Alex Karp: Palantir's Lethal AI & The Truth About LLMs | Full CNBC Interview

Very nice to be here.

Well, first of all, at Palantir, people are currently focusing a lot on our commercial business and there's a lot of interesting things there. But once war zones heat up anywhere in the allied world, the Palantir Maven platform, which takes very valuable LLMs and makes them actually lethal and useful on the battlefield, comes into action. Then the primary focus of our whole company is bringing our American war fighters home safely, meaning that when we engage, they engage in a way that gives them a massively unfair advantage, scares our adversaries, and ensures that when they're in action, they have the best humanly and inhumanly possible chance of coming home safe and without injury. So the whole company really just focuses on the operations of that. Obviously, these things begin well before the public sees, and we have a wide swath of Palantirians working on various platforms, particularly the Maven platform. There are a lot of things that are in the public space. Maven has been discussed and older versions have been shown in public, but I think the American people would be very positively surprised by the advances in technology that really only America has.

There are really two things. Every military force in the world, you see this on the battlefield in Ukraine. One of the big issues the Russians had is how to transform your operations into a modern operation. If you take components of Maven, there's infrastructure which allows you to take older systems and make them newer systems. That allows you to have a completely integrated battle space, which is super hard because of security, clearance, data sources. You can think of it as infant data, some of which is very sensitive, some of which no one else can see, some of which everyone has to see. You have allies working on it. How do you get that into one plane of glass with security? How do you then use advances in AI and machine learning to enhance that while making those decisions not like what's proverbially called loose AI code? It's just not useful on the battlefield. So how do you enhance your ability to see things? How do you enhance your ability to go through very complex systems? For example, in Ukraine, it's not just getting the drone to the end point; it's going through advanced radar systems. Vis-à-vis Iran, it's 100 times more complicated because they have... You look at American performance, which has been, thanks to the bravery of Americans and the ingenuity and I would like to believe our tech, very few losses on our side, very precise, very controlled. You saw this in Venezuela, the ability to dismantle these things. Because the dismantling is happening, we underestimate the complexity that it took to do that. It was not possible to do these things 2 years ago, 3 years ago, and now America can do this with great precision and with very low cost to us in terms of lives and casualties.

One of the many secrets of Palantir: people talk about our forward deployed engineers, they try to copy; talk about ontology, they try to copy; talk about how software that's fine is worthless, but everyone's trying to copy us. Some points of software are not that valuable, but some parts are crucial. There are three kinds of software really. There's infrastructure, which is like hard-coded things written into the enterprise that have to be managed on-prem often to maintain the sovereignty of your own data. That is excruciatingly valuable. Then you have how do you manage that? Ontology, Foundry, and other things are very valuable and durable. Then you have FDEs and large language models. Most FDEs are just throwing people at a problem that anyone else can solve. So if you can manage, some are valuable, some are worthless. Free code is essentially very easy. In any case, what you see in crucial situations in Russia, in the Middle East, or in Ukraine, where you have to bring home people safely and where your adversary has to be dismantled, the Persians, the Russians, and the CCP are exceedingly talented technical people in various different ways with very unique and historically viable ways of fighting. So you're not going up against an adversary that's just going to lay down. When you have to do that, you have to bring your unique way of doing something enhanced with things no one else has. Those things now are infrastructure software, Foundry, Apollo, and large language models that we are producing managed by us. It is the combination that is legal. By the way, you see this in US commercial. The interesting thing about what's going on in the world right now domestically is you have essentially a really important, uniquely American movement called Frontier Labs. The way we think about this is philosophically and technically wrong. Philosophically it's wrong because it's not doomer versus not doomer. It's a hyper religion of hyper optimism. They believe all problems, present, past, and future, including the ones they create and don't acknowledge they create, are going to be solved by them, including human nature, disparities. On the infrastructure side, enterprises are just fed up because they know this doesn't actually work this way and it's not working. That drives our commercial business. On the government side, we are super pro-American, super pro-West.

I don't want to be overly naive. I am trying, sometimes with success, sometimes without, to move people that the AI infrastructure revolution is a unique opportunity for America because we've developed almost all of it first. However, it cannot be discussed in blue-red terms. Currently, the biggest danger to America is people saying, 'Oh, this guy agrees with me politically, therefore he's right. This woman disagrees with me politically, therefore she's wrong.' This is a massive revolution and there is a danger. I've been running around telling people for the last 6 months in private and somewhat in public that nationalization is coming. The reason why it's coming is you can't do a blue-red debate. This is a massive revolution and there are opportunities only America has and there are dangers in this revolution. For example, I got in trouble for saying not saying we would have a draft. I wasn't saying that. I was saying we're in a massive time of dislocation. In dislocation, we have to find ways to come together as a culture. While it is the case in Silicon Valley that everyone thinks we're going to magically like each other because there's abundance and no one has a job, unless we're ending history, it's actually very much the opposite. We're going to have massive resources, but they're going to disproportionately go to people who are already wealthy. That is a political problem. You can't just say, 'Oh, we're all going to have more, therefore we're happy.' We're not at the end.

We have to be frontal about issues that are real and getting worse. Wealth accumulation is totally fine. Even though I am actually a card-carrying progressive, I don't think card-carrying progressives have anything against... The question is not whether I'm progressive. It's whether some progressives are progressive. I am progressive. I want poor people to have a better lifestyle, higher standard of living, all poor people. I got in trouble for mentioning that 37% of our GDP is female. 50% of Americans, roughly 52%, are female. There's a dislocation there. 67% of people who have gone into graduate school are female. These parts of the market are going to be put under massive pressure. It doesn't help to pretend we're not going to have these problems. The progressive position is progressive because you speak about problems. Can we protect our workforce? Can we protect our workforce and have great migration the way we used to? Can we protect our workforce by teaching people the same skills we used to in the past? We have a massive program at Palantir to help neurodivergence. Not because I'm particularly a good person on that score. I am dyslexic and some people think I'm neurodivergent. But because those people are particularly valuable for our economy. What made someone come to America from all over the world? They were neurodivergent. They thought, 'It's not working out for me in my culture. I've got to come.' The point is we have to somehow get everybody around people who totally get it. When Bernie Sanders says, 'I'm going to take 50%,' you know what? If we keep going, in 2 years, they're not going to think Bernie Sanders is progressive. They're going to be like, 'Bernie Sanders, you only want 50%? What is this 50%?' We're absolutely headed in that direction. The most important political decisions in this country are going to be driven by: Do you understand AI? Do you understand what it's doing? And as importantly, are we going to tell the American people, 'Don't believe your lying eyes'? Politicians in the past have done this. It does not work. The American people are really wondering what is going to happen to them. The answers aren't all good or bad.

It's not the job loss; it's the mass needing to retrain and retool and do things differently. By the way, this is an incredibly important American advantage. I spent half my life in Europe. The biggest advantage we have is our dynamism, our ability to change, our ability to rebuild. But we're going to have to use that advantage because the jobs of the past are going to be very different than the jobs of the future. We're all going to have to change, and that's going to disproportionately affect people. We need a common purpose. What is our common purpose? That's something we're really struggling with. When you have these struggles, people go to the far right and the far left, and they're very charismatic. We should not make the mistake Europe has made by not addressing these things. Then the only people who win politically are people who may not know what they're talking about but are very charismatic because they're addressing a real issue. In this case, the real issue is Americans are afraid. By the way, the other real issue, and I'll talk about somebody I've spent 20 years in enterprise: we work with the most important governments. We are controversial, quite frankly, because we power every single Western power that's at war. And by the way, we're also controversial because they have capacities they didn't have in the past, and our adversaries... The commercial thing: enterprises are also unhappy. It's not just the man and woman on the street that is unhappy with the Frontier Labs. In private, every single enterprise we deal with is unhappy. We have this thing inside Palantir that's somewhat controversial, but we keep debating: should we pay people to go to large language model companies before they come to us? Because people come out of there screaming, saying, 'This could never work for me. They don't understand the enterprise. They don't care about my enterprise. They want a token max.'

It's a real question that no one in enterprise factually is worried about. Investors are really into it short-term, have no idea what they're talking about, long-term end up being right. We will just debate this with the facts. I've spent all my life dealing with the most complicated, most interesting enterprises. I'm on the ground floor in that, probably like no one else. Most of them are chillaxing over their latte, reading a report about something they don't understand the technical capacity of. But they're hiring the kind of engineers that you have. Those kind of engineers are great engineers. I'm telling you they don't talk to the enterprises or don't understand the technical challenge. Also, the ability to be a great investor in where large language models are more useful is probabilistic and you don't have to get better than 51%. If you want to manufacture a car and you need a part, or you want to send a rocket to the moon, or you want to put a missile on your adversary's head and bring home Americans safely, that stuff doesn't ship. There's not a single high-end enterprise like that that would ever put... That's before you get to the cultural impasse. When you go to San Francisco and talk to them, their basic vibe is, 'We don't have to solve your problem today because tomorrow you're going to go away and all of your problems are going to be solved.' It's largely religious. Then you get to, 'We're going to replicate Palantir by doing a deployco.' It's a complete farce. The people who go there to solve the simplest, easiest problems sell tokens. They don't understand, honestly, how unlikable they are. The product doesn't actually work the way it's very expensive and you have to overcome the fact that you may not realize outside of San Francisco... Actually, some of the best and most interesting conversations I've had in business are hanging out with some of them, Sam, Dario, others. These are high... You talk to Dario, it's a great time.

Well, most of the things they talk about in public are running on Palantir. At this scale, we're a nation-state basically. They are... I like it. He is the frontier person of the leading frontier model company currently. He's gone from being way behind to being ahead. He's a very, very important person. He believes what he's saying. I would say that he believes all problems of the current, present, and future are transitionary to a perfect future. I believe that we need heaven on Earth, not heaven in 20 years. We disagree on these things. But that doesn't mean we don't agree on other things. They agree on certain national security, we disagree. It's not as simple as agree, disagree, bad, good. These are some of the more interesting, important people in the world. I'm betting heavily, and my customers are pretty much on my side, that it is not that large language models aren't crucial for the world. It's just that the implementation is where the value is, certainly in the next 7 years. If you want to ask me what's going to happen in 10 years, I don't know. But the risk of a pathogen destroying us is probably as high as...

Well, these low probability events that shape the world are often underestimated. They are events that require software infrastructure, implementation, ontology, and large language models. As an example, if you have a deadly pathogen come into this country, the signs of this are going to be very similar to the signs of finding somebody moved a truck by 2 cm and that means they're going to attack. It is only possible with the security around it, which people often underestimate, that we do. That is the bedrock of why Palantir is indispensable. You can't just use data sets in this environment. Each single data set, you can think of each micro part of the data set has to be its own sovereign data set. We do that at the millimeter level: which part of the attribute is sovereign to the person, what can be shared, what can be imputed, what can't be imputed, what can't be generalized and shared, what has to be shared, which experts can see what under what conditions, is the measurement accurate, how long is it accurate, under what conditions is it accurate, who does the implementation, do they have the clearance? All those things have to be modeled and done in a platform that's been doing this. By the way, the cyber threat: you can't just run large language models. How do you patch what you find? How do you patch what you find at scale? How do you... If you could do all that, how do you have the accreditation to work on a secure network if you've never actually done that before? No idea what's involved and you need the accreditation to do it.

I'm very optimistic for the future. I root for every entrepreneur. I'm rooting for Elon. We partner in some areas and probably don't compete in others. When a massive IPO happens, first of all, we're so blessed in this country. I spent half my life in Germany. If there was an IPO, DPO, or anything at the $120 billion range, the whole country would be jumping up and down. Here we have a country where it appears to be one of the largest, if not the largest IPO ever. It's super cool technology, things that no one else has done. I think most Americans, and certainly I hope all entrepreneurs, are rooting for success.

It's very similar to these other debates. People underestimate that bullish on space if the person doing it knows what they're doing. Bullish on Elon and space? Yes. Bullish on space? No idea.

Thank you.

Yeah, you're in great demand.

Great demand.

Glad I got on here. Bye.