Palantir CEO Alex Karp: "It's Like No Glove, No Love. No Ontology, No Love."

Opening and the open-source moment

00:00I'm here with Dr. Alex Karp, CEO and co-founder of Palantir. Thank you for taking the time to do this. >> By the way, I always say this, it's you know, you were you we we've been together since the beginning of our life in the public markets. Um when we first uh DPO'd, um which is why so many people in America and across the world got to participate. But, you know, without you, you know, sending out missives about how we were misunderstood, you know, probably millions and millions of people would not have participated. Um and yeah, it's Palantir's this kind of crazy walled poster child company where people can only understand the Palantir of 5 years ago, not the Palantir that were we have. And so, you know, there was a lot of

00:44work to do on the, you know, front line. I'm happy to have you with us. >> So, 6 years ago, you went on CNBC the day you guys did DPO, and I think a lot of people didn't really understand what you were saying. About 2 weeks ago, you also went on CNBC, and I think you changed in about 20 minutes the entire valley's approach to thinking about front-end models versus open-source models. So, quite simply, why is open source so important for Palantir? >> Well, it what happened, by the way, that was like it was a revolution. It began with like work we were doing with Nvidia, and then going on um on on on television like that, it just exploded in a way no one could have really predicted. I think the

Why enterprises are frustrated with frontier labs

01:29explosion happened because you know, people are being sold this thing they know isn't quite true, like you know, and then then they paid a price. You're paying, you know, a lot of money if you're enterprise for for tokens that are really it's hard to see the value. And then you realize at the same time the alpha of your business is migrating to the person you're paying an absurd sum of money based on value. And then, last and least, they're just behaving in the public sphere in a way that is blowing up the politics of America, making people really afraid, leading to people just wanting to pillory wealthy people but not help poor people. And you're sitting there and saying you're basically like,

02:13"What is going on here?" >> Right. >> So, you have that backdrop. Then, you have And by the way, part of the reason for that backdrop, just to make this a little longer, is these models are excruciatingly effective when used with an application layer. Now, they all know that. When you use it in application layer installed with what everyone's now copying from us called FDAs for X. So, then I think part of the reason this revolution actually happened is technical participants in the market know one thing. Palantir gets it right. You think they're crazy. You think they're insane. Carp is obnoxious. And then it turns out you have to copy him. Right? So, now, investors who are kind of just chillaxing in front of their

Ontology as safety and alpha

03:01screen producing AI slop on their computer and thinking I'm going to get rich and yeah, maybe core people, no one else is going to get rich. They're like, "Yeah, but that's just how the world works. You're good for me." They've been a little bit tone deaf to the technical thing. So, they're very happy to watch the the the like the frontier. By the way, I always feel like I have to say this. You know, >> Yeah, people think you hate the frontier laughs. I don't think that's your point. I think your argument is >> Well, people actually think I'm I'm going I'm like I'm kind of not being like I spend more time defending Dario and more effectively, honestly, cuz they don't even talk to like if you're telling them

03:37the truth about what you think, they don't talk to you, apparently. Because like the views there you know, and like I'm always trying to explain to you, you know, it's a worldview. This is a essentially a world historic figure. He He really did take the LM revolution, bring it to enterprise in a way where everyone found out something is going to be real. Now, of course, they don't talk very much about almost all the values being created in ontology with their models where they're doing good. >> Um so let's get into that discussion because 2 years ago you made it very clear to everybody that the models would be commoditized. The frontier labs have gotten better and better and better, but with the rise of open source, which I

04:16think is kind of in your point and video's point, the cost of tokens is coming down. Which means the application layer, something that processes, orchestrates, governs the tokens is becoming more powerful. >> You know, you would need compute so that Nvidia primarily um and we're very strongly partnered with them. You you would need an application layer because that while the and you need a way of installing them. So, you need FTEs, you need an application layer, people generally call that an ontology now after maligning it everywhere uh because they you need to do it's and there's two points. Simply put, there's safety. So, you know, we've now we're now helping people and I think people are aware of

04:52the dangers of just, you know, like look, they you know, a lot of our clients are running around saying no no ontology, no love meaning no glove, no love. So, like it's probably not exactly what I can say on TV, but >> It's a famous last word. >> It's like they're like, yeah, they're like, "Hey, look, no ontology, no love." It's like it's like it's like gone it's gone viral in corporate America. It's like it's crazy. It's like it's so there's like but there's a safety function like if you're meeting someone you don't know and who may have a history of doing things and leaving certain products behind that you know, you can't get rid of, you probably be insane not to have ontology protecting you. So, that is the

05:31safety function and that's that so that that's crazy important. By the way, with models coming from the whole world, that safety function is going to be even more important. But in the end of the day, businesses need to expand their tribal knowledge, their alpha while protecting it. And so, then there's another term which is now being adopted essentially the orchestration. How do you actually deploy the models so that they can do exactly what they're capable of doing while not not providing like chillaxing swap? Uh so, you get >> Why is open source plus the ontology so important for orchestration? >> Okay. Well, within the Well, first of all, it's because that's where you get the maximum convergence between pro safety because

Open source and orchestration

06:12you the open source model is is is not aspiring in any way, obviously, to export although you do have our, you know, whatever protective layer probably. So, um >> That data is not going back to Frontier Labs. >> So, it's essentially but then you also the it allows you to fine-tune the model. The fine-tuning is very hard to do without dealing with the Frontier Lab. And when you're fine-tuning, you're basically exploiting what is the alpha you're trying to create on top of the alpha that's deep in your business. What is the context of your enter >> What is the context of your enterprise? What is a specialized problem >> sets embedded in your data that you have a unique lever and insight into way of

06:49solving. So, if you're building, I don't know, you're building a drug to reduce the chance of cancer, like you are actually building it on proprietary data and proprietary insights on how the human body works. >> Right. >> Right. And that like so, or you know, in Battlefield's obvious, but every single enterprise that is actually valuable. Like how do you define an enterprise that's valued? It has a specialized way of doing something a consumer wants or in the government context has a specialized way of doing something that the end user does not want that protects your people from the end user. Like this so, that's like this is so, that that to do that most efficiently, you would need compute uh ontology both

07:34protective and alpha augmenting and creating. And you need to make sure it doesn't transfer you that own those weights yourself. So, this by the way, this is not speculative. We we've seen this already in the US government context where in context that is very very hard to use frontier labs or where there are other quite firm. I don't understand this. I'm willing to go like By the way, in what world does the Department of War, the only institution in America that is extremely high credibility, the institution that integrated before anyone else, that protects us every day, that protects, that allows us it to export our technology, in what world are we outsourcing the moral veracity of that to a bunch of people sitting very far

08:18away who do and they're even met a man or woman in uniform and it's presumably don't care if they come home alive at least in practice. So, there's also like who makes these decisions? By the way, Palantir we've always been part of the reason the sovereignty thing actually works is technical users know for whether it was PG, Foundry, Ontology, Maven, uh now sovereign AI, we've always been on the side of protecting individual rights, protecting data, and protecting by logical extension your sovereignty. So, that like we that that's actually where we should all be as a country and as a tech community. And then there's the reason though that this ended up being you know, now I should say I'm wildly

Sovereignty and the Department of War

09:02happy you have a lot of people entering and taking, you know, cuz this is actually where it should be and we don't like just cuz we basically are have been at the forefront of launching many of these revolutions and it's commonly known in tech in technical circles that we do this well. Um you know, building a broad The reason you see this broad phalanx of a coalition is the customers this is what they want and need. They are livid at the option that was provided and this this way of doing things. In fact, currently the biggest problem at Palantir is just how do we scale this? >> So, so what is but there's a lot of customers here. By the way, the backstory here because of this is you get the backstory.

Customer response and scaling

09:43>> Like usually it two three years ago playing AIP like >> Oh, I remember. >> Like playing AIP >> This was like, "Hey, we're doing something really cool. Let's just invite people." And I mean like but but and look at the people out there. >> Yeah. >> It's like they care. They they care. It's like they want to learn. We are actually partnering with them. They're from every part of industry. They're the among the most important industries in the world. >> So, so what has the customer respond? You're on the front lines with customers every day. What is their response been? >> I say Well, it's it's um it's it the response to I would say the open source as a part of the sovereign narrative is a colossal It is what It's not only what

10:23they want and need. It's like Imagine again if someone is you're using this thing you just have a bad feeling. And then And now like now you understand why you have this deep bad feeling and and honestly in many cases that's where I end up defending uh there's anger. So now it's like there's like a valve release. The problem we have as a company is how do we stay very focused like we're going to We just have to I get calls from all over the world. I'm like "Can you come to my office?" Like we We're just like we we're we're rebuilding internal dynamics a lot here. Uh rebuilding infrastructure, you know, this sovereign AIP like requires us to build out certain parts of mythology, to build out our ability to fine-tune

China, open source, and being pro-American

11:07models, to service a demand both in the US government and in US commercial uh that we can't really fulfill. We're also doing a lot of educational stuff. >> Yeah. >> It's Speaking of education, one thing that is I've heard a lot is a lot of these open source models are coming from China. And somehow they're like, "How is Palantir endorsing China?" Can you give a response? >> endorsing we're endorsing what what Well, first of all, we obviously are I mean very pro-American. You see we're partnered with Nvidia, a great American company. Um, the way Look, there's a there's a template of how not to do this. >> Yeah. >> It's called Europe. Europe basically tries to close its borders. I mean, there are all sorts of

11:47reasons how they It doesn't now put this way, but de facto, they're closing their board trying to close their borders to US tech, right? That The problem with that, leaving everything else aside, is they're screwing European tech. >> Mhm. >> You must be able to compete. Like, we I believe that we will we can deliver. I've seen this I in practice frontier surpassing uh alpha creation. Where we're honest about what models do and don't do safely. >> Correct. >> Right? And that's being done with a completely American stack. >> Which means you can download the open-source models, no matter where they were created, but run them locally, which fundamentally changes the game. >> I Yes, but I mainly we have a strong

12:30preference for American models. Some of our clients use models from all over the world, including China. I do believe if you're using any of these models, you know, I mean, I I I this the I can't go and disseminate, you know, no no Palantir, no no ontology, no love, because it may be not be right for the TV audiences, although who knows what I'll do. But um but yeah, it's going to put obviously it creates great demand for our products. So, since I don't want to be out there like I I'm cognizant of how this like is going to help us, obviously. But But But But But again, you you have to be very careful. Like, you can we should obviously scrutinize all models. There needs to be better oversight, no doubt.

13:09All models. But you do not want to do, if you do this, what Europe is trying to do, what you get are weak companies that nobody actually believes can compete. And that's So, you're screwing the people you're trying to help. You know, business is a weird thing. You can you can nudge, you can have preferences. If you do overt like that level without without being able to show something, you end up with industry that is and America is the dominant player in the world. Um, and so I think this will force a lot of honesty because if you are doing compute ontology, uh, actual product-based and technical orchestration of and FTEs with open model models, open way models, that's that is by far the best thing on

13:53the market. And yeah, the other things are, you know, so yeah, it's it's it's going to be very good for America and I don't think it's going to be I think it might actually be pretty good for Palantir, too. >> Last question for you, Dr. Karp. Again, thank you for taking time. Uh, a lot of people are wondering, is there actually going to be a light at the end of the tunnel when it comes to this entire AI revolution? Put broadly, what does it mean to be a enterprise software company that actually delivers value in the age of AI? >> Oh, you mean what does it mean to be Palantir? Uh, look, >> Well, we're we're It's a better way to put it. >> we're we're we're we're Palantir is at the way beginning of a revolution. Like I was I

Palantir's future in the AI revolution

14:32was just uh, two two years ago we're sitting here, we have an offsite, people are like, "Tell me something you believe and no one else believes." And like, we're obviously going to be a trillion-dollar company. Now, I think we're at the way beginning of a massive massive opportunity and there is no, like, it's very hard to know what the end point is, but I can tell you I can see American industry just accelerating forward and I think Palantir is going to play a disproportionate role in that simply because you can kind of fake like you've done these things before FDA application layer. >> Uh, terms that everyone start using now. >> It's like, but it it is yeah, but it First of all, it's it's it's the

15:15orchestration of all these things at the same time. >> Which is art. >> It's art It's much more art than science. You also you need deep access to these data sets and it working you need a knowledge of how to work within these institutions to expand them. They have to have seen your products actually work. Also, the software concept concept is very broad like it includes AI slop. It includes things that are obviously going to be replaced and then it includes things that like most of what Palantir does which are like essentially like primitive they're more like infrastructure you need to do to enhance what these things do and again, you know, the the demand for what we we do is beyond what we can supply.

15:56>> Thank you. Thank you. Thank you.