1 00:00:00,000 --> 00:00:03,837 I want to talk about AI and I would really want to because I think this is 2 00:00:03,920 --> 00:00:06,237 on everybody's mind more than almost any 3 00:00:06,320 --> 00:00:10,397 other subject today um related to 4 00:00:10,480 --> 00:00:11,037 intersection [clears throat] 5 00:00:11,120 --> 00:00:13,437 of business technology 6 00:00:13,520 --> 00:00:17,757 society um so Satia um 7 00:00:17,840 --> 00:00:21,517 you know we're we're moving AI from something that was experimental 8 00:00:21,600 --> 00:00:23,437 something that we always talked about in 9 00:00:23,520 --> 00:00:28,397 the future and now it's it's today um 10 00:00:28,480 --> 00:00:31,997 and it's now more foundational and it's 11 00:00:32,080 --> 00:00:36,237 not just foundational for companies but it it really is becoming now 12 00:00:36,320 --> 00:00:38,717 foundational 13 00:00:38,800 --> 00:00:42,557 for countries and throughout society and 14 00:00:42,640 --> 00:00:45,197 I think you know you have an advantage 15 00:00:45,280 --> 00:00:47,517 over so many other people you know being 16 00:00:47,600 --> 00:00:49,597 at the forefront of this technology 17 00:00:49,680 --> 00:00:53,837 transition. Um so um with that I wanted 18 00:00:53,920 --> 00:00:57,517 to ask a few questions related to that 19 00:00:57,600 --> 00:00:59,757 you have described that AI is a plat 20 00:00:59,840 --> 00:01:02,637 plat uh platform shift 21 00:01:02,720 --> 00:01:04,797 and what does what does that mean 22 00:01:04,880 --> 00:01:07,837 question one where do you see that shift 23 00:01:07,920 --> 00:01:12,077 going in the next few years and importantly the third part of my 24 00:01:12,160 --> 00:01:16,397 question would be fast forwarding a few years five years 25 00:01:16,480 --> 00:01:21,357 what's going to seem obvious in hindsight that feels less clear today. 26 00:01:21,440 --> 00:01:24,237 You >> know, first of all, um it's great uh to 27 00:01:24,320 --> 00:01:27,597 be back here, uh Larry, and it's um I 28 00:01:27,680 --> 00:01:31,277 had a chance, in fact, yesterday when you put out the letter [clears throat] 29 00:01:31,360 --> 00:01:34,557 uh to kick off the forum and read it and 30 00:01:34,640 --> 00:01:37,277 um and in there you sort of had this one 31 00:01:37,360 --> 00:01:41,437 line of uh really I think when it comes 32 00:01:41,520 --> 00:01:43,757 to AI, the real question in front of all 33 00:01:43,840 --> 00:01:48,477 of us is how do you ensure that the diffusion of AI happens and happens 34 00:01:48,560 --> 00:01:52,797 fast? I mean I think you had that line of how do the models, the data and the 35 00:01:52,880 --> 00:01:56,637 infrastructure spread more evenly to create surplus 36 00:01:56,720 --> 00:01:58,317 everywhere. If you sort of think about 37 00:01:58,400 --> 00:02:02,077 it, the the way I come at this is not 38 00:02:02,160 --> 00:02:05,837 that um this has always been the arc of 39 00:02:05,920 --> 00:02:10,317 computation, right? You can sort of take it in the last 30 years or the last 70 40 00:02:10,400 --> 00:02:12,557 years. It's always been about can you 41 00:02:12,640 --> 00:02:14,077 digitize 42 00:02:14,160 --> 00:02:17,997 artifacts on about people, places, and 43 00:02:18,080 --> 00:02:22,157 things and then build analytical and 44 00:02:22,240 --> 00:02:23,517 predictive [clears throat] power. Right? 45 00:02:23,600 --> 00:02:28,237 That's what the mainframes did. That's what the mini computer did. That's what 46 00:02:28,320 --> 00:02:32,397 the client server error did. That's what the web error did. The mobile cloud 47 00:02:32,480 --> 00:02:34,637 error did. So it like it depend 48 00:02:34,720 --> 00:02:37,757 irrespective of which paradigm or 49 00:02:37,840 --> 00:02:41,677 platform. It has been one continuous arc 50 00:02:41,760 --> 00:02:44,957 of saying let's make better sense of 51 00:02:45,040 --> 00:02:46,877 this world 52 00:02:46,960 --> 00:02:51,597 um by reasoning about it in digital form 53 00:02:51,680 --> 00:02:56,477 because in some sense once you have these artifacts in digital form you can 54 00:02:56,560 --> 00:02:59,357 use a more malleable resource like software 55 00:02:59,440 --> 00:03:05,287 >> right >> um which doesn't have the same type of you know I'll call it 56 00:03:05,370 --> 00:03:08,317 >> [clears throat] >> um marginal cost economics associated 57 00:03:08,400 --> 00:03:10,797 ated with it that allows us to then 58 00:03:10,880 --> 00:03:13,517 build more insight and more uh more 59 00:03:13,600 --> 00:03:17,037 capability and in that context AI I 60 00:03:17,120 --> 00:03:19,437 would say is of the same class at least 61 00:03:19,520 --> 00:03:22,797 like the web or the internet um or 62 00:03:22,880 --> 00:03:25,357 mobile or PC or the cloud or and maybe 63 00:03:25,440 --> 00:03:29,597 even greater and so to me right now 64 00:03:29,680 --> 00:03:32,797 where we are is you know let's take just 65 00:03:32,880 --> 00:03:36,797 what's happened with software engineering right which is one you know 66 00:03:36,880 --> 00:03:39,597 is knowledge work um you know you could 67 00:03:39,680 --> 00:03:41,917 say it's elite knowledge work 68 00:03:42,000 --> 00:03:44,877 >> it started off uh you know in fact my 69 00:03:44,960 --> 00:03:49,677 own belief in this generation of AI and its capability 70 00:03:49,760 --> 00:03:53,917 uh really got built up when I first saw GitHub copilot do code completions right 71 00:03:54,000 --> 00:03:58,477 so for the longest time we had the dream that if you're a software developer can 72 00:03:58,560 --> 00:04:00,877 you predict the next sort of word or the 73 00:04:00,960 --> 00:04:04,557 next uh line of code uh and suddenly it 74 00:04:04,640 --> 00:04:07,197 started working with these models u then 75 00:04:07,280 --> 00:04:09,197 You said okay if I can do that then can 76 00:04:09,280 --> 00:04:12,397 I actually go and bring back you know 77 00:04:12,480 --> 00:04:16,797 the flow for a software developer by going to a chat session and asking any 78 00:04:16,880 --> 00:04:21,197 question and it comes back with answers that then uh you can use in your coding 79 00:04:21,280 --> 00:04:24,797 flow right so that was the next thing then you said well if that's working can 80 00:04:24,880 --> 00:04:29,997 I assign it small tasks that was the agent mode uh now you have complete 81 00:04:30,080 --> 00:04:34,557 autonomous agents where you can give it your entire project right it can work uh 82 00:04:34,640 --> 00:04:37,997 you know >> 24/7 >> it can work for 24/7. I mean it's still 83 00:04:38,080 --> 00:04:42,237 we've got some ways to go for these things to remain coherent long time but 84 00:04:42,320 --> 00:04:44,557 nevertheless it's getting better and 85 00:04:44,640 --> 00:04:47,757 better and interestingly enough you look 86 00:04:47,840 --> 00:04:51,197 at it uh the software developer still is 87 00:04:51,280 --> 00:04:54,877 got a lot of agency in it right so that's why I kind of still think that 88 00:04:54,960 --> 00:04:59,517 you know going and thinking of these as somehow living outside of the realm of 89 00:04:59,600 --> 00:05:03,597 human agency is probably not the right way to think about it in fact the way to 90 00:05:03,680 --> 00:05:06,717 perhaps conceive it like it's let's say in early8s 91 00:05:06,800 --> 00:05:10,717 If somebody had come to us and said whatever four billion people are going 92 00:05:10,800 --> 00:05:12,637 to wake up every morning and start 93 00:05:12,720 --> 00:05:15,437 typing you would have said why right you 94 00:05:15,520 --> 00:05:20,077 know we have a like we have a typist pool that's good enough we don't need 4 95 00:05:20,160 --> 00:05:24,157 billion people but we that's what happened right we invented this entire 96 00:05:24,240 --> 00:05:26,797 class of thing called knowledge work 97 00:05:26,880 --> 00:05:29,197 where people started really using 98 00:05:29,280 --> 00:05:32,637 computers uh to go amplify what we were 99 00:05:32,720 --> 00:05:37,437 trying to achieve uh using software I think in the context of AI that same 100 00:05:37,520 --> 00:05:41,917 thing is going to happen. >> Um it's not like you know what is 101 00:05:42,000 --> 00:05:46,077 hardcore coding is going to remain hardcore coding forever. It's just that 102 00:05:46,160 --> 00:05:47,677 the levels of abstraction are going to 103 00:05:47,760 --> 00:05:50,637 change. Uh but we also are going to have 104 00:05:50,720 --> 00:05:53,117 code as output just like documents. In 105 00:05:53,200 --> 00:05:57,997 fact, one of Bill's things at Microsoft from the day I joined in '92 always was 106 00:05:58,080 --> 00:05:59,677 what's the real difference between a 107 00:05:59,760 --> 00:06:02,637 document, a website and an application, 108 00:06:02,720 --> 00:06:05,277 right? It's the lack of sort of software 109 00:06:05,360 --> 00:06:07,917 that can transform itself. Interestingly 110 00:06:08,000 --> 00:06:09,997 enough, AI finally gives us that, right? 111 00:06:10,080 --> 00:06:13,837 Which is I can write a document. I can just say no, I don't want it as a 112 00:06:13,920 --> 00:06:17,837 document. I want it as a website. It'll just transform that document using code 113 00:06:17,920 --> 00:06:20,237 into a website. I say I don't like the 114 00:06:20,320 --> 00:06:22,077 website. I want an app. It'll write more 115 00:06:22,160 --> 00:06:25,197 code to transform it. So that reasoning 116 00:06:25,280 --> 00:06:27,837 and cap reasoning capabilities that 117 00:06:27,920 --> 00:06:30,077 prediction capabilities that ability to 118 00:06:30,160 --> 00:06:33,517 take action remain long-term coherent is 119 00:06:33,600 --> 00:06:38,477 all improving. Um and our job though is 120 00:06:38,560 --> 00:06:42,957 to parlay this like take even what you at Black Rockck are doing right when you 121 00:06:43,040 --> 00:06:46,877 are bring taking something like say co-pilot plus Aladdin and bringing those 122 00:06:46,960 --> 00:06:50,797 things together [clears throat] >> to improve the productivity in the firm 123 00:06:50,880 --> 00:06:54,317 for the decisions you want to make right with your data 124 00:06:54,400 --> 00:06:56,317 >> I could just tell you from in our firm 125 00:06:56,400 --> 00:06:58,637 things that would take 12 hours to 126 00:06:58,720 --> 00:07:04,077 compute now takes minutes for us processing 14 trillion dollars of other 127 00:07:04,160 --> 00:07:05,517 people's money with hundreds of 128 00:07:05,600 --> 00:07:09,357 thousands of different um mandates um we 129 00:07:09,440 --> 00:07:12,797 could do that instantaneous and we you know that to me if it wasn't for the 130 00:07:12,880 --> 00:07:16,877 technology and AI today we would not be able to function to the scale that we're 131 00:07:16,960 --> 00:07:20,797 operating >> that's right and so to me that one firm 132 00:07:20,880 --> 00:07:23,357 at a time one country at a time if we 133 00:07:23,440 --> 00:07:26,477 can really take these tokens and bend 134 00:07:26,560 --> 00:07:30,557 the curve of productivity then there is surplus everywhere and that's really the 135 00:07:30,640 --> 00:07:33,597 goal >> well surplus could be scary too does 136 00:07:33,680 --> 00:07:37,117 does surplus mean fewer workers? What do 137 00:07:37,200 --> 00:07:41,357 we mean by surplus? And so, you know, the I want to tie that into my second 138 00:07:41,440 --> 00:07:43,997 question about AI diffusion. >> Yeah. 139 00:07:44,080 --> 00:07:46,877 >> To me, the the whole realization of AI 140 00:07:46,960 --> 00:07:50,317 for any society and also for a more 141 00:07:50,400 --> 00:07:53,197 balanced world is making sure that it's 142 00:07:53,280 --> 00:07:55,837 diffused and accessible and available 143 00:07:55,920 --> 00:07:58,157 across the world. So what you know can 144 00:07:58,240 --> 00:08:00,877 you describe how this process of 145 00:08:00,960 --> 00:08:06,077 diffusion uh across economies across companies across people and countries 146 00:08:06,160 --> 00:08:09,997 how does that play out? >> Yeah I think 147 00:08:10,080 --> 00:08:14,397 that this is the real question right because one of the uh things right now 148 00:08:14,480 --> 00:08:16,957 the zeitgeist is a little bit about the 149 00:08:17,040 --> 00:08:20,477 admiration for AI in its abstract form 150 00:08:20,560 --> 00:08:24,157 or as as technology. 151 00:08:24,240 --> 00:08:25,837 uh [clears throat] but I think we as 152 00:08:25,920 --> 00:08:28,957 even a global community um have to get 153 00:08:29,040 --> 00:08:32,077 to a point where we're using this to do 154 00:08:32,160 --> 00:08:35,677 something useful uh that changes uh the 155 00:08:35,760 --> 00:08:39,997 outcomes of people and communities and countries and industries right otherwise 156 00:08:40,080 --> 00:08:42,397 I don't think uh this makes much sense 157 00:08:42,480 --> 00:08:44,397 right in fact I would say we will 158 00:08:44,480 --> 00:08:47,837 quickly lose even the social permission 159 00:08:47,920 --> 00:08:50,477 uh to actually take something like uh 160 00:08:50,560 --> 00:08:54,877 energy which is a scarce resource and use it to generate these tokens if these 161 00:08:54,960 --> 00:08:58,637 tokens are not improving health outcomes, education outcomes, public 162 00:08:58,720 --> 00:09:03,117 sector efficiency, private sector competitiveness across all sectors, 163 00:09:03,200 --> 00:09:05,597 small and large, right? And that to me 164 00:09:05,680 --> 00:09:08,637 is ultimately the goal. So therefore I 165 00:09:08,720 --> 00:09:11,677 think really diffusion is everything. 166 00:09:11,760 --> 00:09:14,317 And so the way it happens is let's sort 167 00:09:14,400 --> 00:09:18,077 of unpack this uh on the supply side 168 00:09:18,160 --> 00:09:20,237 what needs to happen in each country is 169 00:09:20,320 --> 00:09:24,157 the tokens per dollar per watt have to 170 00:09:24,240 --> 00:09:29,197 sort of monotonically get more efficient and better right so to some degree even 171 00:09:29,280 --> 00:09:32,797 what we're trying to do with the investments the two firms are doing uh 172 00:09:32,880 --> 00:09:37,117 around the world is to just say that like let's make sure that the supply is 173 00:09:37,200 --> 00:09:39,677 there which is uh everything from the 174 00:09:39,760 --> 00:09:44,397 chips on down ultimately ely to these token factories that get deployed 175 00:09:44,480 --> 00:09:48,477 everywhere. By the way, there's not one token factory. This token factory is the 176 00:09:48,560 --> 00:09:52,157 first thing that's going to be diffused all around the world. It's just like 177 00:09:52,240 --> 00:09:55,117 electricity, right? You just need a 178 00:09:55,200 --> 00:09:59,917 ubiquitous grid of uh energy and tokens 179 00:10:00,000 --> 00:10:02,477 that then will power the rest of the 180 00:10:02,560 --> 00:10:04,237 economy, right? So, that's I think one 181 00:10:04,320 --> 00:10:07,437 side of it. Then the demand side of this 182 00:10:07,520 --> 00:10:10,477 is a little bit like every firm has to 183 00:10:10,560 --> 00:10:13,677 start by using it. If I look back even, 184 00:10:13,760 --> 00:10:16,797 you know, when the PCs first came out or 185 00:10:16,880 --> 00:10:20,957 the personal computing era started, I I loved, you know, I think Jobs had a nice 186 00:10:21,040 --> 00:10:25,917 metaphor. He called it the bicycle for the mind. Uh Bill had a metaphor which I 187 00:10:26,000 --> 00:10:29,117 remember was like information at your fingertips, right? These two metaphors 188 00:10:29,200 --> 00:10:33,037 were great like which allowed us to say that's what it is. It's a tool that I 189 00:10:33,120 --> 00:10:36,957 will use to get information at my fingertips. I'll use it as a cognitive 190 00:10:37,040 --> 00:10:38,477 amplifier. 191 00:10:38,560 --> 00:10:42,157 Now I think that's what we have. But you 192 00:10:42,240 --> 00:10:44,317 know 10x 100x right? So in some sense 193 00:10:44,400 --> 00:10:48,077 you as a as every knowledge worker you 194 00:10:48,160 --> 00:10:50,797 now have access to infinite minds. 195 00:10:50,880 --> 00:10:52,237 That's the way I think about it. Right. 196 00:10:52,320 --> 00:10:56,077 So there's a Turing award winner uh called Raj Ready who had this nice 197 00:10:56,160 --> 00:11:00,797 metaphor of AI and he had this long before uh even generative AI. He said 198 00:11:00,880 --> 00:11:03,037 either either it's a cognitive amplifier 199 00:11:03,120 --> 00:11:05,757 or it's a guardian angel right so if you 200 00:11:05,840 --> 00:11:09,597 think of AI as that um then that in the 201 00:11:09,680 --> 00:11:13,677 global workforce right when a doctor can 202 00:11:13,760 --> 00:11:18,077 get to a patient spend more time with the patient because the AI is doing the 203 00:11:18,160 --> 00:11:20,317 transcription and entering the records 204 00:11:20,400 --> 00:11:23,197 in the EMR system entering the right 205 00:11:23,280 --> 00:11:25,917 billing code so that the health care uh 206 00:11:26,000 --> 00:11:28,077 you know industry is better served 207 00:11:28,160 --> 00:11:31,117 across the payer the provider and uh and 208 00:11:31,200 --> 00:11:33,997 the patient ultimately right that's an 209 00:11:34,080 --> 00:11:36,077 an outcome that I think all of us can 210 00:11:36,160 --> 00:11:39,037 benefit from so I feel ultimately it's 211 00:11:39,120 --> 00:11:41,677 going to require real leadership on the 212 00:11:41,760 --> 00:11:44,237 private sector and the public sector to 213 00:11:44,320 --> 00:11:46,637 ensure that diffusion happens and the 214 00:11:46,720 --> 00:11:48,717 one thing other thing I'll mention Larry 215 00:11:48,800 --> 00:11:52,957 is skilling right so in some sense the 216 00:11:53,040 --> 00:11:56,557 thing that diffusion is very strongly 217 00:11:56,640 --> 00:11:59,597 correlated to one thing alone which is 218 00:11:59,680 --> 00:12:02,717 How broadly are people skilled in using 219 00:12:02,800 --> 00:12:06,317 this? Um, and interestingly enough, I 220 00:12:06,400 --> 00:12:09,437 think if mobile has taught us one thing 221 00:12:09,520 --> 00:12:14,157 is it it's actually distinct from what happened in the PC, right? Uh, I 222 00:12:14,240 --> 00:12:18,477 remember even growing up in the global south, uh, there used to be a real 223 00:12:18,560 --> 00:12:20,957 relationship between learning Excel 224 00:12:21,040 --> 00:12:24,877 skills or word skills and getting a job. 225 00:12:24,960 --> 00:12:27,917 um you know right now um what's the 226 00:12:28,000 --> 00:12:32,317 model in in mobile it's kind of created the same opportunity but it's been a lot 227 00:12:32,400 --> 00:12:37,037 more consumption-led it's these creator economy and what have you but it has not 228 00:12:37,120 --> 00:12:40,717 been about sort of oh wow here is how you get a healthare job or here is how 229 00:12:40,800 --> 00:12:45,037 you get a finance job or here is how you get you know you get ahead 230 00:12:45,120 --> 00:12:49,597 professionally um and that needs to come back right 231 00:12:49,680 --> 00:12:51,517 people need to say I pick up this AI 232 00:12:51,600 --> 00:12:56,877 skill and now I'm a better provider of some product or service in the real 233 00:12:56,960 --> 00:12:58,957 economy. >> So it's it's very [clears throat] easy 234 00:12:59,040 --> 00:13:01,837 to see how mobile and the diffusion of 235 00:13:01,920 --> 00:13:04,077 mobile how it transformed economies 236 00:13:04,160 --> 00:13:06,557 especially in the global south. 237 00:13:06,640 --> 00:13:10,797 How does how does this you know to me I I I just read a research report that 238 00:13:10,880 --> 00:13:15,197 said the applications for AI so far are 239 00:13:15,280 --> 00:13:19,837 heavily weighted towards those who are educated or educated economies. 240 00:13:19,920 --> 00:13:23,997 And so does that create that, you know, more of a bifurcation, a more 241 00:13:24,080 --> 00:13:29,357 polarization? How do we ensure that that that diffusion is spread evenly? How do 242 00:13:29,440 --> 00:13:30,957 we make sure that we're [clears throat] 243 00:13:31,040 --> 00:13:37,037 not leaving major portions of society or the world behind because I think that's 244 00:13:37,120 --> 00:13:39,677 that's [clears throat] going to be the big issue for us going forward. 245 00:13:39,760 --> 00:13:41,117 >> Yeah. So it's it's it's interesting, 246 00:13:41,200 --> 00:13:44,797 right? This is one of those times when 247 00:13:44,880 --> 00:13:48,397 uh by definition and because of the 248 00:13:48,480 --> 00:13:50,237 rails that have been established you 249 00:13:50,320 --> 00:13:53,037 know as you said right which is uh 250 00:13:53,120 --> 00:13:57,597 what's happened with mobile as well as what's happened with um 251 00:13:57,680 --> 00:14:00,317 >> uh you know essentially connectivity >> right 252 00:14:00,400 --> 00:14:04,877 >> you have the ability [clears throat] to sort of deliver the tokens pretty evenly 253 00:14:04,960 --> 00:14:09,917 around the world >> right >> a lot more so than let's say uh the PC 254 00:14:10,000 --> 00:14:13,357 era or even the beginning of the mobile era, right? Because it took a long time 255 00:14:13,440 --> 00:14:15,597 for even the mo smartphone in particular 256 00:14:15,680 --> 00:14:18,797 to penetrate um all of the world. 257 00:14:18,880 --> 00:14:21,277 Whereas now it's not the case, right? 258 00:14:21,360 --> 00:14:24,637 These models and their outputs are 259 00:14:24,720 --> 00:14:26,557 pretty much available everywhere. And so 260 00:14:26,640 --> 00:14:29,837 the question to me is 261 00:14:29,920 --> 00:14:34,077 what's the use cases that make sense, right? One, in fact, one of the demos I 262 00:14:34,160 --> 00:14:35,997 always go back to, I think this was even 263 00:14:36,080 --> 00:14:38,557 in the beginning of 23 was a rural 264 00:14:38,640 --> 00:14:41,517 Indian farmer was able to use a bot 265 00:14:41,600 --> 00:14:45,197 built on I think a very early GPT3 or 25 266 00:14:45,280 --> 00:14:48,637 even uh essentially to reason over some 267 00:14:48,720 --> 00:14:50,717 farm subsidies that he had heard about 268 00:14:50,800 --> 00:14:53,757 in a local language and had it even in 269 00:14:53,840 --> 00:14:56,877 that very early days uh have it even 270 00:14:56,960 --> 00:14:59,037 show some agentic behavior right like go 271 00:14:59,120 --> 00:15:00,797 complete a form for me. So in some sense 272 00:15:00,880 --> 00:15:03,917 it took uh you know it brought back 273 00:15:04,000 --> 00:15:07,917 agency to someone uh who perhaps didn't 274 00:15:08,000 --> 00:15:12,397 have that because the technology was so much more accessible. So I I do think 275 00:15:12,480 --> 00:15:14,717 it's in our hands even in the global 276 00:15:14,800 --> 00:15:18,957 south to use it uh to create I would say 277 00:15:19,040 --> 00:15:21,277 more of that opportunity where there 278 00:15:21,360 --> 00:15:24,787 isn't one. Um but I think what the 279 00:15:24,870 --> 00:15:25,997 [clears throat] necessary conditions 280 00:15:26,080 --> 00:15:29,037 still are do you have uh the capital 281 00:15:29,120 --> 00:15:33,517 investment being put in uh do you even have an environment for capital because 282 00:15:33,600 --> 00:15:37,917 in an interesting way we are for example as hyperscalers investing all over right 283 00:15:38,000 --> 00:15:42,237 including the global south uh so as long as there's an environment which attracts 284 00:15:42,320 --> 00:15:44,557 the capital investment >> and you see the demand 285 00:15:44,640 --> 00:15:47,437 >> and you see and then yeah the demand is there 286 00:15:47,520 --> 00:15:51,757 >> um and so the question is how do you have a set of policies that allow for 287 00:15:51,840 --> 00:15:56,077 both the capital to come in for it to find nexus where there are certain 288 00:15:56,160 --> 00:15:58,317 things by the way private capital can do 289 00:15:58,400 --> 00:16:00,157 certain things that public capital only 290 00:16:00,240 --> 00:16:03,197 can do for example the grid right it's 291 00:16:03,280 --> 00:16:08,317 not I mean grid in most countries is sort of fundamentally driven by governments 292 00:16:08,400 --> 00:16:10,317 >> public >> and public and so if [clears throat] you 293 00:16:10,400 --> 00:16:12,637 don't have a sophisticated sort or 294 00:16:12,720 --> 00:16:17,597 rather if you don't have a a real approach to modernizing the grid uh that 295 00:16:17,680 --> 00:16:21,037 will hold things back I mean there's a lot of talk about behind the meter and 296 00:16:21,120 --> 00:16:23,677 so on and Yes, there's some amount of that we can do ourselves. 297 00:16:23,760 --> 00:16:25,917 >> We can do that in the US. Many countries can't. 298 00:16:26,000 --> 00:16:29,357 >> Exactly. And it's not long-term scalable, right? I mean, like to me, a 299 00:16:29,440 --> 00:16:32,877 long-term scalable solution is to have 300 00:16:32,960 --> 00:16:37,677 uh you know, all of these token factories, part of the real economy, 301 00:16:37,760 --> 00:16:42,397 connected to the grid, connected to the telco network, delivering uh just like 302 00:16:42,480 --> 00:16:44,397 we delivered bits, you have to deliver 303 00:16:44,480 --> 00:16:47,357 tokens plus bits. Um and that's kind of 304 00:16:47,440 --> 00:16:51,757 what's going to drive an at scale whether it's in the global south or in 305 00:16:51,840 --> 00:16:55,037 uh on the developed world. >> So so many people talk about there may 306 00:16:55,120 --> 00:16:57,037 be an AI bubble. I mean the most 307 00:16:57,120 --> 00:16:58,477 important thing that we see as an 308 00:16:58,560 --> 00:17:02,637 investor is the the democratization of 309 00:17:02,720 --> 00:17:05,517 technology is and the diffusion of that 310 00:17:05,600 --> 00:17:08,077 technology really does then transform 311 00:17:08,160 --> 00:17:13,117 the demand and the the companies or the countries that diffuse it fastest are 312 00:17:13,200 --> 00:17:14,637 going to be the ultimate winners. not 313 00:17:14,720 --> 00:17:16,477 the technology creator. 314 00:17:16,560 --> 00:17:20,797 >> That's that's you know it's it's it's 315 00:17:20,880 --> 00:17:23,117 for this not to be a bubble 316 00:17:23,200 --> 00:17:23,757 >> y 317 00:17:23,840 --> 00:17:27,917 >> by definition it requires that the 318 00:17:28,000 --> 00:17:30,157 benefits of this are much more evenly 319 00:17:30,240 --> 00:17:33,357 spread. I mean I think a tellt tell sign 320 00:17:33,440 --> 00:17:36,797 of if it's a bubble would be if all 321 00:17:36,880 --> 00:17:39,997 we're talking about are the tech firms. 322 00:17:40,080 --> 00:17:43,117 Uh right? If uh all we talk about is 323 00:17:43,200 --> 00:17:47,277 what's happening to the technology side that then that's by you know it's just 324 00:17:47,360 --> 00:17:51,437 purely supply side >> right >> uh ultimately if we are not talking 325 00:17:51,520 --> 00:17:55,677 about wow here is a drug comp you know drug that was sort of uh brought into 326 00:17:55,760 --> 00:17:59,757 the market that's super successful because it was uh AI accelerated the 327 00:17:59,840 --> 00:18:01,837 clinical trial it's not even the magical 328 00:18:01,920 --> 00:18:03,757 molecule right it's kind of even the 329 00:18:03,840 --> 00:18:06,877 rest of what is uh needed in order to 330 00:18:06,960 --> 00:18:09,517 make something much more relevant right 331 00:18:09,600 --> 00:18:14,797 um and So the more [clears throat] we have uh and by the way it's happening 332 00:18:14,880 --> 00:18:19,277 right. So I'm not sort of saying that that's why I'm much more confident that 333 00:18:19,360 --> 00:18:21,757 this is a technology that will in fact 334 00:18:21,840 --> 00:18:25,197 build on the rails of cloud and mobile 335 00:18:25,280 --> 00:18:27,037 diffuse faster 336 00:18:27,120 --> 00:18:30,557 >> and bend the productivity curve and 337 00:18:30,640 --> 00:18:33,437 bring local surplus and economic growth 338 00:18:33,520 --> 00:18:36,317 all around the world. Not just economic 339 00:18:36,400 --> 00:18:39,357 growth driven by capital expenses. 340 00:18:39,440 --> 00:18:44,557 uh right because that's it's a narrow point in time uh calculation. right now 341 00:18:44,640 --> 00:18:46,797 that's what we're seeing more >> that's what we're seeing you know you 342 00:18:46,880 --> 00:18:48,397 know in in the developed world in 343 00:18:48,480 --> 00:18:51,917 particular uh but remember my capital 344 00:18:52,000 --> 00:18:55,517 like that the one thing that you know is definitely we're spending a lot of it in 345 00:18:55,600 --> 00:18:58,797 the United States but 50% of it is also all over the world 346 00:18:58,880 --> 00:19:02,637 >> right >> um and so interesting enough it depends 347 00:19:02,720 --> 00:19:07,037 on uh demand all over the world and the demand all over the world will only be 348 00:19:07,120 --> 00:19:11,357 there if there is local surplus all over the world and so that's sort of the way 349 00:19:11,440 --> 00:19:15,357 I see the equation >> so let's drill down a little more. As AI 350 00:19:15,440 --> 00:19:19,437 diffuses, obviously organizations, companies, 351 00:19:19,520 --> 00:19:21,677 governments are going to have to evolve. 352 00:19:21,760 --> 00:19:24,717 I'm now getting to the demand side. So, 353 00:19:24,800 --> 00:19:26,477 how do you think the structure of 354 00:19:26,560 --> 00:19:30,477 organizations is changing in an AI world 355 00:19:30,560 --> 00:19:33,837 across roles, across teams, management? 356 00:19:33,920 --> 00:19:37,357 I'm I'm sure um Microsoft has evolved 357 00:19:37,440 --> 00:19:40,317 itself. So it probably be good to tell 358 00:19:40,400 --> 00:19:42,557 the audience how do you see this 359 00:19:42,640 --> 00:19:47,437 diffusion occur in the utilization at the corporate level or at a government 360 00:19:47,520 --> 00:19:51,437 level that which ultimately then creates that demand which eliminates any fears 361 00:19:51,520 --> 00:19:53,837 and bubbles. >> Yeah. No, I think it's probably one of 362 00:19:53,920 --> 00:19:56,237 the the the big challenges with all of 363 00:19:56,320 --> 00:20:00,717 these new technologies is when uh work 364 00:20:00,800 --> 00:20:04,717 work artifact and work flow changes 365 00:20:04,800 --> 00:20:08,637 uh the means we as firms have to change 366 00:20:08,720 --> 00:20:13,197 how we work. In [clears throat] fact, I remember meeting um uh the CEO of 367 00:20:13,280 --> 00:20:15,277 General Ali, you know, a few years back 368 00:20:15,360 --> 00:20:18,157 and he was describing he had joined um 369 00:20:18,240 --> 00:20:23,197 uh the firm, you know, prec [clears throat] era and uh uh and he was 370 00:20:23,280 --> 00:20:25,117 describing how for example they worked 371 00:20:25,200 --> 00:20:27,837 with their agents in the field uh with 372 00:20:27,920 --> 00:20:31,757 faxes inter office memos and um and and 373 00:20:31,840 --> 00:20:35,757 suddenly the PC showed up and people would then put a spreadsheet in an email 374 00:20:35,840 --> 00:20:39,597 and send it around and the entire workflow and the work process has 375 00:20:39,680 --> 00:20:42,397 changed right so similarly I think with 376 00:20:42,480 --> 00:20:46,237 AI uh you are going to start seeing uh 377 00:20:46,320 --> 00:20:50,397 actual change in how workflow happens right I mean even [clears throat] 378 00:20:50,480 --> 00:20:52,317 in fact for me coming to Davos you know 379 00:20:52,400 --> 00:20:55,117 whatever 50 bilateral meetings I have 380 00:20:55,200 --> 00:20:58,077 preparing for those uh had a particular 381 00:20:58,160 --> 00:21:02,957 workflow right which is uh there is to be my field team would prepare notes and 382 00:21:03,040 --> 00:21:07,037 that would come to my HQ and that would get further refined and nothing had 383 00:21:07,120 --> 00:21:09,597 really changed right since I joined in 384 00:21:09,680 --> 00:21:14,717 '92 to essentially even a few years back. Whereas now I just go to co-pilot 385 00:21:14,800 --> 00:21:18,957 and say hey I'm meeting Larry. Please give me a brief and it comes back and 386 00:21:19,040 --> 00:21:23,037 gives me by the way the one nice thing is it gives me a 360 right? It knows 387 00:21:23,120 --> 00:21:27,437 what we're doing with you as a client what we're doing as a client of yours 388 00:21:27,520 --> 00:21:30,157 and everything in [clears throat] between as an investment. It's so it 389 00:21:30,240 --> 00:21:32,957 captures even information unlike 390 00:21:33,040 --> 00:21:37,277 anything else. In fact, what I do is I take that and immediately share that 391 00:21:37,360 --> 00:21:40,957 back with all my colleagues across all the functions. Right? Think about it. 392 00:21:41,040 --> 00:21:43,757 It's a complete inversion of how 393 00:21:43,840 --> 00:21:48,077 information is flowing in the organization. It's not like this classic 394 00:21:48,160 --> 00:21:51,437 we have an organization, we have departments, we have these 395 00:21:51,520 --> 00:21:53,437 specializations and the information 396 00:21:53,520 --> 00:21:56,637 trickles up. No, no, no. It actually it 397 00:21:56,720 --> 00:22:00,317 flattens the entire information flow. So 398 00:22:00,400 --> 00:22:02,957 once you start having that you have to 399 00:22:03,040 --> 00:22:06,317 redesign structurally uh so the current 400 00:22:06,400 --> 00:22:10,317 structure may not make sense um because 401 00:22:10,400 --> 00:22:14,077 you want people to be able to work in a 402 00:22:14,160 --> 00:22:16,637 way that allows them to have this 403 00:22:16,720 --> 00:22:19,517 information flow freely. So what all 404 00:22:19,600 --> 00:22:23,677 this leads me to if I had to sort of say what's the formula the formula I think 405 00:22:23,760 --> 00:22:25,677 it starts with the mindset. So the 406 00:22:25,760 --> 00:22:29,917 mindset we as leaders should have is we 407 00:22:30,000 --> 00:22:32,957 need to think about changing the work 408 00:22:33,040 --> 00:22:36,157 the workflow with the technology then 409 00:22:36,240 --> 00:22:40,317 that needs skill set. So you can't sort of talk about this in the abstract. You 410 00:22:40,400 --> 00:22:43,837 got to use it like so if I'm not using the >> you have to trust it. 411 00:22:43,920 --> 00:22:47,997 >> You have to trust it. You have to use it. You have to learn even how to put 412 00:22:48,080 --> 00:22:51,997 the guard rails to trust it. Right? So you can't again you can't just be afraid 413 00:22:52,080 --> 00:22:54,877 of it. uh it's going to it's going to be 414 00:22:54,960 --> 00:22:57,677 diffused. So the question is as a firm 415 00:22:57,760 --> 00:23:00,877 you have to use it to learn how to even 416 00:23:00,960 --> 00:23:03,037 uh put the guardrails that allow you to 417 00:23:03,120 --> 00:23:06,317 be able to trust it. So skills uh so 418 00:23:06,400 --> 00:23:08,557 mindsets skills the other big 419 00:23:08,640 --> 00:23:12,077 consideration u really is how do you 420 00:23:12,160 --> 00:23:15,277 make sure you have the data set that 421 00:23:15,360 --> 00:23:18,397 you're feeding like context like it's 422 00:23:18,480 --> 00:23:20,877 kind of like you have a new intelligence 423 00:23:20,960 --> 00:23:23,437 layer but the intelligence layer is only 424 00:23:23,520 --> 00:23:27,197 as good as the context you give it. So 425 00:23:27,280 --> 00:23:31,917 people describe it even as context engineering but that is what firms do 426 00:23:32,000 --> 00:23:35,997 right if you think about what do firms do it's all about the tacit knowledge we 427 00:23:36,080 --> 00:23:41,437 have by working as people in various departments and moving paper and 428 00:23:41,520 --> 00:23:44,957 information so the question is how do you really have this AI also have that 429 00:23:45,040 --> 00:23:47,437 context so these are sort of some of the 430 00:23:47,520 --> 00:23:50,157 new things that have to percolate 431 00:23:50,240 --> 00:23:54,477 throughout an organization uh to take advantage in fact that's why I think you 432 00:23:54,560 --> 00:23:57,117 you're going to see that challenge of 433 00:23:57,200 --> 00:24:01,757 why am I not seeing immediate results in productivity? Because you have to do the 434 00:24:01,840 --> 00:24:04,877 hard work. In fact, that's why it's not going to be some, you know, it's going 435 00:24:04,960 --> 00:24:08,157 to there going to be firmwide differences. They're going to there 436 00:24:08,240 --> 00:24:12,557 could be sectorwide differences, but it's going to fundamentally be because 437 00:24:12,640 --> 00:24:15,997 of the leadership will in an organization. 438 00:24:16,080 --> 00:24:20,957 >> Do you see the applications being used across large companies and medium and 439 00:24:21,040 --> 00:24:24,477 small companies or is it still the domain of mostly the large companies at 440 00:24:24,560 --> 00:24:26,717 this moment? I I think that what you're 441 00:24:26,800 --> 00:24:30,157 seeing is it's easier the because if you 442 00:24:30,240 --> 00:24:34,237 have a green sort of um uh you know if 443 00:24:34,320 --> 00:24:37,037 you start fresh it's easier to adopt 444 00:24:37,120 --> 00:24:41,997 these tools and you construct your organization knowing that these tools 445 00:24:42,080 --> 00:24:43,917 exist. So >> is it a barbell then? 446 00:24:44,000 --> 00:24:48,237 >> It is a barbell. So small companies that are just starting use that platform 447 00:24:48,320 --> 00:24:51,377 >> 100% and I think in fact I would say even for large organization there's 448 00:24:51,460 --> 00:24:54,717 [clears throat] a fundamental challenge right because unless and until your rate 449 00:24:54,800 --> 00:24:57,677 of change keeps up with >> right 450 00:24:57,760 --> 00:25:00,397 >> uh with with what is possible uh you're 451 00:25:00,480 --> 00:25:03,197 going to get schooled by someone small 452 00:25:03,280 --> 00:25:08,397 being able to achieve scale because of these tools. So but I think scale large 453 00:25:08,480 --> 00:25:10,077 organizations have an inherent strength. 454 00:25:10,160 --> 00:25:11,757 You have the relationships, you have the 455 00:25:11,840 --> 00:25:15,517 data, you have um you have knowhow. But 456 00:25:15,600 --> 00:25:19,677 the bottom line is if you don't translate that with a new production 457 00:25:19,760 --> 00:25:24,157 function, uh then you really will be stuck. And so therefore the change 458 00:25:24,240 --> 00:25:28,397 management challenge for large organizations is going to be bigger. The 459 00:25:28,480 --> 00:25:31,837 structural challenge for small organizations of how to overcome scale 460 00:25:31,920 --> 00:25:35,037 issues is going to be harder. So it's sort of the two sides. in an interesting 461 00:25:35,120 --> 00:25:36,317 way. It's going to be a very 462 00:25:36,400 --> 00:25:39,997 competitively intense world uh where 463 00:25:40,080 --> 00:25:44,557 neither side like whether you're a new entrant or an incumbent can't take it as 464 00:25:44,640 --> 00:25:46,957 like I I can just coast. 465 00:25:47,040 --> 00:25:49,917 >> What about country to country? Are you 466 00:25:50,000 --> 00:25:54,477 seeing big differences in how the applications are being used? Is it is AI 467 00:25:54,560 --> 00:25:59,117 still the domain of developed countries or is it becoming rapidly a domain of 468 00:25:59,200 --> 00:26:02,237 all countries? I I I'm seeing there [clears throat] are two things I would 469 00:26:02,320 --> 00:26:04,797 say Larry as I travel around the world 470 00:26:04,880 --> 00:26:09,197 the quality of um whether it's the 471 00:26:09,280 --> 00:26:11,677 knowhow the software developers the 472 00:26:11,760 --> 00:26:13,277 startups 473 00:26:13,360 --> 00:26:17,037 um or even large or large organizations 474 00:26:17,120 --> 00:26:20,557 it's not that different it's fascinating you can show up in Jakarta you can show 475 00:26:20,640 --> 00:26:22,797 up in Istanbul you can show up in Mexico 476 00:26:22,880 --> 00:26:27,837 City it's not that different than showing up even in say Seattle or San 477 00:26:27,920 --> 00:26:29,517 Francisco right it's 478 00:26:29,600 --> 00:26:31,997 uh for the first time just because 479 00:26:32,080 --> 00:26:35,757 access to what's happening uh is there 480 00:26:35,840 --> 00:26:39,597 that said at [clears throat] scale the 481 00:26:39,680 --> 00:26:43,357 commitment to using this the risk 482 00:26:43,440 --> 00:26:45,997 capital being there the large companies 483 00:26:46,080 --> 00:26:49,037 pushing it hard I mean I you know again 484 00:26:49,120 --> 00:26:52,637 the US you know is in fact if I compare 485 00:26:52,720 --> 00:26:56,477 it uh take financial sector 486 00:26:56,560 --> 00:26:59,117 financial sector's adoption of the cloud 487 00:26:59,200 --> 00:27:02,077 versus AI night and day, right? Because 488 00:27:02,160 --> 00:27:05,437 in an interesting way, it's much faster 489 00:27:05,520 --> 00:27:08,077 uh when it comes to AI versus it was 490 00:27:08,160 --> 00:27:09,837 with the cloud and cloud because for a 491 00:27:09,920 --> 00:27:12,717 variety of reasons regulatory issues too 492 00:27:12,800 --> 00:27:18,477 until the regulators allowed banks to bring their data off out of campus that 493 00:27:18,560 --> 00:27:20,707 was a big issue. >> Yeah. So I would say I think 494 00:27:20,790 --> 00:27:22,877 [clears throat] wherever you know so in 495 00:27:22,960 --> 00:27:25,917 the west in particular in the US uh 496 00:27:26,000 --> 00:27:29,357 there is clearly a real I would say more 497 00:27:29,440 --> 00:27:31,997 of an energy around it in terms of going 498 00:27:32,080 --> 00:27:34,957 and using it uh but it's sort of a lot 499 00:27:35,040 --> 00:27:40,157 more uniformly spreading around the world than any technology at least I've 500 00:27:40,240 --> 00:27:45,037 seen but are you [clears throat] are you mentioned about the power the grid is 501 00:27:45,120 --> 00:27:46,637 that going to be one of the determinants 502 00:27:46,720 --> 00:27:49,997 of of the accessibility ility if you do 503 00:27:50,080 --> 00:27:53,517 not have cheap power it the demand is 504 00:27:53,600 --> 00:27:57,597 costly >> 100%. So if you sort of look at the 505 00:27:57,680 --> 00:28:00,077 tokens per dollar per watt right which I 506 00:28:00,160 --> 00:28:01,517 think in [clears throat] some sense I 507 00:28:01,600 --> 00:28:05,197 would claim that GDP growth in any place 508 00:28:05,280 --> 00:28:09,197 will be directly correlated like you if you sort of buy my entire argument that 509 00:28:09,280 --> 00:28:13,037 look we've got a new commodity its tokens >> right 510 00:28:13,120 --> 00:28:17,677 >> and the job of every economy uh and every firm in the economy is to 511 00:28:17,760 --> 00:28:19,757 translate these tokens into economic 512 00:28:19,840 --> 00:28:22,317 growth then if you have a cheaper 513 00:28:22,400 --> 00:28:27,597 commodity it's better uh And so that's sort of what why there's tokens per 514 00:28:27,680 --> 00:28:31,357 dollar per watt. And by the way, there are many many elements to this, right? 515 00:28:31,440 --> 00:28:35,517 Which is uh it's not just um the production side. That's why I think even 516 00:28:35,600 --> 00:28:38,317 having the grid is important. Um uh 517 00:28:38,400 --> 00:28:40,317 construction costs, right? So if you 518 00:28:40,400 --> 00:28:43,517 like if you think about the total TCO uh 519 00:28:43,600 --> 00:28:47,757 everything it's like the how are you a cheap producer of energy? Can you build 520 00:28:47,840 --> 00:28:50,317 the data centers? Uh then what's the 521 00:28:50,400 --> 00:28:53,037 cost curve uh of the silicon and the 522 00:28:53,120 --> 00:28:55,837 systems? uh the and by the way look at 523 00:28:55,920 --> 00:28:58,797 the token pricing right token pricing 524 00:28:58,880 --> 00:29:02,637 basically drops by you know a half uh 525 00:29:02,720 --> 00:29:05,757 every three months uh I mean this is a 526 00:29:05,840 --> 00:29:08,397 so how so that that's why I think you 527 00:29:08,480 --> 00:29:12,237 can sort of really plot how you use the 528 00:29:12,320 --> 00:29:16,717 tokens to create surplus knowing that you have a commodity that's whose prices 529 00:29:16,800 --> 00:29:19,917 are just going to monotonically come down in a pretty fast curve 530 00:29:20,000 --> 00:29:22,957 >> we're sitting in Europe uh and there is 531 00:29:23,040 --> 00:29:25,197 a real fear 532 00:29:25,280 --> 00:29:27,277 because of the co Europe does not have 533 00:29:27,360 --> 00:29:30,317 its own power and has to import mostly 534 00:29:30,400 --> 00:29:35,837 of its power. Um do you have any messages for Europe 535 00:29:35,920 --> 00:29:38,637 related to this? >> Yeah, I mean I think so there are two 536 00:29:38,720 --> 00:29:40,557 sets of things right. is you know here 537 00:29:40,640 --> 00:29:43,437 we are in in Switzerland and I look at 538 00:29:43,520 --> 00:29:45,917 uh the the pharma or the financial 539 00:29:46,000 --> 00:29:51,277 sector you know obviously Switz they're uh they they do do a big job in in in 540 00:29:51,360 --> 00:29:54,237 this country as in in Europe but they're also international brands and 541 00:29:54,320 --> 00:29:58,797 international operations. So one thing that uh whenever I think about Europe is 542 00:29:58,880 --> 00:30:02,957 the Europeans are producing products and services that actually are going 543 00:30:03,040 --> 00:30:05,437 everywhere in the world and so therefore 544 00:30:05,520 --> 00:30:08,077 uh European competitiveness is about the 545 00:30:08,160 --> 00:30:12,477 competitiveness of their output globally not just in Europe. I think sometimes 546 00:30:12,560 --> 00:30:14,077 when you come to Europe there's a lot of 547 00:30:14,160 --> 00:30:16,717 conversation about just Europe. Uh but 548 00:30:16,800 --> 00:30:20,637 European economy is thrives and has you 549 00:30:20,720 --> 00:30:24,317 know thrived in the last whatever 200 years 300 years. the miracle of the west 550 00:30:24,400 --> 00:30:28,157 is fundamentally because of what has happened in Europe uh is because they 551 00:30:28,240 --> 00:30:31,597 were able to produce things that the world needed and so I would say that's 552 00:30:31,680 --> 00:30:33,757 number one and in order to do that you 553 00:30:33,840 --> 00:30:35,997 again I go back to the human capital 554 00:30:36,080 --> 00:30:38,797 here is just fantastic and world class 555 00:30:38,880 --> 00:30:41,757 uh you have to absolutely invest in uh 556 00:30:41,840 --> 00:30:46,477 producing uh you know having the energy and the tokens here which again you're 557 00:30:46,560 --> 00:30:50,317 attracting like as I said we're investing and others are investing uh 558 00:30:50,400 --> 00:30:54,317 the the data centers here so the question is what's that next generation 559 00:30:54,400 --> 00:30:58,637 of output that comes from here. Right? I always think about the German middle. 560 00:30:58,720 --> 00:31:03,117 Whenever I go to a jeweler or a dentist in the United States, I'm surrounded by 561 00:31:03,200 --> 00:31:05,197 German middle. Right. Totally. 562 00:31:05,280 --> 00:31:07,757 >> Um it's just unbelievable engineering 563 00:31:07,840 --> 00:31:11,197 provice of that country. uh and now the 564 00:31:11,280 --> 00:31:12,957 question and by the way that's the point 565 00:31:13,040 --> 00:31:15,517 that they are producing industrial 566 00:31:15,600 --> 00:31:19,277 products which today are built in into 567 00:31:19,360 --> 00:31:22,157 it all the intelligence as well that 568 00:31:22,240 --> 00:31:24,557 data right so I know whenever we come to 569 00:31:24,640 --> 00:31:26,637 Europe everyone's like talking about 570 00:31:26,720 --> 00:31:29,277 sovereignty and data this data that 571 00:31:29,360 --> 00:31:32,157 guess what Europe actually should be 572 00:31:32,240 --> 00:31:36,077 much more concerned about access to 573 00:31:36,160 --> 00:31:38,397 their industrial companies their 574 00:31:38,480 --> 00:31:41,197 financial services companies of data 575 00:31:41,280 --> 00:31:44,077 from you, US and the rest of the world. 576 00:31:44,160 --> 00:31:47,757 Uh as opposed to just thinking that somehow by protecting Europe, you're 577 00:31:47,840 --> 00:31:51,597 going to be competitive. You are only going to be competitive if the products 578 00:31:51,680 --> 00:31:54,237 coming out of Europe are globally 579 00:31:54,320 --> 00:31:58,637 competitive, right? >> Um and so that's I think what needs to 580 00:31:58,720 --> 00:32:00,717 change. Uh you know, Europe has led in 581 00:32:00,800 --> 00:32:03,357 privacy, that's fantastic, has led in 582 00:32:03,440 --> 00:32:08,157 many aspects of even safety around AI and what have you. And that's a feature 583 00:32:08,240 --> 00:32:12,477 uh that's great but you also have to complement it with by building locally 584 00:32:12,560 --> 00:32:14,877 and then also thinking globally what's 585 00:32:14,960 --> 00:32:17,437 the contribution this continent will 586 00:32:17,520 --> 00:32:19,757 make uh to the rest of the world which 587 00:32:19,840 --> 00:32:22,077 it has historically been a leader 588 00:32:22,160 --> 00:32:27,277 >> a leader so do you think the whole idea around sovereignty of data is that being 589 00:32:27,360 --> 00:32:31,197 misunderstood >> I I think that that when people talk 590 00:32:31,280 --> 00:32:33,437 about sovereignty 591 00:32:33,520 --> 00:32:38,557 first of all it's very important clearly And who owns and in a week like this 592 00:32:38,640 --> 00:32:40,157 it's more important. 593 00:32:40,240 --> 00:32:45,837 >> Uh but that said uh it is 594 00:32:45,920 --> 00:32:48,237 you have to kind of think about how what 595 00:32:48,320 --> 00:32:52,877 is sovereignty mean like for example in the AI era the topic that's least talked 596 00:32:52,960 --> 00:32:55,197 about but I feel will be most talked 597 00:32:55,280 --> 00:32:58,477 about in in this uh this calendar year 598 00:32:58,560 --> 00:33:01,117 will be the sovereignty of a firm. Just 599 00:33:01,200 --> 00:33:04,717 imagine if your firm you're not able to 600 00:33:04,800 --> 00:33:08,637 embed the tacit knowledge of the firm in 601 00:33:08,720 --> 00:33:12,317 a set of weights in a model that you 602 00:33:12,400 --> 00:33:15,037 control. By definition, you have no 603 00:33:15,120 --> 00:33:17,357 sovereignty. That means you're leaking 604 00:33:17,440 --> 00:33:22,237 enterprise value through some model company somewhere. In fact, that it's 605 00:33:22,320 --> 00:33:25,197 sort of fascinating that nobody's talking about that, right? It's like 606 00:33:25,280 --> 00:33:26,637 everybody's talking about everything 607 00:33:26,720 --> 00:33:29,757 else that is sort of you know outside of 608 00:33:29,840 --> 00:33:32,637 that whereas the most important thing is 609 00:33:32,720 --> 00:33:36,957 it really doesn't matter if you in fact the data center where it runs is the 610 00:33:37,040 --> 00:33:40,557 least important thing quite honestly but like even there first of all the data 611 00:33:40,640 --> 00:33:44,957 centers all are all over just because speed of light is a real constraint uh 612 00:33:45,040 --> 00:33:49,197 and so therefore the data centers will be spread yes uh you will have digital 613 00:33:49,280 --> 00:33:52,797 you'll be able to encrypt everything you'll be able to have the keys with you 614 00:33:52,880 --> 00:33:54,397 all of these are much more techn 615 00:33:54,480 --> 00:33:56,957 technically solve problems but the one 616 00:33:57,040 --> 00:34:00,237 problem that will only be solved is by 617 00:34:00,320 --> 00:34:05,357 you having much more sovereignty over the you know tacid knowledge uh and 618 00:34:05,440 --> 00:34:07,197 control over the models and it's not a 619 00:34:07,280 --> 00:34:10,237 one-way enterprise value transfer um and 620 00:34:10,320 --> 00:34:12,717 so to me I think sovereignty requires 621 00:34:12,800 --> 00:34:16,077 real thought on what is it um you know 622 00:34:16,160 --> 00:34:19,277 control of destiny means that your your 623 00:34:19,360 --> 00:34:21,517 ability to produce something that is 624 00:34:21,600 --> 00:34:24,637 unique is preserved uh David Ricardo was 625 00:34:24,720 --> 00:34:26,797 not wrong. There's comparative advantage 626 00:34:26,880 --> 00:34:29,277 in countries. Uh there is comparative 627 00:34:29,360 --> 00:34:32,397 advantage in firms that needs to be 628 00:34:32,480 --> 00:34:36,637 preserved even in the AI era. That's what'll give you real sovereignty. 629 00:34:36,720 --> 00:34:40,077 >> One last question. I know we're running out of time. 630 00:34:40,160 --> 00:34:44,077 Um in 5 years or 10 years, is there going to be one dominant model that 631 00:34:44,160 --> 00:34:48,637 we're all going to be using or are and how is Microsoft preparing for this? Are 632 00:34:48,720 --> 00:34:50,397 you going to be are we going to be using 633 00:34:50,480 --> 00:34:53,357 one model for for enterprise, one model 634 00:34:53,440 --> 00:34:55,997 for other other traits? 635 00:34:56,080 --> 00:34:58,237 >> You know, even in the last whatever 3 636 00:34:58,320 --> 00:35:00,397 years, four years that we've been at it, 637 00:35:00,480 --> 00:35:04,557 um the the reality at this point is it's 638 00:35:04,640 --> 00:35:08,157 a multimodel world, right? I mean the in 639 00:35:08,240 --> 00:35:12,477 fact if you think about it the in both 640 00:35:12,560 --> 00:35:16,397 there are going to be multiple models 641 00:35:16,480 --> 00:35:19,597 and the trick is really how do you take 642 00:35:19,680 --> 00:35:22,557 advantage of these multiple models and 643 00:35:22,640 --> 00:35:25,037 in fact build your own model by 644 00:35:25,120 --> 00:35:27,837 distilling these right uh so think of 645 00:35:27,920 --> 00:35:31,117 these models uh that you orchestrate to 646 00:35:31,200 --> 00:35:35,997 build your own model and more importantly you do what is described as 647 00:35:36,080 --> 00:35:37,517 orch orchestration or harness 648 00:35:37,600 --> 00:35:40,317 engineering. So the IP of any 649 00:35:40,400 --> 00:35:43,517 application or any firm is how do you 650 00:35:43,600 --> 00:35:48,317 use all these models with context engineering or your data? Yes. 651 00:35:48,400 --> 00:35:52,477 >> Right. So it's that three part. So can I bring in all the models by the way uh 652 00:35:52,560 --> 00:35:55,117 which is close source, open source, 653 00:35:55,200 --> 00:35:58,877 build my own model, orchestrate them and 654 00:35:58,960 --> 00:36:02,477 feed it my data to change the trajectory 655 00:36:02,560 --> 00:36:05,117 of some outcome that I care about. 656 00:36:05,200 --> 00:36:07,677 That's it. That's the entire picture. So 657 00:36:07,760 --> 00:36:09,837 you can do it in like oh I produce a 658 00:36:09,920 --> 00:36:12,237 particular product or service. Uh first 659 00:36:12,320 --> 00:36:17,037 I got to do better better job in sales or better job in R&D or better job in 660 00:36:17,120 --> 00:36:20,957 finance or what have you. And you take that outcome and then you say can I use 661 00:36:21,040 --> 00:36:23,517 all the models orchestrate them and feed 662 00:36:23,600 --> 00:36:26,077 it my context and then in as a result of 663 00:36:26,160 --> 00:36:29,037 it the reasoning traces are really 664 00:36:29,120 --> 00:36:31,917 leading to some capability and models 665 00:36:32,000 --> 00:36:34,877 that I control as my IP. As long as 666 00:36:34,960 --> 00:36:39,197 firms can answer that question, they're going to be getting ahead. 667 00:36:39,280 --> 00:36:41,677 >> Ladies and gentlemen, let's uh thank 668 00:36:41,760 --> 00:36:48,440 Satia, my friend. Thank you [applause] for 669 00:36:49,200 --> 00:36:53,037 and hopefully this is the beginning of many great dialogue and conversations 670 00:36:53,120 --> 00:36:56,157 here at the World Academic Forum. Thank you everyone. 671 00:36:56,240 --> 00:36:59,400 >> Thank you.