The Pomp Podcast - #1093 Giuliano Giacaglia On How A.I. Will Take Over Your Life

Episode Date: September 20, 2022

Giuliano Giacaglia is the Author of "Making Things Think". In this conversation, we discuss how A.I. will take over every aspect of our life and how it's doing so already. We break down the A.I. in companies like Tesla, OpenDoor and Tiktok, and discuss how every company will utilize Artificial Intelligence. ======================= Amberdata provides the critical data infrastructure enabling financial institutions to participate in the digital asset class. We deliver comprehensive data and insights into blockchain networks, crypto markets, and decentralized finance. Download our Digital Asset Data Guide at https://www.amberdata.io/pomp ======================= The Pod Pro Cover by Eight Sleep is the most advanced solution on the market for thermoregulation. It pairs dynamic cooling and heating with biometric tracking. Even better? Eight Sleep recently launched the next generation of the Pod. The new Pod 3 enables more accurate sleep and health tracking with double the amount of sensors, delivering you the best sleep experience on Earth. Go to eightsleep.com/pomp ======================= This episode is brought to you by Athletic Greens, the best option for easy, optimal nutrition out there. You take one scoop of AG1 and you’re absorbing 75 high-quality vitamins, minerals, whole-food sourced superfoods, probiotics, and adaptogens to help you start your day right. This blend of ingredients supports your gut health, your nervous system, your immune system, your energy, recovery, focus, and aging. Go to athleticgreens.com/POMP to give AG1 a try today. ======================= If you’re trying to grow and preserve your crypto-wealth, optimizing your taxes is just as lucrative as trying to find the next hidden gem.Alto IRA can help you invest in crypto in tax-advantaged ways to help you preserve your hard earned money. So, ready to take your investments to the next level? Diversify like the pros and trade without tax headaches. Open an Alto CryptoIRA to invest in crypto tax-free. Just go to https://altoira.com/pomp ======================= Bullish is a powerful new exchange for digital assets that offers deep liquidity, automated market making, and industry-leading security. Combining the innovations of DeFi with the regulated environment of traditional finance, Bullish empowers users to trade with certainty and earn passively at scale across variable market conditions, in an environment backed by multibillion-dollar liquidity contributions from the Bullish Treasury. Visit bullish.com/pomp today to learn more. Note: Bullish is licensed by the Gibraltar Financial Services Commission. Virtual assets and related products are high risk. Consult your investment advisor and trade responsibly. Bullish is available in select locations only and not to U.S persons. Visit bullish.com/legal for important information and risk warnings. ======================= The bridge between iGaming, online sports betting, and emerging technology, such as Blockchain, NFTs, fintech, GameFi, metaverse, and AI, is loud and clear. The largest global summit of this kind is heading to Malta from November 15 to 17. Over a thousand exhibitors and 25,000 industry leaders will be there, including top executives from Draft Kings, Bet365, Socios, crypto exchanges, betting software providers, operators, gaming affiliates, and more. Log on to AIBC.WORLD and SiGMA.WORLD to see all our upcoming global summits! See you in Malta, November 15 to 17 for the leading global event in Gaming & Emerging Tech! THIS IS SiGMA! =======================

What's up, everyone? This is Anthony Pompliano. Most of you know me as Pomp. You're listening to the Pomp Podcast, simply the best podcast out there. Now let's kick this thing off. Giuliano Giacoglia is the author of Making Things Think, How AI and Deep Learning Power the Products That We Use. In this conversation, we talk about artificial intelligence, machine learning, neural nets, how it will affect various companies and products like Tesla, Opendoor, TikTok, and many others. We also discussed the importance of self-driving cars, how AI will intersect with Bitcoin,

and many other topics that I think folks who are interested in technology and this disruptive innovation will enjoy. This conversation with Giuliano is one that I learned a lot and I hope that you guys enjoy as well. Once you've listened to the episode, go to Twitter and tweet at us

and let us know what you thought, good and bad. We appreciate all the feedback. Let's go ahead and get into the episode and I'll see you guys at the end. Anthony Pompliano runs Pomp Investments. All views of him and the guests on his podcast are solely their opinions and do not reflect the opinions of Pomp Investments.

You should not treat any opinion expressed by Pomp or his guests as a specific inducement to make a particular investment or follow a particular strategy, but only as an expression of his personal opinion. This podcast is for informational purposes only. All right, guys, bang, bang. I'm here with Giuliano.

I thought a great place to start this conversation would be around Lambda, which for those that don't know, there's been a bunch of debate and controversy. Is this computer program sentient? Is it alive? Is it intelligent? Walk us through whether – have humans actually created a computer program that is alive and smarter than us? Yeah, so let's just go over how Lambda works, right? So Lambda is a neural network, and a neural network is basically a representation of how our brains work. But it's just basically simple math.

It's basically matrix multiplication. Okay. You realize matrix multiplication is not simple math to somebody like me? But for computers, it's very, very simple, right? uh um so how lambda works is basically you train with you know a lot of text and you create this big big neural network that's just taking basically you know uh taxes input and generating the next uh word um and and you're taking input to all this data yeah then basically building this neural net and then that neural net is kicking out outputs and in almost like a predictive way

or is it simply trying to synthesize is it sifting through it and looking for something or is it more so trying to predict some outcome yeah so it's trying to predict the next word uh it's usually how how these neural networks work um but the funny thing is that like there was this google ai engineer that thought that okay we can create actually a computer that is sentient because he talked to it he wrote some text and the and the computer wrote some text back and it fooled the engineer it fooled the engineer saying that it was actually sentient so the engineer is literally sitting there almost like if i type into google and it does give me answers but it gives me links to other uh websites or other other services the engineer was almost

like in like a chat and it was like literally like talking rather than you being on the other side of the chat and responding the computer was chatting back and the engineer literally was like the computer has a mind of its own it's alive like holy shit exactly and the funny thing is that the engineer thought that it was sentient because it convinced the engineer that it was sentient because the engineer asked are you sentient that's why the engineer thought that it was alive was because it literally just asked like are you alive yeah exactly and and the funny thing is that It's basically grabbing text from the internet, right, and figuring out the relationship between different words

and seeing which words are connected to others. When you look at text in the internet on Reddit and, you know, like on Google and whatnot, you would see words that are similar to what the engineer was asking. And if you ask the computer to prove that it was not sentient, the computer would also prove that it was not sentient And then the engineer would be fooled thinking that the computer was not sentient, right?

So this brings us to a question of like this idea of being sentient, right? Like what are the ways that engineers think about like when would it be sentient versus, oh, no, it's just predicting. It's just answering with questions off of the internet. Like what is the difference in a computer program between it actually being sentient versus it's just simply really, really good at using machine learning or artificial intelligence? Yeah. Personally, I think it's more a philosophical question than a technical question. There's all these questions of consciousness and sentience, if a computer is sentient. But if you actually look at animals, there's a very clear correlation between the number of neurons in a brain and its cognitive ability.

so if you look at you know mammals in general the number of neurons in the neocortex is very very predictive of their ability to do things and that's why humans are like much better doing tasks than chimps and you know other mammals is because we have so many neurons in our brain and one interesting aspect of humans like the reason why we have so many neurons in your neocortex is because we develop fire. And that's one of the reasons why you see so many people talking about meat, because instead of digesting all these resources in your gut, you know, we can actually break down the proteins outside of our body and actually eat, consume all these resources and produce that many resources to our brain because our brain needs a lot of resources.

so the whole idea is that like as a human being one our brains are highly dense in the terms of neurons uh which is equivalent in a computer program the more neurons or or uh kind of you know uh pieces of the neural net yeah uh the more dense it is obviously the more valuable it could be um but also there's ways to use external forces uh or uh materials to change our body physiologically right if we eat really bad it doesn't just make us fat it also can have an effect on our brain you're sluggish you're tired 100 you lose your intellectual sharpness like all that type of stuff if you eat really healthy there is literally a way to kind of clear out the cobwebs if you will uh from from a brain standpoint uh and perform better with more mental clarity more

kind of high performance intelligence all that type of stuff yeah 100 and if you actually look at other animals right uh the bigger they were the bigger their brains are so if you look at gorillas they have bigger brains compared to other chimps and even like an elephant has a big big brain but most of their brain uh like the biggest part of uh the biggest part of elephant's brain is in their motor cortex and so they can do really good things like they can actually do physical things really well that's why you see elephants in circus right and what happened is that at some point in human evolution and in mammal evolution we we develop fire and we could actually break down proteins and create so many like neurons in our brain and pack them

and actually eat a lot of food and eat a lot of resources and actually use those resources for our brain. The same thing is true for computers. The bigger the neural networks, the better they perform. And it's interesting, our brain has around a hundred billion neurons. And if you look at the best neural networks today, they have around the same number of neurons. Why is it that if we have a hundred billion neurons and these neural networks have a very similar number, Why is it that humans in some capacity is still much more intelligent than the computer systems? And I'm assuming that just computation, right? The computers are better at other things.

But is there just some other way to evaluate their neural nets other than just like the density of the actual net itself? And it's missing something that doesn't actually make it sentient or have the ability to compete with the human brain? Yeah, that's an interesting question. so if you actually look at how we represent our like each neuron uh in the in our computer system they are very very simple compared to our neurons so maybe we don't map one to one it could be maybe a hundred neurons represent one of our brain neurons but if you actually look at different tasks right so if you look at go uh if you look at poker if you look at chess computers perform better than humans at these tasks and when a computer is better than humans at let's say

poker go uh even chess is it because there is a set parameter i kind of think of it like uh bowling yeah right like like it has the bumpers and so like it's kind of hard to go without outside of it and if you give a computer set parameters and even though a chess board as an example has many different moves you can make there's only so many combinations and if the computer understands all of that then the computer is able to just compute at a faster pace than the human and understand what to do your intuition is right okay so one thing is that computers do really well at games And why do I think that's true? It's because games are simulated environments.

So you can play around and train your computer through many years of training compared to humans, right? But the universe of possibilities in a Go engine is really, really big in a Go game. So if you look at a Go game, there are more positions in a Go game than there are atoms in the universe. Wow. Even though it's a very simple game, it's actually, like, there's a huge number of possibilities. There are more possibility, like, possible scenarios in a Go game than there are atoms in the universe? Yeah, yeah. That is a mindfuck.

Like, that is insane. Yeah. Um, and if you look at other games, you know, like Alpha, uh, Starcraft, right? We have game engines, uh, that perform better than humans at Starcraft or Dota. Uh, so these are very complex systems, right? Even though the rules are simple, they're very complex systems. And I would argue the reason why we do much better is because you can simulate the environment and you can play around so many times.

And if you actually look at robots that perform really well, you simulate the environments before. So OpenAI, which is a research company based in Silicon Valley, they develop a robot hand that plays Rubik's Cubes and it can actually solve it pretty fast. And the way they did it is that they simulated the environment before. they trained the hand in this simulated environment many, many times.

And the hard part was to actually transform that into the real world. And I think- So the computer theoretically could solve the Rubik's Cube well before they had built a hand to actually do it. And once they were like, okay, cool, the computer knows how to solve this problem, then you've got to translate it from pure software

into the physical ability to move the Rubik's Cube. Exactly. And that was, according to people that I know work there, it was the hardest part to transform that from software to hardware. And you actually, I think Tesla now will eventually have the best self-driving cars and their self-driving cars will be better than humans probably this year or next year. And I think one of the missing pieces that they had was to simulate the environment, right? They didn't have a simulated environment where they can create all these scenarios that you wouldn't necessarily see all the time, right? You have this long tail of possibilities that you wouldn't see necessarily in the real world, but you still want to train the computer. um and even though there are a million you know teslas out there and they are grabbing data a lot of data is still not enough because there will be there will be scenarios where you didn't see

that scenario before and you want to train and and and figure out what's you know what you want to do and that's a big that's one of the missing pieces for computers to you know reach what they call agi i think that's one like very important missing piece right now when we train computers we need a lot of data right so if you look at um dali uh which is this computer that basically you type it out a sentence so let's say a koala dunking uh uh into a uh a basketball you can transform that sentence into an image right so it's just text to image and this is the engine to do it exactly and the way they did it is that they grabbed 400 million image captions so that's a lot of data right to train the system the way humans work is that we actually

learn through a lot of data, but after some time we don't need that much data. So what do I mean we learn by seeing a lot of data? You take a long, long time to learn basic things. When you're a baby, you're seeing the world and you're reacting to it, but eventually you don't need that much data to learn new things. And one example that people usually refer to is you don't need to drive a thousand times out of a cliff to know that you can't drive out of a cliff and why is that is because we actually imagine that situation right and is that the equivalent that's the human equivalent of a simulation exactly exactly and if we go and we look at uh so let's go back to the chess example if uh i'm a chess player i recently talked to a grandmaster uh and he was

talking about he played two hours a day every day for you know eight ten years read a lot so not just playing but also reading through different scenarios different moves different responses to those moves uh studying historical games all these different things how many games or how long would it take a human to be able to get the same amount of reps that the computer could just simulate it seems like the computer is much more efficient much faster and therefore on a compounding scale it should run away from kind of the intelligence that the human has in a game like chess. Yeah, so that's a very good question.

So if you actually look at Go, I think the Go engine that beats the best player trained for 1,000 human years. Jesus. And how long do we know did it take to train the 1,000 human years for the computer? Oh, I don't... Like months, years?

Yeah, less than months, right? Right. So pretty quickly, we were able to pack a thousand years of one human playing Go into a couple of weeks at most. And that's what people don't realize. Right. We learn through like we pass knowledge through generations. So if you look at Go, for example, there were like three major developments in Go history and they've learned new techniques and that was passed down. So you have this teacher-student kind of knowledge passed down through generations, and that's through history. So we don't only learn in our lifetime, we actually learn what our parents and grandparents passed down to us. And computers can learn much faster than that, because they can simulate all of that. And one of the missing pieces for robots is that they need to actually do the simulation.

And that's what Tesla is doing right now, right? They can simulate all these scenarios and simulate a thousand years of driving. So let's take Tesla as an example. It used to be just in Silicon Valley. Now across the country, people will see the Google car drive by, and it's got the big thing at the top of the vehicle, and it's spinning around. uh and you don't know many times is it like they're collecting data for google maps is it the self-driving car if you're in tune with the differences then maybe okay that's the self-driving car and they're obviously driving with a human behind the driving uh behind the steering wheel

to capture data like that is what that car is doing uh and that was a pretty cool idea it was like let's put a bunch of cars on the road and capture that data we'll have our employees behind it and over time hopefully we can get to the point where we could create self-driving cars makes sense tesla is like well why do we have to have tesla cars why don't we just have everyone who drives a tesla collect the data for us when they're just driving around the road we'll just collect every single piece of data yeah and so i don't know if that's one to a thousand one to a million you know difference but obviously tesla is collecting way more data than google or anybody else but what you're saying is it's not just about collecting the data from the cars on the road

they're then taking that and then they're running additional simulations on top of it to even accelerate the learning faster yeah that's right and they started doing that recently they started doing that i think last year to this year okay so they are running simulations right now and google has been running simulations for a long time but i think it's a combination of both right when i'm when i talked about open ai and the rubik's cube um they the hardest part was transforming like getting the software and then getting that to the hardware right and and i think both parts are really, really important. For Tesla, they have both parts. They have a million cars running in the road and they also have the simulation where they can see scenarios that, you know, they haven't

experienced before and they can actually experience that, you know, in a simulated environment. So they can train a thousand years in, you know, a few months. And when we think of this, how long will it take for Tesla to have fully self-driving cars where I can sit in the back seat? I don't need to be anywhere near the driver's seat and it'll drive me around. And that will be not only possible, like it can actually get me there, but it will be safe and that will become the norm, do you think? Yeah. So we are in September of 2022. I think it's going to be either end of this year or next year. You think by the end of this year, potentially in three or four months we will have the ability for the car to navigate uh the everyday driving for anyone who

wants to sit in a tesla exactly um but they are not gonna allow people to do that so there's the technical capability maybe there but the regulatory approvals or or the uh approval to actually do this on the street won't happen exactly and elon musk has been saying that for years right he has been saying oh we're gonna have self-driving cars that are better than humans by end of this year And he has been saying that for five or six years. But the difference now is that they have these simulated environments that make the car know how to drive in all these different scenarios. And is the simulations trying to get reps because the computer needs to see the situation over and over and over and over again? Or is it if the computer sees a situation one time, it remembers it forever.

and it's more so about the super, super long tail of a kid runs out in the road, then a car backs into the road, then a bicycle goes into the road and it just has to see every single situation. It's the latter. Okay. So it needs to see all different scenarios

that it hasn't experienced before, right? And it needs to know what to do. If it hasn't seen, and maybe it can take some wrong decision, right? So you need to have all dissimilated situations so it can know when that situation actually happens. Is Tesla in these simulations, like, are they making the wrong decision and like hitting in, you know, fake pedestrians or, or like crashing into fake cars? And like, that's part of the learning process is somebody then sits there and says like, no wrong decision.

The car should have done this instead. And it's almost like this like human feedback loop that's correcting the actual computer. Yeah. And they actually have that information also from the cars running in the, in the real world. If there's accidents, obviously, they know, hey, it shouldn't have gone that way. Or if someone, like if the car is turning right and the person driving the car actually takes the wheel and drives somewhere else, they can actually get the data and send back to Tesla. So they actually, when they are gathering that data, they don't send all the video feeds directly to Tesla.

They send selected frames because else it would be a lot of data they're gathering. So if you're just driving down the highway and Tesla car, all the cameras on it is collecting all of this data. If there's almost like no outlier situation, Tesla as a computer has seen a million times driving down the highway with other cars around it or whatever. It's when somebody grabs the wheel and jerks it one way or the other, or slams on the brakes or, or there's some sort of external or outlier type event. That's when maybe they say, Hey, we want to see that. And that may be a situation that we haven't seen before. Exactly. And the interesting part of Tesla is that they have the biggest neural network running at scale in the world. Really? Yeah. So

if you think about it, right, like Facebook, Google, they're a big, pretty, like, pretty big companies, right? And most of the neural networks they're running are very, very simple. So they're running neural networks to look at an image and see, oh, look, there's a beard there. There is, you know this thing in this image or google also has a birth which is a big network a big transformer that actually answers questions so like if you if you search some terms at google some of the the the answers will come from a neural network right um and and this is a dumb question but like the neural network a lot of people be like where does that live is that on like facebook servers or google servers somewhere is that something that's in the cloud and it's

like they're using uh other services is like how do you think about like the neural network where does that actually live yeah they're mostly running gpus okay and that's why nvidia has been doing so well in in the market selling weapons to the uh to the war exactly and and And why GPUs? Because GPUs are really good at doing matrix multiplications, which these neural networks require. So neural networks, they're interesting. The reason why we use neural networks at the end of the day

is that a neural network is a circuit. And if you have enough long neural network, it can simulate a computer. So a neural network is basically a computer. and a neural network can actually represent any kind of function that you may like. They are really good at solving the problems that we like and the way we train them is linearly

compared to the amount of data that we feed them. So they're really, really good at solving the problems that we give them and that's why. So the way I've always thought about it is you have a computer and the neural network is software that resides on hardware somewhere we take these massive data sets we feed them in do we tell the computer what to look for or do we just feed it the data and then later when we're looking for the outputs we're querying that data

set plus the software like how much prep work i guess goes into cleaning and structuring the data that we feed it and specifically telling the uh the neural network learn image recognition learn this and then later we can query it because we gave it that instruction yeah so there are different ways of training these neural networks. Um, Google, for example, had this breakthrough at some point where they fed a lot of YouTube videos and the computer learned what a cat is because cats on the internet is an important thing to know. Yeah, exactly. Um, but most of the systems, the way they work is that you actually feed like both the data and what you want to see out of data so for example for uh dali they fed them images and captions right and a big part of

these systems is actually cleaning out cleaning out that data and that's what tesla does so well right got it like they figure out what data and how to clean that data really well um and the best systems still you need to like actually figure out okay this is the image and this is the caption you need to you know tie them together and if we go and kind of zoom out for a second right we say okay these neural nets and all this is very kind of uh technical it's very nuanced in the larger conversation uh there is artificial intelligence yeah there is machine learning and then another term that people may or may not have heard of is agi right kind of uh the general intelligence walk me through what is the difference between these three things because a lot of times like

when I see pitches from companies or I'm talking to someone, everyone's like, AI, AI, artificial intelligence is going to change the world, right? But there's some nuance and difference between these three things. So describe a little bit as to how people should think about it. Yeah, artificial intelligence englobes everything. So artificial intelligence-

It's the high-level terminology. Exactly. So artificial intelligence are systems that are intelligent, so they can interact with you and whatnot. Machine learning is when computers learn the features by themselves. So it learns by itself these features that we're, you know, interacting with the world. And AGI is basically when computers get better at humans at every single task. Got it.

So AGI is when, you know, a computer is better than humans at go, at chess, at driving, at talking, everything. Thinking, dreaming, sleeping, whatever. So is it fair to say there is no true AGI yet? Yeah. That's kind of a fair characterization, but that is something that people aspire to create. And if and when we get there, that would be a massive technical breakthrough in terms of computer science. Yeah. So one of the interesting things is that when computers get to AGI, right, they will be better than humans at every single task.

and we created these computers so when they get better than us they probably can create computers that are better than they are right so you have this exponential curve that you hit when computers get to agi they can actually get better and better over time should we be scared of that like that sounds like pretty fucking crazy stuff right like if we can create computers that are smarter than us in every dimension not just in kind of mathematical computation but literally it can drive better than us it can play games better than us it can act like a human better than us uh and then those computers can create even smarter computers and there's this exponential runaway of uh of uh kind of artificial intelligence humans very quickly become the lowest on the

totem pole it seems yeah i think we shouldn't worry about it even though yeah even though uh people talk about it and they hype it up yep we are the ones creating these tools and we can control it very well i think when and if we hit that point uh we should cross that bridge when we hit it right uh right now we have many other problems to worry about like there are energy problems you know in in europe in california and you know other areas of the country um are there things that we could do now to set ourselves up better for that scenario or is it just something like right now kind of everyone's talking and and they're worried but actually there's nothing that we could do in the moment to change when we get to that situation yeah so there are organizations

that are actually working on that so there's an organization that came out of like a bunch of people left open ai to create this organization called entropic where they're trying to prevent risks that AI eventually can cause, right? But except that, I don't think there's much we can do, right? Like, it's difficult to create policies about a scenario that can happen. It's really, really hard. This episode is brought to you by Amber Data. If you're a financial institution entering the digital asset class, you'll need access

to granular on-chain and market data from multiple venues to power research, trading, risk management, and compliance. AmberData delivers comprehensive data and insights into blockchain networks, crypto markets, and decentralized finance, empowering financial institutions to apply traditional finance methods to digital assets. AmberData eliminates the infrastructure setup, integrated challenges, and maintenance headaches to access digital assets data, reducing cost and time to market to enter the digital asset class. Learn more and download their digital asset data guide at www.amberdata.io. slash Pomp. Again, that's amberdata.io slash Pomp. Go check them out today. This episode is brought to you by Eight Sleep. Good sleep is the ultimate game changer and the Eight Sleep pod is the ultimate sleep machine. The pod is the only sleep technology that dynamically

cools and heats each side of the bed to maintain the optimal sleeping temperature for what your body needs. With the pod, you can start sleeping as cool as 55 degrees Fahrenheit or as hot as 110 degrees Fahrenheit. What is the result? Clinical data shows that Eight Sleep users experience up a 19 increase in recovery 32 improvement in sleep quality and up to 34 more deep sleep how do i know it works i sleep on it every single night and it works so well that i begged the founders let me invest in the company go check them out today at eightsleep.com slash pomp to start sleeping cool this summer and save 150 on the pod again eightsleep.com slash pomp you get 150 off when you use code POMP. Now, machine learning is here. It exists. We talked about these like neural nets.

Is machine learning similar or different in terms of feeding it information and what it does with that information to be able to be useful to humans? Yeah, machine learning encompasses a lot of other techniques. Neural networks are just one, like one subfield of machine learning, right? And neural networks is is basically a technique to simulate how our brain works basically and neural networks have been doing well so uh and have been solving problems that we like most of the new machine learning techniques that we see today uh they are using neural networks because they perform really really well and i think that's gonna be the case in the future are there other what are the other types of machine learning techniques yeah there there's for example svms they're like

what is that like neural nets we've talked about what is the svms how do they work uh you're you're basically grabbing a lot of data and and you know uh figuring out where this data is in in a graph and creating a line that divides you know uh some of the points to to other points it's basically you know um it sounds like you're mapping connections between different pieces of information and also knowing where it is identified in in the graph exactly exactly and then you have like is it the wetness like i'm gonna be really dumb i already told him that uh all the questions i asked people be like oh my god pop is so stupid um but is that similar like what a web crawler does no it's a little bit different a web crawler usually uses also matrix multiplication so uh

At least I got that part right. A web crawler usually, you know, figures out how many links to a page there are. And you can determine the importance of a page if there are a lot of links to that page. And then you can, like, yeah, that's basically it. Got it. And so when we think of machine learning,

there's neural nets, there's SVMs, there's a couple of these different things. But neural nets seem to be right now at least the superior way to do machine learning. And that's why we're seeing so many people go and use these in these various applications. Exactly. And again, the reason for that is that they can simulate computers really well. And they solve the kind of problems that we see in the world really well. And somehow we're basically simulating how our brain works, right? And somehow we develop this thing, this machine that interacts really well with the world.

So it's not a coincidence that, you know, neural networks and our brains developed. So I think of humans of like the kind of physical world pre what we think of as software today. Humans rose in dominance and we're like the King Kongs of nature, right? Like we can collaborate, we can think, we can dream, we can organize resources in a very unique way. And for all intents and purposes, we have weaponry and like we are the dominators of nature on a global scale. Computers used to be really dumb. Like there was punch cards. We started out real low with the computers.

But they have rapidly been rising in efficiency, efficacy, computing power, all these different things. As we get closer to that being an even playing field between humans and computers, it then brings up this idea of the Turing test, which tests for is the computer sentient? is it not? Talk a little bit about what is this Turing test? How did it come to be and how effective is it? Yeah, the Turing test was developed by Turing, which was a computer scientist that developed this game called the imitation game, where you had in one side of the IO, you had a human and then you had a black box and then you had a human on one side and a computer. So the human would type questions uh to both the uh both the computer and the human and if the human was fooled thinking that the machine was a human or vice versa right uh the the machine would pass

what's called the turing test got it so i'm sitting here i'm typing into a chat box yeah and uh on one side there's a human on another side there's a computer if i can't tell that the computer is the computer then technically what's the difference so that would be passing the turing test and then we could claim that the computer was equivalent to the human. Yeah. And the funny thing is that since 2014, we've been passing the Turing test. Oh, shit.

And this is why the Google engineer got fooled. Yeah, exactly. And thought that, oh, wait a minute, I'm talking to a sentient thing. Holy shit, I got to like raise my hand and caution people. Yeah, exactly. If a Google engineer was fooled by a chatbot, imagine, you know, the average human.

And the funny thing is that humans are fooled by computers much more when they talk about sex interesting yeah and especially males uh-huh because males are dumb and so that's one of the ways that people actually are able to fool humans is they get males talking about sex to computers and the computers trick the men yeah jesus we are so screwed as humanity we are so screwed if they like if the machines get so smart that they realize all we have to do is talk about sex to men and then they're going to do whatever we want that is a bad scenario to go towards exactly and so we've passed the turing test and we've been passing almost eight years now yeah exactly um so we have been developing new tests and new

you know benchmarks and computers have been doing better and better so like for example dota right the game computers play better than humans starcraft which is really really hard or go computers pass humans their poker and we've been doing better and better at the exponential rate so driving driving is another task that humans do right and it's a game to some degree exactly it's kind of a test right of like hey can we figure out this technical breakthrough to actually be able to do this yeah you and i don't think about that when we're in the car and like yo elon make sure i don't crash right but to some degree like that is how the computer looks at it You can play the game of chess with all the different possibilities and rules, and then you also could drive down the street.

There's differences, but not nearly as many differences as maybe people think. Yeah, I'm very optimistic in the way I think, you know, in the world 10, 15 years or 50 years from now. And I actually think these computers are just increasing our productivity, right? So one example of that is that you look at elevators. Before elevators, you had a human actually moving the elevator around, and then we could automate that. And then you don't need humans anymore to, you know, operate these elevators. And the same thing is happening in different areas, right? So you look at farming, for example.

A hundred years ago, I think 50% of the population was working farming, and now you have 1% of the population. A lot of people drive trucks right now. It's like the most popular job right now in the U.S. is driving trucks. Eventually you won't need to drive trucks and then you develop new jobs, right? So talking in a podcast is a pretty new, like pretty new job, quote unquote, right? Investing is a pretty new job. So if you look at the percentage of GDP that the U.S., if you look at the percentage of GDP that entertainment has, it has been increasing over time in the U.S. It was 6% in the 1960s and it's now 10%.

Wow. So we're spending more and more dollars in entertainment. And as you have more dollars, you spend more in leisure, travel, entertainment, right? And that's what has been happening. So as we develop these tools, we increase our productivity, which leaves us to do like more fun things as a society. So one of the things that's interesting to me is if you go back to 1870 or so, about 75 percent of jobs were characterized as kind of not fun, not exciting, pretty tough life. You know, it's farming and that type of stuff. Uh, by let's call it 1970 or so we're down about 25%. So pretty big reduction in, uh, jobs that

were, uh, uh, not necessarily aspirational. They were tough life type jobs. Now, if you fast forward to today, you already mentioned a couple of these jobs that they're getting automated away, or it's easy to see truck driving as one, there'd be many others. How many of the jobs that we see in today's society are not going to be done by humans a hundred years from now. Like when I think of Tesla's manufacturing facility, it doesn't look like, uh, you know, kind of, uh, um, a manufacturing line where people are putting on the car doors and stuff like, sure, there are humans there, but it sure looks like there's a hell of a lot of robots in there. If you look in Amazon's warehouse, there's robots moving things around and there's tape on the floor and you start to

like okay but then also if you go and you look at white collar jobs like it seems like software is coming for that as well and so it doesn't feel like there's really that many areas that are safe from kind of the automation and and uh artificial intelligence that's coming is that fair yeah and they never been safe um so i think a hundred years from now everyone is gonna be a tiktok dancer i'm joking but like we're gonna change the way we work and we always change the way we work right so uh in the northeast of the u.s there a lot of people used to work in uh manufacturing right and now they work in service jobs so like work-wise we've been changing a lot throughout history but if you actually look at productivity we've been changing less and less

over time and that's one of the things that Peter Thiel talks a lot about right so one of the things that has been happening in the U.S. is that we didn't actually increase productivity so much but we outsource jobs so if you look at Detroit right Detroit was the manufacturing place here in the U.S., and we actually outsource that to other parts of the world, so like to China and Mexico. So a lot of the jobs went away here in the U.S. because of globalization and not because of automation. Automation is actually not changing that much, the economic pie of the U.S.,

but it will change. A lot of the jobs will be displaced, but that's the way it works, right? A hundred years ago, most people would be farmers maybe, you know, maybe 200 years ago. And most people are not farmers today. I think it's a good thing. I think it's a good thing that people, you know, are creative. They are like spending their dollars in entertainment and leisure. And I think we're going to be in a much better place a hundred years from now.

This is one of the arguments of, I think, techno optimists, if you will, right? It's like, hey, we freed everyone up from the farm fields so they could go do other things. It's not like jobs disappeared and now everyone is jobless. Just new types of jobs showed up, and obviously we still have food for now in terms of being able to feed the global population. Yes, there are areas that struggle, but we still get farm productivity even though maybe we don't have as big of a portion of the population working in the fields. Now, with that said, one of the areas that everyone has always thought would be safe was the creative jobs. so for example uh what is the hardest thing for a computer to do everyone was like oh like you could be an artist you could be a musician you could do this stuff like that's going to be really

really hard for the computers to ever compete with well fast forward to today like not so fast uh the machines sure as hell look like pretty good artists right with all of this kind of computer generated art or generative art like did we get it wrong as a as like human commenters on the state of automation where like maybe some of those creative jobs aren't nearly as protected as we thought yeah so there's this interesting quote that says like a human asks a robot can a robot write a symphony can a robot turn a canvas into a masterpiece and the the answer was no but can you right you're not mozart you're not picasso but now yes the robots can can write a symphony they can create a masterpiece um there was recent recently there was a state fair where a human created a

piece of ai art using one of the systems stable diffusion and actually won the challenge right so ai was able to create a masterpiece right now that brought controversy yes exactly but it like it's just increasing productivity right like instead of humans creating every thumbnail of a youtube like video you will have a computer doing that but humans still play a role because humans will choose which thumbnail to put into an image right um and but if we go back to the art for a second because i think it's an important uh uh conversation that's happening right now is this artists used ai to create a generative uh piece of art it could not be told if you looked at you couldn't tell the difference between oh that was ai and that was a human that created it and so

went into a competition it won it was selected as the best piece of art the artist community absolutely freaked out they were like what are you talking about this is cheating right it's the equivalent of taking steroids or performance enhancing drugs right for athletes like this was the equivalent in the art world uh and they were going absolutely crazy on the same side the artist uh the uh engineer i'll call them uh to separate from the artist um and others were like well hold on a second you're using tools to create your art uh you are using uh paint brushes or whatever other tools you're using like we're using tools also it just so happens that our tools are superior or yours and so uh what's the big deal that debate is not settled yet and it feels like we go one of

two paths either people say like hey art is art and like whoever can enter it into competitions can do that uh or the second path is like somehow there's a bifurcation either the generative art is not allowed to be entered or there's like a generative art division and then there's like a human art division or something like that like how does this play out yeah so if you allow um um if you allow people to take hormones and steroids they will take it right so so like if you look at people uh driving uh bicycles right like riding bicycles if they can take steroids they will take it and they will do whatever it takes to take steroids right the same will probably be true for the creative creative type right machines will help and make

humans so much more productive that everyone will try to use these tools right and one interesting thing that i think one interesting trend that i think we'll see in the next 10 years is that right now most of like the internet allow distribution to go down to zero and if you look at tiktok instagram youtube a lot of the engine that drives recommendation is done by machine learning is done by computers that select the next image video that you are gonna see right and they're much better than humans at selecting you know what you would like to see when when you do that though can the machine pick what will win or is it the machine says here's 10 options and the human says that's the one that's gonna win both can work right so humans still can choose

whatever you know like you can still have humans handpick but machines will probably be better at at some point yeah maybe even today yeah so like if you look at youtube's uh recommendation algorithm or even if you look at tiktok's recommendation algorithm is really really really good at selecting what video people like to see and and i would argue it's better than humans at selecting videos you know that people would like to agree and we're seeing recommendation algorithms today right but we're gonna see like generated art and videos 10 years from now so most of the media that we consume today is generated by humans right movies images and whatnot i imagine 10 years from now most of the video most of the media that we consume is going to be created by

computers you still have a human determining look i have this script please generate that right but a lot of the media that will consume will be generated by computers i went down a rabbit hole uh i don't know six to eight weeks ago uh and there's a a gentleman uh who i found online um and uh not in a weird way but just like on twitter and he was tweeting about uh generative art and all this stuff and uh he had created a sci-fi um uh mini series essentially called salt s-a-l-t and in this uh sci-fi series my understanding is that he was able to generate not just the imagery but also the video using uh kind of these generative machines the ai whatever he provided his own voice so he did the voiceover but what he was building was this idea

of a pick your own adventure movie yeah and so his idea was like okay normally for us to go ahead sit down uh and create an entire set get the actors and do all this stuff tens of thousands hundreds of thousands millions of dollars in budget to be able to make a movie an actual real movie but if i can sit here with my computer on a saturday essentially and do something that looks pretty good you know close to it in quality uh that's gonna have a profound impact now what he then said was if you drop the cost of creating this imagery and and kind of this movie you then can replicate many different scenarios so rather than watch top gun and every single person who watches top gun gets the same outcome what happens when all of a sudden now you go left and pick

option one, I pick option two, and our friend picks option three. And you have these kind of multi kind of paths to choose. Like that's a whole world that it seems like the content, you know, system or industry is not set up to do. But you can also see some creative people saying, wait a minute, there's a lot of things we can do now that previously weren't possible because of this technology. Yeah, absolutely. And so if you actually look at games, games are bigger in size than movies. They're much, much, much bigger. And the reason why is because people can interact and they can decide what to do next. And generative art, generative media is going to be more important to games than movies, right? Because you can create these characters that have

their own lives. You can create all these different scenarios. All that long tail that we talked about before can be created for games and games will be much more interesting in the future because of AI. Yeah. There's another startup called Osmosis Studio. And what they do is they use the same generative machines and they can do product photo shoots without actually having to do the actual photo shoot. So a lot of people don't know, but these companies will spend tens of thousands, sometimes hundreds of thousands of dollars to literally get two or three models to show up. They bring their equipment, they do the whole setup. They have to hire, uh, the, the photographers or, uh, um, uh, the videographers or whoever, and they go ahead and they do the

photo shoot. And that sounds insane because it takes a whole day or whatever. And it costs, you know, a ton of money. Now, what they're basically saying is you could just sit there and say like, I need, uh, an African-American woman, uh, with a, uh, Brooklyn street background, uh, sitting on a bicycle in the forefront holding, you know, product X. And then it takes a second and bam, there you go. There's the image. And now you can go use that as a product placement imagery on your website or, or whatever, but you didn't have to go find the location, hire the photographer, get the model, bring your product, all this stuff. And when I saw it, I was like, holy shit, like these machines are going to decimate industries.

Yeah, they will create new companies that will be very, very efficient, right? And like the next Pixar, the next Epic Games will be created by machine learning algorithms, right? You think so? Yes, 100%. Like Pixar right now is not a very efficient company. And you have very creative people that have amazing ideas, but it's really hard for them to translate these ideas to an actual product, right?

With these algorithms, you can actually create a script. You can have your idea. You can actually go from idea to product much faster. And so if you look at it, you're just increasing productivity. If you actually look at different industries, You're going from institutions to individuals, right? And that's the same thing that's going to happen to games. That's the same thing that's going to happen to movies.

And if you look at media, Joe Rogan, Russell Brand, they're bigger than CNN, Fox News. Investors, you have individual investors, they're bigger than VC companies. So you have Lad Q, you have Josh Buckley, they're bigger than Benchmark, for example. If you have e-commerce, you have Kylie Jenner, right? You have Tom Brady, you have E or like Kanye West. They're bigger, you know, than, than other brands, right? The same thing is going to happen to games. The same thing is going to happen to movies, right?

So you think the individual creators will actually end up being bigger. And a lot of it is because they're going to have these tools that will give them way more leverage to create this stuff than even a Pixar or whoever. Exactly, exactly right. And you see that in technology overall, right? If you look at the best companies, they didn't have that many people. So if you look at WhatsApp, Instagram, they had 20 people or 14 people when they had significant scale, right? You see unicorns with 10, 20 people. I think 10 years from now, it could be the case that you see unicorns with one person right so you think that we will potentially we will have billion dollar companies created with only one employee absolutely that is

wild i mean like if you look at crypto right yep bitcoin it was a i mean you had a set of people but the initial idea was created by one individual probably satoshi nakamoto who i think was Elon Musk but yeah you think Elon Musk is Satoshi yeah no come on yeah yeah uh wait hold on why do you think that Elon Musk is Satoshi uh there are a bunch of clues why he's Satoshi Nakamoto so for example the code base uh was like is written in c++ which is this programming language and It's basically a block. And that's usually how Elon programs his stuff. Also, the way he used to type his sentences in this forum,

there were like double spaces between one sentence and another. That's how Elon Musk writes. The way he talks, Satoshi Nakamoto writes in the forum. He talks about porn guys in the forum. and then you have pedo guys, you know. There are a bunch of clues that point out to the fact that, you know, Elon Musk, Satoshi Nakamoto, he has been thinking about money for, you know, a long, long time.

PayPal, the original idea was to create like money to the world, to replace, you know, these banks. So hold on a second. This episode is brought to you by Athletic Greens. I take AG1 by Athletic Greens literally every day. I gave it a try a couple of years ago, and now I like it so much that I even became an investor in the business. It is the single best option for easy optimal nutrition out there.

Every time you take one scoop of AG1, you're absorbing 75 high quality vitamins, minerals, whole food source, superfoods, probiotics, and adaptogens. Athletic Greens will give you one free year of supply for immune supporting vitamin D and five free travel packs with your first purchase. Go to athleticgreens.com slash pump to give AG1 a try today. athleticgreens.com slash pomp i take ag1 every day so you should go try it today this episode is brought to you by alto ira they can help you invest in bitcoin and crypto in a tax advantage way that helps you preserve your hard-earned money alto's crypto ira lets you invest in bitcoin and

over 200 other different coins and tokens and it has all the same tax advantages of your traditional ira there's no setup or account fees and it's all you need to do invest in crypto tax-free let me repeat that again you can invest in bitcoin and cryptocurrencies tax-free so are you ready to take your investments to the next level diversify like the pros and trade without tax headaches open an alto crypto ira to invest in bitcoin and crypto tax-free go to alto ira.com slash pomp that's a l t o i r a dot com slash pomp start investing today i disagree with you that elon is satoshi but let's go down this path and say yeah uh if elon was satoshi yeah would they just put him in jail like would it just be like hey man we gotta stop this that's why he can't say

he's satoshi nakamoto right you think so yeah like for the longest time i thought okay elon musk is gonna be the richest man alive and then it's gonna be satoshi nakamoto right because people would have an absolute meltdown if it came out that elon was satoshi but both good meltdown and bad meltdown yeah there's people in the bitcoin community who would not be super ecstatic about it there's people outside the bitcoin community who would uh be ecstatic about like it would just be mass chaos i know uh and i i think he will never say that he's satoshi nakamoto i don't think he has the self-discipline from looking from the outside i think that somebody would say something stupid to him and he would just he's underrated i think people don't realize how

smart he is um one of the things like he he he talks a lot of shit in twitter because that drives engagement yes he knows that that drives engagement that's why he's so funny on twitter right like creating spacex like it's amazing he created rockets that lend themselves in their own fit right he's he's creating self-driving cars this guy is nuts this guy is completely nuts people like underrate him he's like he's steve jobs on steroids yep um anyway satoshi nakamoto has a really strong incentive to take all his coins because he's gonna be the richest man alive or woman or whatever. Elon Musk is already the richest man alive, right?

He doesn't think he wouldn't touch the coins because he doesn't need them. Yeah, exactly. If it was some random guy, right? He has a strong incentive to take out the coins. But if you're already the richest man alive, like taking out the coins, you're still going to be the richest man alive.

So it doesn't matter. All right. Let's put that to the side for a second. When you think of kind of what is actually happening, though, with a lot of these technologies, it is very obvious that we're going to drastically drive the cost of creating some of these products to zero or trend towards zero. It will also unleash immense disruption in some of these industries. what are the areas where you think that we could apply artificial intelligence machine learning even eventually agi most uh um kind of uh quickly but also have the largest impact on society or creating economic value yeah what's ripe for disruption everything but like if you had to

pick one of two industries what would you pick i mean the first one is going to be media like that's the first industry that's going to be displaced and right now we're seeing technology being displaced by remote, right? So we're seeing Silicon Valley basically dismantling in front of our eyes. And the same thing that's happening to San Francisco and Silicon Valley happened to Detroit because of globalization, right?

In the 50s, Detroit was the richest city in the world by income capta. And now it's the second poorest big city. So it's the first income per capita in the country. Yeah. And now it's the second worst. Yeah. Jesus.

Talk about volatility. I think the same thing's happening to Silicon Valley right now. There's a big migration because of remote work. And the same thing's going to happen to Hollywood, right? If you look at the biggest stars in media right now, they're not in LA, right? Mr. Beast. North Carolina. Yeah, exactly.

Logan Paul was in LA. He now moved to Puerto Rico. he's all over the place but the biggest stars in youtube and elsewhere they're moving away because they don't need to be in hollywood right i i was wondering the other day um could you name an actor or actress that comes from the traditional movie world that's under the age of 25 25 28 something like that i couldn't yeah now i've never been into hollywood and kind of entertainment scene and all that stuff uh but i know who mr beast is yeah i know who jake and logan paul are right i know who uh a number of the tiktok stars are right like all

of these other platforms i know who they are i know the people on twitter i know casey nest that right like i know all of these individuals who have built massive online followings I don't think I could name a, uh, a pure actor or actress. Now there's an argument that like many of the people who have starred in some of these movies that are on the younger side, uh, they built their career somewhere else and they cross over. So I think of, um, what is it, the movie, uh, a star is born right with a lady Gaga. I think it's a person in there. Um, like that's a musician that crosses over. And so when you start to see this, it's like either one you've got to be uh multifaceted and kind of in different industries or you actually

got to go to where the attention is yeah like the attention is not in the hollywood studio model yeah the smartest people figure out that they don't need to be in hollywood so you look at joe rogan you look at mr beast they're all outside of hollywood and they distribute through the internet right so that's that's a big trend that we're gonna see right like we're gonna see hollywood studios outside of hollywood the next disney is not gonna be created you know in hollywood it's gonna be created outside uh of hollywood but i still think the west is gonna win like i still think america is gonna be you know the biggest country in the world like a lot people talk about Asia, right? But if you look at the biggest companies, the most innovative

companies, they're still in the US. SpaceX, Tesla, you look at Apple, you look at, you know, Zoom, they are all in the US and they're going to still be in the West, but they're going to be distributed. As we talk about Hollywood and AI disrupting it in this generative media, you have a very interesting background. You were at Genesis Group at MIT. And I have to admit, I didn't know that much about Genesis Group, but I went and I was looking it up and reading a little bit. And then I opened a second tab and a third and a fourth. I think I have like 10 tabs open now in the browser, right? Like, holy shit, this is cool stuff.

So talk a little bit about what exactly is the body of work that Genesis Group at MIT is working on. Yeah. So the group was basically working in trying to simulate the human mind into a computer. A small task. And a lot of it involved studying how the human brain works. That's why I have a fascination about the human brain. And I talk a lot about it in my book, about how the human brain works. So we have to understand, if we're going to try to recreate human intelligence,

we have to understand the thing, the brain, where the intelligence lives, in order to then understand how we can recreate it. Exactly. And at the end of the day, I think with neural networks, we can do that, right? But neural networks, they are basically created based on our neurons. And that's a very, it's the most basic part of our brain, right? And the reason why I think with neural networks, we will be able to simulate brains is because,

like, that's the only thing that we need. I'm thinking, sorry. No, go ahead. so if you look at planes for example right we understand the physics of you know how how air flows and whatnot we don't need to simulate how exactly how a bird looks uh to fly and i think the same thing applies to computers we don't need to exactly know how each part of the brain interacts with each other it would be nice right but we understand the physics how each neuron works and a neural network is basically a network of how these simple parts of our brain interact

and because of that i think that we'll eventually you know get to agi and so genesis group is looking at how do we recreate the human brain with computers um study the brain makes sense then what's kind of the next step in in doing this or what is the body of work that they're focused on right now yeah so so they're basically trying to replicate the human brain and so you have like different parts of it you have attention you have you know uh uh how our motor system works how our new neuron uh sorry sorry our neocortex works and how they interact with each other so they're trying to simulate these different parts of the brain right and when i think of the brain i think of literally the brain but obviously my ability to see is using my brain plus my eyeballs

to synthesize what actually my eyes are seeing, how much of the work to recreate these software-based systems takes into account things that interact with the brain, like the eyes or even the ears or whatever, versus, no, we're just trying to understand how the brain itself works. And so once the information is put into the brain, that's really what we're focused on. Yeah, so one of the interesting parts about the vision

is that most of the data that's actually being transported is from our brain to our eyes. Like, I think 80% of, you can actually see the synapses and see how much of the data is sent from your eyes to our brain. And most of the data is actually sent from our brain to our eyes. That's crazy. Yeah, and we don't understand why is that, right? If you actually look at, for example, our gut bacteria,

the way our bacteria interact with our body can interfere how we feel that day. So we don't understand humans that well, right? And so a big part of what Genesis was trying to do is trying to understand why, for example, 80% of the data is sent from the brain to our eyes. We don't understand that, right? In my personal view,

you don't necessarily need to understand that to actually recreate systems that do better than humans at different tasks, but it would be nice to understand how our brain and body interact with the world. Yeah. And it feels like these machines are only going to get better over time. But what about the engineers behind the machines? So when I think of these neural nets or I think of any of the artificial intelligence, machine learning stuff, like let's put my knowledge at novice level, right?

I have to guess the engineers who are working on this stuff today 20 years ago knew nothing about it, right? There's some gap where we went from, hey, very few engineers know anything about this or are focused on this to now there's a bigger percentage. But the percentage probably is pretty small of the overall engineer population in the United States or globally. Do we eventually have every engineer as an AI engineer

because it's just another tool in their toolbox of computer science to be able to actually do the things they want to do? Exactly, yeah. So we create new tools and we spread these tools really fast. So for example, GPT-3 or stable diffusion, we create a neural network and everyone can use that neural network. You just need to create once, right? Stable diffusion, what they did is that they learned how DALI,

which is this computer software that takes text and creates images, they learn how they did it and they just replicate it. And they distribute this toolbox to everyone. So that's what's happening. How much does it cost to do that? Like to create DALI, stable diffusion, or even GPT-3? Like how much money is being spent to create a neural network? So it depends on the neural network.

So GPT-3 costs around $7 million, apparently. I don't know if that is a lot or a little compared to what I would have expected. Yeah. Like it's $7 million is a lot, but also to create something like GPT-3, you would think maybe they have spent tens of millions of dollars. Stable diffusion costs around $600,000.

Really? Yeah, so it's not that much. And I think the reason for that is because you have a set of GPUs, so for stable diffusion, I think they use 256 GPUs, and they need to train for a certain period of time. And the longer you train it, the better it gets. But people get tired after some points. So you usually train these networks for a month

or around a month before you actually see in the real world. And I think that creates a constraint. Also, these are research labs, right? OpenAI, it's probably generating a little bit of money because they're selling access to their API, but they're not Tesla. They're not Google or they're not Meta. So they can't spend a billion dollars in these neural networks.

It's funny, though, because we've been increasing exponentially the size of these neural networks, the size of, you know, these machines. And that trend is going to continue, right? So to create one of these neural nets, right? Let's say that I wanted to create stable diffusion. Yeah. I put $600,000 aside. Do I need one engineer?

Do I need 50 engineers? Like how big is the team to build something like that? And then what are the steps of what they're actually going to do? Yeah, I've done that before. Not stable diffusion, but I've, I got text and created, like recreated Obama's voice with a neural network. So, so you took text from like a speech or something? Yeah, exactly.

So I got a bunch of, uh, of data from, you know, Obama's speech and then you have the text and then you have the audio. and then you have a lot of data. So you have hours and hours of text and audio and then you have this neural network that you're going to train on this data and you train for two weeks, three weeks, a month in a bunch of GPUs.

And how much does that cost? Let's say a month. Yeah, so it depends how many GPUs you use. So like eight GPUs for a month, maybe it's like 5K, 10K. Okay. So it depends on how much money you're willing to spend, right?

And then you had Obama's voice and you could basically have his voice say anything because of the quality. Yeah. It's not great. The, the, the, like the neural network that I trained, like didn't exactly, uh, recreate his voice because I didn't spend as much money, but you can get one person to do it. Right. Uh, and how long, uh, would it have taken for you to get it good?

Uh, it depends. If I use more GPUs for two weeks, maybe I would get it better, right? So let's say if you spent $50,000. Maybe $100K or $500K would get better. But these systems are getting better because the amount of compute that we can use is getting cheaper and cheaper. So every year, NVIDIA releases a new GPU that's more powerful. And the old GPUs get cheaper and cheaper, right? So what it costs, you know, today, a million dollars, five years from now is going to cost

one or $10. Right. That's crazy. Yeah. That is absolutely insane. Yeah. And so when you think of this, uh, let's say that you had successfully, uh, recreated Obama's voice. Now he's a little bit different for president and all stuff. But, uh, if you were able to recreate that, what are the things that you would do with it in this world that we're heading into? I mean, you can create any media, right? You can create any video, you can create any voice. And I think one of the important things that we will need to do is to cryptographically sign real media.

So whenever we record with our iPhones, your iPhone will probably need to verify that this is real media. Like real because, like it's not real necessarily because they also do much like- Authentic almost. Yeah, exactly. because they have also machine learning algorithms to actually improve the image, right?

Like everything has machine learning embedded in it, but they will probably cryptographically sign, verify that this is authentic because at some point we'll get machines that produce media that's completely indistinguishable from real media. And when you think about something like this, it feels like right now we have a decent understanding

of where we are from a technical standpoint, point, but the possibilities of how people will apply this, we're just the tip of the iceberg, right? It is still got quite a bit of runway to go. Um, how can engineers get to the position where they're able to actually train these neural nets? So let's say somebody is watching this and, and let's start with the person who already has some technical background. Are they taking like, uh, uh, uh, some sort of program online and they watch some YouTube videos and like, they're ready to rock and roll. Are they having to go to like some sort of class? How hard is it for them to go from,

I'm just kind of a quote-unquote regular engineer to, okay, now I'm armed with the weapons of information warfare in the 21st century? Read my book. A lot of the information is available online. So you can actually find a lot of the information online. So for example, Andrej Kaparti was head of Tesla AI, for example he's releasing videos of how to train some different neural networks right

so you can find a lot of that information online like everything right like if you want to be the best at something you can probably find everything online right yeah and what about people who have no uh technical expertise whatsoever i know that you're i think an investor in replit yeah uh i am as well. And they've blown me away with taking nothing to, you have some technical knowledge, but is that the path that most non-technical people are going or are they doing other things? Yeah. Um, so I, I think for most people, um, you're not necessarily going to train these neural networks for computer scientists. I think most people will also not train these neural networks because you just create once and then you replicate, right? I don't need to

recreate stable diffusion to use it, right? And these neural networks are also helping computer scientists write better software. So OpenAI, with GitHub, they release a tool that actually completes your code. So you can type it out, oh, like, you know, can you write this code for me? The machine learning algorithm will actually write. It's not perfect. So it's still, it's like assisting software developers, right?

So these machines will help software developers be more productive. How many companies outside of like Tesla are using some sort of AI or machine learning? Is it every company? Every company, every tech company needs to use AI nowadays. So if you think about Apple, right? They use machine learning to improve your images.

If you think about Spotify, they recommend music for you. If you think about Facebook, Newsfeed has AI. If you think about Opendoor, they're predicting the price to buy your products. Every tech company is becoming an AI company. So talk about Opendoor as a good example. I think people understand the idea of recommending music or the news feed and the ranking algorithm. But something like Opendoor, which in a very general sense for those that don't know, you want to sell your house. You don't want to wait to find the buyer.

And so you can go to Opendoor and I think it's within 24, 72 hours, whatever. They will give you a price that they're willing to buy now. And I think they do it in cash and basically you're done, move on, they'll hold the house and they'll go try to find a buyer at some point in the future. But where is some of this more advanced technology coming into play? So to determine the right price, they are using AI machine learning algorithm, right? So you take a bunch of images, you enter all the information that Opendoor needs to determine the right price. So they basically get a lot of information of different houses and their features and their price.

And based on all that data, they can actually determine the price of a house. So if I ask Keith Raboy how accurate is it, he'll say it's 100% accurate, which is why I love him. But how do you think about like accuracy and especially in a complex market where housing, the price today is not the price three weeks ago or three months ago versus three years ago.

Like it does move pretty quickly. So how do you think about accuracy with something like that? Yeah, I think housing doesn't move as quickly as people think. If you actually look at the 2008 crisis, right, I think housing was moving at 2% every month, down 2%. Housing is going down over time right now because interest rates are going up, right? So I think housing doesn't move as much.

I don't know how accurate you need to be because Opendoor is giving you cash right now, right? So you don't necessarily need to give the exact offer. You may give an offer that's a bit lower, right? Because if you're willing to sell your house really fast, you might sell for a cheaper price than you're selling if it took you a year or like a few months, right? So it doesn't need to be exactly right,

but it needs to be within some reasonable bounds. Yeah. It's fascinating to understand how pervasive this technology has gotten and how good it gets, especially not just in a general, oh, AI is getting better. But when one company takes the technology, applies it, and then they begin to compound that knowledge and that experience, they're getting the feedback of, we bought this house for X dollars. We were able to sell it for Y dollars. Oh, that's a certain margin. Maybe we can actually offer lower later. Oh, no, people won't actually sell at that price. Let's bring it back up. and you just begin to get all these data points to the point where it just becomes a really well-oiled machine. Yeah, and this feedback loop is really, really important, right? So you have companies that focus on transcription.

There are companies that focus on media. There are companies that focus on real estate. There are companies that focus music, right? So because they're getting that feedback loop, they get better and better over time. And there's like a feedback loop And there's a will that, you know, they are getting better and better. And that's why I think Tesla, for example, is going to win with self-driving cars.

There will be other companies that are going to do really well, but they already have a million cars out there. So it's going to take some time for other companies to eventually catch up to Tesla. I'm not the smartest guy when it comes to self-driving cars and the competition between the individual companies. But I do feel like every smart person I know is like, yeah, Tesla is miles ahead than everyone else, which I'll take their word for it. Right. Another company that I hear AI talked about all the time, usually in a negative sense, but still talked about is TikTok. Yeah. What is TikTok doing with artificial intelligence?

Yeah. I mean, they're just predicting the next video. Okay. So similar to Facebook or Spotify or whatever. Exactly. I think TikTok has relevance, but I think people overplay how much influence they have, right? At the end of the day, people used to say, why is the TV showing so much dumb stuff, right? It's because it's dumber.

It's because people want to see that, right? If you don't show what people want to see, people just switch off, right? They transition out of a media channel. And I think TikTok is the new MTV, is the new media channel. Instagram is the same, right? And is TikTok better? I'm going to put better in air quotes because everyone's like, oh, it's so good at recommending. It's addictive.

It's all these things. Is that because they're actually collecting more data? And so more data gives them more accuracy, whereas if you were to take all of the information that Instagram collects and then all the information that TikTok collects, my understanding, again, from two hours of Googling around a few months ago, is that TikTok collects way more information. And then if you're using these kind of AI neural nets, you should end up getting a more accurate predictive engine than if you had less data. Yeah, I think Facebook or Instagram is collecting as much data that is needed compared to TikTok. It's just TikTok was first using this algorithmic feed compared to your social feed. I think a year from now, two years from now, Instagram will have an algorithmic feed that is as good or better than TikTok, right? so in some way i may go on instagram and not see a ton of content from my friends and family or

people i've chosen to follow i could just see like viral videos yeah and people want to see that it's funny uh because your your stated preference is not your reveal preference right you may say oh i don't like that like it was funny i think king kardashian or uh chris jenner was saying i don't like i just want to see my friends i don't want to see these instagram influencers but she's an influencer herself. She wants everyone to see her content. Exactly. People want to see algorithmic feeds.

That's their review preference, even though they say they want to see their friends, right? Some people still want to see their friends, but most people want to see these algorithmic feeds. But in terms of data, I think the company that collects the most data is Google, right? When you're talking about TikTok, You just told me, I Googled this.

You're using your Chrome browser probably to search for things. If you actually download the data that you have on Google compared to Facebook or anywhere else, Google probably has the most data compared to any other company. They can use that data differently, right? But in terms of comparing how much data each one collects, Google collects the most. Yeah. When you think of non-tech industries, right? And tech has kind of permeated into a lot of these industries now.

So maybe we go to more kind of blue collar type jobs or companies. Are there obvious areas where somebody will build artificial intelligence companies to help the non-tech industries or companies? Yeah, and it's happening already, you know. Every industry is being helped by AI. So if you think about real estate, you have OpenDoor, right? If you think about lawyers, there are companies that are trying to predict how a case is going to do, right? Interesting.

Like which argument you could make to increase the odds that you would win. Exactly. And depending on the judge, right? So every industry, if you look at medicine, you have, you know, AI algorithms predicting if you have skin cancer or not. You have, you know, AI algorithms predicting how a protein is going to look based on the genes, right? And so like every industry is going to be helped by AI. And are there major risks to this? So we talked about the AGI risk of like, hey, one day the machines get too big.

But in the shorter term, are there risks that you've identified around the increased use of AI that maybe people aren't thinking about? Yeah, I think the biggest risk is around recommendation algorithms. And again, when we talk about censorship, right, if you look at social media, social media companies are the most powerful companies in the world. And that's why there's so much scrutiny around Facebook, because the most powerful company before that was the New York Times. The New York Times dictated what people thought, right? If you think about the WMDs in Iraq, right? It was a false story, but it determined the future of the US, right? And right now, social media plays that role, right?

So when people talk about, you know, misinformation or even these recommendation algorithms, they're talking about power. why there's so much scrutiny around Facebook is because there's a set of people that want to control what other people see. And so if you look at, you know, the New York Post story, when there was censorship around that story, a lot of people didn't see that story. And that may have determined, you know, the election results.

And so I think this, like, social media and recommendation algorithms are the most important part of AI that we see today. But I think in the long run, if a company decides what media you see and it's not what people want to see, people migrate. If you look at these markets, they evolve really fast.

People complain a lot about Twitter, but they still use Twitter. I think Twitter somehow is the digital manifestation of San Francisco. People like, you know, don't like San Francisco that much, but there was a network effect. People used to live there and people are moving away from there. I think the same things is happening with Twitter, right? You're creating these social media platforms outside of Twitter.

So I'm an investor in Firecaster, which, you know, is a decentralized social network and people are moving away from Twitter to other social media platforms. If you think about YouTube, there's Rumble, there are other social media platforms. If Facebook doesn't serve what people want, they will just move away. This episode is brought to you by Bullish.

They've reinvented the digital asset exchange. They give you access to DeFi features like automated market making and liquidity pools in a regulated environment. It's a whole new way to generate alpha. Bullish's total trading volumes have exceeded $25 billion just in the seven months since it launched. and the industry leading order depth means you can trade confidently when you want at scale with

better pricing and lower risk all within a regulated market environment good reason to be bullish learn more at bullish.com slash pomp and follow at bullish on twitter today this episode is brought to you by sigma the bridge between iGaming online sports betting and emerging technology such as blockchain nfts fintech gamefi metaverse and ai is loud and clear the largest global summit of its kind is heading to malta from november 15th through the 17th over a thousand exhibitors and 25 000 industry leaders will be there including top executives from draft kings bet 365 crypto exchanges betting software providers operators gaming affiliates and more log on to aibc.world and sigma.world to see our upcoming global summits see you in malta november

15th to 17th for the leading global event in gaming and emergent tech this is sigma go to sigma.world today it makes complete sense i think uh dan romero who's building farcaster uh has thought a lot about what does those networks look like what would be something that people would be willing to move away for one of the things i've always said is uh i used to get all these pictures all the time we're gonna build decentralized twitter i'm like okay what's the difference like it's twitter but decentralized and i was like nobody is going to care because nobody actually is going to move just for the decentralization now it's a better product if there's a certain conversation or a certain type of user that they want to engage with and they feel like it's a

superior experience in whatever way. Oh, and it happens to be decentralized. Sure. Maybe they'll be interested, but the idea that we're just going to take every single thing and copy it and then decentralize it, it feels like that is elementary thinking and not really where the world's going to go. Exactly. You need better products. And that is the same insight that Elon Musk had with electric cars, right? I'm going to build electric cars. Nobody cares. Nobody wants to drive a Prius. I mean, some people want, but yeah. But Teslas are the best cars in the market and they happen to be EVs, right?

And that's why people are buying them. And that's the same thing for, you know, decentralized networks. You can't only just create a decentralized network. Nobody cares. Nobody cares. You need to create better products. And that's why people choose one product over another.

But it takes time, right? Like Twitter, there is a network effect there. but you know once you start moving away from from there uh you eventually lose that network effect yeah that makes uh that makes a lot of sense touch me about bitcoin and uh and cryptocurrencies in general like how do you view them fitting into this whole world yeah i i think bitcoin is the biggest invention uh after the internet i think will have a bigger impact than the internet so there's a lot of controversy around social media because all these old companies they lost power right like the new york times and all these other networks like fox news cnn and whatnot they

lost power and now the individuals individuals have power and so like they have more information and whatnot and all these old companies that held so much power for so long they're screaming like please stop stop stop but you can't stop tech you can't stop evolution right the same thing's happening with bitcoin and all these cryptocurrencies you have better money and and and they're private why why was there so much fight over who is going to be the next president because they have the control of the printing press they have the control of the central bank So they can actually print so much money and give to other, like, different states, right? The U.S. has so much power because of the dollar.

And the U.S. is losing power because, you know, other countries are moving away from the dollar. And that's what's going to happen with Bitcoin and other cryptocurrencies. You're moving from, you know, the state controlling money to private, like, having private money. And that changes everything. one of the biggest things that people don't talk too much about is decentralized finance right when you look at interest rates in the u.s they're really really low and that's why i think there's so much craziness in big cities right why why is real estate so expensive in in big cities is because interest rates are really really low and people can leverage up so so so much and then real

state becomes really expensive and then you have homelessness you have people like not being able to have their own families and create their own families in these big cities and that has been happening in the u.s and you have some cities that have this problem uh at the peak right and the whole west is seeing that so like if you look at the fiat system the dollar and euro like european cities are starting to have the same problem right when you have interest rates they're not driven by you know the state so right now what happens you know is that this the government issues that at zero percent interest rate now they're increasing a little bit but they can't increase too much because you know you have the government and even private companies over leverage over

leverage right and when you increase interest rates like they can't pay you know that much debt and they can't borrow as much right um and with zero percent interest rates here in the u.s americans have a huge advantage compared to developing countries so if you go to brazil where i'm from interest rates are much much higher so the cost of capital in brazil is much higher but what's the difference between the average brazilian and the average american It's not that much, right? Like they're pretty similar. And so why are interest rates in Brazil so much higher?

It's because historically the government in Brazil is pretty shitty, right? Compared to the government of the U.S. Historically, sometimes the U.S. goes back and forth, but definitely I think most people say it's more stable. It's more stable, exactly. And so that's why interest rates are so low here. But is that also a thing where when inflation hits globally, US but also everywhere else

Brazilians know what's coming Brazilians understand inflation And they're like oh shit Jack up the interest rate To try to get ahead of this thing Whereas in the US maybe we're a little bit behind the curve Or is that something that Americans are saying Without really understanding the difference between the two countries

Yeah I mean some countries Understand interest rates better than others Brazil is doing much better Than the US right now Economically and also In terms of inflation There's less inflation in Brazil right now than in the US, which is crazy to think about, but it's mismanagement.

But yeah, like Brazilians understand inflation much better, but they also understand interest rates much better. When you have such high interest rates, your cost of money is much, much higher. And I think that's much more important than inflation. Inflation plays a role. But if you can borrow money for much cheaper, why wouldn't you? Right. And that's what crypto enables. Right. When you have Bitcoin, when you have other cryptocurrencies, you can borrow and lend money in a different like the interest rates are much, much different than, you know, your state owned interest rates. Right. and that will drive a lot of development and I think that will change

the world like much more you know like when you have Bitcoin and you can borrow money at much cheaper in Brazil that will have profound effects and so you will have areas in Brazil that will be much much richer

there will be areas that won't be as rich right but the individual will have the choice of having you know different who choose their own money Yeah, that makes sense. How do you see artificial intelligence and Bitcoin playing with each other? Are there things that people will use AI around Bitcoin that are obvious right now? So I think Naval had this quote that was pretty interesting. I think every person is going to be an investor in the future. You're going to have to be.

exactly exactly they will have to be and so when i think about cryptocurrencies in general i think you're you're investing in some technology right and and and cryptocurrencies they enable global investment right in the past to invest in in the u.s market you usually had to be u.s based right to invest in apple to invest in microsoft you had to be u.s based most people that i know in brazil they they didn't invest in microsoft they didn't invest in apple but a lot of people here in the u.s they they invested in these tech companies but right now when we talk about cryptocurrencies when i go back home everyone is talking about it they are putting money into it it's a global uh it's a global movement right ai just increases productivity uh and so uh when you

look at it ai is just gonna transition this world where everyone becomes an investor yeah it's fascinating when you think of on a global scale and then also i guess the only way that i could really see right now where it's like already people are trying this or at least i've seen uh with bitcoin specifically is you have a little bit of artificial intelligence uh that's being used by the energy companies and the miners of when to actually consume the power for bitcoin versus go ahead and return the uh power back to the grid and then also i've seen some people talking about like routing liquidity on the lightning network and trying to understand what the right path is what's the most cost effective that type of stuff um you've got maybe some

trading stuff that people are doing i'm less familiar or intellectually interested in that stuff but uh definitely people looking at tons and tons of data trying to understand what are the different um kind of things that they can pull out a signal and and go from there as well right yeah so it's just increasing productivity when you think about it if it's mining if it's routing money right ai usually you have a heuristic you have a goal and you're trying to improve that yeah and that's what is doing right uh so if you look at different industries it's always helping improve productivity yeah that makes sense talk to me about uh the book that you wrote so this book is fascinating to me because uh it really unpacks a lot of what i think is opaque

to people so to where'd you get the idea to write a book and what was that process like to write this? Yeah. So I think AI is impacting every industry, right? So if you're taking a photo with your iPhone, it's using a machine learning algorithm to like predict what's the best image, right? Or when you listen to a song, it's trying to predict what's the next song. When you're watching a video on YouTube, it's predicting the next video to watch. It's everywhere. It's impacting every industry from real estate to driving cars, to retail, everything. And most people don't understand it right so i explain exactly how it works and the whole history of it right so if you want to understand how ai impacts our life i i recommend this book right um and the

book is called making things think and it's how ai and deep learning power the products that we use and so uh um for the parts that i've read so far it is essentially looking at things that are already in existence right it's it's not a pie in the sky hey one day we're gonna have uh the singularity and there's gonna be a chip in our brain and and kind of all that stuff as much as it's like here is actually understanding the the uh reality of the technology today and where you are interacting with it in your life exactly yeah if you want to you know think about what's gonna happen in the future i give you some hints where we're heading right but i think it's the most important aspect is to know how is it impacting your life,

how is it impacting you today, right? It's going to impact your industry. It's going to impact your life. And you probably should know how it works, right? Yeah, it's pretty interesting to kind of contemplate all of the different ways that the technology will actually be integrated into this. Because we don't understand it all yet, right?

Like there's somebody right now – I think a lot about this. There's somebody somewhere in the world right now who is using this technology and they are tinkering and they're trying to explore different ways to actually implement it in a certain industry or in a certain application or a product. And eventually they are going to come up with a breakthrough that then will spread like wildfire across the entire world and we will all be like, wow, that was amazing. I can't believe no one ever thought about that or had ever built using that technology in that way. Yeah, I think that's great. could you imagine that Bitcoin was created, you know, like before that you wouldn't even imagine about this creation, the iPhone, you know, like there will be new creations that we can't think about it that we'll have 10 years from now that will impact our lives like a ton. Right.

And that's exciting. That's the exciting part of our lives. Yeah, it is. It is absolutely uh, uh, uh, exciting, but also, uh, I think something that reminds you, like there's still a lot of work to do. Yeah. And, uh, if you think about artificial intelligence and its ability to go from where we are to where we want to go, uh, we're going to need a hell of a lot of engineers, right? Um, we're going to need those engineers trained in the latest technology. Um, and it brings kind of the, the conversation full circle, which is, uh, in some way, uh, we are trying to build technology that will put humans out of today's jobs to free them up to do other more productive things and allow the machines to take over some of the repetitive tasks and things that

humans do out of necessity today, but maybe in the future won't have to. Yeah. I think there are two industries that you can have a lot of leverage and one of them is media and another one is you know softer you write it once and then you distribute uh to everyone um and and they impact everyone's lives right like everyone has uh you know facebook or instagram installed in their in their phones everyone has an iphone or android you know they have a smartphone uh and that impacts everyone's lives uh it's great because you're you're basically you know impacting the whole society and it's usually, you know, uh, to the good. Yeah. It is, um, uh, uh, something that fascinates me because it also requires people to be optimistic about the

future. Uh, it is very hard to build these technologies and build these products and be a pessimist, right? Yeah. Yeah. I mean, if you look a hundred years ago and how our lives were shitty right like you you can't like it's hard to not be optimistic about the future right like they see the iphone you know like how our cars evolved if you look at around you know like just look at our cars the cars that how they look in the 70s and how they look now they're so much better so much safer right uh medicine keeps improving like everything keeps improving over time in average right yeah you have some like mismanagement in some areas but overall society improves over time yeah it is um uh it's fascinating whenever i see

uh they say machines can identify like x-rays like problems on x-rays better than uh than human doctors and stuff like just things that you would never have expected uh for machines to be able to do yeah it's great where can we send people to find the uh the book yeah you can go to hallway.com uh the name of the books making things think hallway is a platform where they have different books they have really great books um you can go there and then just yeah buy the digital version of the book awesome so making things think how ai and deep learning power the products that we use go to Holloway, the website, to get that. And then where can people find you online

if they want to talk more about these ideas or learn more? Yeah, you can find me on Twitter. I'm also on Farcaster, which is this decentralized social network, and on YouTube as well. So if you search for Giuliano Giacaglia, which is probably going to be in the show notes, you can find me there.

Awesome. Well, listen, thank you so much for your time today. I learned a lot, and I think that people are very interested in artificial intelligence, machine learning, deep learning, all those things. And so hopefully this will be helpful to them as they try to learn more. Thank you so much for thanking me. Yeah.

Thank you. Thanks so much for listening to today's episode. I really hope you enjoyed this one. Make sure you're subscribed on Apple, Spotify, or your favorite podcast player. And if you're looking to transition into a brand new job in the Bitcoin or crypto industry, we've got you covered. Head over to thecryptoacademy.io. My team and I have been working with the top HR teams in the industry to develop an intensive

three-week training program with over 50 live events. We teach you exactly what you need to know to break into the industry, including live interview prep and resume review. Our students have been hired at over 75 of the world's best Bitcoin and crypto companies. Go to thecryptoacademy.io to learn more. Again, that's thecryptoacademy.io.

If you enjoyed today's episode, make sure you share it with your friends and I'll see you all for the next episode.