FULL: Palantir CEO Alex Karp Warns AI Will Redefine Power, War, and Economies at WEF | AI1G (YouTube Video Transcript)

(00:00:00) Your YouTube transcript will appear here (00:00:00) the World Economic Forum in Davos. It's (00:00:03) my pleasure to introduce Alex Karp. Um, (00:00:06) I want to just start off on something (00:00:09) more intimate between he and I. I'm (00:00:12) pretty proud of what I created at Black (00:00:13) Rockck, but my total return since I've (00:00:16) been a CEO has only compounded at 21%. (00:00:20) Uh, since Alex and Palunteer went (00:00:23) public, uh, his compounded return is (00:00:26) 73%. So, um, (00:00:30) congratulations, Alec. (00:00:33) Uh, but more importantly, um, (00:00:36) we're in the middle of a profound (00:00:38) technological shift. I think we all are (00:00:40) hearing about it, reading about it, (00:00:42) feeling it, being a part of it. [snorts] (00:00:44) And everybody's asking the question, you (00:00:46) know, what can AI do for me or how can I (00:00:49) translate it? What can it do for growth? (00:00:51) What can it do for workers? What can it (00:00:53) do for countries and national security? (00:00:57) We are talking about it that it has the (00:00:59) potential (00:01:01) to grow capacity to modernize industries (00:01:04) expand opportunity. (00:01:06) It also will transform how we worked and (00:01:09) where we work and how we work. And the (00:01:12) question is are governments prepared for (00:01:15) that real transformation of a society? (00:01:19) So we need to make sure that as it's (00:01:21) deployed it deploys in a way that (00:01:23) empowers people, empowers institutions (00:01:26) and builds a more resilient global (00:01:28) economy. Um few leaders (00:01:32) are truly at this intersection though I (00:01:34) I I am certainly not that leader but in (00:01:37) that intersection of technology, (00:01:40) national security and the real economy (00:01:42) the way Alex Karp is. (00:01:45) um as co-founder and CEO of Palunteer, (00:01:48) uh Alex has closely worked with defense (00:01:51) and government (00:01:53) private private organizations to apply (00:01:56) how AI can be used and and in more of (00:01:59) the critical areas um and it's really (00:02:02) important and I must say my (00:02:05) conversations that I've had and with (00:02:07) Alex over over the last year has really (00:02:09) enlightened me. So I'm looking forward (00:02:12) to this Alex. So let me just start off. (00:02:14) Sovereign states have often been early (00:02:17) adapters of advanced technology and I (00:02:19) think we're seeing that really very very (00:02:22) intimately in the United States. But (00:02:24) from your perspective, how is AI (00:02:26) supporting decision-making in defense (00:02:28) and security? (00:02:30) >> Well, first of all, delighted to be here (00:02:32) and with that introduction, maybe I (00:02:34) should just stop. It's good downhill (00:02:36) from here. Would you like to talk about (00:02:37) our return some more? And I'll just uh (00:02:40) I'll just time. I've been as I've been (00:02:44) public now 26 years. So (00:02:45) >> So um uh yeah. No, I I think one of the (00:02:49) things to remember [snorts] uh like the (00:02:51) the background backdrop of your question (00:02:54) I I think for is though you have I in (00:02:57) America and I would say also in Europe (00:03:01) uh historically (00:03:03) um uh industrial development and (00:03:06) military technology were obviously co- (00:03:11) in is a generalization but more true (00:03:13) than not um you developed a product for (00:03:16) the military (00:03:18) that then was dual functioned and raised (00:03:22) the standard of living of the country. (00:03:26) Um and then um and then for lots of (00:03:30) reasons I'll just leave aside for now (00:03:32) that that's not the way at least (00:03:34) technology was built until now there's (00:03:36) all these defense tech startups and (00:03:38) Palunteer. But um the uh um uh (00:03:44) it it was that you had this thing where (00:03:46) you were going to create something that (00:03:48) had to work under the harshest (00:03:49) conditions that was presumably (00:03:53) significantly better than anyone else's (00:03:56) to the point where it would give you a (00:03:58) more than a slight advantage on a (00:04:01) battlefield, especially if combined with (00:04:04) um your way of fighting. And you saw (00:04:06) this um there was a very famous uh (00:04:10) social socialist German historian uh who (00:04:15) was um said well one of the problems (00:04:17) Germany had was the war fighting machine (00:04:19) was so good that they just said well (00:04:21) we'll decide on the battlefield who's (00:04:22) right and that that that led to (00:04:26) obviously lots of disjunctions and real (00:04:28) problems in the in the American context. (00:04:31) Um uh and I would say there you have a (00:04:33) dislocation between what has happened (00:04:36) recently in America and what is (00:04:39) happening in Europe and what I think so (00:04:41) I think America and China are are kind (00:04:43) of very successful and I spent most of (00:04:46) my adult life in Europe and I'm very (00:04:47) pro- European but I think any honest (00:04:49) assessment is it's not that's not gone (00:04:51) very well. Um you built things that were (00:04:56) could be used uh in an adverse (00:05:00) condition. So rough, dirty, morally gray (00:05:04) conditions. So how do you change the (00:05:05) morality to fit with how we fight in the (00:05:07) west is also a big vector. So the (00:05:10) morality is difficult. Um the conditions (00:05:12) under which you use the technology are (00:05:14) difficult in in in software context. You (00:05:17) you're [snorts] not you don't have (00:05:18) you're not directly connected to the (00:05:20) network in many cases. You have (00:05:23) constraints under which you have to (00:05:24) fight even though that's not the optimal (00:05:26) way to fight. and you have very (00:05:28) specialized ways of fighting in each (00:05:29) country. Um, but the the positive side (00:05:33) of that was that you also were building (00:05:36) something that could be deployed that (00:05:38) had an obvious value to the average (00:05:40) citizen, (00:05:41) >> right? Um and um so now you have this (00:05:44) the AI thing is there's [clears throat] (00:05:45) many things that are really interesting (00:05:47) about it but if you start with the (00:05:49) benality that I think um until very very (00:05:54) recently all kind of adversaries of (00:05:59) broadly defined the west assumed that (00:06:01) investments in softwarebased (00:06:04) uh defense were some kind of crazy thing (00:06:08) Americans do for marketing. uh get rich, (00:06:11) company blows up, you're on (00:06:13) [clears throat] your beach in the (00:06:14) Bahamas, but shareholders are happy and (00:06:16) it's gone kind of thing. And um uh and (00:06:21) um what you've seen uh so first of all (00:06:25) you could say well that's changed but (00:06:28) Germaine to your question the learning (00:06:30) process in building how do you build (00:06:33) this for sovereign governments is also a (00:06:35) learning process for how do they adopt (00:06:36) these technologies. So it's it's not (00:06:39) just the technology because if you're (00:06:41) building a tank like you know the (00:06:42) British and then the French and then the (00:06:44) Germans kind of optimized tank (00:06:46) technology it's easy to see how you (00:06:48) would deploy it but how do you deploy a (00:06:50) system that's primary value is (00:06:52) organizing parts on the battlefield (00:06:54) without seeing the parts on the (00:06:56) battlefield and seeing does it work how (00:06:58) much does it work is it much better than (00:07:00) what we had can we do things we could we (00:07:02) couldn't do in the past and then there's (00:07:04) a hidden thing about uh software AI (00:07:08) that a lot of the value interestingly (00:07:10) people always assume the value is from (00:07:13) um where you are to where you should be (00:07:16) and that's obvious but in most sovereign (00:07:18) nations in the world I mean we deal with (00:07:20) almost all in one form or another that (00:07:22) are broadly defined as in a that would (00:07:26) be you know it it could deal with a (00:07:28) company that's you know American (00:07:31) the actual technological (00:07:34) uh uh rigor of the of the enterprises (00:07:37) has significant holes in it. So it like (00:07:40) as you know it's like it's very I'm (00:07:42) dyslexic. It's very dyslexic. There are (00:07:44) like whole pieces of the enterprise that (00:07:47) exist on a PowerPoint that when you go (00:07:49) to battle you will find out do not (00:07:51) exist. And this whatever country you're (00:07:54) in in this room if you are in the west (00:07:56) this is a problem you have. Uh a day of (00:07:59) battlefield you will find out this is (00:08:01) one of the advantages the Ukrainians (00:08:03) had. They essentially started from (00:08:04) nothing. And so there wasn't you didn't (00:08:07) have to rediscover that your enterprise (00:08:08) didn't work after you're in fighting. (00:08:11) And one of the huge advantages America (00:08:13) has is just for you know better or worse (00:08:16) I believe it or not think mostly for (00:08:18) worse I was always against (00:08:19) interventions. I'm not a neocon. Um it's (00:08:23) uh um (00:08:25) uh we had just all this experience on (00:08:27) the battlefield so you could see what (00:08:28) worked and what didn't work. Um, but (00:08:32) pulling it back to the so the first (00:08:34) thing that sovereign nations really (00:08:36) struggle with is can I identify which (00:08:38) tech is better objectively? Can I even (00:08:41) rate it? Can I rate I'm looking at I (00:08:43) guess say y (00:08:45) >> but don't you need to know where you (00:08:46) want to go to ask the right question. (00:08:48) >> Um, you know that's you that's that's (00:08:52) you actually let me reframe it. Y (00:08:54) >> you have to know where you are to know (00:08:56) where you want to go. (00:08:57) >> Okay. So like the the point I'm making (00:08:59) is with this exogesis is one of the most (00:09:02) important things Palanteer's done on the (00:09:04) battlefield is be able to make up for (00:09:06) the fact that half your enterprise (00:09:07) doesn't even work on the battlefield. It (00:09:10) does work in a lab on a powerpoint only (00:09:13) built by country. (00:09:14) >> It doesn't work because of machinery or (00:09:16) human or humans. Well, because the (00:09:19) conditions of a battlefield are rough, (00:09:21) contagious and and like for example like (00:09:24) the modern if you just take Ukraine as (00:09:26) an example, just to make it empirical, (00:09:28) you know, as a everyone like reads this (00:09:30) like, okay, well, how hard could it be (00:09:31) to move a drone from A to B? Uh, well, (00:09:36) actually, first of all, you're going to (00:09:37) need to know where you want to put the (00:09:39) drone. That's going to require (00:09:40) synchronizing all your data. You're (00:09:43) going to need to do that without (00:09:44) transferring that data to your (00:09:45) adversary, which means you're going to (00:09:46) have to know every single person who (00:09:48) touched it. You're going to have to (00:09:49) obuscate it till the final thing. Then (00:09:51) you're going to want to do it presumably (00:09:53) either within the coordinance of your (00:09:54) strategy or with your ethics. So where (00:09:57) does it where does it not go? You're (00:09:59) going to want to be able to correlate. (00:10:00) Do you want to put the drone on your (00:10:02) asset? No one only two people in Ukraine (00:10:05) may know that one of the generals is (00:10:07) your asset. You can't tell people that's (00:10:09) your asset. How do you how do you make (00:10:10) it look like to your should soldiers (00:10:12) that you actually were taking people out (00:10:14) and you just missed your asset? Um then (00:10:17) and then the war advances. So the (00:10:18) Russians which are off they're often (00:10:21) very underestimated for reasons I don't (00:10:23) understand but like they're (00:10:25) mathematically arguably the best in the (00:10:27) world and things they may not have in (00:10:29) the beginning they can kind of cobble (00:10:31) together and so they began jamming (00:10:33) electronics. So now you have a (00:10:35) completely different problem, but your (00:10:37) presumably your enterprise has to be the (00:10:39) same or developed because now it's not a (00:10:41) question of going from A to B. It's a (00:10:42) question of going through a completely (00:10:44) jammed environment where you have no (00:10:47) connectivity while actually collecting (00:10:49) data while you go through that (00:10:50) environment. And every one of those (00:10:52) things is like a dynamic challenge. (00:10:55) >> None of which were foreseen even before (00:10:58) the Ukrainian. every single battle zone (00:11:00) in the world has a By the way, the other (00:11:02) thing is um then people fight (00:11:05) differently. So like if you look at the (00:11:07) big battlefields, I'm sure some people (00:11:10) love our work here and some people hate (00:11:11) it. By the way, we we we welcome all (00:11:14) opinions at Palanteer. Even at inside (00:11:15) Palunteer, we have people who love our (00:11:17) work and people who are unhappy with the (00:11:19) work. Um uh we welcome all. (00:11:21) >> That's a spirit of dialogue. (00:11:22) >> It's a spirit of dialogue with a (00:11:24) somewhat of a leader. uh and uh um but (00:11:29) you know if you look at you know how uh (00:11:33) you know when the Ukrainians it's a (00:11:35) small team of people and very courageous (00:11:37) soldiers that are very technical they (00:11:40) have highly technical people that built (00:11:43) on top of our product things that we (00:11:45) don't understand and proprietary ways of (00:11:47) using the product in Israel they (00:11:50) according to rumors use their (00:11:52) intelligence so most people fight (00:11:53) military to military this was int (00:11:55) intelligence in Iran from what I can (00:11:57) tell from the papers. Um and then in (00:11:59) America you just have massive uh forces (00:12:03) that no other country has but they had (00:12:05) to be integrated and the integration (00:12:07) capacity. So every one of those things (00:12:09) is different. Sure. And so the the the (00:12:11) two-fold role of of of enterprise (00:12:14) software on a battlefield is one to make (00:12:16) sure all the underlying things actually (00:12:17) works and then two to raise the level to (00:12:19) a level no one else has in the world. (00:12:21) >> Let me translate this now. I mean (00:12:24) there's so much technology was created (00:12:25) from defense whether it was the internet (00:12:28) uh GPS. (00:12:31) How do you envision this translating (00:12:33) from defense and military (00:12:36) um to corporations to businesses to (00:12:40) society? Well, first of all, the the (00:12:43) fact that it's basically a purely raw (00:12:46) naked environment means that you (00:12:49) actually know the ground truth of what (00:12:50) could work independent of what (00:12:52) enterprises think can work. And the (00:12:54) generaliz the think at a at a at a high (00:12:57) level, it's almost one to one (00:12:59) translatable, but precisely because as (00:13:01) just to take enterprises in general, not (00:13:03) all enterprises tend to want to over (00:13:06) time become like every other enterprise. (00:13:09) So if you take five, you know, A (00:13:11) enterprise and B enterprise and C (00:13:13) enterprise, they're in the same market. (00:13:15) Their tech infrastructure is trying to (00:13:17) make them into the same enterprise. Uh (00:13:19) they have the same orchart. (00:13:21) >> Sure. (00:13:21) >> They have presumably roughly the same. (00:13:24) They don't have the same data (00:13:25) infrastructure. And what you learn on (00:13:27) the battlefield is that and in life is (00:13:30) that that's not particularly valuable. (00:13:31) What is very valuable is an enterprise (00:13:34) can do something no enterprise in the (00:13:35) world can do. And so that is the goal of (00:13:38) every single military in intelligence (00:13:40) service. In fact, all these intelligence (00:13:42) service and militaries have their own (00:13:43) specialization. And so when we went to (00:13:46) commerce like what we're saying is how (00:13:48) can we make your insurance come the way (00:13:50) you underwrite? (00:13:52) How can we make that to your tribal (00:13:54) knowledge about underwriting? How can we (00:13:56) transform this to knowledge everyone has (00:13:59) to a knowledge only you have and with (00:14:01) efficiencies that no one else has? So as (00:14:03) an example on the battlefield one of the (00:14:06) most important issues I is how do you (00:14:09) acquire data process it and then put it (00:14:11) into a framework where it can be (00:14:13) actioned either from an intelligence (00:14:14) perspective yeah so just obviously what (00:14:18) does a business do in the end of the day (00:14:20) what is a business doing I mean it is so (00:14:22) for example especially in anything (00:14:23) underwriting banking hospital intakes (00:14:26) it's information it's information it's (00:14:28) sorting the information in a way that (00:14:30) you have a a distinct advantage over (00:14:34) other people who are similarly situated (00:14:37) uh and that that advantage can't be (00:14:38) eviscerated easily and so de facto for (00:14:42) what you can't what you're doing on the (00:14:43) battlefield with our products without (00:14:45) getting happy to is like what you're (00:14:46) doing with ontology and foundry in some (00:14:49) cases on the battlefield but definitely (00:14:51) in like commercial organizations across (00:14:53) the world but especially in America is (00:14:55) when you approach the underwriter or the (00:14:58) hospital we power tons and tons of (00:14:59) hospitals right they have an intake (00:15:01) problem they all have an intake problem. (00:15:02) They all have a shortage of doctors and (00:15:04) nurses. They are working in a low margin (00:15:06) environment, but every single one has a (00:15:08) different way of processing their (00:15:10) patients according to what their (00:15:11) specialty is and the kind of patients (00:15:13) they don't do well with. And how do you (00:15:14) manage that? And so the intake flow and (00:15:17) into your enterprise in a way that you (00:15:20) can actually process these things 10 15 (00:15:22) times faster than you could before. And (00:15:24) then the other (00:15:25) >> that saves lives then (00:15:26) >> it save what saves a lot of lives. Um it (00:15:29) also interestingly because you're (00:15:32) processing the LLM in an ontology you (00:15:34) have a structure it all despite what (00:15:36) people may want to believe it also (00:15:38) bolster civil liberties because now you (00:15:40) can say well I mean just simple (00:15:42) questions was someone processed based on (00:15:44) economic uh considerations or were they (00:15:47) based uh were they processed based on (00:15:49) their background like those things are (00:15:51) impossible to see unless you have like (00:15:53) there's a huge civil liberties (00:15:55) betterment side of this that typically (00:15:57) people don't believe we care about or (00:15:59) but it's actually exactly the opposite. (00:16:01) We do care and you know showing is (00:16:03) caring. It's like we can granularly show (00:16:06) uh why someone came in, why they were (00:16:08) taken, why they were rejected and we can (00:16:10) do it in a way that makes business sense (00:16:12) for the b the business itself. Um and (00:16:15) then it it leads to safety efficiencies (00:16:18) >> and probably brings down costs. Well, (00:16:20) the if you want to do a a shorter kind (00:16:22) of financial version, you basically in (00:16:25) the past to do what we can do in in the (00:16:27) full light of a public market, you (00:16:30) needed a pri you need to take the (00:16:31) company private. So, and then you would (00:16:33) take out the cost structure and you that (00:16:36) you probably resell it. Well, now you (00:16:37) can take out the cost structure, make (00:16:39) the workers more important. So, the (00:16:41) actual workers, not the fat kind of in (00:16:43) the middle. And uh um and then you can (00:16:46) you can change the way they go to (00:16:48) market. So the (00:16:49) >> hospitals you can process this (00:16:50) information faster presumably you'll (00:16:52) have (00:16:53) >> much faster and then the nurses the (00:16:55) nurses the one of the the nurses and (00:16:57) doctors are also happier and yeah (00:17:00) >> so you you you (00:17:02) suggested that for companies that (00:17:04) rapidly adopt that they have to best way (00:17:06) to do is go in private then they could (00:17:07) restructure but (00:17:08) >> no no what I'm suggesting is they don't (00:17:10) have to do that anymore (00:17:11) >> they don't have to do that. So what is (00:17:13) the basic um (00:17:16) what is the basic hindrance in the (00:17:18) adaption of of of the of AI? Uh is it (00:17:22) just um legacy systems, legacy issues? (00:17:27) What how do you how do we accelerate the (00:17:29) adoption that it's good for humanity? (00:17:33) >> Well, I mean our adoption accelerates (00:17:35) beyond our capacity. So it it the um I (00:17:40) think if you just buy large language (00:17:42) models off the shelf and try to do any (00:17:44) of this it won't work. (00:17:45) >> It's a commodity. (00:17:45) >> It's a and well also it's not precise (00:17:47) enough. (00:17:48) >> It like you can't do underwriting with a (00:17:51) you can't you couldn't do any of these (00:17:52) things that are regulated. So moving to (00:17:56) like everybody there's not like the the (00:18:00) problem with adoption at this point is (00:18:02) people have tried things that just can (00:18:05) never work like buying you borrow a (00:18:07) large language model you put it on your (00:18:09) stack and you wonder why you know it's (00:18:12) like not working and so what you're (00:18:14) going to see especially in America is (00:18:17) people trying to do something like what (00:18:19) we do withtology maybe by hand because (00:18:22) once you build the software layer to (00:18:24) orchestrate and manage the lounge (00:18:26) language models in in a language that (00:18:28) your enterprise understands, you (00:18:30) actually can create value. And [snorts] (00:18:32) I don't I I think like there's like (00:18:34) there's a lot of discussions like are we (00:18:35) in an AI bubble? What is the meaning of (00:18:37) the bubble? Well, it's like I I think (00:18:40) what if anything we're just in a lag (00:18:42) where there's a lot of AI. Some of it (00:18:44) works. Again, if you go back to the (00:18:46) battlefield context, like it most pe (00:18:49) everybody in the world assumed this (00:18:51) would not work, but now it does work. (00:18:54) And so now the question isn't does it (00:18:56) work, but how can we get it to work for (00:18:58) my country? And this is exactly what's (00:19:00) happening in (00:19:02) >> companies. It's like oh this company it (00:19:04) worked, mine didn't. What are you doing? (00:19:06) >> Like for example, I mean if you want to (00:19:08) just make it like parochial to us, (00:19:10) Palanteer barely has a salesforce. In (00:19:13) fact, it seems to be getting smaller (00:19:14) every time I go see them. And it's like (00:19:18) and it's simply because you're laughing (00:19:19) because you know it says Palanteerians (00:19:21) here, but it's it's getting smaller and (00:19:23) smaller and smaller. And um and it it's (00:19:27) not because we're trying to save on the (00:19:28) unit economics. It's actually because it (00:19:32) is a low trust environment in AI. People (00:19:34) have tried lots of stuff. A lot of it (00:19:36) hasn't worked. But if you've delivered (00:19:38) stuff that does work, why do you need a (00:19:41) salesforce? like you you you just it (00:19:43) sells itself. (00:19:44) >> Well, you have to say, "Hey, don't talk (00:19:46) to us." (00:19:47) >> And then on that's that's that's that's (00:19:48) on commercial in government (00:19:51) >> at this point. It's really we don't like (00:19:55) we don't we it's it's it's very hard to (00:19:57) export simply because we have to train (00:19:58) the people and we have limited man. (00:20:00) >> I was going to say that's your limited (00:20:02) bandwidth, the training once somebody (00:20:04) takes on your your software. Well, in (00:20:07) the government contest, every every (00:20:08) country like obviously has the (00:20:11) equivalent of a security clearance, (00:20:13) right? So to use our to build to build (00:20:16) project Maven or Maven into your (00:20:18) architecture, you're going to need (00:20:20) somebody with the highest level of (00:20:22) clearance that also is technical and (00:20:25) most technical people unfortunately are (00:20:27) not going after the highest level of (00:20:29) clearance. So there are very very few (00:20:30) people like that. So the that that (00:20:32) resource is super scarce. Uh and then (00:20:34) they have to be trained and that that (00:20:37) takes that can take a while and and and (00:20:39) then you also have to have like anything (00:20:40) you really have to believe in this and (00:20:42) really think it's important and you know (00:20:45) not everybody fits in that category. (00:20:47) >> How many people need to be trained to do (00:20:49) this? I mean does it in a corporate (00:20:52) level does it have to be from the CEO (00:20:53) down? How does this work as you (00:20:56) >> um well you're talking about like (00:20:59) insurance underwriting? (00:21:00) >> Well like you know Yeah. So insurance (00:21:02) underwriting um I um the the way it (00:21:06) works is the best case scenario you have (00:21:08) the best and worst case scenario. Best (00:21:10) case scenario um CEO is mathematically (00:21:15) inclined (00:21:16) >> even though they they may not know (00:21:18) nothing about product but you can they (00:21:22) can impute a product working by looking (00:21:23) at the math. Um yeah and uh um and uh (00:21:29) and and has you know and then we (00:21:32) probably need to train five or six (00:21:34) people on there. In the beginning we do (00:21:35) all of it and then we transfer that as (00:21:39) much as we can to them or we're part (00:21:40) trying to partner with people who can do (00:21:42) that with us but you need a small number (00:21:44) of people but we still need more than we (00:21:47) have. (00:21:48) you you suggested repeatedly about how (00:21:51) AI could strengthen foundations for an (00:21:54) economy, especially we're seeing that in (00:21:55) the United States. Um, how rapidly can (00:21:59) AI (00:22:01) change the growth trajectory? Because (00:22:02) you you you mentioned that earlier about (00:22:05) how it could improve the e economies, (00:22:09) the well-being of companies. (00:22:12) Well, um there in a lot of these things (00:22:16) there's a speed question, but um (00:22:21) I think like with a lot of our (00:22:22) companies, it's like we can take out (00:22:26) in the area we go in up to 80% of your (00:22:30) cost and improve your top line (00:22:32) dramatically. But it really depends on (00:22:33) the use case and what we're doing. And (00:22:35) then there's a speed function which is (00:22:37) we can in the past five years ago that (00:22:39) would take us a year. (00:22:40) >> Sure. Now it could take a week. (00:22:41) >> Week. Yeah. (00:22:42) >> Let me further that question though. I'm (00:22:45) sure it's on the minds of some people (00:22:46) here. Is AI going to create jobs or (00:22:49) destroy jobs overall? (00:22:50) >> It Yeah. I think one of the unfortunate (00:22:53) things of the narrative in the west is (00:22:56) it it will destroy humanity's (00:23:00) jobs of like you know you went to an (00:23:03) elite school and you studied philosophy. (00:23:06) Use myself as an example. Um, (00:23:10) >> I did too. (00:23:10) >> Yeah, you it hopefully you have some (00:23:13) other skill. That one is going to be (00:23:16) hard to market. Uh, and (00:23:18) >> we thought it was hard to market. (00:23:19) >> It was hard to market. Very hard. Uh, (00:23:21) >> it was a good education. (00:23:22) >> A very, very strong education. If you (00:23:24) can get a job, you might keep it. But (00:23:26) the hard That's what I always thought. I (00:23:27) was like, if I finally get a job, I'll (00:23:29) probably keep it and do well, but I'm (00:23:30) not sure who's going to give me my first (00:23:32) job. Um and uh um uh uh but like techn (00:23:38) like technicians if you're a vocational (00:23:41) technician (00:23:42) >> or like like we're building batteries (00:23:45) for a battery company and the people who (00:23:47) are doing it in America are doing (00:23:49) roughly the same job that Japanese (00:23:51) engineers are doing and they went to (00:23:52) high school and now they're very (00:23:54) valuable if not irreplaceable because we (00:23:57) can make them into something different (00:24:00) than what they were very rapidly and (00:24:02) those jobs are going to become more (00:24:04) valuable. (00:24:05) Um, I mean, you know, I not not to (00:24:08) diverge into my usual political screeds, (00:24:10) but it there are will be more than (00:24:13) enough jobs for the citizens of your (00:24:15) nation, especially those with vocational (00:24:18) training. I do think these these trends (00:24:20) really do make it hard to imagine why we (00:24:23) should have large-scale immigration (00:24:24) unless you have a very specialized skill (00:24:26) because (00:24:27) >> what about (00:24:29) the foundation for white collar work in (00:24:32) Europe and the United States has been (00:24:33) through the universities but I just (00:24:35) heard you say we're going to need more (00:24:37) vocational men and women and they may (00:24:41) they're going to be but are you also (00:24:43) insinuating we're probably going to need (00:24:44) less white collar? I think like I think (00:24:47) what we need to do is yes, but I I think (00:24:50) we need different ways of testing (00:24:51) aptitude. You know, it's like um you you (00:24:54) know there are a lot of people doing X (00:24:57) that should be doing Y. Like if you (00:24:59) could manage one of our system like just (00:25:01) the person managing our maven system in (00:25:04) the US Army is a former police officer (00:25:07) who I think went to a junior college and (00:25:10) they're doing very very high-end very (00:25:13) complicated targeting globally and that (00:25:16) person actually is irreplaceable and I (00:25:20) think in the past the way we tested for (00:25:23) aptitude uh would not have fully exposed (00:25:27) how irreplaceable that person's talents (00:25:29) are and would they been as talented if (00:25:31) they had not [clears throat] gone to (00:25:32) their college? Yes. Um and but I think (00:25:36) the the a I tend to even inside (00:25:38) Palunteer if you look at inside (00:25:39) Palunteer what am I really doing all (00:25:42) day? I'm want walking around figuring (00:25:44) out what is someone's outlier aptitude (00:25:47) and then I'm putting them on that thing (00:25:49) and trying to get them to stay on that (00:25:51) thing and not on the five other things (00:25:53) they think they're great at like you (00:25:55) know (00:25:55) >> keeping their (00:25:57) >> Yeah. It's like well you know everyone (00:25:58) at Palanteer every every engineer at (00:26:00) Palunteer uh it's it's the most wherever (00:26:03) I go in the world like for for as you (00:26:05) know maybe for 18 years everyone thought (00:26:08) we were like a business joke and now (00:26:09) lots of business people want my advice (00:26:11) you know the only people who don't want (00:26:12) my advice at Palanteer about business (00:26:14) are Palunteer engineers they're like hey (00:26:16) Alex I have an idea about how we could (00:26:18) just be in a much better company and (00:26:20) it's it's always like yeah it's like (00:26:23) it's like literally McDonald's but it's (00:26:25) like we should have some titles (00:26:27) and you should stop speaking in public. (00:26:29) [laughter] And uh yeah, and then I mean (00:26:31) there's probably right about speaking in (00:26:32) public sometimes. I certainly admit (00:26:34) that. (00:26:35) >> I don't think you uh I don't think you (00:26:37) did anything wrong today. (00:26:38) >> Yeah. So uh yeah, thank you for that (00:26:39) high praise. [laughter] (00:26:43) >> One of the keys to success is setting (00:26:44) the bar very low. (00:26:45) >> Yes. No, I don't believe that's how you (00:26:48) uh operate Palunteer. Um one last (00:26:50) question. (00:26:54) Where where is this where will the curve (00:26:57) of AI go in the utilization uh in the (00:27:00) United States and other developed (00:27:03) economies? But what about the developing (00:27:06) economies? How can they participate in (00:27:09) this? I mean, I read a research report (00:27:12) yesterday that said the application of (00:27:14) AI has been so dominant by societies of (00:27:19) high education or companies of high (00:27:21) education and they're seeing a very big (00:27:24) um divergence that is occurring already (00:27:28) and it's so much based on the (00:27:30) application of education and how that is (00:27:32) being utilized. So, is AI going to (00:27:35) create more um a more a greater (00:27:38) imbalance in our in our world in terms (00:27:41) of growth? (00:27:42) >> Well, I think the obvious first (00:27:44) imbalance is it seems like America and (00:27:47) China understand versions of making this (00:27:50) work and they're different. (00:27:51) >> Yes. (00:27:52) >> But they both work and they work at (00:27:54) scale and I think that is very likely to (00:27:57) accelerate way beyond what most people (00:27:59) believe is possible. like the discount (00:28:01) rate I think not in the short term but (00:28:03) in the long term is way too high on what (00:28:06) will be done and how this will impact (00:28:08) every aspect of our society and I would (00:28:10) say especially on military and then I I (00:28:14) tend to be a realist and that I think (00:28:16) you know you have wide divergences it's (00:28:18) going to be hard to have the kind of (00:28:20) discussions people want to have where (00:28:22) two countries are and and with a maybe a (00:28:24) third following of Russia on the new on (00:28:27) like the because they they're so good at (00:28:28) fighting (00:28:30) Um and then and then I I look I spent (00:28:33) and I'll get to the developing world. I (00:28:35) spent a lot of my life my most important (00:28:38) years and my father's family came from a (00:28:42) part of Germany and I I really care (00:28:44) about Europe and especially the germ (00:28:46) parts of Europe uh where I had many of (00:28:49) my best years. I still fantasize of (00:28:51) going back to grad school for not for (00:28:54) the learning reasons. Um uh and um (00:28:58) >> you're going to have more fun. (00:28:59) >> I had so much fun. Oh, we we won't go (00:29:01) into that. Uh but uh it's like endless. (00:29:03) I sometimes when I'm traveling across (00:29:05) the country, I just think of grad (00:29:06) school. But um uh it's uh uh um but um (00:29:13) um uh I I the the tech adoption in in in (00:29:19) Europe is a serious and very very (00:29:23) structural problem. And what scares me (00:29:26) the most is I haven't seen any political (00:29:28) leader just stand up and say we have a (00:29:31) serious and structural problem that we (00:29:33) are going to fix. So that then you get (00:29:36) to the developing world. I would imagine (00:29:39) it also depends what you mean by the (00:29:40) developing world. I would imagine with (00:29:44) not enough knowledge you're just going (00:29:46) to find pockets that go very well and (00:29:48) pockets that go very poorly. As a (00:29:50) generalization like again if you go back (00:29:52) to this somewhat n unsuccessful salopy I (00:29:56) had about the underlying architecture. (00:29:58) One way to look at at the unfairness of (00:30:00) AI is it pentests meaning it it (00:30:03) loadbears on things. So societies that (00:30:06) can and organizations and companies that (00:30:08) can bear that load have a huge (00:30:11) advantage. The problem is if you can't (00:30:13) if you've been pretending you're bearing (00:30:15) a load you're not it collapses and (00:30:18) that's where you have to start. And so (00:30:20) if you go around and just say okay what (00:30:23) societies and micro cultures are going (00:30:25) to be loadbearing here I think you would (00:30:27) find that parts of the developing world (00:30:30) certain communities in that are going to (00:30:31) do very well you you do need a realistic (00:30:35) assessment of the loadbearing (00:30:37) >> and there there's a certain honesty that (00:30:40) is painful for all of us in in in this (00:30:42) technology large language models however (00:30:46) implemented in software it you just (00:30:49) cannot not obfuscate what can bear the (00:30:51) load and what can't. And then political (00:30:53) structures are built to do just that. (00:30:56) Like, yeah, I can't fix anything, but I (00:30:58) can give you some line that you (00:31:00) want to hear that's going to make you (00:31:02) not care about how bad your life is and (00:31:04) how much worse it's going to be (00:31:05) tomorrow. (00:31:06) >> I can give you that for free. (00:31:07) >> And those that that stuff uh is um (00:31:13) that is harder to get away with in this (00:31:16) culture. And you know, I I still view (00:31:18) myself as a card carrying progressive. (00:31:20) And I think it's the single most (00:31:23) important thing a progressive could do (00:31:25) is go around and say, "Yeah, but the (00:31:27) revolution that's coming is going to (00:31:29) expose the actual market value of what (00:31:31) you're doing, whether we want it or (00:31:33) not." Like it's like I don't even want (00:31:35) to know the market value of some of this (00:31:36) stuff. But it is over and over a (00:31:39) relative rapid period of time. So next (00:31:41) three years you're just going to get (00:31:43) market value honesty in all sorts of (00:31:46) character communities and micro (00:31:48) communities and the best thing you can (00:31:50) do if you are in a community whether (00:31:53) that is a large community like Germany (00:31:55) or a large community larger like America (00:31:58) is and you really care for the people (00:32:00) you're representing is to say yeah but (00:32:02) let's (00:32:04) we have to kind of look closely at what (00:32:06) what load we can bear. (00:32:09) Thank you, Alex. (00:32:11) >> Thank you, everyone. [applause] (00:32:17) [applause]