This Week in Startups - Mission Over Money & The State of Defense Tech with Palantir's Shyam Sankar | E1908

Starting point is 00:00:00 There was a colonel John Boyd, very famous military tactician. He came up with the concept of the Uda Loop. Terrain doesn't fight wars. Machines don't fight wars, humans do. And the Uda Loop is observe, orient, decide, act. It's all about decision advantage. And the speed at which you can go through this loop, he initially developed in the context of dog fighting with airplanes, but speed at which you can go through this loop is the determinative factor if you can win or not. So your ability to go through the Udala loop to develop a drone that behaves differently and is resilient in different ways before the enemy can adapt to you. that's the only thing that's going to matter. So I would not place a huge premium on any specific

Starting point is 00:00:34 technologies, although there are ones that are going to matter. It's really gearing up and measuring ourselves, our programs, our investments through adaptability. And how quickly can we iterate on these concepts at the pace of the fight? And the one who does that first, fastest, is going to be the winner. This Week in Startups is brought to you by NetSuite. Once your business gets to a certain size, the cracks start to emerge. Things you used to do in a day, take away. week. You deserve a customized solution, and that's NetSuite. Learn more when you download NetSuite's popular KPI checklist absolutely free at netsuite.com slash twist. Northwest Registered Agent will form your company fast, give you the documents you need to open a business bank account, and more.

Starting point is 00:01:19 Visit Northwest Registeredagent.com slash twist to get a 60% discount on your next LLC. And Imagine AI Live is an AI conference where you'll learn how to apply AI in your business directly from the people who build and use these tools. It's taking place March 27th and 28th in Las Vegas, and Twist listeners can get 20% off their tickets at ImagineaI. Dot Live slash Twist. All right, everybody, welcome back to this week in startups. As you know, the world has been a little chaotic the past few years,

Starting point is 00:01:53 Russia invading Ukraine, flare-ups in the Middle East, a threat of China invading Taiwan, all of these things are bubbling up. And the world always has these kind of issues. Thankfully, we have a lot of great defense tech companies in the United States that are helping defend our nation. This is not always been the case because people in Silicon Valley, probably with good intent, didn't want to be involved in wars, didn't want to be involved in the military, until some folks, folks like Peter T.L. Alex Carr, kind of looked at the situation. The folks are at Founders Fund, of course, and said, hey, defense tech seems like something patriotically we should be doing. Palantir was the tip of the spear here in Silicon Valley when it was founded. Gosh, back in 2003,

Starting point is 00:02:40 again, I was a 20-year-old company to help with counterterrorism. And today, we are super delighted to have Shamsanker with us. And Shom is employee number 13. at Palantir, and he is the CTO of the company. Welcome to the program. Shum. It's great to be here. Thanks for having me, Jason. You heard my little introduction there. Things have changed a lot in terms of how people perceive Palantir, right? At the beginning, I guess, with a little bit of Peter Thiel's iconoclastic, you know, image and Silicon Valley's, I don't know, peace-knit, beat-knit cultural heritage being the Summer of Love here, the anti-Vietnam movement, all this kind of led to maybe Silicon Valley not being super interested in supporting

Starting point is 00:03:27 the government, specifically supporting the military. How has working at Palantir changed over the last two decades? It's changed tremendously. If you go all the way back to, you know, when I joined in early 2006, there wasn't even quite this sense in which we shouldn't be doing government work. I'd say it was mostly people were disinterested in it. And certainly investors thought there was no business to be built here. So it was very hard to actually. raise capital. Then if you fast forward a little bit, I think we started getting into the era where it was politically quite distasteful, working with the government became more quote unquote evil, and that was kind of a different era. And then, of course, things have changed quite a bit now,

Starting point is 00:04:04 I think particularly precipitated by the invasion of Ukraine, by Russia. People kind of realize that there are bad actors in the world, that freedom is not assured that history did not actually end with the fall of the Soviet Union. And in many ways, I think it's a reversion to the mean. It's where the valley came from. It's more authentically what the industrial base used to look like at the dawn of World War II. Yeah. I often say that at the beginning of World War II, there was no defense industrial base. There was just an American industrial base.

Starting point is 00:04:33 We really turned to automotive companies, electronic companies, to build the arsenal of democracy that ended up actually winning that war. In the early Cold War, Chrysler had a missile division. And General Mills, the serial company, used to be. to make inertial guidance systems for ICBMs. They made equipment in artillery that in mechanical division. Wow. I didn't do that. I think this present moment where we have pure defense companies, that's the aberration. The much more normal thing is you have an industrial base that makes things, some of which provide for national security and some of which provide for American prosperity. So let's talk about, I think, the first era of Palantir. And we'll get into

Starting point is 00:05:13 the second and third era as of, you know, broadly speaking here. Yeah. But it was formed after 9-11, when terrorism was considered very asymmetrical type of new war we have to deal with. Of course, it was very similar to guerrilla warfare, and terrorism's always existed. So I'm not sure exactly that was any revelation, but we'd never been hit on American soil, at least the mainland. And obviously, Hawaii got hit in Pearl Harbor. It was very rare for Americans to die on American soil. This led to a really deep scar and then a questioning of what are we going to do to deal with two people, four people, hijacking a planet, flying into buildings and just in a free society, how dangerous things are to live in a free society, how vulnerable we are. And as a New Yorker

Starting point is 00:06:03 and being there on 9-11, we witnessed, my lord, we have been living in a dream that we think we're actually safe. We are actually incredibly vulnerable. And that's what Palantir was born out of, correct? Was that moment in time? That's right. Yeah, exactly. And I think the real insight that I'd say Peter and Alex had, the founders, was that of course terrorism is awful, but also you said, you know, our freedom here. What's maybe perhaps just as bad or worse is the reaction to terrorism. How are you going to ensure that not only do you have security, but you have civil liberties too? And in that era, if we can remember what it was like in 2001, 2002, 2003, just how are you much, how vulnerable we felt and the sorts of tradeoffs we were willing to make. And that's the

Starting point is 00:06:48 role of government. Government needs to decide for a given level of security, what sort of privacy is available. But the role of engineers is to push out the efficient frontier. For any amount of security that society wants, you should be able to have more privacy than you could have had before. And so who's really going to work on technologies that both move the efficient frontier on how much security was possible, i.e., this software would be highly efficacious in finding terrorists, but also it would do so in a way that that honored data protection and privacy so that you could still have functioning democratic society. At that time, I can tell you, any person living in America was more than happy to have their ID checked, to go through a scanner at an airport, to have their bag checked. We would say thank you to TSA for the first five years. Thank you for doing this. Thank you for keeping us safe.

Starting point is 00:07:37 Because people really did feel like the other shoe was going to drop. thankfully, the other shoe, knock on wood here, you know, in some major way here in the United States, has not dropped. We've had lone wolves. We've had, you know, situations that have been thwarted, obviously. But knowing what you know, and you were the 13th employee, so I'm assuming you started in year one or two there, yeah? Yeah. Tell me about what the state of play was in terms of terrorism at that time. And then in your mind, how have we avoided another 9-11? And then what role did Palantir's products? have in helping that and had a 9-11 and that war on terrorism inform the product set. The core observation that people had after 9-11, they described it as it was a failure to connect the dots. And so on one hand, that can seem really easy. You know, retrospectively, we can go back and we can look and we can see that we had little bits of data that if we had just put them together, we could have understood what was happening. But often this data is collected with certain authorities.

Starting point is 00:08:35 The data has to be protected in certain ways. It can only be used in certain circumstances. is when you don't have sufficient technology, you end up having binary information sharing regimes. You can either see all or none of the data. And that's extremely unhelpful. You know, what you need to really have is a granular ability to say,

Starting point is 00:08:51 what data can I see under what circumstances based on the laws under which it was collected, the purposes and uses I have for this data, and that would allow you to connect maximally the dots in a legal and compliant way. So developing that technology infrastructure was absolutely crucial. Then on top of that, you know,

Starting point is 00:09:07 this isn't just about bringing the data together. It's really about helping the people who do this job, the intelligence community, law enforcement community, be able to ask questions of the data in a way that is a thousand times more productive, a thousand times more collaborative, so that you can get far ahead of bang, you know, well to the left of these issues happening and prevent them while the actors may still be overseas before these plots have really developed much at all. Hey, when your business gets to a certain size, the cracks start to emerge. We know that. It's all that stress from your amazing success. Things you used to be able to do in a day. Now it's taking a week.

Starting point is 00:09:43 You got too many manual processes and you don't have one source of truth, which is crazy. If this is you, you should know these three numbers. 37,000, 25, 1, 37,000? That's the number of businesses, which have upgraded to NetSuite by Oracle. You know NetSuite. It's the number one cloud financial system, streamlining, accounting, financial management, inventory, HR, and more. 25. NetSuite turns 25 years this year. That's 25 years of helping businesses do more with less, close their books in days, not weeks, and drive down costs. One, because your business is one of a kind. So you get a customized solution for all your KPIs in one efficient system with one source of truth. Manage risk, get reliable forecasts, and improve margins. Everything you need to grow,

Starting point is 00:10:30 all in one place. So here's the call to action right now. Download NetSuite's popular KPI checklist. designed to give you consistently excellent performance. Absolutely free at netsuite.com slash twist. That's net suite.com slash twist to get your own KPI checklist. NetSuite.com slash twist. And that was the big failure of 9-11. You had people who were aware these terrorists were in the country. You had reports, field reports from the flight school.

Starting point is 00:10:59 And it was a failure of just getting that information and connecting the dots, correct? That is fair. say on 9-11, that's what happened. At its most basic level, yeah. And I think it follows a broader pattern that like the value accretes in the seams between teams. So when you have these teams set up and they have specific missions and mandates and certain data they're allowed to use in that context, you know, connecting the dots would have required all these. Everyone had a different piece of the puzzle. And it requires you to be able to put that puzzle together in a way that's legal and compliant so that you can execute your actual mission. And what were the first

Starting point is 00:11:34 product sets that Palantir put into market, you know, to the extent you can say, because I know you talk about things generally. And then there are specifics that are, you know, tactics and things that you probably are not at liberty to say. But broadly speaking, what was the first product set? How is it received? And then how do you judge its efficacy, you know, in the field there? And how did they judge its efficacy? The first product was the Gotham platform, which was really at the time focused on intelligence and intelligence agencies. And I think that the cynical way to think about Palantir is that it took something as sexy as James Bond to motivate engineers to work on a problem as boring as data integration.

Starting point is 00:12:10 Our core thesis when we started going after this is that there were some really interesting approaches that PayPal had used to fight fraud. We were going to take the intuition for these visualizations and analytics and bring it to counterterrorism. Of course, PayPal was a modern software company. There was no legacy infrastructure. You could simply build on top of this integrated data foundation and provide a lot of value to the human analysts. That's not the case in the intelligence community. You have software that

Starting point is 00:12:36 was written 40 minutes ago, right next to software that was written 40 years ago and everything in between. In order to unlock the value of analytics, actually, you had to create software that would enable you to integrate the data from legacy to modern and everything in between into some sort of common model of the world that represents how humans, not computers, think of this world. So what does your counterterrorism ontology be the word, the frame that we would use for this? And so now I have this consistent view of the world across all of my disparate data and more data is being created every day. And that's the challenge with big data.

Starting point is 00:13:10 It's not just that I have a lot of data. It's that both the volume of data is growing exponentially and the number of data sources is growing. So every day, there's more data that you have to bring. There's more puzzle pieces that you can be bringing to bear to complete the picture. You need to be able to do that at the speed of software, not at the speed of humans. That was the first product we had the big unlock. And we really met our moment around the counter IED fight where you had these.

Starting point is 00:13:32 these networks of bomb makers in Iraq and Afghanistan who were at devastating impacts on U.S. troops and allied forces. Improvised explosive devices for those who don't know the term. Cheap homemade bombs. They could be set off by a pressure plate. They could be set off with a garage door opener. Over time, they're more sophisticated. Exactly.

Starting point is 00:13:52 And so how do you go after that? And really, you have to attack the network, right? It's not about, yes, you want to deactivate the bomb, but you want to get well left of that and figure out who are the bomb makers, what are the expertise, what is their supply chain, how are those humans communicating and moving around? Basically competitive intelligence in some ways, if you think about it from a business information, you know, IT perspective, there's some amount of data out there. And those are taken from field reports, right?

Starting point is 00:14:19 You have CIA in the field, you got Green Berets, you know, Navy SEALs, all kinds of interpreters working with them, collecting information. That information goes into reports. Those reports then go into databases, those databases. Then you have to make sense of them. And just structuring the data probably wasn't happening correctly. Am I right in that era? The structuring of the data problem.

Starting point is 00:14:41 Huge challenge. Yeah. I mean, that was another, one of the big lessons initially. We had really built the first version of the product we had built for structured data because, of course, everything at PayPal was structured, again, to the point of the modern software. And as we started going around to early U.S. government customers, we would show them, they said, this sounds great for structured data.

Starting point is 00:14:58 You know, I don't have any structured data. But that group over there, I think they do. Eventually, we made a whole circle around the building and we realized nobody has structured data. Oh, my gosh. And really, you have to reimagine the underpinning of this of, okay, presuppose that you're starting with human intelligence. This is essentially like book reports, human reports. Plus, on top of that, you want to layer in more structured data.

Starting point is 00:15:19 We would call this all source. So, yes, I have human reporting, human intelligence, but I also have electronic intelligence. I have signals intelligence. I have other things I need to layer on to this to create a, these are more of the puzzle pieces that really matter. So how do I create one platform that allows me to look at all sides of this elephant? And so that was the Palantir Gotham platform. And that's still in production now. And you sell to many different governments, is my understanding. Yeah. Most of Western European intelligence services use our software. Of course, the U.S. intelligence community. Gotham has

Starting point is 00:15:49 continued to evolve quite a bit. I'd say starting with the Battle of Mosul in 2016 is really where Gotham went from being an intelligence platform to an operations platform. So moving from just thinking about how do I organize and understand what's happening with the network to how do I manage myself, the blue forces, the good guys, where are we? How do we do battle management? How do we mission planning? How do we do the after action reviews? How do I do command and control from the software as well? And what we kind of realize is that having unique and deep knowledge of intelligence, that intelligence preparation of the battle space gives you a unique ability then to succeed in your mission. So an example of that might be the intelligence operation brings in the location of the bad guys

Starting point is 00:16:32 and the people supporting bad guys and maybe even their supply chain. Now you got that data. Well, now you need to make a plan. You decide, you know what? Time to take out the supply chain. We're going to do some operations to get rid of supply chain. Well, since you already have that data, Are you saying now when you're making a tactical plan a map, they're going to use Gotham to do that plan? And then it keeps going. So, you know, I did this, I did this plan. It had this effect. What else have I learned?

Starting point is 00:17:00 As I did this operation, you know, I observed new things that give me new intelligence that's happening that I can feed forward into additional operations. A huge number of, you can think about, I think a general one said something about Afghanistan. It wasn't a 20-year-long war. It was 21-year-long wars because we would rotate. people in and out. So what is our institutional knowledge? So many times we would just save lives simply by enabling people to understand where all the historical attacks that ever happened, where have the IEDs gone off? Where are all the times we tried to land a helicopter in this HLZ, this helicopter landing zone before? And have we taken fire or not? And so how do you leverage

Starting point is 00:17:37 the kind of collective knowledge of the enterprise to do more precise, safer operations, reduce risk to life and improve the outcomes of the missions. Interesting is the Gotham platform, if deployed somewhere like Afghanistan, over 20 years, that knowledge base of, you know, all the intelligence, all the tactics that's gone on, that's great for a post-mortem on, hey, what year should we have left? When did things go south? When did things stop having efficacy? Has that kind of thoughtfulness been brought to bear with, you know, rehabping a 20-year

Starting point is 00:18:10 engagement? I think maybe if you come one level down from that, what you're talking about there is I'm going to create, I'm going to help the commanders have intuitions to ask big strategic questions of the battle space. And that is absolutely true. I often think that the most valuable role software can play in defense tech is it is a weapon system for the commander's mind. General Matt has said the most important six inches on the battle space are the six inches between your ears. And so how do we how do we lever up the amazing commanders that we have and give them an ability to see further? into what is true and use that to plan backwards. And so how do I tell if the strategies I'm employing are working or not? And what's different here versus there? Where is it working? Why is it working? And use that to help them with a feel of what's actually going on

Starting point is 00:18:55 and how they're going to change their tactics. Starting a business used to be a pain. You needed a lawyer. They were in fees. It was a mess. Now, with Northwest's registered agent, it only takes 10 clicks and 10 minutes. Northwest provides everything you need to start and maintain your business.

Starting point is 00:19:14 Every LLC, corporation, or nonprofit at Northwest Forms comes equipped with registered agent service, a business address, a website, and hosting, email, a phone number, and this is all covered by Northwest's privacy by default. Again, your full business identity will be live in 10 minutes and in 10 clicks. So here's your call to action for $39 plus state fees. They'll form your LLC, corporation, or nonprofit. often and launch your business in just minutes. Visit Northwest registered agent.com slash twist today. That's Northwest registered agent.com slash twist today. I assume you're only as good as the data

Starting point is 00:19:55 you're able to collect. Maybe you could talk a little bit about how the fidelity of the data and the amount of data being collected has changed. You know, 20 years ago, the iPhone didn't exist. data on phones was very limited, you know, even the metadata, very limited, satellite imagery, infrequently updated. And now, you know, between drones and between, you know, what's transferred over data networks is much larger and then putting all those things together. So maybe you can compare and contrast your, you know, on a battlefield like Ukraine. Yeah. Versus early days of Afghanistan. How is the data different? And what kind of data do you now have access? too. And how does that change? There's probably two dimensions that are pretty interesting there.

Starting point is 00:20:40 The first is that you're drowning in data. I mean, the number of things out there that are actually sensors that are actually giving you something is enormous. And then the rate limiter becomes, how do I fuse those sensors together to make sense of what's happening? That sense making is the bottleneck over the ability to collect, right? So you have tons of data. The other dimension that I think is pretty interesting is how exquisite has the collection had to be over time. You could say, starting with Corona back in the 50s, you know, when we put up the first sort of overhead satellites that could take pictures of the Earth, that capability, we would say, is very exquisite and in the hands of a handful of countries in the world. Now, there's all sorts of commercial

Starting point is 00:21:20 Earth observation that you could be tapping into. And the amount in access of information that even the Ukrainians have organically to themselves, where they don't have to go through other countries with exquisite collection has changed tremendously. And I think that's created for better and worse, a real leveling of what's out there. We should think about that through the lens of the adversary. Bad guys, terrorists have access to this information as well. The next 9-11, God forbid, but there will be another one, you know, whether it's on our soil or another person's, could be informed by a terrorist group with drones, with offensive

Starting point is 00:21:52 drones, with drones, you know, I'm talking quadcopter drones, not military-grade drones, I assume, but they have satellite images. And they would have satellite imagery to plan attacks in a way that the 9-11 hijackers. attackers didn't. Absolutely. And the drones now, I mean, you know, an MQ9 might be $50 million, but a DJI is a couple thousand bucks at most, really. And, you know, we've seen that rate of innovation on the battlefield in Ukraine where there are 3D printing fins you attach to a grenade that turns it into pretty close to a precision munition here that you can draw, a bomb that you can drop. And so the cost performance curve of creating damage is, is changing dramatically.

Starting point is 00:22:30 And that imposed a lot of need for innovation on us to be able to keep up with them and prevent those things. So the question I guess everybody has is artificial intelligence getting into the battlefield, robotics getting into the battlefield, you know, non-soldiers. We had largely drone wars and, you know, very few soldiers actually getting into harm's way in combat in many cases, where, you know, the soldiers came in to keep the peace maybe afterwards, but maybe didn't have to engage every single target, you know, with boots on the ground. Obviously, in many cases they did heroically. What's the state of robotics and AI on the battlefield? And then also, I know you've had to shift the entire company now to Palantir, of course,

Starting point is 00:23:14 and the platform based on customers wanting larger language models incorporated. Or I assume you also see that as the future, but it's going to open up a whole other can of worm. So we can go down either track. You pick which one we go down first. Well, let's start with the government track to begin with. So people tend to go straight to Skynet and their kind of worse fears around. the dystopic potential here. And I think that obfuscates what's actually happening, which is that

Starting point is 00:23:40 there is a rules-based process we go through to identify targets, nominate those targets, prosecute those targets, assess the effect we had on them. You could say, I think in the Gulf War, maybe that whole cycle, the first Gulf War, you could measure that in days. For the sort of fights that we anticipate deterring now against near-pure adversaries, you need to be able to go through that cycle in minutes. And the real advantage of AI here is how do I use algorithms to identify, identify targets that might be on the ground. Look for tanks, just as an example, look for tanks from overhead imagery, look for electronic signatures, bring this stuff together, then put that in front of a human. So you're not bleeding from your eyeballs, scanning many, many, many square

Starting point is 00:24:18 kilometers of imagery, but rather you're looking at key things that are probably relevant for you, accelerating that process. You're able then to more efficiently send that to the fires folks to take care of what they need to do there, and you're able to understand the effects prioritize the targets, The point is to really have, how do I go after high value command and control nodes, not after lower value pieces of infrastructure that aren't going to actually deter the enemy? And so you can think about this as more as driving improved automation, not autonomy, but automation in this process here, to create deterrence, which is the goal. Specifically using AI to analyze data, I saw a demo that you had for Palantir's AI platform,

Starting point is 00:24:58 and I started using a language model and a chat interface. and I thought to myself, clever, but is that actually the right interface for tween planning? And in the example, it was, hey, the language model is like, hey, there's an alert. I'm like, okay, great. System should give alerts. There's a tank coming, you know, from a certain direction. It was like, what do you want to do? And it's like, well, you know, give me some ideas of, you know, ways to counter this based

Starting point is 00:25:21 on, I guess, resources in the area. And it's like, here's three ideas from the language model. And then, hey, can I get approval for these? And I just thought, is that really ready for prime time? And is that a good idea? So how do you balance the fact that these language models are a complete disaster in hallucination and quality? You know, if you ask it 100 times to answer a question, you're going to get 100 different answers. That's obviously not Battlefield ready.

Starting point is 00:25:46 And you can't make mistakes in your line of work. So how do you balance this perception that these language models, and it's probably a correct perception that it's the future, but that it's ready for deployment now? Are these things ready for deployment or, and how are you balancing this tension? Because it's one thing to ask for a recipe and burn your big ziti or have it come out not good. It's another thing when you're deploying military person. At the risk of giving you a very long answer to this, I think we have to back up and kind of unpack. How do we even think about these language models and their optimal application?

Starting point is 00:26:19 Let's do it. Let's do it. Yeah, we got time. Okay. So I think what's interesting about these LLMs is they somehow, they speak in the form that we most often associate with. human thought, natural language, but they don't actually understand what they're saying. And they seem to be instructable in ordinary prose, but they're not good at reasoning. So they occupy the sort of middle ground between human thought on one end and traditional algorithmic reasoning on the other

Starting point is 00:26:44 end. And if you honor that, you can start to really use them efficaciously. So in particular, I would look at our commercial business. 50% of our business is commercial as in non-government and 50% as government. And there, in particular in the U.S. commercial, we have put so many of these use cases to production. A hundred percent of the use cases are an elegant integration of human thought, the existing humans in the business, traditional software, algorithmic reasoning, and LLMs. And so I think that the pitfall is when you're trying to go after every problem with an LLM hammer, you start running up against almost everything you just described there, where it's like, well, this doesn't quite seem right, how do you deal with the hallucination, how do you do this,

Starting point is 00:27:22 it doesn't know anything about my domain. And one of the founders, and one of the founders, is that people are trying to jam as much information into the parametric knowledge of the model. What does the model know about the world? Rather than stepping back and saying, you know what, these models today, they're not as good as my subject matter experts, but what they are, they're definitely better than my interns. And I essentially have infinite on-demand intern capacity right here. Got it. Now, if I could reimagine my workflow, my use case, such that I have traditional software, I have infinite on-demand interns, and I have my humans. How would I solve problems? And how could I array this to be, you know, in the government's case, to increase lethality,

Starting point is 00:28:00 in the commercial case, to improve the automation and efficacy of my business? And I think a big part of why this is potentially hard for traditional software engineers is that these language models are stochastic. This is a little bit like when we first started learning how to predict the weather in the mid-1800s, we thought it was going to be like astronomy. You know, you can tell at this point of the Earth, 10,000 years from now, there's going to be an eclipse. It's dominated by calculus. The math is perfect.

Starting point is 00:28:26 That's deterministic. And actually, when you're writing your Python function, when you're writing traditional code, you're in that domain. So how we think about unit tests, how we think about developing software, how we think about the tool chain, it's all geared up for that. These things are something entirely different. It's more like predicting the weather. It might rain tomorrow.

Starting point is 00:28:42 So now, if you have this fundamental stochasticity, how do you harness this genie in a way that adds value overall? And I think a good example of maybe what we want to try to avoid is something like the self-driving car. Obviously, it's quite compelling, but if we think about the first real proof of concept of the self-driving car was in 2005, the DARPA Grand Challenge. You've seen those videos. It's just rough. It is rough. I would say it's still, especially the ones that didn't work, but, but, you know, the winning car drove 132 miles through the desert. Yeah. It's kind of a serious demo. But, you know, 20 years on, we might just barely be

Starting point is 00:29:19 scratching the reality that self-driving cars could be here. Are you using AI tools every single day? If not, you're falling behind. You know that. In 2024, AI is all about adoption. But here's the hard part. How do you separate the signal from the noise? There are tons of AI tools out there.

Starting point is 00:29:36 We all know that. But some are just parlor tricks. And here's one way you can start to get an edge. Head to Imagine AI Live. Yes, that's right. Imagine AI Live is a conference taking place on March 27th and 28th in Las Vegas. At the conference, you're going to learn how to apply AI to your business directly from the people who have built these extraordinary tools, like the GROC executive, Mark Heaps.

Starting point is 00:29:59 You know, Schmoth mentioned GROC on All In last week, GROQ. And they're going to have the Multion co-founder, Div Garg, which Sunny and I gave an A-plus to when we did their demo on This Week in startups. You're going to see a ton of AI demos from experts, and in those demos, they're going to explain how to use AI to reshape your company. Imagine AI Live is a cross-industry event. It's designed for leaders who want to learn how AI can transform their businesses. So here's your call to action.

Starting point is 00:30:24 The founders of this conference are big fans of this podcast, so TwistListers can get 20% off at ImagineAI. Live slash Twist. That's ImagineAI.com live slash twist to get 20% off your tickets. Then I think you could contrast that. If your objective with self-driving cars was to reduce human death and suffering, you would say Tesla was the company that succeeded most by introducing the incremental self-driving features, advanced automation into the cars themselves. Right.

Starting point is 00:30:53 And that is closer to an elegant integration of the existing human drivers, traditional software, and advances in AI. You're taking an autopilot approach here at Palantier, which is, hey, we know this thing is not going to be bulletproof. It's not going to be perfect out of the gate. But it can do, you know, an increasingly impressive array of things. Yeah, maybe it keeps you from running into the bumper of another car. hey, you know what, that's like a third of accidents on the highway is just fender bender. So if we get rid of those, hey, traffic's going to move better, costs go down. Okay, drifting from your lane because you're texting, hey, if it keeps you in your lane,

Starting point is 00:31:29 those are some of the pretty serious accidents when somebody drifts over into head-on traffic because you're texting. Now you've got real proof points. Of course, Tesla's been barbecued for this by the press and by haters and competitors and bad actors, as they make each mile safer and the distance between, you know, cars safer and safer and create these little mini bubbles where the Teslas don't run into each other as the other cars running into the Teslas. It's taking a while for people to kind of grok that. So are people gawking how you're framing this? And do they understand, hey, great for intern work. If you want somebody to review, I don't know, some satellite imagery. Yeah, it's a good way to look for anomalies and

Starting point is 00:32:07 satellite imagery. But if you're looking to get Osama bin Laden at this location, send you your best people. Exactly. Or bring it all together, right? There is some elegant integration that is better than just humans alone, better than traditional software alone. And there's this, there's sort of pragmatism that you really need to be applying to this. That's the philosophy we've been taking. And as a consequence, I think, to come back to your point, I think chat is a dead end. I think in many ways, you know, Arthur C. Clark has that quote that sufficiently advanced technology is indistinguishable for magic. When Kasparov lost a deep blue, that was like a magical moment. And we thought we were on the precipice of the AI revolution.

Starting point is 00:32:42 yet again. You know, and when AlphaGo was there, Chad GPT was that same moment for us. But in some ways, we anthropomorphized the technology. It is very powerful, but it's not the terminal state here. There's a lot more that you have to do. And I think people are starting to recognize that. So a big thing I've been talking about is as an institution, you need to be looking for proof, not proof of concepts. It's going to be really easy to build a demo that's compelling. You don't want to be on the self-driving car journey in the sense that it's going to be 20, years before you get to proof. What we're focused on is building that tool chain around LLMs with AIP that enables you to get to proof very quickly. What are some of the early wins? Have you had any

Starting point is 00:33:24 early wins in the field yet, or are you just starting to deploy it? No, no, huge early wins. So with one automotive OEM, if you look at their warranty claims management process, right? So if your car's under warranty, you go to the dealer, they repair whatever's broken, they send in a claim to the OEM. The OEM's analyzing that because they're paying for that. So they want to figure out it, do I have a supplier, you know, does the part not work when the humidity is like this or what's going on? How do I root cause this? How do I fix the parts for the next set of cars I'm making? Or do I need to do a recall? Right. So if you think about that process, there's really two steps to it. One is analyzing the claim, which has like a big unstructured field from the dealer writing

Starting point is 00:33:59 what went wrong. And then the kind of statistical analysis of understanding the root cause of what actually happened in this part. This root cause analysis is very much a subject matter expert. You need to understand a lot about the parts. It specialized. The same person doing the root cause and the steering system is not doing it for the air conditioning system, etc. But the categorization of the claims is something that used to take a human half their day to do. And LLMs are actually fantastic at doing this. It's within the parametric knowledge of the model.

Starting point is 00:34:27 As a task, when I look at this and I talk to startups, because what you're talking about that you're doing at Pallantier, which is a very large company with a couple billion dollars in revenue and some of the greatest customers in the world, not just three-letter agencies in governments, you've got, you know, 3Ms and private companies using the products as well. I almost look at it as like, what dollar per hour employee is this? And has it hit yet? And it's very clear that business process outsourcing, occurring in India, you know, in the Philippines, etc., all of that work, they've already been adopting machine learning, AI, etc., to make humans over there making between three and $10 an hour, more efficient.

Starting point is 00:35:12 So that's happening. And then you go up to the next one. Okay, call center employees, people doing tagging of stuff. And then creative work on the margins, now you're getting into the $30 an hour. Feels like that $3 to $30 an hour place is where it's working really well. Let's talk about call center. I mean, that's the next one. So how do you do real-time analysis of the audio on the conversation you're having

Starting point is 00:35:34 in order to surface the next best offer that the call center agent could be presenting to them, provide more assistance to the call center agent to understand what could be going on here, what is the history and context of this customer without them having to operate their CRM software, right? So bringing the intelligence to the point of the company, that frees them up to actually manage the customer relationship, to be a human in the context of that interaction. So you have improved customer satisfaction, you have improved resolution time. You're solving it on the first call. And ideally, if the next best offers are properly tuned, you're improving revenue as well. So many calls, hey, my printer's not working, classic IT phone call. And it's, you know,

Starting point is 00:36:13 have you reset it? The person says, you know, yeah, and I rebooted it twice. I turned it on and off twice. And it's, it's not a paper jam. And I know there's paper in there. And it's just the LOM's looking through the tree of the decision tree and just checking off. Probably not paper. It's probably not being powered off. Okay, let's look at drivers. Let's look at firmware. You know, here are the next things in that decision tree. And that person, could probably feel three times as many calls. So the bionic nature of humans. Let's go to high-end manufacturing here.

Starting point is 00:36:43 Sure. Panasonic Energy, North America, they make the batteries in the Tesla's at the gigafactory. They're co-located there. So it's a pretty exquisite manufacturing process. With them, we built a bot that would augment their level one analyst that have the knowledge and skills of the level four technicians. And so there, of course, their production constraint, right? Every battery they can produce, Tesla can sell.

Starting point is 00:37:04 So it's about these very high-end machines running as quickly as possible, as often as possible, with reliability here. And the powerful insight is that you have pictures, you have PDF files, you have all this structured documentation of these machines. But even more than that, you have the reasoning of your human experts, which is often in Slack rooms. It's in the audio of the video conference calls when they're troubleshooting things. How do you take this data, which is actually the most important data in the enterprise, that we've historically treated as ephemeral? the audio of a conversation. But that's where the most up-to-date present knowledge of reality is that PDF document? That's probably three years out of date. It doesn't reflect reality. It's some thin copy of that, some shallow copy. So being able to wield these sources that historically were beyond

Starting point is 00:37:50 the reach of traditional IT systems has been able to up-level employees massively. And so I think, you know, we're not that far away from it having a big impact on the productivity of even high-skilled labor. This is where things can get truly interesting. whether it's the battlefield or the factory floor. And you start thinking about augmented reality, Apple Vision, Pro. And if you're wearing these goggles or, you know, I think the Green Berets, the Navy SEALs, they all have cameras that are live feeding into a live center. I don't know what the rank can follow.

Starting point is 00:38:23 I'm assuming they have cameras on them. I doubt they're sending video feeds back in real time. But that is obviously the potential. And you could be having an analysis done. in real time of soldiers talking to sources, and you might have four conversations going on and the tone of voice and the ability to know this person's anxious, you know, things, you know,

Starting point is 00:38:48 the Musad and KGB do through their decades of training. They can read tells and stuff like that. Man, the software could start reading tells of informants in the field in real time. And be like, you know what? Turns out, like eight of the 50 informants we talked to today, it seemed a little jittery, more jittery than they normally are because we have their baselines.

Starting point is 00:39:07 Maybe there's something going on here. One of the things we were talking about earlier is how we're just drowning in this information. So triage becomes hugely important. There's so many cases now where you actually had the information but you hadn't been able to process it yet. You don't want that to be the case.

Starting point is 00:39:21 You know, with LLMs, even vision language models where you're combining multimodality, you can be looking at and tipping and queuing the things that are relevant for a human to get on top of because they might be more pertinent, more relevant to what's happening today. So I think it's going to naturally be part of every part of the software tool chain here in terms of sense making. What do you think of this sort of AR in the field and this real-time interpretation of video and audio?

Starting point is 00:39:46 Is that happening yet in the battlefield? We're working on that. I think the real value. So, you know, one of the things that's very different about the future fight from the past fight is that we had such technical superiority. We could transmit as much RF as we wanted. You know, what was the adversary going to do? Right. But that, you know, we had big operation centers. We had big infrastructure. We could protect those things. In a fight with Iran, as we see already, our bases are getting attacked with China. That's not going to be possible. If you look at the battlefield in Ukraine, it is blinded with electromagnetic warfare. And so controlling the signatures you put out are going to be critical. So you're not going to have a big center. Do you explain that in more basic terms for the audience of what this electronic magnetic field?

Starting point is 00:40:32 Is that what it is? Yeah. So let's just say you have your phone. Your phone is communicating to a base station by sending RF, radio frequency, right? So it is transmitting something. And that means with the right sort of sensors, people can see you. They can't see you maybe visually, but they can see your RF signature. Now, that means essentially communications are going to be very difficult.

Starting point is 00:40:56 Anytime you're communicating, you're giving away your location. And so you're going to have to think about how you're going to control your emissions, the RF emissions, in order to conceal your location. It means there's going to be a premium on moving. If you can't move, you're going to be dead. That's where I think AR and VR are going to be disproportionately impactful initially, which is like, we're not going to have big operation center where everyone's get together. You're going to put on the headset in the back of a Humvee or Bradley, and you're going to be collaboratively planning as you're literally moving with the visualization of the battle space, the operation you're going on. You're going to be reacting

Starting point is 00:41:30 to what the sensors are showing you. I assume mobile phones banned in the field, but in a theater like, you know, sending Russian, Russia sending, you know, convicts out of jail, they may not be the most discipline. They may grab a cell phone and think they can call back to their, you know, spouses or loved ones

Starting point is 00:41:49 and do, you know, ill-advised stuff like that. Is that actually happening in the field where people are sneaking devices onto the field or they have them and they're getting unaligned because of it? Do you use the technical term? Yeah. I think you're on to something there. I'll leave it at that.

Starting point is 00:42:05 Yeah. I mean, it's so crazy to think about that. You know, you go to a Chappelle show to give you a Faraday bag, essentially. It's not a Faraday bag, right? It doesn't block the RF. But this is the state of play now, is that people can find you based on those signals. And, yeah, that's bizarre. And so this happens in a number of ways.

Starting point is 00:42:26 So, of course, and now there is, we just talked about, look, I might be, I might be, broadcasting something that gives myself away. But then you want to do things that deny other aspects of capability, right? You want to jam GPS. So can I just flood the spectrum in this area so that I'm overpowering the signals you're getting from space? You don't know where you are. You can't navigate anymore. Wow. You know, your precision munitions aren't going to work anymore. So now you need alternative position navigation and timing technologies to deal with this jamming of the RF spectrum. If they're jamming, they're also. jamming comms. There's lots of ways in which our superiority is based on our ability to communicate

Starting point is 00:43:06 to use these exquisite systems. We have to plan for a world where that doesn't exist. You look at the war in Ukraine, a big part of what people are planning around is like, well, what are the openings I have that aren't jammed? Or where can I go? How can I maneuver in ways where I have the least electromagnetic interference to improve mission survivability? Isn't that incredible? Connectivity has now become not just like incredibly infuriating for teenagers in cars on journeys or getting on a plane that the Wi-Fi is janky or they don't have Wi-Fi, it's incredibly infuriating. But tactically in the field, we went from soldiers using maps and compasses and being able to fight those kind of fights to now, are they still capable of it? I'm assuming

Starting point is 00:43:46 they're trained in it, but maybe they're just not used to it or as good at it as they once were. Well, a lot of these technologies allowed us to reduce risk to life. You know, it's like you think about how just absolutely bloody World War II was. You think about the Korean War. You know, what was the era like before precision munitions? And many of these lessons were learned by our adversaries by looking at how swiftly we won the first Gulf War. You know, precision munitions. That was the first time we used GPS.

Starting point is 00:44:11 In fact, the government didn't even have enough GPS, so they had to buy commercial receivers and they had kind of, they had two basically bans. One was high resolution military precision, and one was they intentionally made it a little fuzzier and less accurate for commercial. They had to flip that switch and make it all the same in that context, which is what the time that the everyday consumer realized how good GPS could actually be. From that point forward, they had to kind of make that available to everyone. Our adversaries saw that. They saw how we use space. They saw how precise and how quickly all of this went and said, we need to deny this

Starting point is 00:44:43 capability to the U.S. or we're going to lose. Let's talk about what we think the next frontier will be. What we'll be talking about in 10 years when we're talking about the theater, God forbid, in Taiwan or Iran, who knows, Iran evading Israel, Israel and I, States invading Iran. I mean, anything could happen. You know, I know there's a lot of people who believe Putin is an angel and he's just taking his just little thin slice of Ukraine. I don't know where you stand on these debates. If you're a Putin group. Not on that end of the spectrum. You're not a groupie of Putin knowing what you know. Not part of the GOP, the groupies of Putin. I heard that term the other day. That was pretty funny. It's pretty funny. It is weird.

Starting point is 00:45:26 What's happened here with, so let me put aside where we're going to be. in 10 years, which is a diversion here. Knowing what you know, how should we look at Putin as an adversary? Very seriously. You know, I have so many thoughts in this dimension here. I think he's not just misunderstood in, you know, NATO and went a little too far and encroaching. You think he's a very serious threat to Western democracies. I think we could have lots of historical analysis of, you know, not one inch and how is our policy plan? I think that there are fair debates to be had there. I don't want to reduce the nuance there. But I think if you look at the present moment, what is our current reality? We have a serious issue to manage here. When or lose in

Starting point is 00:46:06 Ukraine for the Ukrainians, we're going to have to manage Russia. That problem's not going to go away. And in many ways, I think we've goaded them into reindustrialization, you know, that their industrial base is moving at a point in time where ours is not. Europe produced fewer munitions than we forgot we had given to the Israelis, you know, just sitting in a depot. So, you know, you You just think about the scale of what rearmament actually means. You look at what we're doing in the U.S. We have moved, but it's taken a long time. And the part that I feel a huge amount of urgency around here is really the historical lessons

Starting point is 00:46:38 from World War II. So Bill Nudson, a Danish emigre who used to be a very senior executive at GM and then at Ford, I'm sorry, at Ford, then at GM, he went into government in the very late 30s, almost maybe early 40s to start building this arsenal of democracy. And that coincided with Len Lees. So we were not yet fighting. but our allies were. And we started ramping up industrial production to give equipment to really the Brits to begin with. Explain lend lease, because that's what the majority, I think the overwhelming majority of

Starting point is 00:47:08 what we've done with Ukraine is, which I think thwarts this counter argument that, hey, we're just burning through cash here. Ukraine's a very rich country. They've been leased these weapons, and we get them back. We get the money back for them. Lend-lease was the U.S. policy where we were going to lend or lease equipment to allies to fight the war in World War II. Because we were not a participant of the war yet. Pearl Harbor had not happened. But we were the world, America, where's the world's best at mass production? We had just figured out mass production as a technique in the U.S.

Starting point is 00:47:43 We could build like no one else, even Stalin. It was eye-watering him to see the pace of what America could do here. But even then, it took 12 months to build factories. and six months to retool them. So I think the counterfactual of America not doing land lease would be that we would have all lost World War II. We needed that lead up to begin rearmament and reindustrialization in order to seize that moment.

Starting point is 00:48:10 And if we had delayed much longer, it would have been too late. And I think perhaps one of the missed opportunities in Ukraine is we're not even doing Len lease. We're really providing them Cold War era kit. I think it's 80s and 90s era gear. It's still quite efficacious, as you can see. But what we're not doing, and we're taking stuff that's really sitting on the shelf and we're providing it. What we're not doing is taking stuff that modern stuff coming off the factory floor and using that.

Starting point is 00:48:36 Got it. So we're giving them our remnant inventory. We're clearing out the old inventory that we probably were never going to use, which is a great paradox. I mean, of one of the paradoxes of war, and you want to be prepared for it and hope you never use it. And then you're basically throwing things away. I mean, we've built tens of thousands of nukes collectively around the world, and we've used two. And that's kind of the way we want it. What's the state of play in terms of us running out?

Starting point is 00:49:01 You hear these people, oh, my God, we're running out of munitions. The United States can't keep this up. Is that true or not? That the United States can't keep up its production. I think we're in this in-between period where we don't yet have a political consensus that would enable us to pursue re-industrialization. but we're very clearly in a world that requires that of us, even for our own national security. What does re-industrialization look like here in your mind?

Starting point is 00:49:29 We need to get back to making things, making things at scale. Some of my favorite anecdotes from that period of World War II was that the U.S. Army had an idea that, hey, making this piece of munition, I think I'm going to need 2,800 of them. And when I make them as the Army, it takes me about three weeks to make them. When they transfer that over to automotive companies, they got production down to two and a half days, and they made 280,000 of them. Wow. This is one of the profound consequences of how we have managed the defense industrial base since what's called the Last Supper.

Starting point is 00:50:03 You know, after the collapse of the Soviet Union, we didn't have a existential threat. And we were spending quite a bit on defense against that adversary. So it was politically necessary that there would be a peace of it. And that's very reasonable in a democracy here. So we slashed the defense budget 67%. So every dollar we're spending before, we're only going to spend 33 cents going forward. Then Secretary of Defense Les Aspen and his deputy secretary, Bill Perry, who was a Silicon Valley entrepreneur, they got all of the primes together at the time there were 51 primes.

Starting point is 00:50:34 And they said, look, you're not all going to survive. This budget is going to get crushed. And we're giving you permission to merge and consolidate. So that's the only way this is going to work. That proceeded until the government blocked the merger of Northrop and Lockhears. So we went from 51 down to five. Quite a consolidation, yeah. Yes, huge consolidation.

Starting point is 00:50:51 And that's really the birth of the military industrial complex, yeah. I think so, because it's what created the financialization of defense. Before that time period, we think about it today as Northrop Grumman and Lockheed Martin, but it was actually Glenn Martin. It was Jack Northrop. It was Henry Kaiser. There were founders, prolific numbers of founders, innovators who were in this space. And after that moment, it wasn't really possible for that that couldn't be sustained anymore.

Starting point is 00:51:18 It was about more like a private equity mindset of how do we get leaner, how do we get more efficient? It became a spreadsheet process or a task. Hey, how do we consolidate all of these resources to make it more efficient? And so now we have Andro and I guess there's going to be a bunch of Android copycats or just contemporaries. And that is the new industrial complex. that'll do what we're talking about. Yeah. Yeah, exactly. And I think there's a huge opportunity, again, going back to my point on Chrysler, you could say in the fight in China, we're going to need these long-range anti-ship missiles. They're called L-Rasms or these joint air-to-surface

Starting point is 00:51:57 standoff munitions called a jasm. Okay, so today we have our primes who make them, Lockheed in particular, in this case, not picking on them, but their approach to manufacturing is essentially one that's gated on the innovation that's available to that ecosystem. Let's contrast that to Tesla. The Model Y manufacturing process was already eye-watering, but because it could not deliver the scale, speed, cost performance they need for the next-gen vehicle, they did another 10x improvement on top of that. Yep.

Starting point is 00:52:26 We have seen this history before. If you look at Intel, in the 60s, 96% of all integrated circuits were sold to the DOD and NASA, DOD and NASA. But Bob Noyes, the co-founder of Intel, always envisioned a future commercial market. He was not building. He would, in fact, not even let DOD. DoD pay for more than 4% of his R&D. He was going to privately finance it with venture dollars so that he could stay in control of

Starting point is 00:52:48 his roadmap. He had this vision. Because he executed on that vision, he far exceeded the price performance that even DOD could imagine, which led to both a huge amount of American prosperity, but I think relevant to this part, uses in national security that we could not have contemplated. So if you want a step change in production of the LRAZM and the JASM, we need to be thinking about things that look more like using the Defense Production Act to get modern manufacturers like Tesla and the big automotive OEMs and their ilk, new upstarts like Hadrian,

Starting point is 00:53:18 into this business. The opportunity is there because suddenly, defense tech, in large part, thanks to Palantir, and of course, Tesla, SpaceX, and now Andrel, that whole cohort has inspired investors to say, yeah, this is going to take time. But if you do win, it's a pretty big prize. And, yeah, the government's hard to sell into. But if you figure it out, they're pretty long-term customers. and they pay on time, I assume they pay on time, except when we have the government shut down, of course.

Starting point is 00:53:48 So let's go to 10 years from now. What's this all going to look like if something happens in Iran or Taiwan, 10 years from now? What we're seeing, so the drone obsolescence life cycle in Ukraine, how long does a new drone that you make last before the enemy is adapted, their electromagnetic warfare, their EW, and techniques to it? It's about six weeks. Wow. There was a colonel, John Boyd, very famous military tactician. He came up with the concept of the Uda Loop. You know, terrain doesn't fight wars.

Starting point is 00:54:19 Machines don't fight wars, humans do. And the Uda Loop is observe, orient, decide, act. It's all about decision advantage. And the speed at which you can go through this loop, he initially developed in the context of dog fighting with airplanes, but speed at which you can go through this loop is the determinative factor if you can win or not. So your ability to go through the Uda Loop to develop a drone that behavior differently and is resilient in different ways before the enemy can adapt to you, that's the only

Starting point is 00:54:44 thing that's going to matter. So I would not place a huge premium on any specific technologies, although there are ones that are going to matter. It's really gearing up and measuring ourselves, our programs, our investments through adaptability. And how quickly can we iterate on these concepts at the pace of the fight? And the one who does that first, fastest, is going to be the winner. Got it. Yeah. And what about space? There's just a brouhaha that happened. And obviously, we live in such a political time here in the United States. I think people were going back and forth of who is that fault here for not taking these reports seriously. Let's put all that aside because it's meaningless and political nonsense in theater. But the Russians, apparently,

Starting point is 00:55:25 are building anti-satellite technology. And so back to Putin, who you think we should take deadly seriously. This is not a misunderstood dictator, correct? In your mind? Absolutely. Yeah. We need to take him deadly seriously. And in your mind... Space is a contested domain. And I think there's a reason we created Space Force, right? And we need the focus on what's happening there. So much of our capability relies on what's there.

Starting point is 00:55:49 And I think creating a separate service has enabled, I think, the necessary focus on that and clarity around what needs to happen there. You could look at the historical, why are we where we are with space? Because we had space supremacy to begin with. To us, spaces where we could more or less put... big, juicy targets because they were going to be pretty safe. We didn't even contemplate them as being targets. And we were going to have capabilities we depended on there.

Starting point is 00:56:16 Yeah. But our adversaries who never had anything in space, they always viewed it as something to attack because that would deprive us of the capabilities. And so they have an attack first mentality, and we have kind of a defend first mentality there. And so I don't think there's any way around managing the risks that are in space, The risk that has more than just simply risk to military, risk to commerce, risk to prosperity. There's so much of life that actually depends on what's in space and continuing to make sure it works.

Starting point is 00:56:47 And so those threats are real. And that's going to be a big part you think of whatever happens 10, 20 years from now is going to be a space theater will be involved. And space hasn't been involved to date, right? Is there any instance of satellites being taken out or sabotaged even? I know the Chinese were working on some stuff, but, you know, aside from the occasional balloon flying over, doesn't seem like there's been a space theater yet. I don't want to comment on specifics there, but I would say that I think we should take the Chinese hypersonic glide vehicle pretty seriously there, the capabilities that were demonstrated and the implications of that. What does that mean for space domain awareness? What does that mean for the vulnerabilities we have, our ability to detect and deter as a consequence of being able to say you can't get,

Starting point is 00:57:34 away with things, reducing the element of surprise here. We should probably think about all of these domains as being integrated, right? It's not what's happening in the sea, what's happening on land, what's happening in air, what's happening in space. It's really these things all come together. Maintaining this cohesive, integrated view of these things is how you prevent surprise and deter conflict. This hypersonic glide vehicle, these things are 10 times faster than anything that came before them. Yeah? They're extremely fast. They're extremely fast. They can fly in maneuverable ways. which means like an ICBM, an intercontinental ballistic missile, has a predictable trajectory. So if you can detect its launch, you can kind of do physics, you can do math and understand where it might be and react accordingly.

Starting point is 00:58:15 What makes these hypersonic cruise missiles or glide vehicles very difficult is that they fly in unpredictable ways. So your ability to counteract them is, and they fly very fast. So this poses challenges. And these could be armed. And so the idea of it taking an hour or a time. for an ICBM to get to the United States could be 10 minutes or something, 20 minutes. I don't know what the concept here is. And these hypersonics also can run very close to the Earth, too, correct?

Starting point is 00:58:46 And that means it's much harder to see them. You think about the horizon of the Earth, the curvature of the Earth. When will your radar be able to see them? The lower they fly, even then, so not only are they moving fast, but by the time you see them, they're going to be much closer to you. Yeah. When I was at Burning Man, I saw Kimball Musk has this really, cool company that does little tiny drones that light up and they make a drone show. And when you see

Starting point is 00:59:10 it, you're like, wow, that's just incredible fidelity. It's better and better. And how many drones are in that? Hundreds of drones, thousands of drones. You start to think about, well, those don't cost anything, right? They're tiny. They're like, you fit them in the palm of your hand. Now, maybe that's too small to intercept one of these things or to get it off track. But we could be running, you know, sorties of thousands, if not tens of thousands of these mid-sized tiny drones with some kind of capability on them. I don't know if it's explosives or radar jamming or whatever. But has that, you know, we have the Iron Dome. And then you have this new concept of, you know, drone warfare, quadcopter. I'll call a quadcopter just so we're clear here. We're not talking about

Starting point is 00:59:50 military drones. You know, fleets of quadcopters, like, well, how is that going to impact the battle, Phil, do you think? I mean, I'd say in Ukraine, they're impacting it in tremendous ways here. But those are like one at a time, right? And they're like flown by one person? Or are they doing coordinated, you know, dozens of them at a time? They're doing coordinated maneuvers here. What's going to get through the air defense systems? How many do I need? Which ones are decoys? Which ones are active? And so there are multiples coming in to try to hide the active one. Yeah, and to basically exhaust enemy air defenses, right? Because they're going to be shooting at you, but they also have a magazine. Like, they're going to run out of Schlitz at some point. This becomes another sort of cat and mouse game. You're back into a different sort of udaloup here. And so I think the key thing is we will be moving towards a world of cheaper and cheaper hardware that has more exquisite software coordination to drive the effects. This should tilt towards America's advantage. Like, we are the greatest nation in the world at software. There are zero enterprise software companies from China that operate on the world stage.

Starting point is 01:00:59 They're even zero from India. The next best country in the world is Israel, and they build canoes, not aircraft carriers. And so the software advantage skews massively in our favor here. Just as we were the world's best at mass production at the dawn of World War II, we are the world's best at software now. And this is going to be the big offset. And why I think defense tech is so crucial in this moment towards driving deterrence from the conflicts that we may be facing.

Starting point is 01:01:25 Let's end on this capital allocating in defense tech. You did a tweet, but I thought this was interesting in terms of we all understand if you're listening to this podcast, the power law. One investment makes up the massive majority of a venture funds returns. In fact, I had Brian Singerman on the pot a couple of weeks ago and he talked about what a massive return they got from going all in on Palantir. So let's, maybe you could read the tweet for us and what you're thinking here. Yeah, so 5% of the capital invested generates 65% of the return in a venture portfolio.

Starting point is 01:01:58 20% of the capital generates 90% of the return and half of all investments lose money. And that's because VC is a power law business. Right. And so what I think defense tech really needs and what the government, because it's a monopsony, right? There's one buyer. That's what a monopsony is. The opposite of the monopoly where there's. Yeah.

Starting point is 01:02:15 Yeah. There's one seller in a monopoly. There's one buyer here. And so as a monopsonist, you get the market you create. You get the behaviors you incent. And so what we really need as America is an ecosystem where we have a few huge winners here that helps the LPs generate a return, that helps the funds raise future funds, that keeps the $100 billion of venture that has flown, that has been invested in defense tech since 2001,

Starting point is 01:02:41 keeps that flowing to generate this software advantage that I was just talking about here. I think this is not a concept that's well understood, maybe generally, but definitely not well understood in D.C., where the kind of, intuitive responses to try to peanut butter spread the capital they do have around. And as a consequence, you end up creating a lot of zombie companies, none of whom really make it, maybe none of them outright fail, but it doesn't create a return profile that allows LPs to go invest further in this space here. Capital is an advantage. This is one of the things we've seen. And the capital markets here, and one of the reasons I believe American exceptionalism is not at risk. I think we're doing

Starting point is 01:03:22 fantastic right now that we might be fighting with each other over abortion and Donald Trump and he says spicy things and Biden's too old and all this other, uh, Musugina. But if you look at actually what America does really well, we've got incredible entrepreneurs like yourself and your team over there. We've got this incredible risk-taking capital structure that is unique in the world. You know, if you go to Europe and I'm sure you've spent time there, you know, the idea of 50% returning nothing would break people's brains. In Japan, it breaks people's brains. Like, failure is not acceptable. So accepting a 50% failure rate, 60%, 70%, it's 80% at the seed stage. You know, we're talking about the venture series A stage here. The seed stage, it's 80, 90% don't make it,

Starting point is 01:04:08 and don't return. So you do have this massive, insane, hard to understand venture capital ecosystem, which is at the core of innovation, it's a core of companies like Palantir breaking out. Nobody believed in that company, which had a lunatic like Peter Thiel come in and yourself, Alex and everybody who thought, what did you think your chances were? 10%, 20%, at that time was the candid thinking. Real candid view is I thought we were going to fail, but I'd rather fail working on a problem of national significance than, you know, succeed in building a web 2.0 calendar. You know, it was mission over money. I think that sums it up beautifully. Sean, great to have you on the program. We'd love to have you back in a year and check in on the progress.

Starting point is 01:04:55 And I know the company gets derided sometimes. People get confused, you know, about what you're doing and privacy. Honestly, I'm glad you're out there doing this work. It's messy work sometimes. Protecting the country's messy. Nobody wants to be in wars. None of these wars have winners. it's just who loses the least in a war, right? And then maybe how does democracy bend slowly over time to keep more of humanity free than under dictators and communists? The fact is, if you look at all the statistics,

Starting point is 01:05:28 everything in the world's going really well. There's one thing that's stuck in the mud. You know, you read Stephen Pinker or any of that stuff. Yeah, and you do all those charts. There's only one that's troubling. And the most troubling one is that the number of people living in a democracy has decreased in a lifetime, where it's going down, not because there's more dictatorships

Starting point is 01:05:48 popping up or communists popping up. So those countries had a higher growth rate than the free countries. We're at stalemate here, and I think companies like yours are going to help us bend towards democracy winning. It's according to us right now. Let's be honest. We have to work for it. I think a better historical narrative is not that the West won over the Soviet Union, but the Soviet Union lost, and it requires effort to protect and promote freedom and prosperity. It doesn't come free. It's going to be a cost. And so keep that in mind to everybody

Starting point is 01:06:16 when you're thinking about these conflicts around the world. Thanks to the team over there for all the hard work you do, keeping the world safe. Appreciate it. And we'll see you all next time on this week's startups.

Starting point is 01:06:26 Bye, bye, everybody.

This Week in Startups - Mission Over Money & The State of Defense Tech with Palantir's Shyam Sankar | E1908

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.