Palantir CEO Alex Karp on Tokenmaxxing & Taste

Yeah, yeah, get in here, Klaus. This is good. Okay, uh, how is it going? How is AIPCon this time around?

What's changed? Um, well, we're in a phase. Yeah, all each one of these things like marks a time. First of all, you guys are even more baller, more successful.

Thank you, thank you.

Some tendies in your pocket.

I think it might have been part 2. We gotta say thank you.

You know, it's like you blew us up. You're looking bigger and stronger. Hey, are you more attractive in your personal life now, randomly?

Well, here's what we're actually focused on— dead hangs. Yeah, yeah. So you came on last time last time you said your dead hangs around like 5.

Oh no, it's, it's— well, it's plateaued in the last couple months at 5:30. 5:30.

Okay, so we— the thing is, like, people are gonna hear that, they're gonna think hanging on a bar 5 minutes. You gotta go and do it. The audience has to go try to do it. We've started doing it. We're still in the under— we're still between— yeah, between 1 and 2 minutes, somewhere around a minute 30. You feel like your tendons are gonna rip.

30, um, 1:30 dead hang is respectable. Okay. 2 minutes is super elite.

It doesn't feel respectable when you have the 5-minute number. You're looking at the time.

Thanks, Strength Matters. Now, um, the, the thing is, I don't want to go into rabbit hole and training. The single biggest mistake people make is they try to hang every day. You need recovery. It's like anything else. So if you want to mimic and get progress, you just do what I do, which is once a week you hang as long as you can. Doesn't have to be super macho. And then that's your day. So like to say you can do 1:30, but multiple sets. —No, no, no. One day a week you do your maximum. —Max? Wow. So let's say you could do 2 minutes. You try to do at least 1:30. You fight to get to 1:30, but you don't fight to get to 2 minutes. That's your dead hang day. And then you can basically fuck around the next day, do whatever you want. Don't overdo it, but you could do 2 times 1 minute with a long break. And then you just screw around, do less and less and less. 2 days before, if you had 2-minute dead hang, you do 4 times 15 seconds. The day before you take off. And you do that, just keep doing that. And your day— what the mistake people make is they hear my baller time, they gotta go to the Crown.

Fuck that guy. I mean, the mistake you're making is not doing a course. This could be a whole new revenue line of pounds.

You guys have it for you guys. Yeah, so, um, yeah, I mean, the dead hang is that it's like— and also some of it's just genetic, like My other metrics are elite, but that this is somehow alien territory. God-given gift.

What about, uh, what about breath hold underwater?

I don't do that, and I'm not sure I have it. Like, I grew up swimming, and I think I'm weaker at that. Like, I bet you I'd be in your guys' range. I'm a divemaster.

I can hold my breath for 3 minutes.

I'm just saying, I'm just— yeah, so I think, yeah, you'd be like— I think you'd be crushing me on that, honestly. But you have like the lung capacity of a whale. That's true, that's true. I mean, like, yeah, you got like— I'm not moving, I'm not using any oxygen. It's like you're like, like, you're like a, like a like a whale floating out there in the ocean waiting to surface. It's true. So I'll tell you the difference. God, they're always minding me out there. But like, okay, when we first met, it was like, AI, maybe real. Then I would say somehow until about 2 weeks ago, there was like a, holy fuck, this is real, but somehow it's not working, but we're not allowed to say it publicly because we'll look stupid. And then there's a lot of investor hype. There still is investors printing 10Ds. So you have the investors on one side. I think people realize it's real, but it's like you have the whole token maxing. And people are on to that. And then so there's a whole value lecture there. You have a political situation where people who do not understand basic economics are winning the political argument. So you can, you could, we could talk about where AI is.

Yeah, there's a lot here. Let's break it, let's break it up. Let's start with the token maxing thing. Let's start with what's, what's real. Uh, how are you actually thinking about deploying AI?

First of all, what is your, what is, what is Palantir's philosophy around token consumption?

Well, like we, okay, we have a product that will allow you to pay. I mean, internally it's called, uh, something, but externally, but really we call it the demasturbatory get-off masturbation thing internally. It's like people are just sitting there all day kind of like a porn addiction, and enterprises are like, okay, we knew this, we believe this will create value, but we cannot have people just like some people—

Checking the weather with it, just rearranging deck chairs on their personal Titanic.

It's literally like porn. People are full-on—

It feels so— Yeah, tool-shaped objects, right?

Yeah, tool-shaped objects you're looking at more than you want, you hope no one notices, you're kind It feels productive to have every email classified.

Here's what it comes down to.

Business problems can never be— I mean, sometimes they can be solved purely with money and just spending more, but very often— Actually, I think it's the opposite.

So just to give you a weird analogy—

No, and I was going to say very, very often it's more the opposite where it's about figuring out the right way to do something and then you can use capital to fuel that process.

But let me give you a thing that's too generous for you guys. Okay. It's taste plus money. Okay. Yeah. And there is no, like AI, like if you look at, like pick any issue you wanna talk about. Token maxing. Maxing. What's going on with deploy codes? Are other people gonna build ontologies? Why does our political class not understand AI, especially in Europe? It's like, yes, because all these things can be scaled in a very valuable but largely going to commodify way, but you can't scale the taste of like, what is the business problem you want to have to solve and need to solve? At the end of the day, whether it's the Ukrainians fighting, the Israelis, commercial entities, there's somebody sitting there who's like, okay, but this problem is valuable. This problem isn't. And once that value, that problem always has, that problem almost always, but not always has attributes. So there are some problems you could solve with this. I want to write a report on GDP growth in China, right? OK. But if it's a problem that requires a knowledge store, like I want to understand the specialized way I underwrite— we're going to have a guest here— I want to understand the specialized way I drill for oil and gas that's both legal, ethical, and reduces the cost of production. I want to change the supply chain of my industry, whether that's military or whether that's building boxes or whether that's cars. These things require actual precise ongoing processes. They are enhanced by large language models. They are not replaced by large language models. And then you get to security issues. Like for us, like the whole mythos things is it's just a boon because like, yeah, we could take any model, their model, OpenAI model, open model. We can identify, we can now identify vulnerabilities at like 10, 100x. Yeah, but then who patches them? How do you patch them on-prem? How do you patch them on-prem so that your specialized knowledge stays on-prem? Like if you're any business or intel service, a lot of these things are very similar. Sure, sure. Like you're not putting your classified data in a public cloud. Same thing if you're like, you have a special way of farming soybeans. Well, you're not. So it's like, how do you have, how do you get, so all these problems are exposed, identified, And then you always have a thing of where's the charisma, which people really underestimate. And it's not global. There's no global charisma now. So right now, the large language models are very— frontier companies are super charismatic with investors. I'll give you some news. They're super not charismatic with enterprises and the people. Like, in a way—

Really? Even with enterprise?

No, no. I mean, it's— Because I understand with the people. No, no, the enterprise people, I have a secret. Like every company has a secret way of selling. You know what my secret way of selling is? Don't even call. Don't come talk to us. There's a frontier company. Go spend 2 days with them. And if you're lucky, after you're done, I'll let you in my door. They're like clamoring. They're like, hey, I'll take your bad brand. Which we have a great brand in enterprise. But like, it's like secret knowledge because the investors love this. They're like, hey, my stocks are all up. Everything's up. I mean, Palantir has done very well. It's like, and you know, you guys are doing very well, I imagine, right? And it's okay. But I'll tell you what, you go down the street, you talk to a Marine, you talk to a bus driver, you talk to the person who owns the bus driving company, they are not happy. They do not like these people. They're tired of people token maxing. It looks like masturbation that's cost them money. And honestly, then you have something we're not allowed to talk about in this country, likability. Like, at Palantir, I think we have like 50, 100 million global fans. We have like 5 million people that wake up in the morning literally calling me Satan. I didn't know I had that kind of warm hand. But, you know, it's like, that's what they believe. And they really believe it. Okay. What people are not allowed to really address is like, we have fans and enemies. Yeah. Yeah. We're polarizing. Yeah. We're polarizing, which means both sides. These people have one side. They're just— it is. So it's like you— and it's like it's a really—

social media companies too have the same problem. Yeah, everyone uses them, but no one likes them.

Yeah, but then they also live in a circle, and that circle's printing money. Yep. So it's like, you know, when you look in the mirror and you just printed a lot of money, you look pretty fresh.

It's part of it. Is, is part of it that, that some element of the technology, let's just say LLMs, is so magical that the companies involved, that the companies that are making and selling Frontier Intelligence can be bad at a bunch of other things and still great. Well, no, no, no.

They are magical at a certain kind of thing, allowing you to write, for example, code. Now that code doesn't, can't be used as a knowledge store. So if you look at code in like 3 different ways, like just using Pounder's model, we have code that's basically infrastructure. So what are the Ukrainians using? What is the Department of War using? What do a lot of our enterprises, we call that primitives. It's basically hard-coded things that, that understand the world. What do you do? It would take millions of technical hours and an understanding of all these enterprises to do it. So it's much more like how do you build a steel beam? Then you have code that is written by FDEs. Okay, so that's kind of managed. The reason why FDEs work, the secret is it's actually managed in something that we as a product. So you're writing to a code base, we're managing that, we're increasing our product. It's not just random people writing. Then you have, let's call it free code. That free code is, that's magical. Like you can do it very quickly. It's almost right. It doesn't have to be exact. Dashboards. Dashboards, financial stuff. Little flow. Probabilistic stuff where you just have to get— One-off analysis. It was magical. By the way, it's magical. Not only creates, and it's magical in a way. I know people don't like the porn thing, but it's also addicting. It's like, you know it's not good for you, but it may lead to damage. One more dashboard. One more time. It can't hurt that much. I know my doctor says I shouldn't do it. But it's like that, right? And you just keep going. And if you're involved in that thing, you're also making money. And then last not least, in certain circles, if you want to be a researcher or you believe— essentially, it's a religion. And one of the things, it's very charismatic, especially to people who've never had a religion, because all of a sudden, that hole in your heart that was yearning for, I don't know, I would say, hey, established religion, Judaism, Christianity, Islam is like being filled. Yeah. And, and all the answers are there, but it, it's very, very successful at doing things that a company has to do, but it is not actually solving the problem that enterprises are. It is now, it can solve them. That's the trick. It's not, it's not binary. It's not like you can't say they're not valuing the, they're totally putting our business on steroids. Mm-hmm. Like without LLMs, Nobody would be talking about our ontology, about Apollo managing secure exploits, about our ability to manage an enterprise essentially, turning all these companies into FDEs. These deploy codes, we love them because now every company wants a deploy code. You know how you do that? You re-platform on Palantir. And it actually works. It's not somebody with no taste who's never done enterprise, who has no earthly clue how these things work, who's done something else and is just imagining they know how to do it. Right. Yeah.

As part, as part of this moment, quite entertaining for you because you guys have been working on understanding businesses at a deep fundamental level, creating, you guys have effectively been doing the work that it, that people are promising AI could do for 20 years now, but actually doing it, finding all the really rough, finding all the really rough edges. And, and not— and being at a point where you don't have to oversell the technology, you can sell both things. But now there's maybe— here we go. We got it together.

Now there's maybe— oh, it's the wrong side.

That's why.

Flip it around. All right. Go. Dyslexic. Yeah.

There you go. There you go.

There you go. Hopefully we got that. I don't know. All this stuff was recording.

I trailed off, but— No, I understand. Is part of it entertaining to you that it feels like, you know, Palantir has always been in some ways, um, not had competitors because there's nobody with Alex Karp running a company that does what Palantir does besides Palantir. But at the same time, there's been tens of billions of dollars deployed now to effectively do what Palantir does, but just selling the intelligence part, not selling all the underlying kind of infrastructure that—

well, they're doing two things. They're selling— they're trying to sell the intelligence part, and they're trying to pretend if you just hire a bunch of people and let them run around their FTs. Now, the, the very cool thing is when you've been in your basement doing your thing and everyone kind of uses it as the freak show, it, it's really interesting and, and great to have adoption. The pretty ironic thing is half the people adopting now don't even know they're copying. But now the copying thing, it helps and hurts. Where it hurts is in the beginning it puts clutter in the market. Yeah. Um, and there's, there's no doubt about it. Where it helps— and then we saw this with defense tech, honestly. So like in defense tech, we were the only people, we were the first people, despite what— it's— I, I love these, honestly, other podcasters, they're interviewing people who are parroting things I said 20 years ago, they don't know it. And it's like, oh, that's so insightful. It's like, yeah, of course it's insightful. Karp said 25 years ago. Uh, and like, but it says— but so that kind of— that part is super weird. But, uh, and like, but, um, but it's— but what really happens when we see is like it expands the market. So like in defense tech, we would not be doing this well in just purely in government unless there weren't 50 companies that were doing similar things, because then the people are like, okay, first of all, you view it as like off-balance sheet sales resources where other people are basically— well, it's off— well, no, that, that's the large— like, it's— they do treat 2 things. They increase the size of the market because de facto nobody wants to find an underwriter market where there's only one person. Sure. So like, if you're the one person, the percentage of the defense budget you can get is much smaller. And 2, they set up a comparator. It's like, you know, you may not like the freak show, okay, if you like, but have you noticed the people who are serious buy it? And then, then it changed. And then 3, it changes the standard. Now What you're seeing now is like that times 100x. And it does change recruiting, retention, and how you build a company. And we're always thinking you have to think about how to— being dyslexic, huge advantage there because you don't have a playbook and now you need things to shift and we're doing that. The central thing though that it just cannot be developed, even if you understood the playbook, a lot of these things appear like it's like, LM code appears like Palantir code but isn't for a deployed thing, appears like Palantir, it isn't ontology. You could theoretically copy parts of it, but they're essentially structures that are built deep into organizations that we own. And by the way, take you 3 years and then 3 years we're in a completely different world. But there is this magical thing called taste. In the end of the day, the reason why you guys have done so well, of course there's aptitude and diligence Showing up and all those things. Yeah. But you have to be able to differentiate between two people who are in business, one of whom is saying something that sounds weird, that is insightful, one of whom is parroting something that sounds weird and that's all they're doing. Yep. And a lot of people, very few people can do that. And you do the same thing, like the enterprises that succeed, there is a taste arbiter. And at Palantir, we have taste, we have taste in every product. Taste in every deployment, taste in every casting. Who puts the people there? How do you put them there? How do you organize the thing? Our ontology then does that technically. How do you manage the whole org with taste? Who should be in charge? What data sets should come in? What are the ways in which you protect? What should you push into the public crowd? What should be on-prem? What should, I mean, leaving aside the law and like wars, war ethics, what do you want to protect? What should you protect? What should you not protect? Quite frankly, you want that to be out there so you can get more data. All those things are arbitred by taste, and then you have to have the credibility of having taste. That's a real problem for a lot of these places, because they don't have— they're popular with their friends. They really don't understand how unpopular they are in enterprise. They think it's like, oh yeah, it's like the way I think I have a problem with professors at Columbia. It's like, no, it's a real problem. They think I'm Satan. It's like, I think we grew up in the same community. Let's talk about Heidegger. They're like, they don't want to talk about Heidegger. So it's like, it's like, yeah. And so that's just a, it's a weird thing. It's going to be a super, the one thing I would say for anyone listening, if you're listening to this and you're chillaxing and not active, I'm not saying you have to agree with me politically or anything. Yeah. They're the, like, partly because of this dynamic and very self-inflicted, because I, I tell you, I can't name names. I called many of the titans of this world. And like started this 6 months ago, like every couple days we're going to be nationalized.

You call them every couple days?

Like some of them I'm like, yeah, we're going to be— I mean, you know, it's like, honestly, they're like the bad— they find me very entertaining. Like, I'm not sure, like, so they call because, yes, like, yeah, it's like, oh yeah, this is going to be entertaining. You're going to pick up. Yeah. So any case, so I've been telling them for 6 months, we're going to be nationalized. Yeah, yeah, you talk about this. We're going to be nationalized. And they're like, why would anyone nationalize? Never happened in America. It's never, "Why would anyone nationalize us? We're so likable. We're creating so much value." Like, okay, I'm not going to debate that. I know how likable I am. I'm not going to tell you how likable you are. But I am telling you, and you know the momentum on this is on the side of people and nationalizing. And we don't get our act together and figure out ways we can say, "Hey, look, there are problems here we're going to deal with. These things are not going to— yes, they are going to create opportunities. You have to talk openly about how these things are valuable because we have adversaries." You can't just say these things. All that stuff. So the primary risk, honestly, to Palantir and a lot of these other countries is— and then it's going to be nationalized. Before nationalized, it's going to be regulated by people who don't understand this. And now they'll tell you in private, I'm working on this, I'm da da da, and this lobbyist is like, not going to work. So like, that's something like, if you're listening to this and you're like, look, you know, you don't have to agree with me on all my proclamations. I got a lot of— by the way, there's some people who think I'm saying we should have a draft. Too lazy to read. I'm just saying we should, like, in a world where everything is changing, everything is changing, don't we have to find some communal structure to remember we're American? You don't like my, you don't like my idea of like we all do a week in the park? Great, come up with some other idea. We can have no idea, you know? And like, and then they're like, well, I'm saying I do not want a draft, just to be explicit. They're like, oh, that's pro-war. No, honestly, you know what, most of our wars are fought because no working-class person is making a decision. You start making sure everyone is involved in everything, I'll see you have a few wars we fight. It's actually the anti-war position. But any case, disagree with everything. If you're— we have on the right and on the left people, uh, people who have no earthly clue what they're talking about. Right and left. All they're talking about is how much they hate us. And those of us who are sensible in the middle, you know, too many of us are chillaxing. Like, oh, like nationalization, it can't happen. America would never do that. Sleepwalking. Sleepwalking into— and you guys have tendies to protect now. You guys should be on the front line of this. Like, you got full— oh, I'm sorry. I have a full-on, very impressive corporate leader coming on. So I got to turn it down. Last question.

Last question if we have time. How are your conversations going with Fortune 500 CEOs around headcount planning? There's been so many layoffs this last year that people were saying, Hey, we're getting so much out of AI, we're able to cut back here or there. People inside tech often know, like, these— maybe there's just reduction because there needs to be a reduction. The org got bloated. Maybe they do need to fund some AI. Or it's a declining business model, like getting outcompeted by something. Yeah, the business just doesn't have momentum. But how are those conversations going? What does it look like?

By the way, I talk to Fortune 500 companies. I talk to unions. I talk to soldiers. I talk to fire— it— if you upscale somebody, they're more valuable. Sure. And like all these, whether it's people working on batteries, people driving trucks, people, corporate leaders. And again, this is where I think we have to be very careful to be more disciplined on the corporate side. Like if you run around saying AI allowed you to fire two-thirds of your workforce and you did it because maybe your competitor's kicking your ass. Yeah. That could— that is a really like, You might as well just go sign up for Bernie Sanders Manifest. And part of the thing is they really believe that can't happen, so they're free-riding on the fact that it could. Like, we have— and it just cannot work anymore. These things are very, very explosive. The American people sense that there is something dangerous here, and when people are playing with that fire, it's like they assume the fire won't burn their hands. That's not the world we're in. That fire is going to consume us. What we see, again, the warfighting example is just the most neutral, not for everybody, but like the soldiers at the bottom have gotten much more valuable. And I don't even just mean the special operators, which obviously they're in a different league, but like every— the people doing a lot of the operations now are doing our product. They're high school vocationally trained. You see this everywhere. The modern enterprise is going to have like we have a true like He very, very, very smart, uh, person coming on. And it's like, you're gonna have a very smart executive. He's much better at hiding it than I would be if I were him, but that's— you can talk to him about that. But, um, uh, um, and, uh, and then very talented creative people with taste all up and down the stack. Any case, I think this is time for me to— I think this is time. We, uh, thank you so much.

Yeah, great to catch up. It was fun.