L’ancien CEO de GOOGLE me raconte ce que Steve Jobs lui a appris - Eric Schmidt

That was fun. What I learned over the years is you're better off making the wrong decision quickly and quickly reversing it than to be paralyzed by it. Every failure I've had was a failure that I didn't understand quickly enough and didn't terminate it. Every success I've had could have been stronger if I had accelerated it. How do I grow faster than everyone else? Literally scale. Traditional business logic is basically maximize your profits and so forth. Bad strategy. A much better strategy is to simply maximize the number of users you have. So just get the users and then you can monetize it. I think one of the things that we should try to do in society is to get the joy of discovery, especially in the high school period where a lot of that happens, and get people to say, you know, I'm good. Progress in society occurs through innovation. And we're at the beginning of an innovation revolution that is hard to describe at its scale.

When I showed up at Google, I used to say to our employees, "Do you understand how big a megabyte is?" And everyone would say, "What are you talking about?" Because I started where one megabyte was the size of the house we're in. So extraordinary progress that we've seen in my career. It's a gift. I'm just jealous of this next generation of the power that they're going to have to do that. You are fantastic. I look forward to coming back.

You're very kind. I've wanted to visit for a long time. You guys are pretty famous.

So, I've been advising Ukraine for years on trying to modernize their military and so forth. They started off with just an army, no navy, no air force. It's hard to describe what modern war is like. We all have an idea of what war is because we've all seen it in the movies. Real war is much, much, much worse. Here's a note from my friend, a woman who was on the train that I was not on yesterday coming out of Ukraine when the Russians hit them at the Polish border. That's the train from Kyiv to Helm. Right. Right.

And the airspace over Ukraine is shut down for obvious reasons. So you cannot fly there. You have to take a train. The trains are lovely and the people are great and so forth. This is their report. Feeling quite vulnerable after feeling firsthand what modern warfare feels like. The experience that you, meaning me, Eric, had with ballistic missiles was horrible. But there's comfort in knowing the loud noise itself is scary, but that you're hearing it, you're not dead. Seeing a first Giron drone flying and hitting was visually scary to us, but not new. Running after it hit and being told to hide in the field and under a tree was something different. I was under a tree with a Ukrainian mother with her 10-year-old son. I cannot yet properly explain to you the fear of hearing the next set of shaheds buzzing overhead. The most comparable analogy I can make until I've processed is maybe how it would feel treading water in a dark ocean and you see shark fins circling. You just can't do much. You truly feel the thought that if you run maybe they can see you crying as I write this to you. These are my friends. They hit the train and the train next to the group that I was part of a delegation to talk to Ukraine about how to help them. It's real war. People are really dying. It's really horrible. We need to figure out a way to stop the war. The West has not provided enough support, although people claim to, but Ukraine needs money and they need technological help. They can fight the war but they need our help or they will ultimately lose.

The prime minister, the former prime minister was in that train, wrote about it in a very eloquent way.

I think one way to understand this is that Russia is testing the resolve of the West. Russia is very committed as best I can tell to taking over all of Ukraine, not just the limited war that they talk about which involves historical things which predate all of us. And I think it's important that the West support Ukraine, a young democracy who want to be free. They don't want to be under Russia. They want their own country. It's not a perfect country. They have all sorts of problems, but so does the West. We need to fight for freedom. One of the things that I learned better now is that the earlier generations of Europe, your parents and grandparents, they fought for freedom. And we're free in Europe and in France because of the hard work of your grandparents.

And American soldiers and British soldiers too, right? Yeah. But let's talk about Europe. Europe fought this for a reason. And the principles of Europe apply to all of humanity. Europe is a particularly thoughtful human rights, equality, liberal thinking, classic liberalism is at the core of European values. Those are my values as well as an American. It's important to support them in Ukraine. And one of the problems with war is that people's attention spans are very short now. Wars go on for a long time. The Afghanistan war in America went on for I think 12 years. Iraq was five or six. World War II, as you know, was roughly six, maybe seven. They go on much longer than social media cycles.

So, I've learned a lot about war. I don't have a military background, but I'm just trying to be helpful. And when we started and when I started advising, it was pretty simple. You had these simple so-called FPV drones that were numerous. They held the line. They helped push the Russians back. The Russians adopted that. It was sort of a form of parity. The canonical thinking at the time was that Ukraine was the young underdog innovator. And Russia was big, old, slow, and powerful. That was the thinking. It's wrong now. And it's wrong because Russia is on a full war footing. They're very smart and they understand how to scale.

And well, Ukraine continues to fight. They fight seven days a week, 24 hours a day. Wars have an ebb and a flow and there are new strategies that are offensive and then they're defensive strategies. An example was that Russia invented so-called fiber optic drones which use fiber optics which hunt down Ukrainian soldiers. Ukraine copied that strategy. You can't jam those, right? And so they're unjammable. Now that's old news, but six months ago it was very big news. Russia has been adopting and improving drones called Shahed drones which they originally got out of Iran, but they're really a Russian product now. They're big, relatively slow, relatively loud. They were relatively easy to hit and so these drones carry 90 kg. They take out an apartment building. 90 kg is a lot of explosive. It kills a lot of people. And Russia recently made two changes. One of which I anticipated and one of which I missed, so I blame myself a bit. The first thing they did is they bought a series of Chinese small jet engines. I didn't know this, but you can buy online for about $5,000, something like that, a very small jet engine, proper jet engines, a single impeller, like on Alibaba or. And it's roughly, I don't know, a third of a meter long. Good. And they put out enough power to move these small airplanes very, very fast. So they fly higher and they fly faster. And the current interceptors, the ones that Ukraine mills, and they have quite a few companies that do this that are quite good. I work closely with a company called Skyfall, which has the number one drone interceptor called P1 Sun. So in any case, these new things are too fast. Okay. But they did 700 km per hour. That's correct. Yeah. And historically the max is about 400 km. And at 700 km they're moving very very much quicker. So the other thing that they did is that Russia did not have the equivalent of Starlink over Ukraine, which meant that they had to know what they were going to hit when they launched. In other words, they could only hit fixed targets. The most interesting targets, which include trains, are moving targets, but you have to have a continuous video feed either from satellite, which they don't have, or over some kind of mesh network. Well, the Russians took Chinese radios and built meshes that are two hops. So you sit in Belarusia, you launch the first and the second repeater drones, and then the third is the attack drone. And as the attack drone goes forward, its video goes to the repeater drone and the repeater drone. Why do you need a repeater drone? Curvature of the earth, right? It turns out the distances are such that you have to be high and so forth. All the physics require it. They have released this in the last month to great effect. And now another example is that there was a hit in Kyiv right next to the hotel that I normally stay in, right in the window of the commander of the SBU. And that commander of the SBU is roughly the equivalent of the KGB. In fact, it dates from the KGB. So that's an example of Russia saying we can hit anything anywhere if we want to and you can't stop us. A few weeks ago there was a private meeting that was publicized by President Zelensky where he called the entire cabinet and called all of the Ukrainian companies in to say how are we going to solve this? What are you going to do? What are you going to do? What are you going to do? Now they're working really hard on this and they will solve it. The reason I tell you this story is that at the moment a new weapon and a new communications thing is terrorizing my friends in Ukraine. Soon there will be a counter, but then there will be another one. I now have having briefed on it. It looks like the future of war is not going to be these little electric battery drones, which is what I thought, but much more likely these powerful small jet engines that go 700 km an hour. They're very hard to stop and they have much longer range. The cost of these things is also going up. It was estimated that these things cost about $100,000 or euros if you will. And the historic shaheds cost about $30,000. So what I originally thought, not to go into the details, I thought that it would be millions and millions of tiny things. This is my vision. Yeah. 400 euros a drone. That was my vision. And I was wrong. It looks like the majority of the war is going to be in these small jet engines and then ballistic missiles. For those of you who are not experts in missiles, missile technology, I never thought missiles were very important, but they're obviously very important. When I was in Kyiv in August for a different meeting, I was coming back on the train just as I was done. It was midnight and I get a text saying ballistics coming to Kyiv. Get off the train now. The train is moving by the way. So I go, "Oh my god, oh my god." You know, it's like midnight and what do I do? So I put my shoes on and then I thought, grab my computer and my phone. I have my clothes on and then I thought, oh, I should wake everyone up on the train. So every door I throw the door open, turn on the lights and say, "Get up." Right. And in one minute, everyone on the train is at the vestibule getting ready for the train. The train stops. We run in the vestibule. I hear boom. Now, I'm used to hearing bombs, I'm sorry to say, because I've been in the front. This is a different bomb. This is a bomb like right next to ballistic, right? It's ballistic. And then I step on the train and I see the bomb shelter and what happens when a ballistic missile is about to hit you, everyone runs. At the time, I had the KGB equivalent protecting me. So the secret agents and I run as fast as we can. Bomb, bomb, bomb. I'm never scared and I was scared. I go into a bomb shelter with 500 Ukrainians. And I thought about the blitz, kids crying, you know, the whole bit. It's midnight, you know, it's just terrible. And of course, eventually, we're cleared. We get on the train. The train leaves as fast as possible. That night was seven hours of that kind of stuff. So it turns out I think the future of war as practiced by humans is going to be these very very fast jet drones and ballistic missiles of one kind or another. If I look at Europe, who I desperately want to help, Europe is short. Europe is still very short ballistic missiles, very short anti-ballistic missiles, very short the systems to protect them against Russian or other aggression. I'm not suggesting Russia is going to attack NATO. I don't, it's not my area of expertise. But the entire doctrine of Europe is to protect itself. Europe is vulnerable to this. This has to get addressed. Well, and vulnerable is not good when you're facing Putin anyway. I mean, you have to be strong at least and show it somehow. Yeah. And I can make the same argument about the United States. I worked with the US military for quite some time. The US military is very focused. The military is focused on a potential war with China. Well, do you think the Chinese are going to use similar technology? Of course they would because it's their stuff. The Russians are using Chinese parts. Plus Chinese, I mean, can prevent Russians, well not Russian but Europeans or Americans to use some parts that are, I mean, very useful that you can't. So one of the things that is I think the saddest of this is that I'm old enough to have seen in my lifetime the benefit of the general term is Pax Americana, that post World War II 80% of the globe was essentially American money, American economy. As you know, we helped rebuild Europe. Very proud of that. And I had assumed that globalization and the general sort of calmness since 1945 was permanent. I think I'm wrong. I think we're moving to much more great power competition, much more regional conflicts. And it's unfortunate to say, speaking as a person who's been studying war now for a while, I'm a scientist, so I speak with an academic view, is humans were evolved to fight wars and we've not overcome our tendencies. I'm very worried that the Iran war is not going well. I'm not part of that, but I want to be helpful to the United States. The Iranians are tough. They know what they're doing. They're in it to fight to survive. You have to take these things very very seriously.

There are laws about war. And when I worked with the Pentagon, I did this in my spare time when I was running Google. Google allows you to spend 20% time on something interesting. I thought, and again, I don't have a military background. I was trying to be helpful. We wrote an AI policy which has been adopted by. Yeah. And that AI policy says the following. You can't just build an automatic weapon to go look for something. That's illegal. You have to have real intelligence that in this boundary, in this corner, in this time period, there's a real threat. And if you have real intelligence and a real threat in a bounded time and space, you can go look for it. I was comfortable with that because I think that that's a reasonable limitation. And there were versions of this already in precision munitions in the West. And I continue to believe that that's the right answer. If you allow hunting, automatic hunting, I think it drives a kind of terror and a likely set of war crimes that ultimately would occur with you. Ultimately the people who did it would be tried in the same way the Nuremberg trials were used against the Nazis. Those men were ultimately hanged. So there is justice although maybe not contemporaneously in war and if you look at Milosevic in Serbia, Serbia, Serbia, he was also tried in a similar way so it's very important that the people who are fighting on both sides appreciate that there is justice at the end. Now with respect to AI in war, the simple reasoning that people have is they say if the other side is unconstrained, my side will also have to be unconstrained. That's not how the law works. You're not allowed, if the Russians, I'm not saying that, use a biological weapon that does not allow the West to use it in return. Right? We have agreements about this and it's important that we behaved as civilized people. With respect to AI and war, most people think that AI and war is going to ultimately be the thing you've seen in the movies where you have a thousand soldier robots carrying guns. That's precisely not what's going to happen. It makes no sense. What will really happen is AI will be used for complex decision-making and my hope is ultimately the following. So, let's imagine you and I are enemies. This would never occur. And that you're ahead of me and you have 10,000 drones and I have 5,000 drones and you have a great AI team and I have a great AI team and we're enemies again. Not going to ever happen. Would I be willing to attack you if I don't know your algorithms and you have more? No. Would you be willing to attack me if I have half the weapons that you do, but I have different algorithms? I think the answer is no. One of the things about military doctrines is when somebody attacks someone, they have certainty as to the response. They've spent months saying if I attack this target, what will the response be? And the doctrine in military thing is proportionate response. So if I'm not minimizing this. If there's one bomb on one target, then the response is one bomb on one target. And you've seen this in the Middle East until recently. I believe that the arrival of AI creates such an uncertainty in war planning, this is my opinion, I have no way to prove this, that it will be a great deterrence that people will not be willing to take the bet. My best friend was Dr. Kissinger for a long time and he would tell me these very funny stories about meeting in the Kremlin in the 1980s where they would basically count weapons like you have this many. Well, we have this many. Well, no, no, we have this many. No, we have this many. And it was, again, I'm not trying to make light of something very serious, but it was sort of almost academic. But imagine that you and I again were enemies and we're meeting and we're trying to have diplomacy. And I'd say, "Well, how many do you have?" And you say, "I'm not going to tell you." And how many do you have? I'm not going to tell you. What are your algorithms? You're not going to share them. So unless you know what your opponent is going to do, I think you're not going to fight these wars. And that's an important fundamental assumption I think going forward. If you believe I'm wrong, then you could end up with horrible, horrible escalation tactics because the moment the one side starts, the other side will have to launch a. It's the same as the nuclear threat somehow. Yeah. And those are called dynamically unstable scenarios. So basically what happens is if I think you're about to attack me, I will attack you. If you look at the 67 war in Israel, yeah, they had intelligence that the Arabs were going to attack them in one day. They formed a unified government and then they decided to attack first. They ultimately, as you know, won that war against a much larger opponent, which you know changed history and so forth. But the important thing is they had the benefit of timing. So, one of the dangers in war is you want to be first on the attack because these things are offense dominant. Anyway, long story short, AI and war is going to be very new because of the swarming and coordination aspects of it that if they're done by computer, they could be done at a scale that humans cannot do. This is to be worried about. You asked about the most recent employee who left one of the tech companies. There is a changing subjects and you've done a lot of interviews with tech AI people so you understand this really well. There are a set of people who believe they're all seem to be in San Francisco by the way that recursive self-improvement which is the rate at which you can improve systems on their own words they're self-improving recursive means it keeps going so here's the way you would say it say learn everything start anywhere you want and the computer says oh I want to start with French literature so it reads all of French literature and it discovers that there's war so it says I'll learn all about war and then I'll learn all about suffering then I'll learn all you see how it's recursive it just keeps going.

That's an example of it. But recursion really means learning the way you do. So in the morning when you get up, you read the paper or a book or something, you've learned something for that day. Yeah. So historically, computers couldn't learn this way. This is in general, this is the recursive self-improvement and the development of deep reasoning are the new news. Now, I don't know about you, but most people still think that AI is the equivalent of chat GPT. Yeah. You know, please write my paper so I don't have to write it myself. That was years ago. In other words, four. Yeah. Two years ago, the industry figured out a way to integrate tools and time. So, what happens is you can now say, can you make a plan for me? And now the tools are so powerful that they can do reasoning. So for example, I have a friend who's organized a conference at Stanford which I'm attending and he was trying to figure out a new compact for America. Okay. And a new compact of politics because America's politics are broken and so he used in this case Claude and Gemini extensively to get the proposals based on all of the writing and so forth and so on. Now of course it's human judgment. The belief is among some people in San Francisco that there will be an explosion in learning that will accelerate past what humans can understand. So the simple example I'll give let's imagine there's a problem in physics called dark energy and dark matter. We don't understand where it is. There are many theories. Let's imagine for purposes of argument that there is a math proof in physics that answers the question what is dark energy that is so complicated no human can understand it but we know it's true. So we know the proof exists we no human can understand it and it's true. What is that? How do we know it's true? Because the machine told us. Because you can verify it because science is about verification. So I write a document which has an end conclusion. You don't understand the language, but you can verify the conclusion is correct and you'll never understand the language. It's in a language that no human can understand. Insane. Now is what happens in that scenario? Well, on the one hand, the great news is we would have figured out what dark energy and dark matter meant. So we could use that to have far better science. Understanding the physics of energy would help us a great deal with climate change. It would make us have less expensive and more powerful cars, better airplanes, you know, better jet engines, also better tools for war. All sorts of things would come from that. But what does it mean if we humans cannot understand it and will never understand it? Is this a situation where we are the dogs to some other human? Right? These are the questions that I've explored in two books which I wrote with Kissinger and written about. We have them here. Yeah. Genesis. Genesis and Age of AI. Yeah. And these are deeper philosophical points than people appreciate because we've never had an intelligence that's smarter than our intelligence on something. I'm by the way I'm not making the claim that these things are sentient. They're not human intelligence. They have different kinds of strengths and weaknesses. We as humans, we all understand our own natural. We understand emotional structures and people get upset and so forth and so on. These things have a different psychology if you will, whatever term you want. So in the danger for AI is that this stuff gets ahead of our understanding of it. This is called the alignment problem. Keeping that activity in the alignment bucket is crucial. There are many people who believe that we need to slow down progress so that we can not that we want the progress but that we need to make sure we're checking it. One of the more interesting results is that it looks like you can take the previous model and check the new model. Now remember the new model is smarter than the previous model of course. So how does a dumber model check the smarter model? It looks like you can instrument in such a way that it can trigger these things and then the dumber model if you will plus humans can say it's okay you know it's within the bounds of acceptability that's a big result if it's sustained.

So let's talk about the OpenAI example what appears to be true is that a team within the company was testing a newer algorithm and the algorithm got somewhat ary. Yeah. And the algorithm there were 1500 or so GPUs involved. There were many agents. Yeah. 700 or something. And you know, you understand this very well. And the agents were given a task of achieving a goal and they broke through many security systems and ultimately got to their goal. No one was killed, no one was harmed, of course. But it's a warning. And imagine if the same thing were true with weapons. Yeah. With the 700 drones like 700 drones or so forth. So for all of us who care about this, and there's many, many, it's just a warning shot. Now, when I looked at it, I thought, okay, so for purposes of argument, you have your OpenClaw and your Clawed Code thing. You forgot to say in your instructions, by the way, follow all laws, follow all human conventions, don't embarrass anyone, don't harm any humans, don't harm any animals, don't pollute the environment, don't overrun your stay, don't break into any systems, even if you have permission. And in other words, there are all of these assumptions. So I for example am sitting here with you. We're operating under in our case British law. We assume that we're legal by sitting here. We conform to societal ethics. These AI systems don't have the benefit of the decades of learning what the rules are.

So this is the alignment problem. So I'll give you another example. There are quite a few people who have discovered the following and I think you're going to be one of them. You don't need as many people as you thought in your company. No. So I have another friend who explained to me that his vision is that future companies will have exactly one employee which is himself. And the reason is that everything can be automated. Now let's think about is that really true? If it's a digital product, right, where there's no production, right? You're making something digitally, you're charging for it. You could probably build the entire business. And imagine a company with no employees. You'd have no HR issues. You'd have no labor issues. You wouldn't have to have real estate. Now, you'd have a great deal of money going to the AI, Nvidia or what have you. So I think the more like, and this is a thought experiment, that's not really going to happen. Another example is look at a large bank. Pick a big American bank and they're big. They have 50,000 people. I think the European banks are of a similar size. HSBC. They're just huge. Now, what do you really need in a bank? Well, you need people to talk to the regulators because the regulators are human. You need people to build the product and you need people to sell the product to humans. Why do you need all the infrastructure? All of that can banks basically do risk calculations. It makes sense. Another example, I'm a professional CEO and I've been a professional CEO for more than 40 years and I do understand being a CEO. I spend all day reasoning and what happens is people hand me documents. I read them and we have a discussion. Is this right? Is this wrong? What is the decision? What do you need now? Why do I need someone to present this document to me? Why can't I have the computer anticipate my question and show me what's going on? And then if I have a question, I'll call a human to say, I don't understand this. How could this be true? So, an example would be that, and this is now common, Claude has a thing called Claude tags. The way to understand it, again, I'll put it in corporate speak, is there are a set of agents which are watching. So you have an agent that's watching for security. You have an agent that's watching for cash. You have an agent watching for manufacturing problems. Sub agents or Yeah. Okay. But just think of it as just in their case they used claws which is the same principle. So maybe you have 10 business functions and by the way they're running 24 hours a day and you can't run 24 hours a day. So that agent talks to your agent and you communicate through Slack. Why Slack? Because Slack is used heavily in businesses. So you just have another Slack participant and that Slack participant knows what the permissions are for each of the players. I have another company that I helped found which is a startup which is trying to build an ontology which will work with the auditors and so forth. Okay. So that the auditors can say all of the AI reasoning in this company to financial rules, business rules, legal rules and so forth. So imagine a situation where the people who are presenting these reports are just not needed. Now you might say, well there not going to be any jobs. That's not true. The people will do something different. But this sort of let's use finance as an easy target. Finance is largely computerizable. I get daily and weekly finance reports from human beings. Why does the computer not produce that automatically and the humans can oversee it? It makes more sense. Furthermore, the computer can do it 24 hours a day and it's not going to make mistakes. Now, why do I mention this? The majority of the revenue in AI is going to be workplace agents of the kind that I'm describing. Why? Because businesses will pay very large amount of money for better management, better customer invention, better strategy and so forth and so on. It's going to be huge. Another way to say this is why is Anthropic's revenue growing so quickly? Partly it's because they are the dominant or the current leader in software development. But the other is in enterprises are using them for these kind of purposes. OpenAI has now come out with a new version, Astra, which they claim is bigger, better, and cheaper, which is great. And my friends at Google have told me, "Oh, yes, ours is coming, and it's going to be much, much better." Which is great. That's called competition. That brutal competition is going to ensure that these platforms are safe, usable, and so forth. It may be boring to some of our audience here, but the money in businesses to automate them, make them global, make them more efficient, make them stronger, make them more all of that is going to be large. The other issue to know is that the Chinese models are about six months behind. And you know this, but maybe not everyone understands. The Chinese models are open source, which means they're free to use. Open source and free is a problem for the western company. Of course it is. Who now Mistral is interesting. What does that mean? Sorry. It means that open source and free if you buy like a iMac M5 Pro, let's say, you'll have an Nvidia chip in it. You take a Chinese model, you install it on your computer and then it works for free. Like no more tokens, no nothing. Yeah. Just free. The truth is I'm actually doing this myself. You need a large a lot of memory to run these things on your computer and it's quite interesting. Mistral is, I think, the best example of this in Europe and they started with an open approach and I won't speak for them although I'm obviously a big investor and big supporter of Mistral. But what they figured out is if you have open software, you still have to the customer has to have help. And they figured out a way to make a great deal of money, right? Which I'm very proud of for them, which is funding their company. And they are essentially, in my view, the dominant European player. We'll see how long that goes because now they're going to face competition from the Chinese as well. It's a highly competitive market and that competition is good for consumers. My concern if I can just anticipate one obvious question is if the Chinese models are free and the western models are not free how does the geopolitics work? Yeah. Interesting, right? Well, if it's free, there'll be a lot more users. If it's expensive, it might be better. So you might end up in a situation and this is very important where the vast majority of humans on the planet use Chinese models simply because of cost. I don't think that's a good outcome.

Oh, I'm Eric Schmidt and I'm a computer scientist.

I have a PhD in computer science and I was fortunate to be very early in the computer industry. I started programming essentially on time sharing machines and I was a heavy time sharing machines. It's an early kind of technology and I was a very good programmer in my humble opinion of myself on the original IBM 360 architecture. So this shows you how dated I am. When I showed up at Google, I was the CEO for a long time. I used to say to our employees, do you understand how big a megabyte is and everyone say what are you talking about? Because I started where one megabyte was the size of the house we're in. Right? So the extraordinary progress that we've seen in my career is it's a gift. I don't know how to describe it any other way. I'll give you another example. When I was 21, I was at Berkeley graduate student and I worked on a computer called a vax 11780. We loved it. I mean, we obsessed with it. There was one we stayed up all night. You know, it was that important to our research. And that computer was the prime that single computer was the primary source of the software that is now used many decades later on, for example, the Mac you're using and on the servers you're using. That single computer. So my friend who was the leader, I was his protege if you will. His name is Bill Joy did an analysis. And he said that today the same computer would be a particular Nvidia chip. It's an Envy 72 with a particular set of hardware called B200s and GB300s. A single one of those computers costs about the same and it's two to the 42nd times faster. And I said, "Okay, Bill, you can do the math." 4 trillion times faster. 4 trillion times. So if you're a 22 So imagine there is a smart 22 year old today that's smarter than I was and I had the ego of a smart person. They must be incredibly smart now. That young man or woman and great that there's equivalent number of women doing this today. What could they do with four trillion times more power than my little group had? It's extraordinarily tantalizing. And so in my case for example I decided to create and we've not announced the details but I'll say in general we're going to create a philanthropic set of science data centers for leading researchers in Europe and in the United States to replicate that. That's my gift back. Okay. For all the success I had that's related to Lafayette. Yeah and the Lafayette program is a good example of this. But the important thing, the important thing to understand, the details are interesting to me and you and maybe 10% of your listeners, but 100% of your listeners need to understand that progress in society occurs through innovation. And we're at the beginning of an innovation revolution that is hard to describe in its scale. So, here are some examples. We can probably solve most human diseases within 10 or 20 years. Not all of them but most of them because most of biology is still not understood. We still don't have a digital model of a cell. I already mentioned new chemistry, new physics, all of those lead to do you like your shampoo? We can make more shampoo cheaper. We can make it more effective. We can have it change your color of your hair whenever you want. That kind of stuff. These are trivial examples but there are far more the most important thing ultimately is human physical health human mental health human freedom and so forth we can access we can accelerate that. I'm just jealous of this next generation of the power that they're going to have to do that.

Well, I used to say biology because I thought that the next frontier was biology. What I think now is it's deep reasoning. And deep reasoning is getting the computer to do the thinking that you as a human are unable to do because it's smarter in some dimension than you. So for example, every high school in the world including for your kids should have a class of what the technical term is called prompt engineering. Mhm. Learning how to use the tools. Now you are already you yourself are doing this with OpenClaw and others. But if you think about it, your 14-year-old his or her life will be determined by how effectively they can use these powerful tools.

Okay. One of my friends is a very famous mathematician. And I said, "What was it like?" And he said, "Well, I was 15 or so forth and I was upset and confused about who I was. And then I discovered that I could discover things." And it gave me, he didn't say it, but what he meant was street cred. He, you know, he looked hot or powerful or whatever he cared about to his classmates. Classmates, you know, he was like, I'm cool. Yeah. So, I think one of the things that we should try to do in society is to get the joy of discovery, especially in the high school period where a lot of that happens, and get people to say, you know, I'm good. Another example, I have a large because of my success, I have many assets and I have a large organization that manages them. Philanthropic donations, personal assets, that kind of stuff. And I've been running around asking everyone to show me their latest ideas of how they will use these to make their jobs better. So, an example would be that I have a group that works on a house and they said, "Well, we thought we would just check all the bills." And I said, "Okay, cool." And it turns out there are so many bills that they had not been tracking what was going on. And they were very proud of themselves. It doesn't matter whether I thought it was good or bad. They were proud of themselves because they discovered it on their own. The person who runs my finances, who's very, very talented, called me up one day.

and said, "Do you realize that I can get my financial reports in an hour instead of waiting a week?" And I go, "Yes." But she discovered it.

So, so the key thing about AI in terms of teaching it is that the individual has to get their head around why it's good for them. If you treat it as a "you have to do this, you have to do this," you're going to lose. But if you tell people the joy of discovery, another example, I have a friend who's decided that all of the architectures that we're using are wrong. Now, he's a scientist and I have no idea if he's right or not,

but he makes an assertion. Excuse the technical stuff for a minute. That the way all of AI works is it's Gaussian. It uses the most probable next word.

So technically what this means is that you have a very many parameterized space. You're picking the most likely point in each of those spaces and somehow that predicts the right word and it produces great prose. But there are plenty of systems in the world that are nonlinear, non-Gaussian, and deep inside of these systems there's a presumption of a Gaussian world or statistically, you know, everything you and your audience that's mathematical will know what I'm talking about. So his assertion is that's not true. And I said, well, what are you doing? He said, well, I'm writing a new one. And I said, well, who are you writing with? Oh, me. And I said, well, okay, how many people? He said, no, just me. I said, what do you do? Well, I get up in the morning and I give it a set of questions and I've been writing a language and a way of expressing this and I'm going to rewrite all of the scientific libraries in the world. I said, "You are? I mean, you're smart, but do you know every science?" No, but I can have my system read the books. I said, "Okay, so you're having your system read all the textbooks to produce all the scientific libraries using your new idea for non-Gaussian distributions?" And he said, "Yes, of course." In other words, he had become so into it, he didn't realize how outrageous the claim is. And I said, "Well, how much money do you need?" "Oh, I need millions of dollars." "For what?" And he said, for in his case, he's using Claude. "Using Claude?" And I go, "Really?" And he goes, "Yes." And I said, "Well, what are the agents?" This was lunch yesterday. "What are their agents doing?" "Oh, I gave them a one-hour task." And I said, "Why?" "It's a one-hour lunch and I'm almost done. I have to go home in order to check my agents." I said, "Well, why do you have to go home?" He said, "Well, no, my agents are in the cloud." He said, "Well, I prefer to work from home." And I said, "The agents are working without you." We have this hilarious conversation of why he has to go home because the agents are working without him and they don't need him. So I said, "Well, what do you do at night?" And he says, "Well, I have a wife and a family and all that kind of stuff." He says, "Well, I get tired in the evening, so I give my agents an overnight task. Maybe you do the same thing, Matt. You know, maybe you give them a hard problem." I said, "Well, how do you decide?" He said, "Well, it gives me an estimate of how hard it will be." And I said, "Okay, so you give it the hard problem at night and the easy problems during the day." So you're tougher on your agents at night than you are during the day. He said, "Of course." I said, "Well, do you wake up when they're done?" I said, "When are they done?" And he goes, "Like four in the morning." I said, "Do you get up?" "No, of course. I wake up naturally." I said, "What do you do?" "I have coffee and I see what they did." You see how the... And again, this is an extreme case. That's the power.

And so if you... and the same stupid story that I told you, which may be incredibly important or may be unimportant at all, can be replicated in art, in writing, in history, in any pursuit. You could build a new constitution, you could build a new framework for how people think, you could try to understand how consciousness works. I mean, when I look at the interesting scientific questions, every one of them is informed by this notion that a small group of people — my friend's case is maybe extreme, two or three people — can use these tools. We're no longer in the laboratory and pipette phase with a microscope looking for, you know, E. coli. We're in a very different scientific pursuit. This is very powerful.

if she masters the scenario I just described, her impact on humanity at a statistical level will be infinitely higher than the expectation that you as her father have. Now, her ability to achieve her goals, but more importantly change society for good — in her case, I'm sure — will be unmatched by the current previous generations.

Yeah. Yeah.

So this is a collective action problem. I think it's best informed by the specific problems rather than the general fear. I'll give you an example. It is not okay for teenagers who are on social media using AI to kill themselves. There's no world where that's okay. It's just not okay. It's not okay for girls. It's not okay for boys. It's not okay for parents and so forth. Most people can agree that adults should be able to sort of be relatively free, but we need to protect our children. That problem's got to get solved. Another problem we have to solve is that in the competition between China and the United States in models, we have to win that competition without losing our values, without violating the principles of liberal thought, freedom, freedom of assembly, all the things that are Western values. And the way to understand that is if the majority of the world uses Chinese models, the Chinese government systems, censorship systems are just different. And we know that these systems change human behavior because people have trouble distinguishing between physical reality and virtual reality. So that's one. Another thing we have to do is we have to make sure there will certainly be jobs lost, but there will also be jobs created. We have to get through that transition in a way that does not harm society. I can go on. So each of these is manageable if we have the time to deliberate them, because democracies are deliberative and they should be.

100%. The Trump administration is largely free market, libertarian, doesn't particularly want to regulate AI, and yet these things are on their desk. And I'm told by my friends in the president's White House and other groups that there is a big disagreement over these principles. So sometimes when you're in a government, these problems are brought to you even though you didn't want to, right? You didn't want to have to face them. So we have to face them now. My own...

Well, thank you. There's a set of groups that are informal that are talking about this. There in Vienna, there was a meeting that I sponsored between Western and Chinese thought leaders.

To discuss this. This was started by Kissinger, by the way, you know, before his death, because he anticipated this conflict. And these are social, polite conversations. They're not adversarial. Trying to make sure that everyone is talking with the same language. We need to make sure, for example, I'll go back to a military example. Let's imagine, I'm making this up, that there was a Chinese lab and the Chinese lab had an escape of an agent that looked like it was going to harm the West, but it was false. But the West is so triggered by it that the West begins a conflict because of an accident, right? Because we're so scared. That's to be avoided for obvious reasons. And you can imagine treaties in that area. So I think what you're going to see is a large set of agreements. The first agreements will be among the tech companies, and that's what's going on right now. There are informal meetings with all the tech companies, of which I know at least five different groups, and they're trying to get to the following: Where is the line that you can't cross? We know this is coming. So what do we voluntarily agree is the point? That's a metaphor for slowing down. We're not going to go in this area. We're only going to pursue this. That's an appropriate discussion. There's a long history in this. There was a conference in 1975 in Asilomar where the scientists, not the government, but the scientists agreed to restrictions on recombinant DNA, which was very much feared at the time. Those restrictions are still in place and they work well. Right. So you have to start with the science because the scientists understand the stuff better than the politicians. I'm sorry to say.

Then the science stays and the politicians change over. The second thing is that there will be agreements among Western countries. There will be a rough agreement, compatibility in these approaches between the AI safety groups of the UK and in Europe and America. The strongest AI safety group is actually the one in the UK. It was the earliest, the one that is sort of the best functioning, but I'm sure that all of them will get there. And then the third is that there will be agreements that are essentially treaties between countries that are not particularly friendly. At the end of the day, the China-West relationship will require such treaties because you can't have a situation where China is putting an evil agent in every chip that's going into every car. If that were the case, then we would have to completely replicate the Chinese ecosystem, which would be very expensive. And similarly, the Chinese, from their perspective — I'm not defending them — would say you can't take your Western ideas and embed them in things you're shipping into China. You can't put them in American cars, for example. The same will be their concern.

Well, recently, well, the most interesting thing that has changed is the arrival of these reasoning systems that has occurred since October of last year.

Opus, I think 4.6 came out and it was a material improvement in reasoning. Now everyone has it, so it's not just...

My job is to talk about what's coming.

And so...

Ask your question again. I'm sorry.

So I think some of the generic rules for business are that you have less time every year to do the same thing. So I wrote a book a long time ago entitled Longitude, which made the thesis that essentially everything is related to the speed of communications. Right? And it made an argument that TV arrived and then radio, TV, and so forth. And I think that's correct. And so the speed with which you can make a decision, the right decision or the wrong decision, is now much more compressed. Partly this is because the internet itself is pervasive and now universal. Everyone presumes the use of internet. When it started, you obviously couldn't presume it. And so if you have a free digital distribution mechanism for information, it's essentially ubiquitous and free. That's not true for physical goods, but it's true for the internet. And so you're seeing these reasoning systems that can essentially take all the public data, open source data, as well as your proprietary data, and combine that to make better decisions. That is a profound change in leadership. Most companies are not doing it. If I were recommending actions to CEOs, I would do exactly what I just said I did with my own team. I would go to each function and I'd say, "Show me your latest idea about using these new tools, no constraints on you, to make our business happier, employees happier, our customers happier, better business, more revenue, what have you." And you'll be extraordinarily surprised by the creativity. In my case, I was surprised. I know my employees and I have opinions about them. This one's smart. This one's not so smart. This one works hard. So forth. I found that this upended my biases. That once you gave people permission to get excited about something, they became so passionate, they were the leaders, they were the internal champions. So the advice to a CEO and to your audience that are business people is start with asking people the question, what can you do with these tools that makes our lives, our company, and so forth much better? You'll be shocked. Now, my training as a CEO was over, as I said, decades, and I had the luxury of time. Which now I had a coach named Bill Campbell. I wrote a book.

For Steve Jobs at the same time, which is insane. We wrote the book and we called it Trillion Dollar Coach because he was the coach of the two companies that got to a trillion dollar valuation. Today the title would be $7 trillion coach. That's a pretty impactful coach. Think about it. And he died unfortunately of cancer at a young age. But the important thing about Bill is that he had a set of principles, and I really believe in these principles when you're managing humans. The first is you manage the team, not the individual, because organizations and CEOs operate through teams. So working the interpersonal conflicts, which are always present, is really important. The second thing is that he said get to the best decision, not the consensus decision. So I was taught and learned to spend the time to debate an issue quite thoroughly and say at the end, is this our best idea? And then somebody might say, you know, I've been thinking about it, this sucks. You know, this is the wrong... we need a better answer. And then that drives the discovery process. When you work with creative people, you're much better off asking them questions than telling them what to do. Now, if you're angry, you can say, "Get over there and get that done. And what is wrong with you?" If you're saying that you're losing, you're winning, you can say, "We have a problem. Do you have an idea? Can you get this fixed? And by the way, we've got to get it done in 24 hours or we're kaput." Much more motivating. One of the things that I find most frustrating in war, now that I've been studying it, is that human organizations are full of politics. The presidents are political, the corruption, you know, whatever, special interests and so forth. I was reading the history of World War II and Winston Churchill here in London, and there were all sorts of politics during the war, right, which weren't told to me when I was studying this when I was younger. So trying to figure out a way to manage humans in a difficult situation and doing it coherently is a key skill for a new CEO. This is not taught. It has to be learned. A lot of it involves human charisma, human trust, human integrity. Those are the standards. You're not going to get rid of them. So to summarize the answer, the biggest change is that you can see everything and you can reason about everything, but humans themselves have not changed at all. You have to have a leadership style. I've also learned that there's no single leadership style. I have a particular leadership style which has worked for me, and frankly, I don't have another one. This is like what you get. And there are plenty of bombastic, loud CEOs, which is not who I am, who are also equivalently successful, right? Steve Jobs, for example, was extraordinarily successful. He was a very close friend, and I studied him very carefully. He was so charismatic, but he was also so emotional that when you were in trouble, you would get a dressing down that was quite painful. But on balance, he was so extraordinary as a leader that he could lead even without, in my view, tough behavior. So it's not required that you have the same style. It is required that you have a style that is successful and that empowers people. However you do it is fine.

Well, because I'm a programmer, I get to do things most people don't do. Programmers, you know, programmers used to build the house. Now we design the house and the computer writes the code. So I've been writing various things similar to what you've been doing, but that's more for as a hobby. The way the steps work in a corporation is the first thing you have to do is get all the data connected. There's a library called MCP, Model Context Protocol. My new question, and I've been in the enterprise business for 40 years. MCP, which is essentially a server that talks to the historic databases, allows the large language model, and it could be anyone, to essentially synthesize the data. That is a major change in architecture because it means you can sit and you can literally see all the data that used to be segregated from you, and you can look at things. I own a rocket company, as you know, and the rocket company, I'm walking around the factory floor and I'm thinking, what's in all these machines? These immense rocket machines, every one of them has a big computer in them and none of them are talking to each other. So we agreed that we would interconnect them. And the first thing they did, again, this is using a variant of Anthropic. They, and the company generally uses Gemini and GCP, but for this one, Anthropic was the right answer, is just go look at what they're all doing using MCP connected to them. That telemetry monitoring allowed us to understand and save money and capital. Each of these machines is like $5 million. Saved me $5 million. I'm a pretty happy guy, right? And you can save it because you can figure out efficiency and predictive analytics and things like that. So the first step, not to go on, but it's important to hear it, is just get everything interconnected. If you don't have everything interconnected, and the database of choice is Odo right now, which is taking off, but you can do it with traditional ones as well, and you build connectors for them. Then the second thing you do is you start doing your current job in the current way, just reproducing what you're doing. So that takes another month. Everybody has to kind of figure that out. So the computer can now make the report that you would type in, it makes for you. That's it, labor saving. Then the third stage, which I'm literally dying to see in these businesses, is the reasoning part. And the reasoning part is where you succeed. What are the questions? First question of a CEO is how do I have more revenue? Revenue solves all known problems. How do I minimize my expenses while maximizing my revenue? That's called profits. What is the best financial source of cash for work and essentially work financing, work product financing? It can analyze all this. What's the correct split between equity and debt in today's market? Now, these are all questions that you would ask a banker. Well, now you can just ask the computer and the computer can give you the answer.