Speaker 1: Alright, so DeepSeek R1 just came out, and it looks like it has pretty much entirely caught up to OpenAI's O1 at least. So, kind of one of the mantras that we've been saying out on the internet is, there is no moat. So, what do I mean when I say there is no moat, and what does that mean for us? So, no moat in artificial intelligence basically means that there is no barrier of entry, or there is very little barrier of entry. So, for instance, intellectual property or IP law doesn't really seem to be slowing any other companies down. Microsoft, Google, Meta, Anthropic, DeepSeek, all these other companies are catching up to OpenAI, and that means that there is no real secret sauce. There is no barrier to replication of what it is that they're working on. There are capital requirements. So, for instance, it does take money, it does take data centers and compute. But as we've been seeing with the Chinese companies, is they're often duplicating what OpenAI and others are doing, and they're doing it much more cheaply. And part of the reason is just because they don't have the resources that we have, and by we I mean Americans. They don't have as many data centers, they don't have as much compute, and so that is actually forcing them to actually be even more clever and even more creative. And we actually have a historical precedent for this. The Soviet Union often had to duplicate what America was doing on a shoestring budget, which forced them to be even more clever. Now, ultimately, it still didn't help because they just did not have the capacity to keep up with us. I think the same will be true of China. But anyways, my point is that the capital requirements are not necessarily an insurmountable barrier considering how many startups are getting funded. We haven't even seen anything come out of Ilya's SSI, Inc., but he got a billion dollars of funding or something like that. And then the technical know-how. You might sometimes think like, oh, well, only 10 people in the world know how this works. That has demonstrably been proven false. We have people around the world and across many companies, even here in America and Europe, that clearly have the technical know-how to advance this down the field. And so, yeah, you end up with, basically, the long story short is there's no barrier of entry. There's a little bit of friction, but there's nothing intrinsically keeping it from going. Also, if you hear my dogs trotting around, we got two new dogs, so I apologize. There's nothing I can do about that. Okay. Now, with that being said, you might say, okay, well, Dave, if there is no moat, then why is OpenAI still out in the lead? And that is because of first-movers' advantage. So first-movers' advantage basically is they're the first out of the gate. ChatGPT was the first shot around the world back at the end of 2022. Here we are a full two years later. No, two years later. What is time? Because we had 2023. That was like the year that everything took off, and 2024 was last year. Okay. So we're just over two years after that initial shot, and even just two years after that initial shot, a lot of people have caught up. So that's why OpenAI is experiencing a diminishing lead. So initially their lead was 12 to 24 months, but now it's down to one to two months before other shops replicate whatever it is that they have achieved. We have seen this with Google, which is about to release their reasoning model, and so on and so forth, and, of course, the Chinese DeepSeek. Now, one thing that I will say is a big, huge caveat, a giant asterisk around anything that the Chinese do. It's entirely likely that what they have done is literally just stolen. This is one of the primary concerns that I've heard echoed again and again, which is that the Chinese just exfiltrate all the data and the research. Why do it yourself when you can just steal it from someone else? Now, however, that being said is that this lead has allowed OpenAI to maintain commercial dominance just because they were there and they were doing stuff for years before anyone else was. So just by virtue of the fact that people got used to OpenAI means that OpenAI is still the market leader, even if other companies are catching up and overtaking them in terms of cost, quality, those sorts of things. Now, it's still hotly debated which model is best. A lot of it, honestly, when people argue about which model is best, honestly seems to come down to personal preference and taste, which is really interesting. The fact that there is room for nuance where it's like it's not just one model to rule them all. It's I like this model because it's better for this or I like this other product because it's better for that. That actually shows a huge amount of market opportunity for diversification, which is great. It's just like there's not just one car like you can have a car. You can have a hatchback. You can have a pickup truck. You can have a van. There's all kinds of form factors that are ideal, which means that there's plenty of room for market diversification. Also, I apologize. It's really dry. The winter is making me itchy. I apologize for all the interruptions. I'm getting back used to this. Anyways, so the existing user base of OpenAI means that there's lots of people that are just going to keep using it just because it's what they're used to. They say, this is what I got used to. This is what I grew up with. I'm going to keep using this. The market dynamics, though, mean that even though OpenAI has first mover advantage, it's not going to last forever. Now, the other thing is open source and market competition. One of the longer-term impacts of the fact that there is no barrier of entry means that you might be able to use models completely fungibly. A fungible model means that you can use them interchangeably. You can use an open source model. You can use Llama. You can use DeepSeek. You can use Claude. You can use OpenAI. From a business perspective, you don't care. You just want it to work. If it's completely free and open source, great. If you have to pay another service provider, great. You really don't care. In the long run, this is really, really, really bad for OpenAI. The reason is because they are what I call a one-trick pony. OpenAI offers literally one product. They offer one product, one service. They have zero diversification, which means that if some other company or even an open source model comes along and eats their lunch, every company out there that's paying OpenAI right now, they're going to make the rational choice like, well, this other Chinese company can do the same thing, but it's 50 times cheaper and 5 times faster. Or maybe even this open source model can do exactly what I need, and it's completely free. Now, what I will say is that OpenAI still has that first-mover's advantage, but that's literally their only moat is the cadence at which they have released. That's their only moat. So then you end up with this process of what's called market democratization, which means that there is no possibility for monopoly. The data is out there. The data is basically free. The research is basically free, which means the water's warm. Come on in. Anyone can join this, and that means that, again, no moat. This is really, really good. So there's two reasons that this is good for everyone. So number one is price dynamics. Hey, dogs. Number one is price dynamics. Competition drives down price. This is basically going to be like trying to do price control on nuts and bolts. And I mean the literal metal things because literally anyone can make nuts and bolts. You just need the right equipment. Likewise, already today, anyone can host models. You just need the right equipment. It's not a big deal. It's going to be commoditized very, very soon, and it's just going to be ubiquitous. It's going to be everywhere. This is good for consumers because it means that prices will be down and access will be ubiquitous. And also, if someone tries to jack up the price, guess what? You're just going to go to someone else because you just swap out models. You just say, you know what? You're too expensive. I'm going to go over here. So it's a very, very cutthroat market, which is good for us. And it's also good for innovation because that means in order to differentiate your product on the market, you need to do something better. You either need to be cheaper, you need to be faster, or you need to be smarter. Ideally, you're all three of those. But if you're not at least one or two of those, why is anyone going to use your product? It's just that simple. Now, what I want to talk about because I always do this is I want to tie this back to historical examples because while history doesn't exactly copy itself, it always rhymes. So first and foremost, I compare this to three previous technologies. Sorry. The printing press, radio, and internet. In all three of these cases, the printing press, once Johannes Gutenberg invented it, it started spreading very quickly. Within a decade, it had jumped borders. And within 100 years, it was literally all over Europe and America and starting to spread other places. Now, what's really interesting is the printing press was actually banned in places like the Ottoman Empire for three centuries. There's a reason that the Ottoman Empire no longer exists. Put it that way. So the lesson there is that any nation or company that resists these technologies will cease to exist eventually because eventually you get so far behind that it's like the inverse of the first mover's advantage. It's like if you get too far behind on this, you lose. And when I say you lose, you lose forever. You cease to exist. So that was the printing press. Radio, again, once the radio was figured out, it was too democratic. You can't put the genie back in the bottle there. And likewise, the internet itself is literally designed to be used by everyone. And so the printing press, the historical impacts of the printing press are very difficult to overstate because everything from the Enlightenment, which led to democracy, and the fall of the hegemony of the church, all of that can be traced back to the printing press. And the reason that that is is because it dramatically raised literacy rates and it dramatically democratized the marketplace of ideas. The internet has taken that to its maximal conclusion where everyone can share every idea, which includes flat earthers and anti-vaxxers and QAnon and all that fun stuff. Everyone has an opinion and everyone can share their opinion. And whatever opinions most resonate with you, you can adopt those opinions, whether or not they are supported by empirical evidence. So we have reached maximal democracy of information with the internet. And the radio also helped with that because radio is instant broadcast and long-distance communication. So the reason that I cluster these three together is because they are fundamentally information technologies. The printing press was the first form of really mass-produced information technology, followed by radio, which led to newspapers, then radio, and then internet. So these are basically three generations. And now we have the fourth generation information technology, which is you can compress literally all human knowledge into a single model. And it's not... Hey, be nice. Sorry. We just had the dogs. We've had them for a week and so they're still getting used to each other. They're very sweet, but sometimes they have miscommunications. Go on. Be nice. Anyways, so the fourth generation of information technology is artificial intelligence, which it's not just a repository of knowledge and facts. It is also a cognitive engine. It is also a problem solver. And yeah, so it's really difficult to oversell the magnitude of this new technology because it's not just the book. It's the book that is interactive. And that's a really oversimplified way of characterizing artificial intelligence. But anyways, all right, now taking a bigger step back, what are the global impacts of this democratic access to artificial intelligence? Number one is that every nation will have access. When you look at the fact that you can start running some of these models on MacBooks, even the poorest nations on the planet can afford a MacBook, which means that literally every nation on the planet can afford to run artificial intelligence. Now you might say, okay, well, is that good or bad? Because you don't necessarily want some chaotic rogue actor like North Korea to have access to AGI, which I would tend to agree, and we'll get to that. But on the other hand, this will reduce conflict because basically when there is no gradient, when there is no technological gradient, there's not really any advantage. Now I need to have a gigantic caveat to that because, again, in this world, the number of data centers you have is basically that's your military base. That's your economic base. And the United States has literally exactly 12 times the number of data centers as China. So in this geopolitical competition, America has already won. That's just sight unseen because even if China releases a better model, guess what, we can run more of them. And yeah, so that's a done deal. But in the longer term, the long view of humanity, the fact that this technology is intrinsically democratic, oh, and by the way, it actually benefits more from cooperation. The more you share your data, the better the AI gets. And so this creates a really interesting attractor state or Nash equilibrium where the Nash equilibrium is everyone pools their resources. It's like I'm not going to say socialist, but it's very collectivist in terms of what this technology affords us because the more data it has, the smarter it gets. The more you share computational resources, the smarter it gets. Now that's not saying that the maximally thing is just everything should be open source and free forever because there is still some competition. But I am not joking when I say in the long run, I suspect that artificial intelligence will be the unifying force that creates a unified global humanity. Now I'm not saying like a one world government that's going to control everyone forever. I do think eventually it would make sense to have a layer above federal governments where you have a global government. But I think it should be again federal where it's basically all the nation states are treated like states here in America where there is some level of autonomy. Anyways, yeah. It's really, really difficult to oversell the long term geopolitical, social, and technological impact of the fact that artificial intelligence seems to be intrinsically democratic. That is just how the math has played out. No monopolies and the fact that there are no monopolies applies also to nations as well as companies. So we're heading towards what I call cognitive hyperabundance. All right. So let me just put it into perspective. Right now, the global estimates are about 15 to 20 million doctorates. So that's PhDs, MDs, EdDs. If your title ends with doctorate, PhD, that's doctorate of physics, or doctorate of philosophy, or MD, medical doctorate, then there are only 15 to 20 million of you in the world. That's not enough. And it takes time to make new human PhDs. So if you want to cure cancer, guess what? You have to find people that are smart enough to help you cure cancer. If you want to get to Mars, you have to find people that are smart enough to help you build rockets. And by the way, one PhD is not fungible for another PhD. If you have a PhD in cell microbiology, then that is not going to help you to get to Mars unless you need biofuels or something. But as we develop and deploy these technologies, we're going to turn this up by a factor of 1,000. Because imagine if basically probably by the end of this year or next year, we have the equivalent of 15 to 20 billion PhDs on the planet, and they can do anything. Again, that's why I keep saying, and I'm driving this point home, you cannot oversell the long-term scientific and economic impact of what is happening right now. So just cognitive hyperabundance, that's what we're heading for. Let's talk about some of the risks, though, as promised. So the risk of this is, number one, cyber risks. These things can already write code, and you can jailbreak them, and they can write malicious code. So basically, that means any bored teenager or disgruntled federal employee or whoever can write great software viruses. Now, what I will say, one mitigating factor, though, is that just because one disgruntled person has AGI in their pocket or superintelligence in their pocket, so does every IT department, so does every firewall eventually. So I'm not as worried about cyber risks because there's always an arms race like, oh, you get a faster computer. Well, guess what? The cybersecurity experts also got faster computers. You get artificial superintelligence. Guess what? The other, the red, the blue team, so it's red team and blue team. The blue team also gets the same tools. Now, what I've mentioned before, though, is that to me, the greatest risk is always the bioweapons because state actors, rational state actors, they know like, hey, I'm not going to benefit from releasing a bioweapon that kills everyone. We already saw that with COVID. Don't do that. That's just nobody. Nobody wins in that situation. That's a lose-lose. So that's basically the same game theory as mutually assured destruction. So like nobody launches the first nuke because if you launch a nuke, everyone loses. It's kind of like, you know, someone just like if you get angry and like you just throw the Monopoly board, everyone loses. Likewise, rational state actors are probably, after the COVID pandemic, probably not going to release bioweapons. That doesn't mean that an irrational actor like a terrorist organization or a rogue state like North Korea won't just say, ha, ha, ha. If I can't win, nobody wins, you know, like a little finger from Game of Thrones, like chaos is a ladder, right? Psychopaths think like that. So there might be some psychopath out there that says, you know what? I'm going to use this technology to make a bioweapon just because I think it'll be funny, right? Some people think like that. So those are the biggest risks in my view. Now, some people will say, well, what about, you know, upsetting the jobs and blah, blah, blah. Look, if we have cognitive hyperabundance, your needs are going to get met. That's long story short, like you're not going to starve. You're not going to go homeless. We will figure that out. I have that much faith in humanity. Now, everyone's role in AI advancement. By this point, you might be saying, well, okay, there's nothing I can do. You don't have to do anything, right? Like I tried to do that for a little while. I was like, you know what? I'm just going to let it play out. And eventually I got bored and I was like, you know what? I still have a role to play. Actually, it was Julia convinced me that I still had a role to play because I wrote a tweet. This is Julia McCoy, by the way. I wrote a tweet saying like, I can just kick back and relax and just let it play out because I have no contribution to make. And she's like, that's not true, Dave. And I'm like, all right, you're right. We call her team mom. I call her team mom. She's like, no, that's not true. Anyways, so what can you do? Not everyone is going to be doing core research. Not everyone needs to be working at open AI. Basically, once the thing exists and the thing already exists, we have all these AIs out there and there's more and more coming every day. If you can get into core AI research, great, go do that. But that's only the first piece of the puzzle. Education. So if you can teach people, you teach friends, family, coworkers, if you can teach a business, if you can teach on the Internet, teach everyone that you can about AI, how to use it, what is it good for, what is it not good for. Then you can deploy it. If you actually use it in your life or your business, share that. The more you deploy the stuff, the harder you push, the faster it will get adopted. So it's adoption, acceleration, deployment, and integration. That's everything. Also, my dog is trying to initiate some wrestling with me. I know. Making videos is boring. We should just play with the dogs. Also, if you want to know what to do, get dogs. I'm serious because dogs don't care about AI. All they care about is love, affection, food, good walks, and quality time together. They are so immediate. I need to take them for a walk after this. Anyways, so deployment. And then best practices. So best practices, basically, I kind of break best practices down. There we go. Into two overarching categories. Number one is, what is it good for? How do you use it? And then number two is, how do you not use it? What is it not good for? What problems can't it solve? So capacity and constraints is kind of the two Cs. So capacity constraint, what can you use it for? What can you not use it for? Knowing what something can't do is just as valuable as knowing what it can do. And then finally, network effects. So this is where I'll end. And basically saying, you won't necessarily know the impact that you're having. I don't even know the impact that I'm having. But every now and then I get someone messaging me saying, Dave, this video that you made a year ago inspired me, and now I'm doing all this other stuff. Or Dave, this thing that you published solved so many problems for me, and it generated $10 million of revenue. And nobody has given me a commission check for that yet. Anyways, just saying, if you want to pay me for any help that I've rendered, that would be great. Yeah. So the network effects, you often don't see the impact that you're having. But that's how it works. So anyways, thanks for watching to the end. And I'll check you next time. Bye.