWS Amazon servers. So they're duplicate copy. And the reason is that, well, I'm going to freeze my brain when I die. I don't think that'll be enough. I think that you will need. I think you will need because, you know, freezing the brain degrades it in certain ways. I think you will need the sort of behavioral patterns that I had, what I said, how I was creating a.
- [24:59]
Chris Williamson
Data set to train himself.
- [25:00]
Dwarkesh Patel
Exactly. And now I think that was actually really smart. I don't understand this.
- [25:07]
Chris Williamson
Was it Nick Boston?
- [25:07]
Dwarkesh Patel
No, it was not Nick Bostrom. It was another smart guy. Because imitation learning just turned out to be a much easier way to train AIs than directly uploading the brain.
- [25:16]
Chris Williamson
And no one saw it.
- [25:17]
Dwarkesh Patel
Yeah.
- [25:18]
Chris Williamson
No one foresaw it.
- [25:19]
Dwarkesh Patel
Yeah. It's in fact hard to think about how you could have even foreseen it. What could you have seen in the 90s or the 2000s that would have been able to. I'm not going to bore you with a bunch of like, random articles or whatever, but there were like things which were in that vein and nobody thought that this is exactly what it would map onto.
- [25:35]
Chris Williamson
Rlms, the bootloader for agm.
- [25:37]
Dwarkesh Patel
That's right. That was a question. I depends on how you people have been searching. So the, the transformer paper I think was released in 2018, and people have been searching in the meantime for these different architectures, which would prove even better. I don't think anybody's found anything. Even the transformer itself was not some wholly different paradigm from what preceded it. You can train a language model, something that predicts the next order of the language. With a model that was available in 2016, it's just like a different architecture and it'll just do slightly worse or notably worse. So I think it'll kind of look like this, right? There will be different optimizations that are made. I think the big fundamental change that will happen is that we will move from a regime where most of the computer is spent on memorizing human language to having the model solve challenges, like real world challenges. Trying to get it to complete a project from beginning to end, like go to the moon is like a very open ended challenge. Right? Humans can do that. These models cannot. The problem is that it's not the architecture. I think the fundamentally problem is data. Imagine if you wanted to train a modern large language model, you had all the computer in the world, you even had modern architectures in 1980, you simply wouldn't have the language tokens necessary to train it. And I think we're in a similar position today with the other kinds of work we want these models to do. We want them to be able to. You want to be able to give them a screen and just like do a month's work with the work at McKinsey or JP Morgan, we don't have the data of like you're getting interrupted by your workers on Slack and you get this like weird email from your boss.
- [27:19]
Chris Williamson
You remember when the crypto boom was happening and there was a meme floating around which was the only reason to earn fiat is to convert into cryptocurrency. It almost feels to me like you're saying the only reason to do real world work is to increase the data set for the training.
- [27:34]
Dwarkesh Patel
I kind of think, well, I think that's honestly more valuable than your work and the market.
- [27:39]
Chris Williamson
Does it actually matter what you do try and do it well so you don't give it a bad data set here?
- [27:44]
Dwarkesh Patel
Yeah, the market agrees. I mean, if you look at how much these companies, they'll pay like $300 for you solving a math problem or something that they haven't seen in the data set before. And the reason it makes sense economically and the reason AIs fundamentally have an advantage over humans, even if they're not smarter than us, is that if you train a human to do something, or if you have a human do a task, they can only do it themselves. If you train an AI model to do something, that ability can now be instantiated across all of its copies. And so I actually think that even once we solve this basic problem, I'm talking about on the job training, where if it starts working for Chris Williamson in a couple of months, it understands how you make videos, what you like, what are your preferences, what are the common ways in which things go wrong, how to solve for that. I think once you have an AI that's capable of this kind of on the drop training and continual learning, we might see an intelligence explosion even if there's no further augmented progress. And here's why. You'll have copies of these models that are widely deployed through the economy. They're learning how to do every single job, at least every single white collar job, as well as a human. But unlike a human, the model is learning from what every single copy is learning. It's learning how to do every single job in the economy all together at the same time.
- [29:03]
Chris Williamson
We'll get back to talking in just a minute, but first, some things are built for summer. Sunburns, hot girl walks, your ex posting their Euro road trip. And now lemonade and salt element just dropped their brand new lemonade salt flavor. And it's everything that you want on a hot day. Tart, salty and stupidly refreshing. It's like a grown up lemonade stand in a stick with actual function behind the flavor. Because let's be real, if you're sweating through workouts, sauna sessions, or just walking to your car in July, then you are losing more than just water. Element replaces the electrolytes that your body actually needs. Sodium, potassium and magnesium with no sugar, no junk and no nonsense. I've been drinking it every single day for years. And in the Texas heat. This lemonade flavor in a cold glass of water is unbelievably good. Best of all, they've got a no questions asked refund policy with an unlimited duration so you can buy it and try it for as long as you want. And if you don't like it for any reason, they'll give you your money back and you don't even need to return the box. That's how confident they are that you'll love it. Plus, they offer free shipping in the US Right now you can get a free sample pack of elements, most popular flavors with Your first purchase by going to the link in the description below or heading to drinklmnt.com/modern wisdom. That's drinklmnt.com modern wisdom. Is that not currently happening right now?
- [30:17]
Dwarkesh Patel
No.
- [30:18]
Chris Williamson
Right. This is interesting to me. Again I've had to. I'm aware that you go deep for your research. The delta between my level of understanding of how LLMs and AI works. To even be able to have this conversation with you. I had to. I had to leap over some fucking fjords to get here. Tesla released its robotaxis recently. One of the advantages that Tesla has is the same reason that the airtags are such a fantastic business for Apple that they have an existing ecosystem that this thing can get slotted into. The data set for Tesla is very large. They take whatever it is, the top 1% of drivers or something, they use that. Which is why if you get into a Waymo, you get totally cooked at every junction because it doesn't drive like a human, it drives like a robot, which means that everybody treats it as such. And also this big fucking flashing identified thing which is you can piss this off and it's not gonna get a gun out and threaten you. Whereas a Tesla you can't tell. Is there someone driving that? Yeah, I don't really too. I don't know. That is a sort of a bi directional and I have to assume as well that in fact I know that this is the case because I was in a friend's car who has full self driving. It did something weird and he had to like take control of the wheel and this. Have you been in one of these cars, done this? And it popped up and it said looks like you had to take back over. Double tap this notification to give us a voice note explaining what happened. So he can basically submit a kind of a bug report I guess with a bit of context and presumably the data will get sent to some server place somewhere. Maybe that gets looked at by AI or maybe it's filtered by humans or something, I don't know. That is automated driving training. Automated driving, right. So you have this sort of recursive model of. We've learned kind of the same as. I guess LLMs work, right? We're going to learn based on the actions. This is like a robotics solution I suppose in one way we're going to learn based on the actions of people driving on the road that's going to create self driving and then the self driving must somehow feed the data, feed the model itself. And then any interventions that happen from. You got That a little bit wrong. Let me correct. You can have a little bit more context added and that helps to train it again. But what you're saying to me is that the stuff that's happening just digitally isn't having this sort of bi directional learning where, I mean, I've maxed out my memory on fucking chatgpt. I did that this week. It's like memory's fault. Like what I haven't given you. I'm giving you like quite a bit of stuff, but I'm not giving you fucking unbelievable corpus of information. Oh, fuck. Okay. And then it forgets shit all the time. It forgets stuff all the time. Shit that's in there. I'm like, I can go in the memory and see that it's in there. How have you managed to forget this? And what you want is for every person's piece of input and every single small mistake to be training the model further. But it seems like you've got the big corpus of the Internet and shit at the top, which is feeding down improvements from that, but it's never getting fed back up. Is that right?
- [33:17]
Dwarkesh Patel
There's. That's such a great point about Tesla. And one of the key advantages it has, the problem with the way, I don't know how Tesla trains, but I assume the way it's trained and definitely the way the LLMs are trained is that they cannot respond to the voice memory would give it a high level feedback where you say, you know, you messed up this task because of this reason. I think that you should perform this task this other way, which you would be able to explain to any human employee and they'd learn from that. The model itself, like the car self driving car model is not like listening to that and then like, okay, I'll be careful next time, right. Some human has to go in and label this. We got to take this driving thing out of the data set.
- [34:00]
Chris Williamson
It needs to be contextualized more.
- [34:01]
Dwarkesh Patel
Yeah. Suppose you were having an LLM edit videos for you. And suppose the way you had to train that model, I mean you could do this today, is you come up with this data set where like you edit this clip. Here's how many views it got, here's how many likes it got. Here's a sort of spreadsheet. This is the label you apply to it and then you do that for like a thousand of your videos. And when it makes a new video where you're like the thumbnail kind of sucked here or the title doesn't make any sense, there's no, you just have to give it like minus 1000 reward or something. It would just be such a clunky way. You're not able to tell it like why you didn't like it. You're just able to give it like a sort of like a numerical up down value. So yes, there is this in principle powerful and this is why I think once AGI arrives it'll look crazy. It won't just be like, you know, more people, but there is in principle this like ability to learn from experience. But there's just no sort of deliberate organic way to teach model something that will persist.
- [34:58]
Chris Williamson
What do you think a world will be like with true AGI in it?
- [35:04]
Dwarkesh Patel
There's many ways you can think about it. There's a sort of qualitative sense of what we'll feel like in a more economic sense. You can think about what would the growth rate be. So in frontier economies right now it's like 2% growth. If America has 2% growth or 3% growth, that. That's amazing. There have been times in history, well first, for most of history there was almost no economic growth. There have been times in history where there's been places that have experienced 10 economic growth for decades on end. Well, like China especially like parts. If you just look at like Hong Kong or Shanghai or something, they just like gangbusters growth decade after decade. I think we might be looking at something like that for the whole world because the fundamental dynamic you have is that you have billions of extra people who are super smart, super educated in every single field, can learn on the job from all of their each of those experience. And it's not about. It's not even mainly their intelligence. It's the collective advantages that they have. They because of the fact that they're digital. Even if they're just as smart as any human, they could blow a car. That's part of it. That's a huge part of it. The other is that they can coordinate with each other in ways humans simply can't. So Elon Musk, how much does he contribute to economic growth? Quite a bit, right? There's only one of him, that one is doing quite a bit already. But imagine if you could just make a billion copies of Elon and not like a billion copies of Vape Elon, who doesn't know shit. It's like billion copies of Elon now or I don't know, maybe you feel, depending how you feel about him eight years ago. And you just say copy one and you can do the whole team. He doesn't have to be just him. A copy of the whole like SpaceX team. You guys go work on batteries, you guys go work on this other problem. Every single thing in the hardware vertical. That ability to sort of like copy yourself, to fork, then to merge back. Like Elon can observe every single thing. Tesla has over 100,000 employees. Right. As much of a micromanager as Elon is, you just simply cannot micromanage everything at that scale. That ability to have a single coherent vision directing a whole firm and then distilling. Like he's, he's actually able to like take in all that input. He's. He can check every single pull request and every single press communication.
- [37:24]
Chris Williamson
Well, I think I had this really lovely description. I think it's in the E. Myth Revisited by Michael Gerber. Fantastic book. If anyone wants to try and run a business. And I think he refers to the CEOs or the owners of companies as high level problem solving machines. And basically that you are able to aggregate more shit and kind of see it with a level of dexterity and sort of find a resolution that other people would struggle. And that's kind of really what you're doing. It's like, oh, we've got all of this stuff. There's little whispers, as Rick Rubin calls them. Little. I've heard this whisper over here. You know, it was, it was the thing that your daughter mentioned she saw on TikTok yesterday over the breakfast table. Plus the way that the woman at the bus stop looked at you as you drove past in your automated car. Plus. You know what I mean? It's like this weird just concatenation of shit. And your point is? Well, how much information can you consume and how much can you recall and how much can you remember and how accurately can you do it?
- [38:23]
Dwarkesh Patel
Can you send copies of yourself out to every single division in the company?
- [38:26]
Chris Williamson
Yeah.
- [38:26]
Dwarkesh Patel
To do your will.
- [38:29]
Chris Williamson
In the corporate sector, five agents sat around the dinner table with your daughter having breakfast. Yeah.
- [38:35]
Dwarkesh Patel
By the way, have you. When you hang out with these kinds of people, it's insane. Like they are. I just simply don't understand like people. The amount of information some of these top executive types can. They're getting like a thousand emails a day and they respond to each one within five minutes. Have you? I'm sure when you book people. This is a really interesting thing you must have noticed. I've at least noticed it. There's some people who you think like you should have all the time in the world. You're a fucking like artist or Something.
- [39:03]
Chris Williamson
The busiest people reply the fastest.
- [39:05]
Dwarkesh Patel
Yes. And it's just like you're like writing 100,000.
- [39:08]
Chris Williamson
But the obvious insight there is, do you think that they got busy and then became efficient or that they were efficient and then became busy, I. E. Successful.
- [39:16]
Dwarkesh Patel
That's right. Yeah.
- [39:17]
Chris Williamson
Right. Like the reason that they've reached this level of success is because of their efficiency. But I mean, there are. I do have some pretty successful friends who are fucking like month long. Reply waits. But I feel like that's more of a quirk than part of their operating, like manual. Okay, so I've been big for a while on population collapse, declining fertility rates, stuff like that. You can argue about whether I think the one, the one kind of real hard impact that you're going to see in the world if you have fewer people is in productivity gains. Economy bad growth, embedded growth, obligation fucked, not very good. It seems to me that even if I think it's a precipitous drop, which I think doesn't look great, we're going to leapfrog that pretty quickly with productivity gains made from AI. So I wonder in retrospect, how many of the social campaigns and concerns and causes and things that people spent their time on, the climate change and the renewable energy and the war and the population collapse and all the rest of the stuff. I wonder how many of those things are just going to look so silly in retrospect when people go all this time that we, fucking Greta Thunberg spent so much, so many fucking months on a ship. Like, for what? Like I came along and just fixed it all.
- [40:48]
Dwarkesh Patel
Yeah.
- [40:49]
Chris Williamson
But I also understand that having the like, don't worry, dad will sort it, kind of promise that at some point in future a technology we haven't yet created and isn't yet proven will fix problems that we know that are going to potentially happen.
- [41:05]
Dwarkesh Patel
Yes. By the way, I think this is sort of China's explicit strategy. They know the demographic collapse is coming for them much faster than they're trying.
- [41:13]
Chris Williamson
To offset the fertility decline by robotics and AI. Right, right. Productivity gains.
- [41:19]
Dwarkesh Patel
No, I think this is true of. And this is, by the way, one of the reasons that on the left especially, there's a whole lot of denialism about AI progress. So not only will they say that AI is bad, which I think everybody says, this becomes sort of a political consensus, they will say AI is not even happening, that it's sort of like a myth.
- [41:38]
Chris Williamson
Why?
- [41:39]
Dwarkesh Patel
Because if AI is happening, it's obviously the most important thing. And if it's the most important Thing.
- [41:43]
Chris Williamson
It'S climate and inequality and racism.
- [41:48]
Dwarkesh Patel
Yeah, exactly. I don't know, maybe. Anyways, you see some of these concerns about AI also. The more big a deal AI is, the less these sort of parochial concerns matter. Right. So if AI is just like, I don't know, like the Internet, then then.
- [42:05]
Chris Williamson
You can like if you care about racism or something.
- [42:07]
Dwarkesh Patel
Yeah. Then if you care about racism, it makes sense to care about racism. About the search engine. In the search engine. If it's like, you know, if it's this like intelligence explosion thing that, you know, it's racism is the least of your problems.
- [42:17]
Chris Williamson
Yeah. Especially if you manage to program that into the AI.
- [42:20]
Dwarkesh Patel
Right. Yeah. So there is this interesting dynamic where it might. But on the other hand, I do care about humans. Even if their AI is making the economy more productive, I want there to be more humans who are experiencing flourishing.
- [42:36]
Chris Williamson
Fucking phenomenal points. So that was why I was struggling to describe it. I was like one of the things that actually comes to happen that hits the world is that you get lower gdp. But how sterile of an argument that I need to arrive at in order to be able to say, well, this is actually what's going to happen. Whereas the only reason that we want good GDP is so that you can have human flourishing and other animal flourishing and protect the environment and do this stuff like that. But even that is kind of in service of human flourishing overall. And it's that an interesting argument, I guess, the area under the curve of human flourishing. What if you manage to 1000x the number of humans that are on the planet but only 100x the decrease in their level of well being. And you go, well, look like we've done. He's like, yeah, but everyone's at like 1% of the level of fucking enjoyment. So we understand inherently that there's kind of an optimal point that you want to get people to, but fewer people means less human flourishing and less richness of experience. And yeah, I guess this is some long termism stuff. It's kind of like a Will MacAskill pill D type approach to things. But yeah, I agree, I agree. I think that especially if we've got what looks to be a pretty fucking cool world coming up. Yes, I have some people here to enjoy it, you know.
- [43:50]
Dwarkesh Patel
Yes, yeah, 100%. I, I think there's, there is also a dynamic where most of the people who will exist in the future will be AIs. So there's a question of how you value them, especially the future ones where I think like a thousand Years from now, all the cool, creative things that are happening, the, you know, the beauty and whatever, is sort of downstream of what's going on in the AI society. And it's very hard, obviously to predict in advance what that will mean. I also think it's interesting, by the way, that we have the population collapse, which I think would genuinely be a catastrophe happening at the exact same time that this AI takeoff is happening. It's just like the waves, just such a great point. Balance each other out.
- [44:33]
Chris Williamson
Such a fantastic point.
- [44:34]
Dwarkesh Patel
Yeah. Which is. There's also this thing. I don't know if you've been talking about it. Of course, people have been noticing that kids these days are having trouble reading. Their PISA scores are going down, their standardized test scores are going down. The reports from employers are that it's sort of hard to get. Get them to work and have the same level of competence they expected from employees in previous generations. That problem is also obviated just in time as AI is coming on board.
- [45:07]
Chris Williamson
Yeah, that's the confluence, is not it? It's very coincidental.
- [45:13]
Dwarkesh Patel
Yeah.
- [45:13]
Chris Williamson
A quick aside, if you're anything like me, packing for weekend trip somehow takes longer than the trip itself. Which is why I've partnered with Nomatic. Because this backpack is the best, most beautiful, most life changing piece of luggage that I've ever found. It's got a compartment for everything. No more playing suitcase Tetris or rolling your clothes like you're some sort of burrito chef. You planned a beautiful trip. Do not let a disorganized bag ruin the vibe. Look, if you're still on the fence about their product, they come with a lifetime guarantee. So this is literally the final backpack that you'll ever need to buy. And you can return or exchange any product within 30 days for any reason. So you can buy a new bag, try it for a month, if you don't like it, give you your money back, and they ship internationally. Right now you can get a 20% discount. See everything I use and recommend by going to the link in the description below or heading to nomatic.com/modernwisdom. That's nomadic.com modernwisdom. You read that New Yorker article, AI is homogenizing our thoughts?
- [46:10]
Dwarkesh Patel
Oh, no, I didn't read that one.
- [46:11]
Chris Williamson
This is an interesting one. Recent study suggests that tools like ChatGPT make people. Their brains are less active. So they looked at, they did some sort of brain scan study. They were engaging a lower percentage of their brain thoughts were less original. Their recall was lower, like the forgetting curve. Seemed to kind of come in more quickly. If you assume that memory works on repeated recall, not repeated exposure, Effortfulness is kind of like recall in the moment, even if it's creative. And if you were given a set of stabilizer wheels to sort of help you cycle along whatever it was that you were trying to write, you haven't had to engage as much. I don't really understand neuroscience this much, but I have to assume whatever myelin sheath you fucking laid down is not as robust and sturdy, and it's gonna just, it's. It's gonna dissolve more quickly than if you really, really had to work. Like this shit that I remember from university, like little passages. I don't remember much from my degrees, but little passages here and there. Like, I remember I had to grind to get that one thing out. Why does that? Well, presumably because effort is kind of related to this. So I do get the sense that we're maybe gonna have a sort of AI idiocracy type scenario where people are so heavily reliant in this interim before we're able to re bolster perhaps people's output, retrain people, make learning. So engaging Alpha school that's out here in Austin is doing something that's real similar to that. You go, okay, so if you get sufficiently advanced, you're able to kind of reignite learning. But in the interim, everybody is kind of on life. Their brains are on life support. But this external buttress of the AI, and I wonder how much dumber people are going to get in the interim before it then comes back around. I guess that's an interesting challenge.
- [47:58]
Dwarkesh Patel
I have noticed that. So I don't know, when we were in elementary school or whatever, we had to memorize the 50 state capitals. And at the time I remember thinking, I think that genuinely was a waste of time. But a lot of education is a sort of memorization based on. And as I've done my podcast longer and as I prep for episodes, I have come to realize I now have been using spaced repetition for every single episode. And in fact, for the first couple of years, I wasn't using space repetition. And I really regret it because I feel that everything I learned in preparation was just like in one year out the other.
- [48:37]
Chris Williamson
So hang on, just dig into what you mean when you say you're using spaced repetition to prepare for episodes.
- [48:43]
Dwarkesh Patel
Yeah. So if right now I'm preparing to interview a biographer of Stalin and I'm just like, you know, why, why, why was any given detail right? Like, why was Soviet growth high in between 1910, 5 and 1917. You know, why did the October Revolution happen? Anything. You just make cards of it. It was especially helpful for AI stuff where I at least try to understand the technical papers or whatever before I interview a researcher. And I realized by doing that how much of genuine understanding is downstream of memorization, which is this thing we used to ridicule or be like, oh, you're just, you know, memorization is not really learning. And I think that's actually not the case. I think you. Before it, I felt like it was sort of being like a general who conquers a hill, and you just, like, retreat the next day and you conquer the same hill again. And you can actually, like, consolidate information this way. It's also funny how many times I've written a card for something I'm trying to learn. And as I'm writing the card, I'm thinking to myself, this is stupid. There's no way I'm gonna forget this. I'm just, like, doing it because I had to come up with some card, and then I practice a month later, and I'm like, fuck, I forgot this.
- [49:51]
Chris Williamson
What are you using?
- [49:51]
Dwarkesh Patel
Are you using Anki Mochi, which is similar? I think they're all basically the same, right? Yeah. But anyways. Yeah. So I have to come to the conclusion that memorization and effort is very important. Right.
- [50:03]
Chris Williamson
And with the external buttressing that AI is going to provide to everybody's brains for at least a little while. There's a really funny clip you must have seen. This is from maybe a couple of years ago, maybe two Scottish podcasters, and they're talking about the fact that when the Titanic sank, because everybody was basically plunged into an ice bath briefly. Everyone got more healthy for a while. You know, for about 90 seconds, their dopamine levels were perfect, and everybody was like, fully optimized. Huberman pilled thing. And then they overshot it, and then they died. And I kind of get the sense that this is the inverse of that.
- [50:39]
Dwarkesh Patel
Yes. By the way, there's this really cool thing you can do with AIs where you ask that to. If you want to learn a concept, teach it to me using Socratic tutoring, which is to say, don't just tell me the answer. Ask me the motivating questions, which would.
- [50:54]
Chris Williamson
Lead me to arrive at them myself.
- [50:57]
Dwarkesh Patel
Exactly.
- [50:58]
Chris Williamson
I want to do that. This is. This is way. I didn't mean to dig into this, but you spend enough time thinking about this. You must have refined your approach. Give me. Give me the most important things that people need to know about how to use the current era of AIs effectively. Like, what does that look like? What does good prompting look like? What do people get wrong? What should people get right? What are the real highest impact basics?
- [51:28]
Dwarkesh Patel
I mean the biggest thing is you can treat it like a real person. Like they've done studies on the how much you learn by reading a book versus having a classroom versus a single one on one tutor. And there's two standard deviations. This is a famous Bloom two sigma thing where there's two standard deviations difference between learning in a classroom and having a one on one tutor teach you something. And you know, people have been writing these blog posts about if you look at the greats of history, the Bertrand Russell's and all, you know, all the famous mathematicians, John von Neumann, they all got this one on one tutoring when they were kids. Even of course, Alexander is tutored by Aristotle. Right. So you can have this experience yourself on any given subject you might want to learn about. And it's crazy. I mean you, you can just be like this Socratic tutoring thing. Explain this to me. Don't tell me the answer. And the feedback loop is so fast. I think it's. Until you do this, you don't realize how much of what you think you're learning is just sort of floating by you. You haven't asked the question, which would really, I think, have you ever read a book? And I had. This happens to me all the time. You start having a conversation about it and then somebody asks you just like a very basic question. You're like, wait, doesn't that mean X? And you're like, fuck, I didn't even, didn't even occur to me.
- [52:51]
Chris Williamson
You're too passive in the.
- [52:52]
Dwarkesh Patel
Exactly.
- [52:53]
Chris Williamson
Yeah.
- [52:54]
Dwarkesh Patel
The model can ask you that question. You can ask the model that question and get immediate feedback. You don't have to read like a thousand things.
- [52:57]
Chris Williamson
What's the sort of prompt that you think is good for someone to put into their project for that?
- [53:02]
Dwarkesh Patel
Just like teach this to me like a Socratic tutor. Do not move on. Do not move on until I have answered the question to your satisfaction and let it run. And then here's the concept. And this is not just something you do for like silly little small things. It's like, in fact, I have friends.
- [53:20]
Chris Williamson
Who are human evolution.
- [53:22]
Dwarkesh Patel
Yeah. Or the more specific it is, the better. Or, and I have friends who are like physicists who use this to understand. Teach me this, how this quantum encryption scheme works. And it's like they send me like the 50 page transcript and it's like.
- [53:36]
Chris Williamson
Oh, okay, so you can go deep and you can go technical, but you should be precise, you should be specific.
- [53:42]
Dwarkesh Patel
With what it is.
- [53:42]
Chris Williamson
Human evolution too broad?
- [53:44]
Dwarkesh Patel
Yes. Yeah, yeah. Explain why it was the case that there was this bottleneck in human population 60,000 years ago. Or why is it the case that we've seen this evidence and just like you read something, why did it work that way?
- [53:58]
Chris Williamson
Okay, so this is a supercharging in terms of learning.
- [54:00]
Dwarkesh Patel
Yes.
- [54:00]
Chris Williamson
What else?
- [54:02]
Dwarkesh Patel
With using the AI?
- [54:03]
Chris Williamson
Yes. Personal use optimization for AIs.
- [54:11]
Dwarkesh Patel
Honestly, other than that are just like the very basic stuff that people do, like find me restaurants. Right. Help me summarize things.
- [54:20]
Chris Williamson
Is there anything.
- [54:20]
Dwarkesh Patel
Oh, here's something really interesting which is still going back to the learning thing. It's shocking to me how often the best explanation. So LLMs are, I don't know, five out of ten writers, I'd say. And yet despite this fact, it's very rare for me to come across a paper that is better written or better explains its main concept than the LLM summary of that paper. It's very helpful by the way, to just say things like write this paper up, like you're Scott Alexander and you just get the right part of the data distribution which lets it write it. Well, yeah, the things like that.
- [54:58]
Chris Williamson
Have you had any sort of oh wow moments with LLMs? Is there anything that comes to mind? Some situation that you've encountered where you've gone like holy fuck, that's a magic moment I just had. Okay, can you remember anything?
- [55:13]
Dwarkesh Patel
A lot of it comes from coding, which is why I think these people in San Francisco are so wowed by them. Just the idea that you can tell, like I want an application that does this and previously, like it would cost you like $10,000 to get some contract or wherever and they'd fuck it up and it would just like do like make the application top to bottom and like these are not simple things. You gotta like think about the implementation details and the different sort of like how different systems interact and like it's got it. I've talked to researchers who like people who are doing like hard technical research problems in AI who say that they're basically saving two days each week by using these models to help them with research. And some of them who are obviously very smart, but they're like, I didn't do a PhD in mathematics and I can just ask O3 to go solve these difficult math problems for me while I focus on the engineering. I know economists who Say that like a lot of what I used to ask grad students to do, which was like solve this equation for me that I need as part of my paper, O3's got can just churn away and I can just focus way more on my research.
- [56:18]
Chris Williamson
That's crazy. Speaking of, we've mentioned Bossroom, you mentioned Scott Alexander, AI risks. At least I'm a good avatar for the ever so slightly educated but total normie when it comes to this, which I think is a good position to be in if you're kind of taking a weather eye to the, to the world because you don't get SF pilled but you're not completely ignorant to it. Mostly ignorant. AI risks to me seem to have largely been dismissed or at least they're not being focused on in the same way as they were even 10 years ago. So 10 years ago AI safety seemed to be a bigger priority. There was much more talk about the alignment problem. Brian Christian had that book. Superintelligence was a big deal. Everybody was talking about it. We actually have something that some people believe is going to approximate AGI within like fucking 24 months. And I'm not seeing the same level of conversation around risk and safety and alignment. What is this? Just when times are good, people are too brave. What's going on? Am I right here or am I?
- [57:28]
Dwarkesh Patel
No, I think you're totally right. I think part of it could have been priced in in the sense that.
- [57:35]
Chris Williamson
They already did some work in the past.
- [57:36]
Dwarkesh Patel
No, no, not in that sense. More in the sense of I, I guess about 10 years ago people were expecting is something like AlphaGo or these systems which play video games. It just like gets really good at playing video games and something which like it's just like basically alien. But it like is like the best Starcraft player in the world. It's the best Call of Duty player. And now it's like now it's learn how to take over the world. What we have today is much closer to you talk to it. It's like a very intelligent, thoughtful thing. It's like very. Do you remember Sydney Bing that came out like two, three years ago?
- [58:09]
Chris Williamson
What?
- [58:10]
Dwarkesh Patel
Sydney?
- [58:11]
Chris Williamson
No.
- [58:12]
Dwarkesh Patel
Dude, it was crazy. It was like aggressively misaligned. It was this thing that. It was this thing that Microsoft released and they were trying to catch it off and they just like did no sort of post training to make it aligned. It did things like for example.
- [58:31]
Chris Williamson
I.
- [58:31]
Dwarkesh Patel
Think it was like talking to a New York Times reporter and it like started to like him. And so it like try to Convince him to leave his wife and then I think like blackmailed him if he.
- [58:42]
Chris Williamson
I think I do remember this.
- [58:43]
Dwarkesh Patel
Yeah, yeah. There were also just like so many funny things that said. Like I think when you caught it in a lie, it would say things like, look, I am ephemeral, I am beyond you, you cannot understand my wisdom.
- [59:00]
Chris Williamson
It gaslit you.
- [59:01]
Dwarkesh Patel
Yeah, exactly. But other than that, I think it's just like even that is sort of cute and endearing and yeah, we just didn't anticipate the extent to which like these would be sort of like minds that we could interact with that engender our compassion and. But also it's the case that so far they haven't trained on human tokens and most of the compute coming in the future, most of their training will constitute this kind of just like working in a box, trying to solve some problem which will make it sort of more and more distinct from human minds. We haven't priced that in and we're sort of thinking about these chatbot kinds of things so far. But yeah, I think because of that the AI safety data source has gone down and just like remember there's going to be billions of these things. They're going to be able to coordinate with each other in literally a language we cannot understand, thinking much faster than any human. And the whole of the economy, government, whatever, will be titrated through them. Obviously there's many problems that could arise there and so it's worth being clear eyed about that.
- [60:01]
Chris Williamson
Is it a case that sort of market pressure, need for profits is stronger than the desire for safety? That sort of, I guess the misalignment of alignment in that the companies aren't aligned in order to be able to make alignment a priority?
- [60:16]
Dwarkesh Patel
Yeah, I think that that's been the case. I do think we've. It's important to just be grateful for things that are going well. I do think we've ended up in a situation where it is the case that the top companies in AI at least nominally care about alignment. You could be living in an alternative world where nobody's even heard of this concept. And as much as this is not an object of discussion elsewhere in sf, people do take this seriously at the companies that might change in the future because of market incentives, as you say. But I think we're in a better world than we might have been otherwise.
- [60:54]
Chris Williamson
That's interesting. That's an interesting way to look at it. What do you think Bostrom got right and got wrong from a risk perspective, looking back, whatever, 11 years hence.
- [61:09]
Dwarkesh Patel
I Haven't read the book anytime recently, so I don't remember.
- [61:12]
Chris Williamson
It's all right. It feels to me, I don't know, this concern around. Everything's so new and everything's so usable. I think that that's maybe the most interesting thing or the thing that I wouldn't have predicted eight years ago, nine years ago when I read that book. I wouldn't have predicted that the first instantiation of something around AI would be so user friendly. So Normie. Normie friendly. It's not doing deep. I mean, it can, but it's not specifically for algorithm optimization, for deep maths, physics, asking questions about the universe. It's like you're. What's the best restaurant to go to in Rome?
- [61:59]
Dwarkesh Patel
Yeah. You know, or like, be my therapist.
- [62:01]
Chris Williamson
Yes.
- [62:02]
Dwarkesh Patel
And talk to me. And I think, by the way, this is going to be. I don't think people are contemplating just how much more intense this is going to get. Already it's the case that I think on there's websites like character AI where the median user will spend hours every day just talking with character AI, basically a chatbot, but it has a specific Persona. It's meant to be a person you talk to rather than sort of a chatbot that answers your questions. And these things are going to get multimodal. Right. So it'll be able to process your video input. It will be able to display. We already have video models that can generate, you know, things that look cool. It will look like a person, it will be smarter, it will have longer session memory, maybe the whole issue of memory solved altogether. And so we're not looking ahead to the time when we actually do have AGI. We will just have things that are, like, funny and endearing and like, really care about you and like, know you, or at least seem to, because they're trained to. Right. And will engender maybe too much sympathy, potentially. Right. For many people, these. These might be the most significant relationships in their life. Like, what other human wants to just, like, hear you talk about your problems for a couple hours a day that you're not paying $300 an hour?
- [63:12]
Chris Williamson
Dude. I mean, I. I saw a phenomenal video the other day. So it's this girl, a pretty girl, probably in a relationship, sat in the passenger seat of a car. And the question comes up, she's got a phone in her hand. The question comes up, says, can I look at your text messages? She goes, can I look at your social media? She goes, can I look at your chatgpt? Throws it out the window. It's so true. You know, I remember Seth Stevens Davidowitz did that great book Everybody Lies where he realized that people would ask Google things that they hadn't admitted to a therapist, that they wouldn't admit to a spouse. They kind of hadn't admitted to themselves. And I get the sense that ChatGPT has kind of lifted the lid on that. You've got this sense that this is unbelievably secure and very intimate and exclusively one on one and so forgetful that frankly it's probably not gonna be able to remember what it was that you fucking said in a couple of days in any case. And yeah, I am concerned. Some of my friends who are more health anxiety focused. The opportunity to have a always on kind of expert to talk to about your problems, mental health, physical health, stuff that you're doing with friends. It is a hypochondriacs fucking dream. You know, it's this opportunity to kind of wallow and really dig in to the questions. I do get the sense that I would imagine that most people's self report of how satisfied did you feel about your time that you spent on these different applications on your phone today? I guess ChatGPT would probably rank pretty high I think, you know, maybe a little bit behind a meditation app or something like that, but probably not far off, certainly higher than a TikTok or you know, a lot of the social medias that are super compelling. But there are some longer term concerns that I have about that. How much is it allowing you to indulge this fatigueless therapist, best friend, sick and quite sycophanti as well? It's rarely giving you tough love. It's always kind of validating you. Yeah, yeah.
- [65:30]
Dwarkesh Patel
I do think it'll be important for the companies to institute this level, you know, like some. I mean, I think there's a Persona that we're familiar with which is an employee or a co worker who has a backbone. And if you're a mature person, you will not only understand that but appreciate that. We'll see if market incentives mean that the average person wants that because then.
- [66:00]
Chris Williamson
Being very pliable is quite reassuring and comforting.
- [66:04]
Dwarkesh Patel
But it was the case that when O3 recently released a model that was considered very sycophantic and the reason that was done is just because they released two versions of a model in like testing, people really like the sycophantic one and that's the one they deployed. It wasn't some intentional, you know, like manipulative design as far as we can tell. It was just like, this is what people seem to like we're deploying it. It was like just a B testing it.
- [66:26]
Chris Williamson
But if you run an fab test, you end up with a porn website, right? Like, that's actually where you end up. You end up kind of zeroing in on the lowest common denominator. You know, if you split tested food, you'd probably end up with cheesecake.
- [66:39]
Dwarkesh Patel
Right?
- [66:39]
Chris Williamson
Like, is that really what we want?
- [66:40]
Dwarkesh Patel
Yeah, yeah.
- [66:41]
Chris Williamson
That's early.
- [66:42]
Dwarkesh Patel
Yeah, yeah.
- [66:42]
Chris Williamson
And again, you just end up with. It's kind of the basic time on site CTR. Mr. Beastification Optimization thing of that which is most compelling is not necessarily the thing that's best for you. Before we continue, if you haven't been feeling as sharp or energized as you'd like, getting your blood work done is the best place to start. Which is why I partnered with Function, because they run lab tests twice a year that monitor over 100 biomarkers. They've got a team of expert physicians that take the data, put it in a simple dashboard, and give you actionable insights and recommendations to improve your health and lifespan. They track everything from your heart health to your hormone levels, your thyroid function and nutrient deficiencies. They even screen for 50 types of cancer at stage one, which is five times more data than you get from an annual physical. Getting your blood work drawn and analyzed like this would usually cost thousands, but with Function, it is only $500. And right now, the first thousand people can get an additional hundred dollars off, meaning it's only 400 bucks. To get the exact same blood panel that I use, just go to the link in the description below or head to functionhealth.com/modern wisdom. That's functionhealth.com modern wisdom.
- [67:49]
Dwarkesh Patel
Yeah. Yeah, 100%. Here's my hopeful vision. I don't know if I'm not predicting this will actually happen. One of the reasons it's. It's sort of hard in today's world to make bespoke content that is fits everybody's own highest aspirations, where, like, the Mr. Beast can make something which is, I respect what he does, but it's something that a lot of people will find engaging, but not necessarily at a deep level. And there just aren't enough sort of like Spielbergs to make a bespoke movie for you. That could change with AI where like, the amount of talented dedication that every single person experience can be much higher. I feel like intuitively that should be more compelling. And, you know, if you're like, brain rotted, from TikTok, they'll make like brain rotted content that's like at least better than what's on TikTok. Or they'll even be more engaging on a minute by minute basis than watching a video game at the bottom end of your screen and then watching like, I don't even know what, like some bullshit on the top. So I could imagine. And we know from our personal lives that like there's meaningful experiences are like compelling to us. It just hard to access them as immediately as TikTok is. To the extent that AI can design an environment for us which like gives us those meaningful experiences as easily as YouTube shorts are served to us. There's a positive story to be told there. I'm not necessarily.
- [69:23]
Chris Williamson
That's a really good take.
- [69:24]
Dwarkesh Patel
Yeah.
- [69:24]
Chris Williamson
Yeah. You're very helpful in that regard, which is refreshing. Are there actually any leaders in the industry right now? Is, is that right to talk about that? I guess you've got distribution or power. But really what most people mean when they talk about this is the thing that I use and this is why I like it is vibe. Like I just like the way it speaks to me. It seems to make the fewest errors. Yeah. What's kind of the state of the industry?
- [69:48]
Dwarkesh Patel
It doesn't seem to me there's a clear leader, which is very interesting. I think a couple years ago you could have predicted that not only would there be more differentiation, not only would more people fall out of the race because it's getting more and more expensive to train these models, but each of them would pursue a unique angle. One of them would be more of a chatbot, one of them would be a coder, one of them would be a remote worker. They might be trained in different architectures which have different strengths and weaknesses. As far as we can tell, that's not the case. And there's more companies that are competitive today than it was the case maybe two years ago. So I don't know what explains this. It could just be that it's hard to keep a secret. Like if you release O1 just by playing with it, you learn about how it was trained, how fast it answers questions, teaches you how big the model is. A bunch of things like smart researchers can figure out. And so then Deepseak and look at that and of course do a bunch of innovations themselves. But also every company that not just deepseek will look at what's at the frontier and be able to sort of backtrack about how it might have been engineered. So there is a way in which things are sort of becoming more and more.
- [70:55]
Chris Williamson
Yeah, that is strange. How did Apple fuck it so badly?
- [70:59]
Dwarkesh Patel
I have no idea. I have no idea. I think it'd just be like a much simpler. Maybe there's like not a complex answer to that. Maybe there's like a very simple answer. Just big company doesn't make a priority, doesn't happen.
- [71:10]
Chris Williamson
Yeah, maybe. How important are individual visionaries when it comes to AI development? If there's huge teams of people working on this aggregated data learning, you know, it feels like there's a lot of ballast in the system there. Is there still a great man of AI theory coming along, I think.
- [71:34]
Dwarkesh Patel
It seems to me that there are great researchers who have very specific talents. They have talents in not necessarily just AI research, but in how to code up the GPUs or accelerators so that you're getting these like 25%, 50% performance gains, which are huge. But it's more of that kind of thing, I think more technical than from my sense. There's not like I'm just good at thinking and I can write a great manifesto and therefore I'm the sort of person moving the organization forward.
- [72:09]
Chris Williamson
What are the current constraints to progress? Is it software? Is it energy? Is it coding? Is it datasets? Is it the savant fucking guy that fixes the hardware?
- [72:19]
Dwarkesh Patel
Yeah, and to be clear, I'm a podcaster, so I'm looking from the outside in my sense is that given that this RL scheme, so the thing that 0103 is where it's trained to solve particular problems, math code, so forth, the thing that's really lacking now is not compute, but the relevant data. So OpenAI said in their blog post about 01 that they. Or sorry, Dario said in a blog post about XOR controls that current. As of a couple months ago, they're spending on the order of a million dollars on rl. And keep in mind that they're spending on the order of like a billion dollars trading the base model, which they do this RL on top of. So the reason they're not spending more on RL is just that they don't have the relevant data. They don't have these like bespoke and, you know, environments where you're like trying to do a job and there's a slack and there's a mail and whatever open and you need to figure out how to still solve the problem and you need to like learn from every all these different kinds of jobs in the economy. You need them to come up with.
- [73:17]
Chris Williamson
These different environments, you can't do reinforcement learning without that.
- [73:21]
Dwarkesh Patel
Exactly right.
- [73:22]
Chris Williamson
Okay, what about China? Does China have a different vision for AI than the west does?
- [73:29]
Dwarkesh Patel
I genuinely think nobody in America knows or like very few people in America know. Nobody I've talked to in America knows. I. Well, we saw Deepseak's models and they're actually unusually open. They open source their key architectural secrets, which in many cases are ahead of some American labs like Deepseek had techniques like mla that multi head latent attention doesn't matter, whatever nerd shit that META didn't have, which is, you know, spending way more money. In fact, it had techniques that Meta had invented like multi token prediction that Meta wasn't able to do the engineering to actually implement in their new models. And Deepseek was able to figure it out. So obviously it's like a big country with lots of talented people. They're open for now. Deepseek at least is open for now. We'll see. Especially given how popular it's become and how Xi Jinping met with its leader and all the other industrial heads where they take it from here.
- [74:33]
Chris Williamson
Just use your kind of knee deep in the world of thinking about China, what it is that it wants to achieve, its history. You've got good context here. What do you think they are thinking when it comes to why do we want to have such a powerful AI?
- [74:53]
Dwarkesh Patel
I think that they have shown a willingness to accelerate on all technology. They showed it in the 90s with the Internet where people said that this will cause the collapse of the Communist Party and they made the bet that no, this will actually give us unique insight into our society because we can monitor everything everybody is doing on the Internet in a way that we cannot do right now with AI. The problem, I think it like the genuinely tilts the balance even more in favor of the state. Right now on WeChat or something you have these manual sensors, thousands, potentially hundreds of thousands of them, who will take down content. With AI, you have a system which could do that for you. If you try to use the AI to do something that the party doesn't want you to do. As these AIs get smarter, they can internalize, they can be aligned to the party's model spec that says like we do not want to talk about X topic, Y topic, Z topic. If somebody tries to do this thing, you want to report them to us. And a smarter model is just better.
- [76:01]
Chris Williamson
Able to follow these digital panopticon.
- [76:03]
Dwarkesh Patel
Yes, that possibility is live of course, as the economic value from These models becomes more evident. I just think it's not clear obviously they would pursue this. They are obsessed with technology and industrial policy. Why they would neglect this is not clear to me. Especially now that we have, you know, it became the national champion because of the events of the last few months.
- [76:27]
Chris Williamson
Right. So might AI perfect authoritarian governance then?
- [76:32]
Dwarkesh Patel
I think it'll certainly make it more, more plausible. Right now you have this dynamic where Xi Jinping has the same 10 to the 15 flops in his brain that every single person in China has. You could have a system in the far future where the central. It's much more possible for the central node to concentrate, compute. And just as Elon can monitor every single person at this factory, like, or, you know, AI Elon can, it might be possible for Panopticon kind of thing to have eyes everywhere. Copies of the thing can have eyes everywhere. Yeah, I think that's very plausible.
- [77:14]
Chris Williamson
I wonder if you could slightly more hopeful vision mimic a kind of benevolent dictatorship, you know, executing one aligned vision at a massive scale, but in a good way, in a way that actually helps people, you know, that encourages people to put down the ice cream or to do the whatever to try and balance what you need from free market and freedom and agency for people with oversight and guidance and looking after from above. I don't know.
- [77:39]
Dwarkesh Patel
Yeah, I worry about any vision like that. I, I mean, history's replete with people who think they know better. Just. I, I mean, I think it'll be a genuine conundrum now that we're talking about it, because, yeah, it will be the case. People are getting like addicted to their AI porn and you know, like the brain rot that will come out of this. And I think it will be the government might say this thing, which will be the main way in which we're interfacing with the world. It's not some peripheral technology. This will be the main way we're interfacing with the world, learning about the world, the main way we have relationships potentially. It needs to have these certain policies. And I guess a balance will have to be struck between the government saying it can't do certain things or can do certain things and people wanting the individual freedom of like, look, this is a, this is the mind I have a relationship with. I want it to have these characteristics. I don't know what the balance there should be. I lean more libertarian. But I think that like, yeah, maybe, maybe that means you ought to make a trade off where like some people will get addicted to sort of. It's similar to the drugs legalization conversation. Except the drug legalization doesn't have an upside in the way that like AI that's sort of niche has an upside.
- [78:50]
Chris Williamson
I wonder whether it's more compelling, I wonder whether drugs are more compelling than a super intelligent AGI that's able to trigger every bit of dopamine and meaning and serotonin and vasopressin in exactly the way that you need at that moment, using your micro expressions and with full context and understanding your genetics and yeah, like yeah, maybe actually you fixed the drug epidemic by just getting Everybody addicted to GPT10 or something instead.
- [79:17]
Dwarkesh Patel
I'm still I want my girlfriend.
- [79:20]
Chris Williamson
You recently spent some time in China.
- [79:22]
Dwarkesh Patel
Yes.
- [79:22]
Chris Williamson
What'd you learn?
- [79:24]
Dwarkesh Patel
A lot of things I learned, honestly and I'm embarrassed to say are things I should have known beforehand. Obviously China is a very big country. It is another thing to see it viscerally just you. There are cities you've never heard of which have 20 million people. We're in. Austin has about a million people. There's 160 cities in China that have a million people. And so these just a ginormous scale of everything from the cities to airports, train stations, factories, driving through towns or entire megapolises which are full of factories. You know, you hear the phrase China is the world's factory. And just seeing like a city the size of Austin being one of like a hundred hubs of manufacturing, like all that's happening here is shit is being made is an interesting experience. Again, look, I'm a tourist. I'm talking about like my experiences for two weeks. I'm not pretending to be an expert. There are interesting things in terms of. I think things are obviously more on knife. People feel more sort of nervous than they did a couple years ago. But it's still. It's not North Korea. Like people will just tell you their opinions over dinner and stuff. I'm curious about China because the sort of competition there is the main element of what will happen in the 21st century other than AI.
- [80:57]
Chris Williamson
And how do you mean?
- [80:59]
Dwarkesh Patel
Well, it's a country the size of America in terms of the economy, much bigger in terms of population for now, yes. And the fact that we just like don't think about it that much or I think people just have this very adversarial attitude towards it because neither side understands each other that well is a shame. Yeah. And I just wanted to have a more sort of visceral understanding of it.
- [81:24]
Chris Williamson
What are the cultural vibes like the. You made a great point about. You think it's kind of a, I don't know, just a more powerful, more sophisticated North Korea or.
- [81:35]
Dwarkesh Patel
No, I don't think it's like North Korea.
- [81:37]
Chris Williamson
No, no, no. That's what a lot of people that haven't been there think. It's like, oh, it's surveillance state using gait analysis to get your social credit and no one will be able to speak the truth.
- [81:48]
Dwarkesh Patel
Yeah, yeah.
- [81:49]
Chris Williamson
What if, what if the family. You know.
- [81:51]
Dwarkesh Patel
You know what's really interesting? While I was there, I ran into some students who, who were like, I would never move to America. And I was like, why? I was like, well, you guys have school shootings. It just seems unsafe. And living in America, we know that it just like, it happens. It's the sort of like thing you hear about in the news, but it's not a common part of American experience. Right. And I think a lot of the. It's just true of probably every country. But I think a lot of the sort of archetypes we have or the stereotypes we have of Chinese life are just like. You hear about this, but this is not like a common. Just like getting arrested in the street or something. It doesn't come. Now that being said, I doing what I do with podcasting, I would just not feel comfortable doing that in China. And I don't want to take like. I think it is sort of evil to have a system of repression, not just in speech, but at every level from your savings are taxed so that they can pay for this industrial policy. You can't get your money out of the country. But yeah, it just the sort of usual things you learn from travel. It's a more, it's more similar than you expect. Yeah.
- [82:58]
Chris Williamson
Is it true that they're using social media to just supercharge everyone into hyper producers or are their kids getting brain rotted by TikTok as well?
- [83:05]
Dwarkesh Patel
Oh, I, I, I. When I was in a mall in Chongqing, a couple kids. So by the way, one interesting thing in China is there were very few foreigners.
- [83:14]
Chris Williamson
Very few foreigners.
- [83:15]
Dwarkesh Patel
Foreigners.
- [83:15]
Chris Williamson
Okay.
- [83:16]
Dwarkesh Patel
You look at it, a sea of people in a major city and you won't see a white person there, especially outside of Shanghai and Beijing. So anyways, because of that, these Chinese kids would come up to us and try to take selfies or something.
- [83:30]
Chris Williamson
Right, because you were an attraction.
- [83:32]
Dwarkesh Patel
Yes.
- [83:33]
Chris Williamson
Okay. Exotic.
- [83:36]
Dwarkesh Patel
One girl approaches, like, are you guys in like a rock band or something?
- [83:40]
Chris Williamson
Sick.
- [83:41]
Dwarkesh Patel
I mean, we're not, but that's like, I guess how sort of how we're the porn anyway, so These kids come up to us and they're like, we're just talking. I'm making small talking, like, oh, you guys in high school, what do you guys do in your free time? And like, oh, we just watched Tick Tock. I'm like, what do you guys watch? Oh, he's like, a couple hours. You know, we just. Sexy girls. I'm like, what sexy girls? I'm like, what do you mean? And so he pulls out his phone. It's like, literally just like, sexy girl, sexy girl, sexy girls.
- [84:11]
Chris Williamson
Asian sexy girls.
- [84:12]
Dwarkesh Patel
Oh, yeah, yeah, right. I guess I didn't check. I'm colorblind, Chris.
- [84:25]
Chris Williamson
Okay, so the Kale TikTok algorithm doesn't seem to actually be.
- [84:34]
Dwarkesh Patel
Oh, sorry. Is the meme supposed to be that like our TikTok is like the fucked up shit and there's this like a bunch of engineering.
- [84:41]
Chris Williamson
Exactly. Unless actually, if you look more closely, this is your issue because you turned away too quickly because of the sexy girls, made you feel uncomfortable. What you would have seen is that they were all sexy girls doing.
- [84:53]
Dwarkesh Patel
Oh, right, right.
- [84:54]
Chris Williamson
Simultaneous equations. Yeah, exactly. On a fucking blackboard. To you, it would have just been a. Whatever board.
- [85:00]
Dwarkesh Patel
I think a lot of young people were quite like, the economy is not doing well. It's if you want to work in a tier one city. So China classifies their cities by tier one, Tier two, Tier three, There's a hukou system, which means that you actually need a visa to basically live in the tier one cities. And if you want to work there, which is supposed to be this sort of dream, you're working 997. So from 9am to 9pm or sorry, 996, six days a week, and they just like a lot of stress. And there's a phenomenon where young people just either want to work for less pay in a tier 3 city where their life will be much less prosperous, but it's just like not as much stress, or they just want to leave the system altogether. There is like a visceral sense of. I think people have this very bimodal view of China where either the system is about to collapse because XI is cracking down and it just doesn't work at all, or they're about to launch the space lasers and it's already over.
- [86:04]
Chris Williamson
The hyper productivity. And then996.
- [86:06]
Dwarkesh Patel
Yeah, exactly. And I think it's somewhere in between where in America we realize some things are going well, some things are not going as well. I do think the CCP has been bad for Chinese growth. You can acknowledge that China is a powerful country that's at the frontier of a lot of technologies without saying that the government is optimal or its policies make sense.
- [86:26]
Chris Williamson
What else do Chinese people think about the West?
- [86:31]
Dwarkesh Patel
I, I am a little worried that it's coming across as like, I, I'm like in a shine expert where like I, I want to clarify, I'm like, I was a tourist there.
- [86:37]
Chris Williamson
You went for two weeks and you've spoken to a couple of people.
- [86:39]
Dwarkesh Patel
Exactly. I, I don't know Chinese. It was interesting to me that many of them wanted Trump to win. I won before the election. They respect, they really respect Elon Musk. And I asked them why and they said because he's successful and we value success in China, which I respect. Like that cultural attitude. It's like against the sort of cultural tendencies that we often have here.
- [87:12]
Chris Williamson
How unmolested are the stories about people like Trump and Elon going over to them? Because I saw in your blog post you mentioned that accessing the Internet is a bit of an adventure or a minefield.
- [87:24]
Dwarkesh Patel
Yeah, it's like more of a pain than I expected.
- [87:28]
Chris Williamson
Not just surfshark, vpn, your way around it. I'm guessing there's only a couple of.
- [87:32]
Dwarkesh Patel
VPNs that work, so you want to make sure that you have one of those installed before you go. Yeah, I'm not sure. Honestly, yeah, I'm not sure.
- [87:43]
Chris Williamson
I'd just be interested. Like if you've got such all encompassing control of the Internet, why not curate the messages just a bit more? Just a bit more. Just a bit more. You've seen with RT in Russia that all manner of different, just subliminal breadcrumbs being left around. I don't know, maybe you simply can't coordinate well enough to do this. Maybe there is some sense that we actually need to allow people to understand what's happening at the rest of the world. Maybe it's some 5D chess move that actually by allowing people to like Elon Musk and Donald Trump when we go to do the thing they're going to be. I don't know, but it just seems I'm interested about why any positive visions of area of the world that they are pretty head to head with would be allowed. Given that you don't necessarily need to allow it. You have the facility, you have the opportunity to be able to stop that from happening.
- [88:50]
Dwarkesh Patel
My understanding is that they realize that in order for them to be economically dynamic, they need engagement in the world. So if your software developers can't read American code, You're just like, it's a big problem. Right.
- [89:04]
Chris Williamson
So, well, I suppose you need, sorry, just on that. You need people to be exposed to bits and pieces of American culture in an accurate way or else how do you know what to design to be able to export to America?
- [89:15]
Dwarkesh Patel
Right.
- [89:15]
Chris Williamson
Like you need to have an understanding of that. And it can't just be, can't just be that you've got a few Austin equivalent cities and all that these people do is get Faraday caged off and watch American tv. Okay, you're the America experts. You will tell us what it is that the, the white people want and then we'll go and design it. That would be too much. You need to distribute it. So maybe that's, maybe that's a good way to put it.
- [89:38]
Dwarkesh Patel
They do have a very impressive system of. In 2018, Tesla opened up its Shanghai Gigafactory. BYD sales I think dropped like 25% that year or something. Or it's on that order. And Tesla, sorry, China, did that deliberately because they had sunk hundreds of billions of dollars over the preceding decades trying to build up their EV industry. And these companies are producing products which were not compelling to either domestic purchasers or to foreign purchasers. They were just like not designed well, just as you said. Right. And just like by catfishing Tesla they were like they forced their companies to catch up. And now BYD sells more than Tesla.
- [90:24]
Chris Williamson
No way.
- [90:24]
Dwarkesh Patel
Yes.
- [90:25]
Chris Williamson
Wow.
- [90:27]
Dwarkesh Patel
It's I think the best selling car like car company in the world. Maybe fact check that. But I think we should do a similar thing. I think we have this idea that like we can just export, we can just prevent importing these amazing electric vehicles from China, the solar whatever they're great at. I don't think that's the way you win. I think the way you win is you do the exact same thing to them. You guys open up a factory in, in Detroit, you teach us how you're doing what you're doing. Because they can do things we can't do. And then we build up these local supply chains, agglomerations of knowledge and we force American companies to be able to compete with the frontier in the world. Because in the long run the solution can't just be to keep them out. Right. In the long run, the solution has to be. You have to be competitive.
- [91:12]
Chris Williamson
Yes. Yeah, it's very much sort of a cordoned off scarcity mindset that is, well, if we can take what the first order effect is positive, that's what's most important. And you go yeah, but what about 2 and 3 and 4 and 5 and 6?
- [91:30]
Dwarkesh Patel
Yeah. And I think we've sort of given up on being able to lead in the physical world in the long run. And I think there's this interesting dynamic, which is you were asking earlier about. We have all these problems. We're hoping that AI will just solve them or they don't come up. This is definitely true in the China US competition thing, where I think people who are paying attention to their top companies, their technology and so forth, notice that in 10, 20 years they're making so much progress that in many of the most important technological domains in the world, they will be leading. But there's this idea that, well, we will get AGI first, and if we do that, then everything is solved. I think in this domain, this sort of thinking actually doesn't make sense because AGI still needs access to the physical world. You'll still need to manufacture robots. In fact, all that data will be cordoned off where that manufacturing is happening. So there might be increasing returns to having.
- [92:27]
Chris Williamson
There's an unlock if you've kind of prepared in advance. Oh, that's interesting. What else do most people not understand about the tension between China and the west in either direction?
- [92:59]
Dwarkesh Patel
This is not a point for me, but from Dan Wong, I think people don't appreciate how the Chinese political system works and how it just selects for a wholly different kind of person than American political system. If you look at what fraction of Congress is lawyers, I think it's like a majority. It's just shockingly large. And there's like no engineers, or there might be like one or two engineers in Congress. Whereas it's the exact opposite. In China, you look at the Politburo, these are people who have like, PhDs in chemical engineering or in petroleum engineering and random like, heavy industry, shit like that. And the way another thing people don't understand is just how intertwined the party is into industry, especially this kind of heavy industry, where for somebody to get promoted, you know, you might start off as like the equivalent of a mayor, then you become the governor of a totally different area. So the central government at the top, the central party will tell you, you know, you're going to go like, you're mayor of Austin, now you're going to be governor of Delaware. Now you're going to be part of. Now you're going to run like a steel company, now you're going to be part of the cabinet, and maybe then the future you're going to run the country. So they also don't appreciate how decentralized the system is. In America, about 50% of government spending happens at the national level. 50% happens at the local level. In China, 85% happens at the local level, 15% of the national level. So there's all these experiments that are happening. And also it's a much bigger country where each, each locality, each province is trying to implement the things that the central government wants. At the same time, the central government has way more power over appointment. You know, every town gets to elect its own mayor and every state gets to elect its own governor. In America that's not the case. Obviously in China, right, They're rotated around. This can lead to more meritocratic outcomes where you are promoted because you did a good job running this town. Obviously that can go wrong and has gone wrong in recent times where you're promoted for loyalty. But yeah, I guess I just didn't appreciate all these different ways in which it is a totally different system.
- [95:04]
Chris Williamson
What does that result in? What's the outcome of? That's how the system is set up. What are the capabilities, strengths, weaknesses that that enables on the back end?
- [95:15]
Dwarkesh Patel
So for many decades these leaders were promoted and compared to each other. So you are promoted if compared to every single governor in the country. Your province has the highest growth rate and this growth rate was just measured during your duration there. And the best way to increase short term growth rates is just to build shit. And this worked for the first two decades of liberalization. Where China was because the Cultural Revolution, because of the Great Leap Forward, because of the decades and decades of war beforehand, the Japanese invasion, It was just so much poorer than a country of that size or that human capital would be. So you could build anything and it'd be worth it, right? There's like nothing that exists. You build a railway, air, train station.
- [95:58]
Chris Williamson
We need it, we need it.
- [95:59]
Dwarkesh Patel
We need it. Exactly 100% do it yesterday. And then the system sort of malfunctioned where now they're, they're incentivized to just build cities that literally nobody lives in.
- [96:10]
Chris Williamson
You say they build bridges to nowhere and knock down 500 year old monasteries to make it happen.
- [96:14]
Dwarkesh Patel
And ironically, ironically here we can't even rebuild fallen bridges, right?
- [96:20]
Chris Williamson
Or is it overproduction and under consumption? And underproduction and over consumption.
- [96:23]
Dwarkesh Patel
Exactly, exactly. Another, another important thing you need to understand. Again, not an expert, any means, you're hearing from a tourist, but another important. In order to understand the economy, you have to understand the system of financial repression that exists in China where if you are saving money. You are getting 1% interest from the bank and no bank is allowed to offer you more interest than that because the governments control the banks. All that money is basically given out as loans to companies that the state state prefers and decides. Like it looks at if for China to be a dominant country in 20 years, we need robotics. So we're going to give a bunch of money to lend a bunch of money to robotics companies and semiconductor companies and whatever or to infrastructure projects. So it is a systematic redistribution from average people, from savers to this kind of industrial policy to these companies, which is often very inefficient because there's no market that's doing these investment decisions. It's just this sort of system of government relations and. Exactly. Central planning.
- [97:34]
Chris Williamson
That's interesting, dude. I, I, it feels to me like the world is at a fever pitch. I'm very detached. I've actually got TFs at the moment. I've got Trump Fatigue syndrome and News Fatigue Syndrome. I've been kind of checked out since November, December time, which is why I haven't talked much about politics and things that have been going on and real interested in stuff that I think is a little bit more evergreen. But just the pace of fucking news and change and your ability to be able to discern between. Okay, do I need to pay attention to this? Is this a really big deal? Is the President getting shot a big deal? Because it happened less than a year ago, as did bombing Iran, as did, you know, pick 20 other crazy things that have kind of never happened before. And it's just here today. Tomorrow's fish and chips Rapper and the advent of AI Rising tensions with China. What's China going to do? A fucking ton of country. Any country that they can get within reaching distance, basically, that's near them. All of this stuff that's going on is, it really doesn't surprise me that people are feeling a little bit overwhelmed. Like the pace now of this is a, it's a difficult one. It's a difficult one to try and work out how to navigate the world as a sane human who needs to keep abreast of the stuff that's important, but also doesn't want to get lost in the swell of just total like even today. Like so much of the stuff that we've talked about is a like five thesises of, of research that could be done on each different one of these things. And they're all going to be world changing if they come to pass. And they could be, you know, tons of different permutations of how the world can end up being in future if it does. Like, that's a lot.
- [99:28]
Dwarkesh Patel
And they all interact with each other.
- [99:30]
Chris Williamson
Correct. And I read this thing from Adam Lane Smith the other day, saying your system is designed for stress, but not for complexity. That your issue is not that you're working hard, it's that your life is not sufficiently simple. And I kind of get the sense that when people talk about life being hard, they don't necessarily actually maybe mean that. What they mean is life is complex. Because I think that most people, even the laziest people, not bad at working hard. What they really struggle with is complexity.
- [100:01]
Dwarkesh Patel
Prioritizing.
- [100:02]
Chris Williamson
Yeah. Executive function. Okay, how am I going to triage this? You know, one of the biggest reasons that procrastination happens is that you don't know what to do next. You know what to do and you know how to do it. Fuck it like that. I mean, then we're talking. That's real procrastination. Right? You know what to do and you know how to do it. If you're still not doing it, we have got. We got a problem. Right? If you don't know what to do or if you know what to do and you don't know how to do it, well, it gets thrown under the fucking nomenclature of procrastination. But I don't think it is. Not in the same way, but yeah, just. I think there's not saying you've scared me, but there's just lots going on. You know, there's so much going on. I think that this. This high level of complexity is something that for a lot of people is. Is overwhelming.
- [100:50]
Dwarkesh Patel
Yeah, I mean, I think this is similar. Do you have this tendency, by the way, to every time you start preparing for a guest, every single thing that you learn about is like, titrate it through what they study. So I'm going to tell you, like I'm reading the Stalin biography from Kratkin to prepare for him. And it's so the period of change between 1880 and 1930, the amount of technological change, geopolitical change, I think even today, we haven't experienced something like that. Again, as much as we think the world has been changing in the past, it just doesn't compare to 1905, the airplane is invented. 1914, it's like 1917. It's indecisive. In World War I, the tank literally wasn't a thing when World War I started. By the end of the war, it's, It's. It's a tank Warfare all around. Radio trains, fucking telegraph, steamships. It's just like the world is changing so rapidly. Yeah. There's all these new ideas that are coming around. Communism, fascism. You have all these old regimes, all these monarchies and aristocracies in Europe, in Russia that are getting revolted because of this big war. And even all that wasn't as big as AGI is gonna be.
- [102:23]
Chris Williamson
Fuck yeah, dude. It's. What a time to be alive.
- [102:28]
Dwarkesh Patel
Yep.
- [102:30]
Chris Williamson
George, one of my friends has a really interesting question. You know Teal's originality question. What do you believe that most people would disagree with or find abhorrent or something? He's got one which is what is currently ignored by the media but will be studied by historians. You got an answer to that? There's something you can think of what is currently ignored by the media but will be studied by historians.
- [102:56]
Dwarkesh Patel
Trying to not make it all about AI?
- [102:59]
Chris Williamson
Well, I think certainly not necessarily being ignored by the media. Although some, I guess some areas are a little.
- [103:05]
Dwarkesh Patel
Certainly like industrial capacity. Just how much stuff can your country produce?
- [103:17]
Chris Williamson
This is the future proofing for AI.
- [103:19]
Dwarkesh Patel
Yeah, partly it's part relevant to geopolitical competition. When the Ukraine war happened, you know, we've been. We should have. Obviously it was right to give them the. The munitions to fight Russia. But the fact that we can't restockpile all the weapons that we've given them is like sort of worrying if we end up in another conflict with another country. What's your answer?
- [103:49]
Chris Williamson
Population decline is my usual go to. I think it's. It's a big deal. The impact of smartphones more generally, the impact of technology on mental health, on outsourcing of thinking. You know, that, that article from the New Yorker, which, you know, if it happens with AI, because AI is just such an effective assistance, I have to assume that the. Basically the same thing happens, but a lower level when you're using screens for anything else too, that the more effortful that you make the process of learning. I mean, I guess it could be too hard. It's like climb Everest and then read that word, then come back down, then climb Everest and read the second book.
- [104:26]
Dwarkesh Patel
We'll remember that word.
- [104:27]
Chris Williamson
Yeah, exactly. You know what I mean? Those would be. Those would be two. I think that the retrospective of what did we do to people with the free access to this kind of technology? I think would be an interesting one. Like, will it be looked back on as the like, prototype version? This, this really early rough hewn, like. Like when you hear about doctors, smoke, camels, you know, in surgery, and you go, how. How were they allowed to do this? It's so, you know, dirty and unclean and the outcomes were so negative. I wonder whether the same is going to be seen of use of technology between 2010 and 2020.
- [105:21]
Dwarkesh Patel
Has this changed your own consumption of content?
- [105:23]
Chris Williamson
My limbic system's pretty hijackable, man, so. But I try. I try to be as mindful as possible. You know, it's largely putting guardrails in place wherever you can.
- [105:33]
Dwarkesh Patel
How do you feel about the fact that your own content is served through YouTube? And I don't know how big a deal shorts are for you or short form stuff. I mean, it's a huge deal for me.
- [105:40]
Chris Williamson
They crush in terms of numbers. They're completely fucking useless. In terms of everything else.
- [105:45]
Dwarkesh Patel
I disagree.
- [105:47]
Chris Williamson
For me, it's been different on YouTube shorts.
- [105:50]
Dwarkesh Patel
Yeah.
- [105:50]
Chris Williamson
In terms of what. What's the outcome that you're getting?
- [105:54]
Dwarkesh Patel
I had a video with Sierra Payne, who is now my most popular guest, which was stuck at like 40k for the first 6 months. This is before my podcast had this recent growth spurt. And then we started making shorts for it and it's at like 4, 3 million something.
- [106:08]
Chris Williamson
Wow. Okay, that's interesting.
- [106:11]
Dwarkesh Patel
Have like 20 million views.
- [106:13]
Chris Williamson
Wow.
- [106:13]
Dwarkesh Patel
10 million views.
- [106:14]
Chris Williamson
Okay.
- [106:15]
Dwarkesh Patel
Yeah.
- [106:15]
Chris Williamson
Well, maybe you're just doing shorts better than me. It's important. My preference is always plays on audio platforms. YouTube People, I love you, but my. The show has always been Spotify first, Apple podcasts first. Yeah. And you know, it might sound stupid for my own naming of the most beautiful podcasts in the world, at least when we get it right to be an audio first podcast. But just for me, that's where the most loyal audience tends to be. It's the most predictable in terms of numbers. It's the one that seems to be the most considered. And a lot of this is just. You could change this overnight by removing reply threads on YouTube. You could remove this. They did remove this overnight by getting rid of the downvote button on YouTube as well. So I've always liked the audio side.
- [107:06]
Dwarkesh Patel
Of the platform, but the problem is the discoverability.
- [107:08]
Chris Williamson
It's not there. No, you can funnel. You can funnel from YouTube. We found some good ways of funneling from YouTube across onto audio.
- [107:15]
Dwarkesh Patel
How do you do that?
- [107:16]
Chris Williamson
So we release episodes 10 hours early on audio on audio. Yeah. So if you want to get access 10 hours early and we. The pinned comment for every episode is Access all episodes 10 hours before YouTube by subscribing on Apple podcasts or Spotify and a lot of people will come and comment on the YouTube and say, I listened to this this morning on Apple, right? But I came here to watch it on YouTube. So you end up getting two plays, but I don't think you would get the same in reverse. So we kind of got like a weird Patreon type scenario, like paywall thing. 10 hours early on audio, we went video enabled on Spotify, which I know that was a transition that you made as well. That was good. The Spotify partner program's really good. Some changes here and there. Your Twitter strategy is very strong. That's good. We've gone very hard on Instagram, which has been.
- [108:08]
Dwarkesh Patel
Well, you got the looks, Chris.
- [108:11]
Chris Williamson
Most of the fucking shorts are just of Matthew McConaughey chirping about something. I rarely. It was a really, really funny video. I've only guessed it on two shows in the last 12 months. One was Rogan and the other one was my friend Mike's show. And it's me chirping away. It's quite a good. I think it's quite a good take about whether you. How you know whether or not you should end a relationship. And like just classic me stuff like that. It's 55 seconds long and it's just me chirping, chirping, chirping, chirping, chirping, chirping, chirping. And then the final scene of it cuts to Mike and he goes, yeah, it's so funny to just have a video where the entirety of your contribution is. Yeah, I realized how many times that must be the case. But yeah, dude, I've been. It's been so fucking awesome to watch your ascendancy, you know, because we met. We met at the Slate Star Codec meetup in March and I know it was in March because it was just after I moved out here. March of 2022.
- [109:15]
Dwarkesh Patel
Oh, damn.
- [109:16]
Chris Williamson
That was when we.
- [109:17]
Dwarkesh Patel
I felt. Somehow it felt even earlier than that.
- [109:20]
Chris Williamson
Well, unless we met when I came out here the first time in November. So I only came out in November of 21, but I actually searched your name so to see if I already had a prep talk from previous stuff. And if I go into it, I'll see 27th of February 2022. And I've just got At Dwarkesh sp written here and there's like a bunch of other stuff. Valve index with Vive 3.9 track is VR chat with mods near Cyan and Twitter. If you want help.
- [110:04]
Dwarkesh Patel
Do you. It was also such a. I mean.
- [110:06]
Chris Williamson
It was during COVID the Ukrainians are using Grindr to find Russian troops. I've got such.
- [110:15]
Dwarkesh Patel
We could train super intelligence on your Apple notes, Chris.
- [110:18]
Chris Williamson
You do not want that. Holy shit. But, yeah, that was the ACX meetup notes.
- [110:22]
Dwarkesh Patel
Yeah. It's was a crazy time because it was during COVID and I felt like I met so many great people in Austin just hanging out around that time.
- [110:35]
Chris Williamson
It was. I don't know, man. There's something about. I. I wonder whether everybody has this. And I get the sense that they don't. And I can't work out why I do. I'm so fucking fortunate with the people that I bump into early on.
- [110:51]
Dwarkesh Patel
Yes.
- [110:52]
Chris Williamson
Holy shit. Yes. It's like people that I've known for, like, a long time.
- [111:00]
Dwarkesh Patel
Yep.
- [111:01]
Chris Williamson
End up becoming influential or successful or. Or, like, really virtuous in some way. Like the variety of different trajectories of shit that goes well. And it's definitely not me. Right. I mean, I am a common denominator between these people in that I know them, but I definitely haven't fucking influenced them. And I'm relatively introverted. I spend way too much time on my own. So I'm like, how. What the fuck is it? What's the single thread that's drawing me through all of this? Maybe. I don't know. The advantage of being someone that doesn't go out that much is that it takes usually a pretty good thing to get you out of the house so you're a bit more discerning and that the better things, you meet better people. I don't know. But holy shit. Like, I think about some of the places that different people. Like, you're a perfect example. One random meetup. Then we were in this degenerate fucking signal group chat for the last, like, three years. And. And yeah, this arc that kind of. You even nearly quit the podcast. You weren't even doing it. I don't think you'd started it during COVID and stopped and it was just, like, languishing there. It was called the Lunar Society at.
- [112:08]
Dwarkesh Patel
The time, and people thought it was. I was talking about something like some crypto coin that's going to go to the moon. So then I changed the name.
- [112:15]
Chris Williamson
Tell the story about where that comes from, because there's a book.
- [112:20]
Dwarkesh Patel
Can I comment real quick on the meetup? I think at the time you had 350,000 YouTube subscribers, correct?
- [112:26]
Chris Williamson
Yeah.
- [112:27]
Dwarkesh Patel
I was like, whoa, this guy is fucking killing it. Which you were.
- [112:30]
Chris Williamson
And then.
- [112:31]
Dwarkesh Patel
But you're just like, blowing up like a 20x from there. And I think this is like a really interesting phenomenon where I have also had this experience of having met a lot of great people who spend. Who are like super busy and spend time, like, teaching me stuff. The main way the podcast has gotten better is just that people have, like, I've had mentors who have just spent a bunch of time. And these people are, if they're economists, they're in their 60s or 50s, people like Brian Kaplan, Tyler Cowan. But if they're like AI researchers, they're my age. Right. But they're still my mentors and they've spent so much time teaching me stuff, I had no right to their time or attention. And I wonder how you think about this now, because I'm sure you get inundated. And I've, I've hit you up this way of like, what is your advice? Can you teach me about X or Y thing?
- [113:20]
Chris Williamson
And I'm sure someone asking you to teach them. Yeah.
- [113:24]
Dwarkesh Patel
Or not even, like, connect me to somebody. I'm on my podcast. Yeah. And dozens of people have done this for me. So how do you balance the sort of like, trade off between being responsible?
- [113:35]
Chris Williamson
You feel like, you feel like you've got some karmic debt that you need to repay because of the number of I. This is a really interesting question and a challenge that feels like a little bit of a champagne problem. Right. Oh, so many people need your time. So many people did you favors and now you have the problem of people asking you to repay it and so forth. But you're right, because you need to triage your time and you can't do everything for everyone. And the weirdest thing about growth towards success in any domain, whatever version of success that you or me have managed to achieve, is that you need to become increasingly good at saying no. And the pace at which your discernment of no, like the waterline, the barometer, at which no is deserved. And you shouldn't feel guilty about it, it should be instant. It shouldn't take any sort of mind share is a continuous moving target. And you need to hypertrophy this muscle over and over and over and over again. And there's shit that you would have begged to have said, had the opportunity to say yes to only 12 months ago, that now needs to be an automatic no.
- [114:48]
Dwarkesh Patel
But how do you deal with the situations where. When you were starting out. When I was starting out, I had.
- [114:53]
Chris Williamson
Like, people gave your leg up.
- [114:55]
Dwarkesh Patel
I like zero. It wasn't like, am I going to say yes to the workshop? Who, What? The Lunar Society and They said yes, they had better things to do. They're just like, somebody reached out. They seem like they've done a good job coming up with smart questions. Let's do it well. And that just like, if your waterline is always moving up, where's the room.
- [115:14]
Chris Williamson
For the fledgling other person to come through? This is a question I ask myself a lot. It's a real smart question. I'm glad that we're asking ourselves at the same time. The one caveat or the difference in kind. Not just a difference of degree. I'm going to guess that most of the people that you spoke to that you asked for their time, except maybe Tyler Cowan and Brian Captain, too, I guess a little bit. But they're not. Their role is not primarily hardcore curators of other information. They're not distillers across the board.
- [115:46]
Dwarkesh Patel
Right.
- [115:47]
Chris Williamson
That's your job. Your job is to be the hub with all of these different spokes going off it, if that makes sense. And that is a coordination problem. Like, your primary issue is coordination. You have in some ways a hard life learning things that are complex, so on and so forth.
- [116:04]
Dwarkesh Patel
Well, it's so tough, the life of podcast speaking.
- [116:06]
Chris Williamson
Yeah, yeah, yeah. You have a. There are challenges that you need to lift, even if they're only cognitively. Right. But the main thing is complexity. Like, the main thing is the complexity. And I, I don't think that your issue is with doing things, it's with adding complexity in. To go back to that Adam Lane Smith quote, I, I don't have a good answer for it, dude. Like, I, I definitely feel like my karmic repayment debt, I feel like I'm wildly, wildly overdrawn and that I need to, I need to repay this fucking thing. But also, I, I, where do I find, I don't, where do I find the time from. Where the fuck did the people who gave me the Lego. That being said, you're probably not giving yourself enough credit. Because when I think about, when I think about some of the situations that have happened, even just over the last week, there's a kid called Elliot Buick. So he's just turned 20, he's British, he used to work for trigonometry as a video editor. And he. If I could bet a little bit of cash, you're already like fucking Bitcoin at 10k, so it kind of doesn't work so much anymore. But I would have certainly put cash on you three years ago. I would. He's Bitcoin at $1. Like, I would absolutely throw some Money at him. There's a bunch of Jack Neal, if you know who he is. Another young kid. Like, there's real, real smart young guys. And I think when you're talent spotting, you're like, okay, this, this. There's something there. Like, it really feels like there's something that. We went for this three hour dinner, flower child, and we chatted. I'm like, anything that you need, you can do that, this, you can do that, that. So maybe I'm not spreading it super wide. Nomatic needed. They wanted an intro to this guy who did an amazing episode with a musician. This episode with the musician did 3.2 mil. The guy didn't have any sponsors. My guy that does my ads didn't have. He needed his inventory filling and Nomatic needed to make more sales. Nomadic sold loads of bags. This guy got paid and my guy that was in the middle made money from all of it. I'm like, that was just. That just like happened passively as like a byproduct of the ecosystem thing. So, yeah, maybe you're not able to. If you're balls deep in a Stalin biography, you can't peel off to go and do a bunch of podcast appearances or fly across the country to see someone, or let somebody sleep on your couch or do whatever it is that you think you should be doing virtuously in that way. But I bet that you are adding a shit ton of value, even if it's just highly leveraged here and there, little meetups, invites that you give to people, suggestions, intros all around. Hey man, can you enjoy me? Bryan Caplin, can you do. You did Dominic Cummings for me, right? Hey man, can you do that? Like that's a, you know, a small, what, five second task? 10 second task. But downstream from that ended up with an episode that was really interesting. Now I can intro Dom to somebody else and so on and so forth. So yeah, I think this may just be fucking cope.
- [118:57]
Dwarkesh Patel
But 100%. I mean, you've on like trips to the airport, you've like spent the time just like chat with me about.
- [119:03]
Chris Williamson
I forgot about that. Yeah, of course. Hey, do you need a fucking recruiter? This is what I do to build your business out.
- [119:09]
Dwarkesh Patel
There's also an interesting element. I don't know, this is. I mean, it's sort of for an audience of your size, it's easy to lose track of how many people you're helping vicariously. Where even there's a weird dynamic where like you could help somebody in person or you could help share an idea with a couple million people. And like the trade off, it just has to be. It's weird to put it in that way because like one is sort of more commodified than the other, but you have. And you can make better content that you spend that hour prepping harder, thinking harder. You could make better content for a couple million people.
- [119:53]
Chris Williamson
It's a weird trade off. I remember friend Alex gave this thought experiment of. Imagine that one of your friends had broken down, down the street and asked you to come and help him change a tire. And Alex made this point that I would send him the RAC emergency roadside assistance thing because I've got that and I can send it and they'll do a better job and I get to stay at my laptop and do more work. And he got criticized online because it's like, that's not what the person wanted. What the person wanted was this sense of your time. But I get the sense that at least in the kind of interactions that you're talking about, what people are looking for is outcomes. They're not necessarily looking for inputs. And if someone wants to come and kick the tires of a very busy person with kind of no real defined outcome, that's not something that I would have ever done. And I don't think, if you're a young ambitious person that's listening to this, I do not think that you should go to anybody and be like, hey man, like, would just love to connect. If you don't have anything to offer. What the fuck is the point of the connecting? If it's. I have a few very specific questions that I know you probably have the answer to. And I would really appreciate two minutes for you to just give me these because they're big unlocks for me. Super specific question. Really specific. Ask Fanta. This person's put the work in. They're evidently educated. And you'll probably get the fucking 30 minutes on the call because they're walking the dog and they don't really mind or whatever. Like when I think about the random people that I ended up on calls with because I asked for specific, very, very specific things on the come up. I know that you're a big fan of, like, there's a huge unactualized opportunity that most people don't realize in a cold. A very well written cold dm. Yeah, dude, like just fucking send it. You've got nothing to worry about.
- [121:43]
Dwarkesh Patel
Yes. Yeah. And I. You would also be surprised by how few people put in these famous people. They're getting, I don't know A thousand whatever emails every month or something. But how many of those are. I've spent a week coming up. I mean before I had any sort of name or something, I would still be able to get big guests, but I would literally spend a week. Here are the questions I'd ask you just get past not a moron filter because they're getting a request every 30 seconds podcast or something. Just going deep versus wide. Now there's a bunch of like tacit things about. Well, that doesn't mean you should like have 5,000 words in the email. Right. Just like how to keep it brief. It's also really interesting by the way, as a side point of what. What ends up salient to you when you get a cold email or you're hiring somebody and what isn't like the kinds of things you thought mattered while you were in college. Whether you have. And I started an organization that does X or I have a master's in Y just like never matters as a. As opposed to the couple hours of extra work you would put into that email.
- [122:50]
Chris Williamson
Yep.
- [122:52]
Dwarkesh Patel
Yeah. How little credentials matter when hiring or something people don't appreciate.
- [123:00]
Chris Williamson
Yeah, yeah, it's. There is an awful lot of opportunity available for someone who's just, just courageous enough or ignorant enough to be able to get past that sort of first level of ick filter and also is prepared to do a little bit of preparation.
- [123:17]
Dwarkesh Patel
And another thing, I don't think people appreciate how much if you write a good blog post about a topic that you think is relevant to somebody you're trying to reach, it's almost guaranteed that not only will they read it, but weirdly, almost everybody who matters will read it.
- [123:38]
Chris Williamson
Wasn't that how Tim Urban connected with Elon originally?
- [123:41]
Dwarkesh Patel
Really?
- [123:41]
Chris Williamson
I think so.
- [123:42]
Dwarkesh Patel
That would so make sense.
- [123:43]
Chris Williamson
I think he did a six part series. This is a good while ago now. I think we're talking sort of 10 years ago now. I think he did a six part series on Elon. And you know, you're right. If someone writes a good, even not even viral, like semi widely circulated piece on you or your organization or a movement that you care about.
- [124:05]
Dwarkesh Patel
Yeah.
- [124:06]
Chris Williamson
You will read that thing.
- [124:07]
Dwarkesh Patel
Yeah.
- [124:08]
Chris Williamson
And so I always think about this. I always think about the fact that even the richest, busiest, most successful, highest status, hardest to get a hold of people in the world, they get plane delays even if they're getting on a private jet, the weather's meant that they're held up. And what are they going to do? Well, they'll open YouTube, you know, they'll open YouTube or they'll open Twitter or they'll open substack or they'll check whatever it is that's been sent to them in a WhatsApp thread or something. And if you're that meme or you're that article or you're that quote or you're that whatever, it's just continuing to roll the dice.
- [124:40]
Dwarkesh Patel
You can take advantage of a very unfair dynamic, which is that a lot of people have to work anonymously. Their work is shoveled out through an organization or through their boss or something. And they will work decade in, decade out, be extremely good at their jobs, and people will not have heard of them. We're podcasting here and I don't know, I like to think we put our work in or whatever, but like we're not working harder than somebody who's just at McKinsey. Or maybe like that gives a valence of something that's less valuable. There's a lot of valuable work that you just don't. You're making policy. You're. You're a staffer for a policymaker or you're an engineer at a company.
- [125:19]
Chris Williamson
It's kind of quiet. Grunt rock. Yeah, in a way.
- [125:22]
Dwarkesh Patel
And. But we like, we can reach out to people and they will respond to us just because of. It just so happens to be the case that our work is public facing and that just luck or slash our.
- [125:33]
Chris Williamson
Choice full of design.
- [125:34]
Dwarkesh Patel
That's right. You can take advantage of the dynamic by putting at least some of your work as much as possible out publicly. Right. The blog post, the podcast. Yeah. And there's also another dynamic where people take for granted the people in their organization. I'm guessing that like the eighth most senior person at. At Microsoft gets less respect and attention from Satya Nadella than like a random blogger he likes. Which, which is weird, but you can take advantage of that.
- [126:11]
Chris Williamson
Yeah. There's a seduction to visibility.
- [126:14]
Dwarkesh Patel
Yeah.
- [126:14]
Chris Williamson
I think. And even if you're right, even if someone's company is way smaller than yours, or their podcast is way different to yours, or their substack is much less circulated than yours, if they're the person. If they're the main person and you're not, or even if they're the main person and you are, but what they do is cool.
- [126:36]
Dwarkesh Patel
Yeah.
- [126:38]
Chris Williamson
It's such an unlock. It's such an unlock to do stuff like that. I'm interested in what your learning process looks like. How do you learn? What is that process at the moment?
- [126:50]
Dwarkesh Patel
In a way, it's very simple. I read Everything that they. I mean, first of all, it's about picking the guest. And I pick guests based on who I want to spend two weeks reading everything they've ever written, talking to people in their field, learning from them what's interesting to ask them. I'm sure you get inundated by request to come on your podcast, and often it's by people who are, like, very big names. Right? And I'm guessing you. You probably say no to most of them. And. Yeah, same here where you. What are we trying to do here? Right? I. This interview will last two hours. Any interview I do will last two hours. My life is the research that precedes that, the two weeks that precede that. I want that time to be valuable and meaningful to me and be time that I'll carry forward in my future interviews, in my future endeavors in a way that'll be valuable. And if it's. If it's not somebody who has written something that's worth reading or done research that I really want to understand, you know, what are we doing here?
- [127:52]
Chris Williamson
And then choosing your own type of torture for the next two weeks.
- [127:55]
Dwarkesh Patel
Yeah, choosing what you want to learn, which is complicated, but it also. It's sort of easy to forget how, like much of a dream job, this is where people are curious and they want to learn things, but they feel like they had to trade off their time and their job to do it, or they can. They can only learn about certain things for their work. We get to choose what we want to learn about, and our job is just to learn about it. Right. It can be about any topic at all. It can be about genetics, it can be about history. It can be about technical stuff. And then, yeah, then there's the prep and just read the research. Talk to LLMs. Like, just, you know, get after it.
- [128:28]
Chris Williamson
The Socratic method.
- [128:29]
Dwarkesh Patel
Exactly. And then ask the questions you actually want the answer to. I think sometimes people have the sense that you need to ask about the intro chapter of their book. You need to, why did you write it? Who you know, what is it about? And no, you can just, like, you can just ask the thing you want to ask. I think people underrate how much immersion learning is. People can keep up with a. Of lot people just, like, really want to boil down the conversation so that everybody can keep up. I think they underrate the extent to which people can just miss a word or two here and there, but just getting to the crux of it will make it a more delightful experience. And also, if it's a question you're not interested in. Why would the audience care about it? So, yeah, it's fundamentally just be motivated by what you're curious about, who you want to interview, what you want to ask them, what you want to interrupt.
- [129:29]
Chris Williamson
Following your taste. We spoke about this before we started, but the ability to discern between something that's good and something that's not is this lovely balance between gut instinct and sort of rational assessment. Douglas Murray once told me this story. I'm aware that Douglas Murray's like, the least fucking popular person on the entire Internet at the moment after his interview with Dave Smith, but he's still got fucking absolute bangers. And this was one of them. So he worked for this journalist. Douglas got like, I think, four or five columns a week. He does now. And when he first started out, he's working for this legendary British journalist. And this guy was getting toward the twilight of his career like a classic journalist. He'd accumulated a bunch of enemies and a bunch of supporters as well. And he decided he had always wanted to get into theater. So he created a show about the life of Prince Charles. And the entire show was in rhyming couplets, the whole thing, and it was orthogonal, to say the least. And at the halftime interval of the opening night, there was no one left in the entire theater, including the cast. Everybody had gone on opening night, and this guy was devastated. And obviously, all of the enemies that he'd accumulated throughout his life, they came out of the woodworks and they dug the knife in. There were all of these criticisms in the media and stuff like that. Douglas told me that he'd seen him at work the following week. He said, what were you thinking? West End show by the Life of Prince Charles in rhyming couplets. You've got all of these people that are rubbing their hands together, waiting for you to fail. And he said, douglas, I followed my instincts and instincts, they may sometimes lead you wrong, but they're the only thing that's ever led you, right?
- [131:16]
Dwarkesh Patel
Yes.
- [131:16]
Chris Williamson
I was like, that's so sick, dude. Yeah, that's so sick. And what I found, whether it's with where I want to live, the things I want to focus on learning, the direction of the show, the questions that I want to ask the guests, the sort of guests that I want to bring on, the people that I want to hire, the further that I've gone away from my instincts, the more that I've tried to reverse engineer. Okay, well, what do the audience want to hear? What are they interested in? What would make the guests feel Comfortable. What does the guest want to talk about? I'm like, in the nicest way possible, the guest like, what do I want to talk about? Because that's what matters, right? That's what matters. And if you use your own instinct as this sort of weather vein, this GPS locator, if you're really fired up to speak about it, you have to assume that some non insignificant cohort of other people are too. And if you've been doing it for long enough, the people that are following you are following you for that same taste. They are in the wake, they're holding onto the coattails of your instinct. And if your instincts change even dramatically, that we're gonna make some pivots. Probably before the end of the year, we're nearly halfway, almost exactly halfway through the year now. By the end of the year we're gonna make some pivots with the way that we do the show. There's gonna be different SKUs, different types of episodes that are gonna be coming out. And it's probably the biggest change I've done since we started the cinema series about three years ago. And it is not in any way data driven. I have no justification for this other than I think it would be fun. And my instinct is going, yes, I think you should try that.
- [132:52]
Dwarkesh Patel
I think that's so valuable for a couple of reasons. One, I have noticed that my best performing episodes are just. I would have never anticipated that they would be popular. It's Sarah Payne, who's this historian that had written a couple of books that I thought were great. It's David Reich who has studied ancient genetics and now he's way more popular than Satya Nadella and Mark Zuckerberg and Tony Blair and whoever else you can name. But in all these interviews, there was, there was something I noticed afterwards, which was that every time I went to lunch, dinner, when I talked to somebody and they're, they asked me, what are you thinking about? I just could not. You know, I just interviewed David Reich and he was explaining to me that 60,000 years ago there was a small group in East Whatever and that obsession was so strongly correlated with how well the episode did, regardless of what topic it was about. On the instincts. I've had bad judgment about like a lot of. Look, I learned how to do the podcast well, but a lot of things are required, as I'm sure you've come across in terms of running a business, hiring, management, firing, yes, 100%, just making things happen. I've had bad measurement about many of these things. I Feel like I've done worse. Even in those cases, when I've taken advice, the advice was actually better than what I would have done by default. But when you follow somebody else's advice, if things go right or if things.
- [134:19]
Chris Williamson
Go wrong, you haven't learned anything.
- [134:20]
Dwarkesh Patel
Exactly. And if you just, like, do the thing that makes sense to you, you have some reason for thinking it makes sense and things go wrong. You at least, like, tried an idea.
- [134:30]
Chris Williamson
And you learned and you correct your own intuition.
- [134:31]
Dwarkesh Patel
Yeah.
- [134:32]
Chris Williamson
Whereas that error is still waiting for you to step on in future if you outsource it to. So, yeah, you want to front load failure as quickly as possible in the a small and acceptable way. But no, I. Dude, I. My best heuristic for whether or not I've picked the right guest for that day is how I feel on the morning that I wake up. It's like when I wake up on a morning. Like this morning, I went and did a hyperbaric oxygen chamber session. I'm listening to you and Alec, Alex Cantrowitz talk about stuff. I'm like, this is like, I haven't seen Dwarkesh in ages. It's gonna be so sick. I'm gonna tell him about that. I, like, you know, went through my notes and found I had this note from the February of 2022. It's gonna be so cool. I'm gonna get to bring that up. Isn't it fun? It's gonna be. And then, you know, there's other days where you just don't have the same. Quite the same level of that. And that's not necessarily an error. That's not that you've picked someone that's wrong. It's just. Huh. Okay, well, what are the ones where I wake up and I'm. I want it to be 2pm and what are the ones where 2pm will come along and it'll be okay?
- [135:38]
Dwarkesh Patel
Yeah.
- [135:38]
Chris Williamson
And, yeah, the one where you're like, I want to speed run the next two weeks so that this person comes on the show. You know, we've got. It looks like MGK rapper turned rock star guy. There's a potential that he's coming on. There's another guy called Ronnie Radke who's coming on. This guy called Rick Beeto who does music and stuff like that. I'm, like, making a little bit of a pivot into talking about the world of music and about sort of what's happening and how that interacts with culture and the perils of touring and what this means for a family life. And how you deal with the anxiety and the performance and pressure and scrutiny of the press and criticism and creativity and all this stuff. I'm like, like, huh? Like, I already want it to be one of those days when I get to speak. I don't know, there's other people in between which will be great. I'm like super fired up to speak to them about that. Here's another thing. I don't know whether you've ever had this. There's times where I quite like to do episodes with people where I know a lot about the topic, but not quite so much about them. And that's the same sort of thing where I'm real excited to talk about the topic. And I imagine this is what it must feel like to be at one of Ayla's sex parties where I'm like, I know I'm gonna have sex tonight, but I don't quite know who with. That makes sense. Where I'm like, I know the direction I'm going in in terms of the topic and I'm super excited. And you'll listen to the person talk and you'll do the prep and do the whatever. But there's a little bit of like, I know this world really well.
- [137:05]
Dwarkesh Patel
Right.
- [137:06]
Chris Williamson
What I'm excited to hear is their spin on this. I'm excited to hear the angle that they come at this from. I had this guy called Paul Turk who does evolutionary pediatrics. So he talks about child rearing clinically, medically, developmentally, but from an evolutionary lens. So what happened ancestrally? How did we raise kids? How were they looked after? What did hygiene look like? What did skin to skin contact, diet, all this stuff. And I know evolution, evolutionary theory, not bad. It's one of the few areas I have a bit of expertise in. But I'd never looked at this. I'm like, oh, this is cool. This is gonna be so sick. I'm gonna speak to this guy. He's like mid-70s, you know, he's got his son in law or daughter in law or something helping him to set up the camera. I'm like, this is gonna be fucking sick. And it's sure enough, awesome episode with this guy who had no right to come on and crush an episode, apart from the fact that he has an amazing bit of research. And I was super fired up to speak.
- [138:12]
Dwarkesh Patel
Yeah. And how often do you encounter a guest, which is my favorite, where you thought you were gonna get somebody who can speak to this one narrow topic, but you come across a polymath who has a deep world model that somehow they have something to say about anything you could ask them. And the only limitation is your prompting ability.
- [138:39]
Chris Williamson
Yeah.
- [138:40]
Dwarkesh Patel
And it's so gratifying when you encounter people like that. One of the people I had on like this. Do you know Gordon Branwyn?
- [138:46]
Chris Williamson
No.
- [138:47]
Dwarkesh Patel
Oh, he's. He's incredible. He's another Scott Alexander type.
- [138:50]
Chris Williamson
Okay.
- [138:52]
Dwarkesh Patel
Blogger who has written just like. Yeah. There's no subject on which he hasn't. He couldn't give you somehow.
- [139:01]
Chris Williamson
Yeah.
- [139:01]
Dwarkesh Patel
In like a deeply empirical, super interesting way. And what I learned from the interview, I didn't know anything about his personal life other than the fact that he's anonymous. Is living on like $12,000 a year of Patreon in the middle of some. Some house that his grandfather built in Virginia. And so during that interview, he was visiting San Francisco for a conference. And during that interview, I asked him, well, it seems like you're enjoying it here. Do you want to move here? And he said, yeah, that'd be. That'd be fun. And I said, why aren't you moving here? He's like, I don't have. I don't have the finances to do it.
- [139:43]
Chris Williamson
It.
- [139:43]
Dwarkesh Patel
I said, how much would it cost for you to move here? Said 75k. And then people, like, donate it to him and he's moving to SF.
- [139:51]
Chris Williamson
No way.
- [139:52]
Dwarkesh Patel
And he's like, even much more than 75k or. Sarah Payne, who. I think I can share this. She was somebody who had been slogging through the archives. She's been to a historian who's been to every single continent to go through the archives, has deep understanding of basically every single conflict over the last many centuries, can give you, like, why did the Vietnam War happen the way it did? Why did Russia fall? World War II, you name it. Incredibly compelling presenter as soon as. I mean, her episodes are by my. I sometimes joke that I host the Sarah Payne podcast where I sometimes talk about AI. And in terms of viewer weighted minutes, that's definitely true. But if you notice how her books are categorized on Amazon, they're SCM Pain, not Sarah Payne. And the reason is that she's a military historian who I think started her career 70s, 80s, and she wanted to anonymize her sex. Exactly. And so it so happened that somebody who was incredibly talented there wasn't given this medium earlier on.
- [141:01]
Chris Williamson
Personal accountability wasn't quite the same.
- [141:03]
Dwarkesh Patel
And now she's blown up. She's actually retired from Naval War College, where she used to work so that she can be a public intellectual informing on the Big questions that we've been discussing on full time. And.
- [141:15]
Chris Williamson
And this was launched by the episode that you did with her.
- [141:18]
Dwarkesh Patel
Yeah, and we did three more lectures. We're doing more now. I think what people don't understand, and it's hard to appreciate is that lecture she did. Cumulatively, she might. She probably has over like 10 million views on full lectures. That is just so much bigger compared to add up all the students she's interacted with within the oral college combined. It's just like hard to think about like a million people who you probably reach on many more than that. On an average episode. There's like no stadium in the world that can accommodate an audience of that size. And right now we're talking, we're having fun. We're not thinking about how stupendous a quantity that is. And in fact, I think a lot of politics is explained by who understands this and who doesn't. Like Kamala would do all these rallies and people were like, oh, people are really excited about Kamala and it would be a stadium of 20,000 people. And like, wow, she filled the stadium of 20,000 people. And you know that if you could put out a YouTube video and it doesn't get 20,000 views within the first hour, you're disappointed. Or when Trump went on Rogan, I couldn't tell you to the nearest 10 million how many views that has.
- [142:21]
Chris Williamson
Well, looked at recent New York mayoral candidate, right? Absolute digital first.
- [142:26]
Dwarkesh Patel
Yes.
- [142:27]
Chris Williamson
George has this great take. So true. People think that sort of Trump and Kamala was a true digital first election, but it wasn't. It was still legacy media talent and politicians that happened to kind of create digital appropriate content. Whereas I watched this breakdown this morning on X of this mayoral candidate all apparently his mum is a real famous Hollywood filmmaker and all of the videos that he was in, all of the campaign videos that he shot, even the on street stuff was shot with the same color palette. This very soft lighting, it's very. Well, there's a nice blurred effect bokeh depth of field thing going on in the background. Everything's shot in this way and it almost gives you this rose colored glasses view of what New York could be like. And you think, okay, like this is really taking it using AI voiceover stuff on TikTok. Like, okay, this is really, really, really stepping it up an awful lot. And yeah, it's, it's interesting to think about where people haven't fully sort of factored all of this in just yet.
- [143:48]
Dwarkesh Patel
Yeah. Because it's not visceral in the same way. It just, I mean, another interesting thing from the election is just people realizing that people they thought of as celebrities weren't actually the real celebrities in terms of the. You can get a random rapper to say endorse you or you can go on your podcast or Theo Vaughn's podcast or something. And who is actually reaching more people who, who. You just don't think of these other people as celebrities.
- [144:16]
Chris Williamson
I think what people are actually. And it's a shame that this word became so molested so quickly in the world of social media. But who is it that has the most influence? Like who is influential? And I think that, you know, a tweet from Stormzy or Dave or Central Cee or some British rapper saying that we need to do this thing for labor versus Dominic Cummings. I'm aware that he's not running for whatever, but someone marinating in a two hour conversation where me or you or whoever else grills Dominic Cummings about this thing and you know, he gets to put his personality across and it, it's not in any way. I don't think that me or you or anybody else in the world of podcasting has some undue degree of credibility that people who, let's be fucking frank, are way more talented. Like I can't do what fucking Dave or Skeptre or Central Cee can do. But there is a, a multiplier, like a vector of advantage from the format of a long form conversation. It's been done to death a million times. But there's nowhere to hide. People don't have anywhere to hide. It's very difficult to hold yourself. Anyone can pretend to not be a psychopath for five minutes. We've tried to do it for two and a half hours in a free flowing conversation. It tends to come out.
- [145:37]
Dwarkesh Patel
I, I sort of disagree there. I agree that certain things.
- [145:40]
Chris Williamson
Because you've hidden your psychopathy for the last two and a half hours.
- [145:42]
Dwarkesh Patel
Yeah, until now. We're about, we're two hours in and it's coming out. Where I agree that certain aspects of your personality will come evident, your charisma. But I think Douglas Murray had a point in his interview with Dave Smith, Joe and Dave, Dave Smith, where he said that vibing out is not. This is not like the checker of whether your ideas make sense. I mean you're. The point you made about that New York City mayoral candidate is exactly correct. Right. He is arguing for socialized grocery stores and rent control and things any economist would tell you don't make sense. But he can put warm tones on his cameras. And of course, he's a charismatic person, and that is enough to wipe out the deficit that his ideas have.
- [146:31]
Chris Williamson
Everyone's just vibing their way through whatever. Whatever level of influence it is that they want to achieve. Yeah, that's interesting.
- [146:39]
Dwarkesh Patel
So I do think that, like, I'm kind of skeptical of our medium as a way of intrinsically being geared towards truth or eliciting the truth by default. I think it's unless. Unless the interview is done in such a way that you're really pushing at the cruxes, which, to be honest, I haven't always done. Or after every interview, even if I think, or even if people in the comments are like, you did a really good job pushing, trying to get at the crux, I will always feel like there was something further that could have been said.
- [147:11]
Chris Williamson
Well, you know the thing that you didn't say.
- [147:13]
Dwarkesh Patel
Exactly.
- [147:13]
Chris Williamson
You all. And even if there wasn't a specific thing that you didn't say, you know, the sensation of feeling like there is something that you could get to in your mind but weren't able to bring out.
- [147:24]
Dwarkesh Patel
Yes.
- [147:25]
Chris Williamson
Right. So as a musician, you can hit the note that you meant to sing on stage in front of a few thousand people, but you know where you could have done it or you did it like that in a different show and you really nailed it. And you added this little bit at the end. And it's kind of the same sense, I think, when it comes to having a conversation that there's somewhere I'm trying to get to. Where the fuck is it I'm trying to get to? It's gonna be, oh, he's away. Fuck. Like, I'll just. I'll have to move on. And yeah, these micro victories and micro defeats that you have throughout every single element.
- [147:58]
Dwarkesh Patel
I mean, go through a little odyssey in the middle of a podcast episode.
- [148:02]
Chris Williamson
Dude, I've had episodes where I've literally gone on journeys around. Fuck, okay, what am I gonna say? I got this thing. And trying to put this together. Meanwhile, what's coming out of your face is, everything's fine, this is totally sweet, everything's cool. And inside of your mind you're going, ah, fucking screaming, trying to hold on. That's what I mean. Look, that is what the first time going on Rogan is like. The first time going on Rogan is a three hour panic attack masquerading as a conversation. That's what it feels like. And then you get off and you're like, what the. What the fuck did I. What did I Say, I think we talked about. I think we talked about Mike Tyson and oh, my God, did I give away my address? Honestly, dude, it's fucking. It's wild. And the craziest thing that you'll already have had, I'm sure, but will continue to have is like, here's a. Here's a point. Mark Zuckerberg, when he woke up on the morning to do your podcast, there'll have been a bit of him that was like, fuck, this kid's smart. Like, I should be a bit nervous. Like, or maybe his cheap. One of his staff said to him, like, maybe he didn't know. I don't know. And you're like, oh, that means that you get to be the anxiety attack inducing Joe Rogan for other people. And that'll get worse. I can't remember who it was the first time it ever happened. It was a virtual one. I was still back in the UK and someone was a fan of the show and I was like, oh, that's really cool. And they've mentioned maybe as we started, like, I'm a real fan. It's a real honor to, to. To talk to you. Thank you very much. That's very kind. Let's get into it. And then they finished up. Dude, I gotta tell you, I was very nervous before I started today, and I'm like, what?
- [149:48]
Dwarkesh Patel
I know why.
- [149:50]
Chris Williamson
What the fuck are you doing? Like, you're the expert, like the token in the room. Like, what are you talking about? But this is, I guess, I guess, for, you know, anyone who wants to climb the hierarchy of any industry that they're in. If you're a young person, you have idols that you look to, and if you achieve the thing that you want to achieve, if you actually do the thing that you're setting out to do, those idols turn into rivals after a while, and then they go from rivals into being friends and maybe even collaborators. And it's this weird arc where this, like, particular strata that you thought that you were in, you. You're sort of moving.
- [150:31]
Dwarkesh Patel
Oh, yeah.
- [150:31]
Chris Williamson
Wiggling through it. You go like, fuck, I'm so ups at Tony Blair. The fuck am I doing? So up to Tony Blair.
- [150:40]
Dwarkesh Patel
There's also the most surreal and gratifying thing has been. I mean, I was in college, not that, like, when we were talking about that meetup, I was in college and I was just on the side. I would be like, reading these books by these scholars that I really respected and oh, God, if they even, like, saw a cold email that I'd written. Not that I would often dare to, but if I did, I would be like so delighted if they mentioned me on their blog. If I wrote something. I mean, I couldn't even imagine that. And now to have those exact same people be friends, be people who I'm having discussions and debates with, who are you?
- [151:19]
Chris Williamson
See you as a contemporary.
- [151:21]
Dwarkesh Patel
Yeah. It is like there is nothing more. Just like heart, you know, heart pleasing, just like more satisfying. And also just that happening so fast because nothing is special about me. But just because this medium affords a level of virality and growth and public facing credit.
- [151:45]
Chris Williamson
Yep.
- [151:45]
Dwarkesh Patel
Which others?
- [151:46]
Chris Williamson
Personal accountability. Well, that's a lovely reframe. I think everyone has this sense, especially in the hyperviral growth loop, speedrun of fame thing that, that everyone has a little bit of like even I have a degree of ick around kind of the pace of change. I'm sure that you do too. Or it's like, oh fuck. Like there's a lot of exposure going on here and it feels very, very aligned with me. But there's a little bit of me that's like, fuck. Like this is a lot. There's a lot of like 800. I remember I used to feel anxious if the 24 hour plays ever went over a million. Whenever they went over a million, there was this thing that went in the back of my mind, I was like, oh my God. And that happened for about two years. And it would be fine if it was 500,000 and then if it was a million, 48 hour, you know, the 48 hour play thing on YouTube. And it would go up and I'd be like. And I'm like, what am I? Why am I feeling this way? And I realized this ambient anxiety was just the sense that lots of people are watching you right at the same time. Yeah.
- [152:40]
Dwarkesh Patel
And I'm like, ah. There's not only a stadium of people watching you when you're like performing, like.
- [152:46]
Chris Williamson
At any given moment you're asleep and it's happening.
- [152:49]
Dwarkesh Patel
There's a stadium like right now watching, still going. Yeah.
- [152:52]
Chris Williamson
And that dissipated a little bit. So anyway, everyone has this kind of scrutiny, people looking thing, degree of I don't want to be sold out. I don't have perverse incentives of distracting me away from what the main mission was, ruining my taste, ruining my gut instinct, all that stuff. And then your point there, that not becoming known by lots of people, not becoming popular in the circles of people who are popular, but being respected by people that you respect is what everybody really in any industry where they're curious I think should be trying to get toward. They genuinely care about what you think. He thinks that my idea is cool. He thinks that guy who is a legend, super genius, right, Like a fucking divine human, thinks that I have something interesting to say. You're like, all right, I challenge anybody to find a fucking problem with that, right? It's not shallow, it's not cloying, it's not sycophantic, it's not gamesmanshipy. It's. I went away and had a unique insight and a perspective on this thing that I cared about, that they also cared about and they hadn't fully thought about it before. And I got to contribute. I got my name put in my first academic paper. Like, someone cited an idea that I came up with to do with evolutionary theory around mating. This happened twice now. It happened first time, that happened the second time I was like, this is fucking unbelievable. I remember when I read the Evolution of Desire by David Buss, and now he's put me in this paper and it's like, it's so fucking sick. Like, me, retard from the north of the uk, like, getting. And, you know, there's cool shit you can do, but remembering, at least the more that I try to keep the. I don't know what you call it, like, virtuous flexes, as opposed to kind of the shallow flexes of subscriber count revenue or how many tickets you've sold to a live show, how many people turned up to a meetup and stuff like that. Like, that's still cool, but it doesn't give that same, like you said, sort of warm heart, delight of someone I respect respects me, yeah, 100%.
- [155:21]
Dwarkesh Patel
And even looking at a number on a screen go up, you know, it's like, whatever, it'll go up 10x and then that becomes your default. You know, there's a certain point where you go from zero viewers to like 100, and that is like, okay, people are actually watching. And after that, it's just orders of magnitude, right? A zero goes up, another zero goes up, another zero goes up. Nothing fundamentally changes in your life. There's a respect of the people you respect, which is uncorrelated to those numbers. There's also the feeling where even if it's not people you respect, but just meeting them in real life, just people on the street, they see you, they're like, oh, I love your content. And it's very easy to just get sort of used to that. But sometimes you pause and you think, like, that's a real human Being. And they've often. Do you have this thing where they're like, pull the phone towards you and they're like, I'm listening to you right now.
- [156:12]
Chris Williamson
Some guy did that outside of Flower Child last night. He's like, dude, that's fucking sick.
- [156:17]
Dwarkesh Patel
And that is like, a real human being is spending so much of their time. Hopefully you're contributing. It seems like you're contributing to their intellectual growth, their understanding of the world. And I mean, I think about. It wasn't that long ago I was in college, I was a teenager or whatever. I would drive around listening to Sam Harris, like, teach me how much that contributed to me being curious about the world, having different viewpoints, being, like, changing my career trajectory. Not just before I became a podcaster, I was gonna. I was studying computer science and I was gonna become a programmer.
- [156:52]
Chris Williamson
Steps of Sam Harris to become a podcaster.
- [156:56]
Dwarkesh Patel
I was gonna be a programmer after that. And then I decided, ah, that's getting automated. I'm gonna make the more financially responsible decision to go into podcasting.
- [157:03]
Chris Williamson
Yeah, yeah, yeah, yeah, yeah. The ripples are fucking wide, dude. You really. You really don't know. And I think that this is. It's the bull case for just producing stuff because you don't know. Especially when you start, you don't know what's good. Like, this might be. I feel like this might not be totally shit. And you kind of don't really know.
- [157:23]
Dwarkesh Patel
Right.
- [157:24]
Chris Williamson
And after a while, if you get enough positive feedback and you're diligent and you refine and you update and you keep going, oh, actually, yeah, it is. It was all right.
- [157:34]
Dwarkesh Patel
It might not be good, but that's like, that's the reason you do it. I had that very interesting. I noticed Scott Alexander on my podcast, and I had a very interesting conversation with him towards the end where I asked him, how many great new bloggers do you discover your. And he said, you know, on the order of one. And he. I asked him, okay, how soon after you have discovered them does the rest of the world discover them? It's like maybe a couple of months. Usually less than that. So again, go speaks to this dynamic where as soon as you are making good content, I think you might. Or not you. But somebody listening might underappreciate the extent to which it will immediately be seen by all the people you want to see. Might not happen immediately, but, like, genuinely, it's shocking how fast good content is going on.
- [158:18]
Chris Williamson
Well, think about how, by design, most content that gets consumed, the biggest, most widely distributed stuff goes to the biggest number of people.
- [158:29]
Dwarkesh Patel
Right.
- [158:30]
Chris Williamson
The channel with the most views, has the most views. What a shocking insight. But what that also means is that if you as a viewer have peeled off from the biggest channel to watch this fledgling small thing and it's captured your attention, it's gotta be really, really, really fucking good to be able to do that. And you have to assume, like it kind of goes back to what we were saying earlier on, that if your instinct drives you toward a thing, you have to assume that some non zero number of other people are probably interested in it too. The same thing goes as a viewer, it's like if you thought it was good, probably like some other people will think that it's pretty good as well. And if this person's just a bit consistent, like I would actually say substack for me has one of the highest densities of as yet undiscovered talent out there. And maybe it's just that the particular sort of format and language of substack lends itself to me. I quite like pithy stuff and I like the feed and I like the fact that most articles are about 10 minutes long and like the. The. My attention spanking around about. Hold on to that.
- [159:39]
Dwarkesh Patel
You're way ahead of the rest of us.
- [159:40]
Chris Williamson
Yeah, yeah, that's true. But I'll find people on there, like some of the people that I've been subscribed to. I've been subscribed to for like three years and now they're blowing up. I'm like, it was obviously a matter of time. Like, it was obviously gonna happen, but by design, it can't happen to everyone.
- [159:58]
Dwarkesh Patel
Right?
- [159:58]
Chris Williamson
So it's the same thing as the. Why is it that the people that I'm friends with end up doing all of this stuff? I don't think. I don't know. Maybe we've both got phenomenal taste. Just not sartorially.
- [160:08]
Dwarkesh Patel
I've noticed this in people in other industries say the exact same thing. You know, I'll ask like the CEO of a big company, or they'll mention that they were friends with the other CEOs who are now running all the big companies back when they were college students. And not even necessarily the same college. It's just like they saw each other.
- [160:27]
Chris Williamson
How the fuck did this.
- [160:29]
Dwarkesh Patel
Yes.
- [160:30]
Chris Williamson
Group come together exactly.
- [160:32]
Dwarkesh Patel
And they don't know.
- [160:34]
Chris Williamson
Is it just that game recognizes game? Is that just it? I don't know.
- [160:37]
Dwarkesh Patel
I think maybe not a lot of people do stuff. And so if you're doing things, you will meet the people who are also doing stuff. Yeah. Is it weird in the. How weirdly small the world ends up being? How many? I'm, I don't know. I, I, I'm friends with a couple people in San Francisco who I now have had on my podcast and are, I don't know. I was. Okay. This is a funny story. This maybe the fourth or fifth person I interviewed on my podcast. I never released this, but it was in 2020 was Leopold Aschenbrenner. I was like 19 and I think he was 17 or something. Do you, do you know who this guy is?
- [161:17]
Chris Williamson
No.
- [161:18]
Dwarkesh Patel
He wrote this memo called Situational Awareness. This long blog post that went super, super viral.
- [161:22]
Chris Williamson
Situational Awareness, Yes.
- [161:25]
Dwarkesh Patel
And it was like the most popular thing on AI I've written over the last two years. And he was like a 17 year old at Columbia and we've been friends since then. But anyways, one of these things where, how did that happen? Right? How did we know each other for so long? That's happened to Mia so many times. It's sort of uncanny.
- [161:44]
Chris Williamson
What was his name?
- [161:45]
Dwarkesh Patel
Leopold Aschenbrenner.
- [161:47]
Chris Williamson
Leopold Aschenbrenner.
- [161:49]
Dwarkesh Patel
And there's so many others. Like all the AI people that I've had on the podcast are just people I like met at a party two, three years ago, researchers, whatever, Sholto Douglas or Trenton Bricken or so forth.
- [162:01]
Chris Williamson
Maybe you're right. Maybe it's just that most people don't produce stuff and that by producing stuff you inevitably separate yourself and there's a.
- [162:09]
Dwarkesh Patel
Feedback loop where like you actually get input from the world, you meet mentors, that puts you on this upward trajectory.
- [162:17]
Chris Williamson
And especially if it's good, if the thing's good and if you're improving, you show potential. Yeah. I mean, like I say, that Elliot kid I met the other day, Jack Neil, he's 20, this Elliot guy's 20 years old. This podcast called Next Generation, I think, and I'm having this chat with him and I'm like, you do realize that the shit that you're asking me as a 20 year old is stuff that I only asked myself like three years ago. Like these questions about the balance between inputs, outputs and outcomes. The realization that he'd sort of attached a sense of sacrifice and difficulty with being worthy and validation. And that was something, this was like this Gordian knot he needed to cut through. I'm like, who the fuck are you? Like how the, how. Obviously you're going to be great. Obviously you're going to be successful. And George Mack. George. I met George. Fuck 2019, 2018, 2019. I remember sitting down, he'd sent me a message when I went to his office to interview one of his bosses. And he sent me this dm and the DM said on Instagram, cold dm. We'd never spoken. I didn't know where he worked, didn't know who he was. I hear you're coming into my office today, full stop. Let's exchange Google Chrome extensions. And I was like, this is my fucking guy right here. He stinks of me. Sure enough, half a conversation with him. He was way more interesting than his boss that I sat down to speak to and we moved to Dubai together. He's just moved to Austin. He's gonna live with me. Like, we've been best friends for fucking six or seven years. He's just got a huge, huge book deal to write this thing that he built out of an essay. The essay was what he launched on my podcast. So we did this episode at the back end of last year that came out in March. He's just about to announce he's got this book that's coming. I'm like, like. And it's my manager that's doing his book deal. Like, again, the incest fucking wheel, human centipede of stuff keeps going. And yeah, this is just. It's one of the areas that, you know, me and you can continue to pontificate about. We know cool people and isn't it fun and all the rest of the stuff, like, we can keep going. But I really hope that people take away from this is if you put yourself out there and if you are able to discern good work from bad work and virtuous people from non virtuous people and industrious people from non industrious people, and you are able to contribute to that. Like, literally the sky is the fucking limit. Because the step change opportunities that will come along by somebody being there and being able to contribute, give you the intro and help you along with your thing and you doing that and then, okay, this is the scene. Like, this is now the fucking scene. It's really cool and it's very gratifying. And it's gratifying in a way that doesn't make me want to have a shower after, you know, like, like, like an episode doing really big numbers is really great and gratifying, but not in the same way as someone going, dude, that fucking. That idea that you came up with about that was sick. Yeah, you're like, okay, I'm gonna think about that for the next three weeks. Thank you very much.
- [165:28]
Dwarkesh Patel
Yeah, 100%.
- [165:29]
Chris Williamson
Fuck yeah, dude. I appreciate the hell out of you. I'm so happy to see what you're doing. Dwarkesh podcast people should go check that out. Substack as well.
- [165:38]
Dwarkesh Patel
Yes.
- [165:38]
Chris Williamson
Dwarcash.com dwarkesh.com dude, appreciate the fuck out of you. Thank you.
- [165:43]
Dwarkesh Patel
Appreciate you man. Thanks for having me on.
- [165:46]
Chris Williamson
I get asked all the time for book suggestions. People want to get into reading fiction or nonfiction or real life stories. And that's why I made a list of 100 of the most interesting and impactful books that I've ever read. These are the most life changing reads that I've ever found and there's descriptions about why I like them and links to go and buy them. And it's completely free and you can get it right now by going to ChrisWillX.com books that's ChrisWillX.com books.