WEBVTT 1 00:00:01.199 --> 00:00:06.200 Welcome to the Sentient Code, where intelligence is engineered, autonomy 2 00:00:06.280 --> 00:00:10.439 is emerging, and a line between human and machine grows thinner. 3 00:00:10.800 --> 00:00:15.359 Each episode, we decode the algorithms, explore the robotics, and 4 00:00:15.439 --> 00:00:21.920 examine the ideas shaping the future of artificial minds. 5 00:00:24.000 --> 00:00:27.640 Usually, when we think about the future, there is this 6 00:00:28.359 --> 00:00:30.480 comforting illusion of distance. 7 00:00:30.800 --> 00:00:33.520 Right, Like it's always just over the horizon. 8 00:00:33.119 --> 00:00:38.799 Exactly, flying cars, artificial general intelligence, radically restructured economies, those 9 00:00:39.000 --> 00:00:43.240 milestones always seem to be perpetually, you know, twenty years away. 10 00:00:43.399 --> 00:00:45.880 Yeah, it's a horizon that moves backward as we. 11 00:00:45.799 --> 00:00:50.240 Move forward, right, And that perceived distance is well, it's essential, 12 00:00:50.280 --> 00:00:50.759 isn't it. 13 00:00:50.759 --> 00:00:53.920 It absolutely is. It's what gives societies time to adjust. 14 00:00:54.000 --> 00:00:57.000 I mean, it creates this generational buffer where we can 15 00:00:57.039 --> 00:01:00.679 slowly adapt to new realities, plan out thirty year careers, 16 00:01:01.280 --> 00:01:05.120 and build traditional business models without the ground completely falling 17 00:01:05.120 --> 00:01:06.120 out from under us. 18 00:01:06.640 --> 00:01:09.359 But you are tuning in today because you probably suspect 19 00:01:09.359 --> 00:01:10.400 that bedrock is shifting. 20 00:01:10.519 --> 00:01:11.799 Oh, it's definitely shifting. 21 00:01:11.959 --> 00:01:13.719 Yeah, And you want to know what happens when someone 22 00:01:13.760 --> 00:01:17.799 suddenly hits fast forward, collapsing that twenty year horizon down 23 00:01:17.840 --> 00:01:19.000 to just what twenty. 24 00:01:18.719 --> 00:01:20.680 Months basically, Yeah. 25 00:01:20.359 --> 00:01:23.120 Because the reality is the timeline for the future just 26 00:01:23.239 --> 00:01:28.760 radically accelerated. We are barreling toward a transformative, nonlinear explosion 27 00:01:28.760 --> 00:01:33.159 in artificial intelligence, and the target date for impact is 28 00:01:33.200 --> 00:01:34.760 the first half of twenty twenty six. 29 00:01:35.000 --> 00:01:37.319 And if we connect this to the bigger picture, we 30 00:01:37.359 --> 00:01:40.280 aren't talking about incremental software updates anymore. 31 00:01:40.439 --> 00:01:41.159 No, not real. 32 00:01:41.359 --> 00:01:43.519 We are not looking at a new app that writes 33 00:01:43.599 --> 00:01:47.239 like slightly better emails or generates amusing images. This is 34 00:01:47.280 --> 00:01:50.719 an event that redefines how societies and economies function at 35 00:01:50.719 --> 00:01:52.400 a fundamental level completely. 36 00:01:52.799 --> 00:01:56.879 And the anchor for this timeline is this sweeping new 37 00:01:56.959 --> 00:02:01.319 forecast from Morgan Stanley that currently has walls in Silicon Valley. 38 00:02:01.280 --> 00:02:02.920 On edge, totally on edge. 39 00:02:03.079 --> 00:02:05.760 So our mission today is to break down exactly what 40 00:02:05.879 --> 00:02:08.199 is driving this leap. We're going to explore the raw 41 00:02:08.240 --> 00:02:12.080 fuel being poured into the system, the staggering new benchmarks 42 00:02:12.080 --> 00:02:15.199 that have already been shattered, the dual edged economic shockwave 43 00:02:15.280 --> 00:02:17.879 coming for our jobs. And this is the crazy part, 44 00:02:18.080 --> 00:02:21.360 the physical real world bottleneck that could crash the entire system. 45 00:02:21.400 --> 00:02:24.080 Yeah, the physical limits are the wild card here exactly. 46 00:02:24.199 --> 00:02:27.879 So whether you are planning your career path, managing a business, 47 00:02:28.000 --> 00:02:30.120 or just trying to figure out where the world is heading. 48 00:02:30.599 --> 00:02:33.280 You need to internalize this. The rules of the economy 49 00:02:33.319 --> 00:02:35.759 are being rewritten right now, and I mean most of 50 00:02:35.759 --> 00:02:37.680 the world is completely unprepared. 51 00:02:37.319 --> 00:02:40.439 Completely And to understand why twenty twenty six is the 52 00:02:40.439 --> 00:02:43.560 inflection point, we have to look at the engine of progress, 53 00:02:43.719 --> 00:02:45.639 the raw fuel, right we have to look at the 54 00:02:45.759 --> 00:02:49.360 raw fuel of modern AI, which is accumulating at a 55 00:02:49.439 --> 00:02:52.960 rate that just it defies historical comparisons. 56 00:02:52.960 --> 00:02:56.639 Okay, let's unpack this because for years, the biggest constraint 57 00:02:56.639 --> 00:03:01.520 on artificial intelligence hasn't actually been the software or the algorithms. 58 00:03:01.639 --> 00:03:04.280 No, it hasn't. It's been the physical hardware, the sheer. 59 00:03:04.199 --> 00:03:07.719 Volume of processing power available to train and run these 60 00:03:07.800 --> 00:03:12.479 massive neural networks. But that constraint is rapidly evaporated, evaporating yet. 61 00:03:12.560 --> 00:03:17.039 I mean top us labs Open Ai, Anthropic, Google, DeepMind, 62 00:03:17.280 --> 00:03:22.560 XAI Meta, they are aggressively stockpiling graphics processing units. They're 63 00:03:22.599 --> 00:03:26.240 hoarding custom silicon and building data center clusters at a 64 00:03:26.280 --> 00:03:27.479 scale we've just never seen. 65 00:03:27.759 --> 00:03:31.759 They are essentially hoarding compute. And you know, when we 66 00:03:31.879 --> 00:03:34.599 use the term compute as a noun. Here we are 67 00:03:34.639 --> 00:03:39.039 talking about the physical processing capacity required to perform trillions 68 00:03:39.080 --> 00:03:42.680 of mathematical operations per second, And the reason these companies 69 00:03:42.680 --> 00:03:45.599 are pouring hundreds of billions of dollars into physical hardware 70 00:03:45.800 --> 00:03:48.719 is due to what the industry calls scaling laws. 71 00:03:48.960 --> 00:03:50.800 Okay, but I want to push back on that concept 72 00:03:50.800 --> 00:03:54.560 for a second. Sure is intelligence really just a math equation? 73 00:03:55.199 --> 00:03:55.280 Like? 74 00:03:55.439 --> 00:03:59.240 Can you genuinely just stack thousands of computers together in 75 00:03:59.280 --> 00:04:03.080 a giant ware house, feed the more data, and automatically 76 00:04:03.080 --> 00:04:04.000 get a smarter machine. 77 00:04:04.000 --> 00:04:05.360 It sounds too simple, doesn't it. 78 00:04:05.520 --> 00:04:08.520 Yeah? Doesn't that hit a ceiling eventually, like a point 79 00:04:08.560 --> 00:04:12.680 where throwing more processors at a problem yields diminishing returns. 80 00:04:12.719 --> 00:04:15.680 Well, what's fascinating here is that, empirically the ceiling simply 81 00:04:15.719 --> 00:04:16.519 hasn't been found. 82 00:04:16.639 --> 00:04:16.920 Wow. 83 00:04:17.120 --> 00:04:20.920 Yeah, The Morgan Stanley forecast confirms that these scaling laws 84 00:04:20.920 --> 00:04:24.839 remain perfectly intact even as these models push into entirely 85 00:04:24.920 --> 00:04:28.920 uncharted territory. When we talk about scaling, we're really looking 86 00:04:28.920 --> 00:04:32.600 at three pillars, right, more data, more compute, and more parameters. 87 00:04:32.920 --> 00:04:35.399 Let's define parameters for a moment, because that term gets 88 00:04:35.399 --> 00:04:37.759 thrown around a lot. Please do think of parameters like 89 00:04:37.759 --> 00:04:41.319 the artificial synapses in the model's digital brain. They are 90 00:04:41.360 --> 00:04:44.879 the numerical values that the network adjusts as it learns. 91 00:04:45.439 --> 00:04:48.560 So the more parameters of model has, the more nuanced 92 00:04:48.639 --> 00:04:51.120 multidimensional connections it can make between concepts. 93 00:04:51.720 --> 00:04:54.879 Exactly. And when you scale up those parameters and feed 94 00:04:54.879 --> 00:04:58.240 the system exponentially more compute, you want just making a 95 00:04:58.319 --> 00:05:03.120 larger database, right, The model literally gains a higher resolution 96 00:05:03.279 --> 00:05:07.240 understanding of reality. Think of early AI like a city 97 00:05:07.279 --> 00:05:10.639 map with only the major interstate highways drawn in. You know, 98 00:05:11.480 --> 00:05:14.759 scaling up the compute doesn't just make the map physically bigger. 99 00:05:15.120 --> 00:05:18.519 It adds the side streets, the alleys, the traffic patterns, 100 00:05:18.639 --> 00:05:22.720 the topography. It develops a deeper, more complex representation of 101 00:05:22.759 --> 00:05:24.160 how the world actually works. 102 00:05:24.399 --> 00:05:27.600 That makes total sense. And Elon Musk recently broke down 103 00:05:27.639 --> 00:05:30.279 the math on this, and it's staggering. He pointed out 104 00:05:30.279 --> 00:05:32.959 that applying ten times more compute to a model's training 105 00:05:33.000 --> 00:05:36.879 phase effectively doubles its overall intelligence. Doubles it. 106 00:05:37.040 --> 00:05:40.439 Yeah, and the crittle detail is that these companies are 107 00:05:40.480 --> 00:05:42.759 not increasing their compute by a factor of ten. 108 00:05:43.000 --> 00:05:43.759 Wait they aren't. 109 00:05:43.879 --> 00:05:47.560 No, they are scaling by factors of hundreds and soon thousands. 110 00:05:47.759 --> 00:05:51.439 Oh wow, the curve is purely exponential. I mean, insiders 111 00:05:51.439 --> 00:05:55.720 at Morgan Stanley's recent Technology, Media and Telecom conference were 112 00:05:55.879 --> 00:06:00.399 privately warning attendees to physically brace themselves for the coming months. 113 00:06:00.480 --> 00:06:04.199 Brace themselves. That's intense language for a finance conference. 114 00:06:04.759 --> 00:06:07.160 Very because the progress is going to shock even the 115 00:06:07.160 --> 00:06:09.000 most bullish observers. 116 00:06:08.439 --> 00:06:10.639 Which means we don't actually have to wait until twenty 117 00:06:10.680 --> 00:06:12.879 twenty six to see if these scaling laws hold up. 118 00:06:13.560 --> 00:06:16.399 Concrete evidence of this acceleration just quietly dropped a few 119 00:06:16.399 --> 00:06:16.759 weeks ago. 120 00:06:17.000 --> 00:06:17.560 Yes it did. 121 00:06:18.079 --> 00:06:21.519 Open Ai released its flagship GFT five point four model, 122 00:06:21.759 --> 00:06:25.639 specifically a thinking variant optimized for advanced reasoning and multi 123 00:06:25.720 --> 00:06:26.759 step problem solving. 124 00:06:27.000 --> 00:06:30.519 And the benchmark this model hit is what completely changes 125 00:06:30.560 --> 00:06:31.199 the conversation. 126 00:06:31.360 --> 00:06:31.600 Right. 127 00:06:31.639 --> 00:06:34.480 It achieved an eighty three point zero percent score on 128 00:06:34.519 --> 00:06:36.040 the GDPVOW benchmark. 129 00:06:36.319 --> 00:06:38.879 And we really need to pause and unpack the weight 130 00:06:38.920 --> 00:06:42.000 of GDP VOW because this isn't a standard academic test, No, 131 00:06:42.040 --> 00:06:44.319 not at all. We aren't talking about the SATs or 132 00:06:44.360 --> 00:06:48.920 a bar exam where a model just regurgitates memorized. 133 00:06:48.360 --> 00:06:53.040 Facts, right, GDP VOW evaluates performance on real world economically 134 00:06:53.120 --> 00:06:56.680 valuable tasks. It spans dozens of professions in the specific 135 00:06:56.720 --> 00:07:00.639 industries that drive the majority of US GDP heavy hitters. Exactly. 136 00:07:01.040 --> 00:07:05.319 It tests the ability to synthesize information, makes strategic decisions, 137 00:07:05.360 --> 00:07:10.120 and produce usable output and feels like law, finance, consulting, engineering, 138 00:07:10.120 --> 00:07:11.319 and medicine. 139 00:07:10.839 --> 00:07:12.839 So practical application, not just theory. 140 00:07:13.000 --> 00:07:16.680 Yes, imagine a prompt that requires the AI to review 141 00:07:16.720 --> 00:07:20.000 a five hundred page medical history, cross reference it with 142 00:07:20.079 --> 00:07:24.399 the latest pharmacological trials, and recommend a personalized multi phase 143 00:07:24.439 --> 00:07:25.079 treatment plan. 144 00:07:25.439 --> 00:07:27.519 That is I mean, that's incredibly complex. 145 00:07:27.639 --> 00:07:30.439 It is. And a score of eighty three percent places 146 00:07:30.480 --> 00:07:33.399 this digital model at or above the level of highly 147 00:07:33.439 --> 00:07:37.360 trained human experts who have spent years, sometimes decades, mastering 148 00:07:37.360 --> 00:07:38.519 these specific domains. 149 00:07:38.759 --> 00:07:41.519 We are looking at a single piece of software replicating 150 00:07:41.560 --> 00:07:45.160 the cognitive output of professions that generate trillions of dollars 151 00:07:45.199 --> 00:07:46.000 in economic value. 152 00:07:46.040 --> 00:07:46.480 Trillions. 153 00:07:46.600 --> 00:07:49.879 Yes, think about what that actually means for your daily operations. 154 00:07:50.560 --> 00:07:53.839 It means having a senior law partner, a chief structural engineer, 155 00:07:54.079 --> 00:07:56.680 and a top tier financial consultant sitting in a little 156 00:07:56.680 --> 00:07:57.800 digital box on your. 157 00:07:57.720 --> 00:07:59.680 Desk and they don't sleep right. 158 00:08:00.279 --> 00:08:02.639 For pennies on the dollar, you can ask the system 159 00:08:02.680 --> 00:08:06.480 to structure a complex corporate merger, write the underlying code 160 00:08:06.480 --> 00:08:10.240 for the software integration, and draft the airtight legal contracts, 161 00:08:10.759 --> 00:08:13.639 and it performs at an eighty three percent expert level 162 00:08:13.800 --> 00:08:17.360 across all of those distinct disciplines simultaneously, simultaneously. 163 00:08:17.600 --> 00:08:20.720 And Morgan Stanley stresses that GPT five point four is 164 00:08:20.839 --> 00:08:24.680 merely the new baseline. Wait, the baseline, Yes, it is 165 00:08:24.720 --> 00:08:28.519 the starting line. If you apply the unbroken scaling laws 166 00:08:28.560 --> 00:08:32.679 to this new foundation, the capability curve gets drastically steeper 167 00:08:32.679 --> 00:08:33.720 over the next twenty. 168 00:08:33.480 --> 00:08:34.879 Months, which is terrifying. 169 00:08:34.879 --> 00:08:37.039 Honestly, when an AI can perform the work of a 170 00:08:37.080 --> 00:08:40.799 senior professional, you aren't just rolling out a better software tool. 171 00:08:40.960 --> 00:08:44.399 You are fundamentally dismantling and altering the global labor market. 172 00:08:44.559 --> 00:08:46.960 So what does this all mean for the economy. If 173 00:08:47.000 --> 00:08:49.600 we now have an expert in a box and that 174 00:08:49.679 --> 00:08:53.679 capability floods the global market, it triggers a profound double 175 00:08:53.799 --> 00:08:55.240 edged economic shock wave. 176 00:08:55.519 --> 00:08:56.120 It really does. 177 00:08:56.279 --> 00:08:59.960 Let's start with the upside, which is massive historic deflake. 178 00:09:00.480 --> 00:09:02.919 Yeah, transformative AI will act as one of the most 179 00:09:02.960 --> 00:09:07.279 powerful deflationary forces in modern history because it's so cheap. Exactly, 180 00:09:07.919 --> 00:09:11.480 by replicating human cognitive work at a fraction of the 181 00:09:11.519 --> 00:09:16.480 cost and executing it at superhuman speed, the price of 182 00:09:16.559 --> 00:09:21.000 knowledge work collapses. The marginal cost of intelligence plummets. 183 00:09:21.279 --> 00:09:23.840 Put that in perspective. If you are starting a business 184 00:09:23.879 --> 00:09:27.000 today and you need a marketing department, a legal team, 185 00:09:27.080 --> 00:09:30.240 and a data analysis division. Historically you had to hire 186 00:09:30.279 --> 00:09:32.399 twenty highly paid individuals. 187 00:09:31.919 --> 00:09:32.840 Right with all the overhead. 188 00:09:32.919 --> 00:09:37.039 Yeah, you had salaries, healthcare, office space. Now the marginal 189 00:09:37.120 --> 00:09:40.000 cost of adding that intelligence to your company is just 190 00:09:40.080 --> 00:09:42.600 the cost of electricity and an API call, which is 191 00:09:42.720 --> 00:09:45.799 practically nothing right, And an API call is simply the 192 00:09:45.879 --> 00:09:49.000 digital request your computer sends to the AI servers to 193 00:09:49.120 --> 00:09:52.559 process a task and return the result. It costs fractions 194 00:09:52.559 --> 00:09:53.000 of a cent. 195 00:09:53.240 --> 00:09:56.600 And Sam Altman, the CEO of OpenAI, preticted the logical 196 00:09:56.679 --> 00:09:59.639 conclusion of this dynamic. We are going to see new 197 00:10:00.279 --> 00:10:04.000 billion dollar businesses built and run by teams of just 198 00:10:04.159 --> 00:10:05.399 one to five people. 199 00:10:05.679 --> 00:10:08.480 One to five people building a billion dollar company. 200 00:10:08.600 --> 00:10:14.360 Yes, these ultra LAN teams, armed with expert level AI 201 00:10:14.480 --> 00:10:19.960 agents will be able to completely outcompete massive lumbering corporate 202 00:10:19.960 --> 00:10:23.759 incumbents that are bogged down by huge, expensive workforces. 203 00:10:24.320 --> 00:10:27.960 But I have to challenge this narrative for a second before, 204 00:10:28.159 --> 00:10:31.320 because every time we face a technological revolution, we hear 205 00:10:31.399 --> 00:10:34.240 that the machines are coming for our jobs, and every 206 00:10:34.320 --> 00:10:36.720 time technology ends up creating new jobs. 207 00:10:37.080 --> 00:10:39.080 True the classic ledite fallacy. 208 00:10:39.240 --> 00:10:42.519 Exactly when the tractor was invented, farmers became mechanics. When 209 00:10:42.639 --> 00:10:45.840 computers came along, type it's became software engineers. Are we 210 00:10:45.919 --> 00:10:49.039 just talking about entry level spreadsheet pushing being automated here, 211 00:10:49.519 --> 00:10:52.399 or are the highly paid experts genuinely in the crosshairs 212 00:10:52.399 --> 00:10:52.840 this time? 213 00:10:53.360 --> 00:10:57.200 This raises an important question about the limits of historical precedent. Okay, 214 00:10:57.279 --> 00:10:59.919 the comparisons don't perfectly map onto what we are facing 215 00:11:00.080 --> 00:11:03.159 now because the nature of the automation is fundamentally different. 216 00:11:03.360 --> 00:11:07.639 How SOO, well, a tractor automated physical labor, a spreadsheet 217 00:11:07.639 --> 00:11:12.360 automated mathematical sorting GPT five point four, and the models 218 00:11:12.440 --> 00:11:17.360 arriving in twenty twenty six are automating cognitive synthesis, complex reasoning, 219 00:11:17.840 --> 00:11:19.440 and strategic decision making. 220 00:11:19.519 --> 00:11:21.440 Ah so it's the thinking itself. 221 00:11:21.559 --> 00:11:24.879 Exactly when a digital model can perform at an expert 222 00:11:24.960 --> 00:11:29.720 level on nuanced legal drafting or complex medical diagnostics. Entire 223 00:11:29.799 --> 00:11:33.559 categories of high level knowledge work become candidates for immediate automation. 224 00:11:34.000 --> 00:11:37.679 Wow, and we're already seeing a corporate decoupling taking place 225 00:11:37.759 --> 00:11:38.480 right now, aren't we? 226 00:11:38.639 --> 00:11:39.080 Yes, we are. 227 00:11:39.440 --> 00:11:42.960 Revenue growth for major companies is rising, but employment growth 228 00:11:43.000 --> 00:11:44.679 is separating from it, installing out. 229 00:11:44.840 --> 00:11:48.240 Companies are discovering they can deliver outsized, record breaking results 230 00:11:48.240 --> 00:11:50.399 with significantly smaller human teams. 231 00:11:50.600 --> 00:11:52.559 So the jobs aren't growing with the profits. 232 00:11:52.639 --> 00:11:55.240 No, they aren't. While new roles certainly will emerge, I 233 00:11:55.240 --> 00:11:58.840 mean we will need AI infrastructure managers, oversight boards, and 234 00:11:58.919 --> 00:12:03.039 specialists in novel applications of the technology. The transition period 235 00:12:03.080 --> 00:12:04.440 is going to be highly disruptive. 236 00:12:04.600 --> 00:12:07.960 We're looking at a scenario where overall productivity booms, but 237 00:12:08.120 --> 00:12:10.799 human labor is decoupled from revenue generation. 238 00:12:10.759 --> 00:12:12.720 Which is a paradigm shift completely. 239 00:12:12.919 --> 00:12:14.679 Societies are going to have to figure out how to 240 00:12:14.720 --> 00:12:18.600 manage an economy where the vast wealth being generated is 241 00:12:18.679 --> 00:12:23.440 no longer intrinsically tied to mass human employment. It chased 242 00:12:23.480 --> 00:12:26.799 the foundation of how we value work and distribute resources. 243 00:12:27.200 --> 00:12:31.200 And you know that deflationary software miracle is currently colliding 244 00:12:31.279 --> 00:12:33.919 with a hard physical reality. 245 00:12:33.679 --> 00:12:36.519 Right, because having a senior partner in a digital box 246 00:12:36.679 --> 00:12:38.960 requires massive physical infrastructure. 247 00:12:39.000 --> 00:12:40.240 It does, it really does. 248 00:12:40.320 --> 00:12:44.399 We have the software, the unbroken scaling laws, and businesses 249 00:12:44.480 --> 00:12:47.799 clearly have the financial incentive to automate. So what is 250 00:12:47.840 --> 00:12:50.039 the bottleneck we're running out of? Electricity? 251 00:12:50.120 --> 00:12:54.360 The power grid? Yeah, Morgan Stanley's proprietary Intelligence Factory model 252 00:12:54.440 --> 00:12:58.399 outlines a massive structural limitation. They are projecting a net 253 00:12:58.519 --> 00:13:01.759 US power shortfall of nine to eighteen digglewads through. 254 00:13:01.639 --> 00:13:03.600 Twenty twenty eight nine to eighteen gears. 255 00:13:03.600 --> 00:13:06.000 I want to put that abstract number into perspective. That 256 00:13:06.080 --> 00:13:08.320 is a twelve to twenty five percent deficit in the 257 00:13:08.360 --> 00:13:10.960 electricity required just to power the next immediate wave of 258 00:13:11.039 --> 00:13:11.679 data centers. 259 00:13:11.919 --> 00:13:14.679 That is staggering. We aren't talking about a few blown 260 00:13:14.759 --> 00:13:18.320 fuses in a neighborhood. We're talking about a bottleneck on 261 00:13:18.360 --> 00:13:22.080 the scale of entire regional grids failing to meet. 262 00:13:21.919 --> 00:13:23.480 Demand, failing completely. 263 00:13:23.600 --> 00:13:26.799 Yeah, but wait, why does AI need so much more 264 00:13:26.840 --> 00:13:30.960 power than regular computing? For decades we've been running giant 265 00:13:31.039 --> 00:13:34.679 data centers for cloud storage and web hosting without crashing 266 00:13:34.759 --> 00:13:35.200 the grid. 267 00:13:35.480 --> 00:13:38.919 The difference lies in the mechanism of the computation. Traditional 268 00:13:38.919 --> 00:13:41.799 computing is mostly about retrieving data. When you load a 269 00:13:41.799 --> 00:13:45.480 website or stream of video, servers are essentially fetching existing 270 00:13:45.600 --> 00:13:47.559 files and sending them to your device. 271 00:13:47.679 --> 00:13:49.519 Right, it's just finding it and sending. 272 00:13:49.240 --> 00:13:53.279 It, exactly. But AI computing is generative. Every single time 273 00:13:53.320 --> 00:13:56.080 you prompt an advanced model, it is generating net new 274 00:13:56.240 --> 00:14:00.799 text images or probabilistic math from scratch. It is actively 275 00:14:00.840 --> 00:14:04.960 calculating billions of parameters in real time. That requires the 276 00:14:05.000 --> 00:14:09.120 hardware to run constantly at maximum thermal limits, drawing exponentially 277 00:14:09.159 --> 00:14:11.320 more wattage than traditional data retrieval. 278 00:14:11.639 --> 00:14:16.559 It is the ultimate irony. The seemingly limitless ethereal potential 279 00:14:16.600 --> 00:14:21.200 of artificial intelligence is hitting a hard physical wall made 280 00:14:21.240 --> 00:14:24.720 of copper wire, transformers and concrete power. 281 00:14:24.480 --> 00:14:27.000 Plants, literally a physical wall. 282 00:14:27.120 --> 00:14:30.240 Yeah, data center developers are getting so desperate for energy 283 00:14:30.279 --> 00:14:33.919 that they are literally scrambling to convert former bitcoin mining 284 00:14:33.960 --> 00:14:36.360 operations into high performance AI clusters. 285 00:14:36.480 --> 00:14:40.919 Oh. Absolutely, They are slapping down natural gas turbines and 286 00:14:41.120 --> 00:14:44.120 massive fuel cells next to their buildings just to keep 287 00:14:44.159 --> 00:14:48.080 the server's humming because the municipal power grid simply cannot 288 00:14:48.159 --> 00:14:49.039 handle the load. 289 00:14:49.279 --> 00:14:50.200 That's insane. 290 00:14:50.279 --> 00:14:54.320 The forecast actually describes the underlying financial dynamic driving this frenzy, 291 00:14:54.360 --> 00:14:57.399 and it illustrates why the bottleneck is so critical. They 292 00:14:57.399 --> 00:14:59.360 call it the fifteen fifteen to fifteen dynamic. 293 00:14:59.440 --> 00:15:02.399 Oka break that down because the numbers behind it are wild. 294 00:15:02.519 --> 00:15:05.960 So the fifteen fifteen fifteen dynamic refers to fifteen year 295 00:15:06.000 --> 00:15:07.600 commercial leases on data centers. 296 00:15:07.679 --> 00:15:09.200 Okay, fifteen year leases which. 297 00:15:09.039 --> 00:15:12.200 Are currently delivering fifteen percent yields to their investors, which 298 00:15:12.240 --> 00:15:15.360 is a huge return, massive while generating roughly fifteen dollars 299 00:15:15.360 --> 00:15:18.840 of net economic value for every single watt of power consumed. 300 00:15:19.039 --> 00:15:22.960 I mean, think about that final metric, fifteen dollars of 301 00:15:23.159 --> 00:15:26.720 direct economic value for every single. 302 00:15:26.480 --> 00:15:27.919 Wat, every single watt. 303 00:15:28.159 --> 00:15:31.519 Data centers are no longer just passive real estate housing servers. 304 00:15:31.879 --> 00:15:36.679 They are active intelligence factories, generating more direct economic value 305 00:15:36.679 --> 00:15:40.440 per square foot than almost any traditional heavy industry on Earth. 306 00:15:40.519 --> 00:15:41.799 Ye's unprecedented. 307 00:15:41.960 --> 00:15:45.399 When you multiply fifteen dollars a want by gigawatts of power. 308 00:15:46.000 --> 00:15:50.080 The financial incentive to build these centers is astronomical. It 309 00:15:50.159 --> 00:15:52.559 is creating a terrifying gold rush. 310 00:15:52.600 --> 00:15:55.639 But the math of the physical world is unforgiving. Right, 311 00:15:55.799 --> 00:15:58.200 you can have all the financial incentive imaginable. But if 312 00:15:58.200 --> 00:16:01.200 you don't have the physical infrastructure to generate, transform, and 313 00:16:01.240 --> 00:16:04.320 transmit that electricity, the entire build out stalls. 314 00:16:04.399 --> 00:16:05.120 So what's the fix? 315 00:16:05.360 --> 00:16:08.320 Solving this requires hundreds of billions of dollars in new 316 00:16:08.320 --> 00:16:11.960 power generation and deep grid modernization. If we fail to 317 00:16:12.039 --> 00:16:15.759 upgrade the grid, the AI revolution risks hitting a sudden wall, 318 00:16:15.799 --> 00:16:19.080 precisely at the moment its cognitive capabilities are exploding. 319 00:16:19.200 --> 00:16:21.799 Here's where it gets really interesting, though, Oh yeah, because 320 00:16:21.840 --> 00:16:24.399 let's say they do solve that power grid issue. Let's 321 00:16:24.399 --> 00:16:27.440 assume they throw enough capital at natural gas, nuclear energy, 322 00:16:27.480 --> 00:16:30.440 and grid upgrodes to keep those GPUs running at full throttle. 323 00:16:31.080 --> 00:16:33.360 We have to look slightly past twenty twenty six to 324 00:16:33.399 --> 00:16:36.639 see where this exponential curve inevitably leads. 325 00:16:36.879 --> 00:16:38.679 And that's where things get strange. 326 00:16:38.960 --> 00:16:42.399 Very According to AI executives cited in the forecast, the 327 00:16:42.440 --> 00:16:45.240 next major milestone arrives in the first half of twenty 328 00:16:45.240 --> 00:16:46.039 twenty seven. 329 00:16:45.919 --> 00:16:48.559 The emergence of recursive self improvement loops. 330 00:16:48.840 --> 00:16:53.600 Yes, this is where the trajectory shifts from rapid technological 331 00:16:53.639 --> 00:16:57.360 progress to something resembling a sci fi reality. We are 332 00:16:57.440 --> 00:17:01.759 talking about advanced AI models autonomously designing better, more efficient 333 00:17:01.840 --> 00:17:02.840 versions of themselves. 334 00:17:03.080 --> 00:17:06.599 Right, and to understand the mechanism here, imagine an AI 335 00:17:06.640 --> 00:17:10.079 that has achieved expert level proficiency in software engineering and 336 00:17:10.119 --> 00:17:11.680 hardware optimization. 337 00:17:11.400 --> 00:17:13.480 Like we just talked about with GDPVEL Exactly. 338 00:17:13.640 --> 00:17:16.359 Instead of a human engineering team spending six months tweaking 339 00:17:16.359 --> 00:17:19.960 the architecture of the next model, the AI itself analyzes 340 00:17:20.000 --> 00:17:20.799 its own code. 341 00:17:20.880 --> 00:17:21.880 It looks at its own brain. 342 00:17:22.000 --> 00:17:26.640 Yes, it identifies inefficiencies, rewrites its underlying algorithms, and optimizes 343 00:17:26.680 --> 00:17:29.680 its neural pathways. And it does this iterating as speeds 344 00:17:29.680 --> 00:17:31.839 that no human team could ever match. 345 00:17:32.160 --> 00:17:33.319 It's just NonStop. 346 00:17:33.480 --> 00:17:37.279 The AI conducts its own AI research, effectively removing the 347 00:17:37.359 --> 00:17:40.000 human engineers from the upgrade loop entirely. 348 00:17:40.359 --> 00:17:43.519 Jimmy Bay, the co founder of xai, has been very 349 00:17:43.599 --> 00:17:47.880 vocal about this specific concept. He describes a future where 350 00:17:47.880 --> 00:17:51.839 intelligence compounds exponentially. The moment this threshold is. 351 00:17:51.839 --> 00:17:53.400 Crossed, it's a runaway train. 352 00:17:53.480 --> 00:17:57.240 At that point when an artificial intelligence can autonomously build 353 00:17:57.240 --> 00:18:02.799 a smarter artificial intelligence between generations shrinks dramatically, goes from 354 00:18:02.920 --> 00:18:05.920 years to months to weeks to days. 355 00:18:06.160 --> 00:18:08.599 Pure intelligence becomes the coin of the realm. 356 00:18:08.720 --> 00:18:12.319 Exactly. It becomes the ultimate infinitely scalable commodity. 357 00:18:12.680 --> 00:18:16.440 And this is where traditional economic forecasting completely breaks. 358 00:18:16.200 --> 00:18:18.119 Down because there's no precedence none. 359 00:18:18.400 --> 00:18:22.200 Markets, regulators, and societies have entirely failed to internalize the 360 00:18:22.279 --> 00:18:26.200 velocity of a recursive self improvement loop. When intelligence begins 361 00:18:26.240 --> 00:18:29.519 manufacturing intelligence, you can no longer predict the capabilities of 362 00:18:29.519 --> 00:18:32.680 the system based on historical rates of human progress. 363 00:18:32.359 --> 00:18:34.480 Right because humans aren't making the progress anymore. 364 00:18:34.759 --> 00:18:37.559 Exactly, you enter entirely new territory where the rate of 365 00:18:37.599 --> 00:18:39.400 acceleration itself is accelerating. 366 00:18:39.519 --> 00:18:41.960 So to pull all of this together for you, the 367 00:18:41.960 --> 00:18:44.400 core reality you need to prepare for is that the 368 00:18:44.440 --> 00:18:48.160 twenty twenty six breakthrough isn't some distant pipe dream. 369 00:18:48.279 --> 00:18:49.200 It's not science fiction. 370 00:18:49.440 --> 00:18:53.720 No, it is a mathematical certainty, powered by an unprecedented 371 00:18:53.720 --> 00:18:58.279 accumulation of raw compute and unbroken scaling laws that dictate 372 00:18:58.319 --> 00:19:02.119 a simple truth. More hardware equals more. 373 00:19:01.960 --> 00:19:05.559 Intelligence, and we already have the concrete proof with GPT 374 00:19:05.759 --> 00:19:08.640 five point four hitting an eighty three percent score on 375 00:19:08.680 --> 00:19:10.599 the gdpval benchmark. 376 00:19:10.160 --> 00:19:13.039 Proving it can rival human experts in the most economically 377 00:19:13.119 --> 00:19:14.519 valuable professions on Earth. 378 00:19:14.640 --> 00:19:18.079 Yeah, the resulting deflationary boom is locked in the marginal 379 00:19:18.119 --> 00:19:21.400 cost of complex knowledge work will plummet, allowing tiny teams 380 00:19:21.400 --> 00:19:23.720 to build massive global companies. 381 00:19:23.319 --> 00:19:27.000 But the corresponding workforce disruptions are equally locked in. The 382 00:19:27.079 --> 00:19:29.960 decoupling of revenue growth from human employment will force a 383 00:19:30.000 --> 00:19:33.400 deeply painful transition for millions of white collar workers. 384 00:19:33.599 --> 00:19:36.160 The only real speed limit on this runaway train is 385 00:19:36.160 --> 00:19:37.640 the physical world, right. 386 00:19:38.359 --> 00:19:41.960 The race for power infrastructure mitigating that nine to eighteen 387 00:19:42.000 --> 00:19:46.200 gigawatt energy shortfall is the final hurdle before we reach 388 00:19:46.240 --> 00:19:49.480 the event horizon in twenty twenty seven, where AI begins 389 00:19:49.559 --> 00:19:51.319 autonomously upgrading itself. 390 00:19:51.720 --> 00:19:55.519 Governments, corporations, and individuals who treat this as just another 391 00:19:55.799 --> 00:20:00.160 standard tech cycle, like the transition from desktop computers to 392 00:20:00.519 --> 00:20:03.599 mobile phones, they're going to be left entirely behind. 393 00:20:03.359 --> 00:20:06.079 Completely behind. The twenty year horizon we talked about at 394 00:20:06.079 --> 00:20:08.720 the beginning, it doesn't exist anymore. The clock is ticking 395 00:20:08.839 --> 00:20:10.480 in twenty twenty six is right around. 396 00:20:10.319 --> 00:20:13.599 The corner, and as you prepare for that collapsing horizon, 397 00:20:13.880 --> 00:20:16.480 there is a completely new thought you should be pondering. 398 00:20:16.599 --> 00:20:19.960 What's that If digital agents will very soon replicate the 399 00:20:19.960 --> 00:20:22.799 cognitive output of our highest paying professions at a fraction 400 00:20:22.880 --> 00:20:26.200 of the cost. What happens to the value of uniquely 401 00:20:26.319 --> 00:20:30.720