1 00:00:00,000 --> 00:00:02,291 Sixteen developers. That's it. 2 00:00:02,291 --> 00:00:05,452 Sixteen. Every one of them experienced. 3 00:00:05,452 --> 00:00:10,674 Average five years working on their own open source projects. 4 00:00:10,674 --> 00:00:15,696 These aren't tourists. These are people who know their code bases the 5 00:00:15,696 --> 00:00:19,298 way you know the layout of your own kitchen in the dark. 6 00:00:19,298 --> 00:00:23,839 Code bases averaging over a million lines of code. 7 00:00:23,839 --> 00:00:27,621 Meter paid them a hundred and fifty dollars. 8 00:00:27,754 --> 00:00:30,196 dollars an hour to do something very simple. 9 00:00:30,196 --> 00:00:36,240 Do your real work. 246 real tasks. 10 00:00:36,240 --> 00:00:39,001 Half of them use the AI tools. 11 00:00:39,001 --> 00:00:44,405 The good ones. Cursor Pro running Claude 3.5 and 3.7 Sonnet. 12 00:00:44,405 --> 00:00:46,406 The state of the art. Half of them, 13 00:00:46,406 --> 00:00:51,209 no AI. And here's the thing. Before they started, 14 00:00:51,209 --> 00:00:55,352 they were asked to predict how much faster will AI make you. 15 00:00:55,352 --> 00:00:56,383 They said 16 00:00:56,639 --> 00:01:01,923 24% faster. The economists who study this predicted 17 00:01:01,923 --> 00:01:07,258 39% time savings. The machine learning experts predicted 38%. 18 00:01:07,258 --> 00:01:11,982 Everyone agreed. The AI would make them faster. 19 00:01:11,982 --> 00:01:14,224 Then they actually measured it. 20 00:01:14,224 --> 00:01:17,727 Like actually measured it. Not a survey, 21 00:01:17,727 --> 00:01:23,192 not a vibe, a gold standard randomized controlled trial. 22 00:01:23,192 --> 00:01:24,513 We call that science. 23 00:01:25,228 --> 00:01:30,411 The developers using AI took nineteen percent longer, 24 00:01:30,411 --> 00:01:36,766 slower, not by a little, nineteen percent slower doing their own work with 25 00:01:36,766 --> 00:01:40,178 the best tools money could buy at the time. 26 00:01:40,178 --> 00:01:46,812 Then, after they finished, after they were objectively measurably 27 00:01:46,812 --> 00:01:48,943 slower, they were asked again, 28 00:01:48,943 --> 00:01:54,267 how did the AI do? And they said it made them twenty percent faster. 29 00:01:54,996 --> 00:01:57,937 They were slower. They felt faster. 30 00:01:57,937 --> 00:02:00,457 They could not tell the difference between the two. 31 00:02:00,457 --> 00:02:05,388 Sixteen people who write software for a living on their own coat could 32 00:02:05,388 --> 00:02:09,219 not perceive that a tool was slowing them down. 33 00:02:09,219 --> 00:02:13,080 They were paid to notice. They were experts. 34 00:02:13,080 --> 00:02:18,321 And their own senses lied to them in the same direction at the same time. 35 00:02:18,321 --> 00:02:24,113 Now hold that thought, because right now across this country and across 36 00:02:24,113 --> 00:02:25,113 the country I'm doing. 37 00:02:25,023 --> 00:02:30,711 currently recording this, companies are laying off tens of thousands of people. 38 00:02:30,711 --> 00:02:35,678 And the reason they give is that the AI makes everyone faster. 39 00:02:35,678 --> 00:02:38,202 They can feel it. Stick with me. 40 00:02:53,201 --> 00:02:55,773 This is The Overlap, the show about systems of power, 41 00:02:55,773 --> 00:02:57,835 labor, and American injustice. 42 00:02:57,835 --> 00:03:00,236 I'm Joshua, and I'm very glad you're here. 43 00:03:00,236 --> 00:03:05,540 I am ~ actually recording this in Canada right now. 44 00:03:05,540 --> 00:03:07,272 Don't worry, I haven't moved yet. 45 00:03:07,272 --> 00:03:13,557 I don't think I'm going to, but I am recording this ~ remotely. 46 00:03:13,557 --> 00:03:15,949 So the audio might sound a little bit different today. 47 00:03:15,949 --> 00:03:19,001 The ~ content, however, is not going to be. 48 00:03:19,408 --> 00:03:22,430 Because we still talk about systems of power and we talk about labor 49 00:03:22,430 --> 00:03:25,402 and we talk about money and politics and how it relates to you. 50 00:03:25,402 --> 00:03:27,723 Quick word for Will. He's still not with us. 51 00:03:27,723 --> 00:03:32,747 We're sending our love and hope and all of those things to him out in the universe, 52 00:03:32,747 --> 00:03:34,167 putting out some positive vibes. 53 00:03:34,167 --> 00:03:35,839 I hope he's drinking a cup of coffee right now. 54 00:03:35,839 --> 00:03:37,350 And listen to this. You earned it, 55 00:03:37,350 --> 00:03:39,752 buddy. We love ya. So here's where we are. 56 00:03:39,752 --> 00:03:44,115 The topic today is AI. Or more precisely, 57 00:03:44,115 --> 00:03:45,115 it's the gap. 58 00:03:44,866 --> 00:03:48,888 The gap between what we're told artificial intelligence is doing to 59 00:03:48,888 --> 00:03:52,530 the economy and what we can actually measure it doing. 60 00:03:52,530 --> 00:03:55,151 And I want to be clear from the top. 61 00:03:55,151 --> 00:03:58,793 This is not an episode about whether technology is cool. 62 00:03:58,793 --> 00:04:02,095 We all know it's cool. Some of it is really, 63 00:04:02,095 --> 00:04:05,336 really impressive. This is an episode about proof. 64 00:04:05,336 --> 00:04:10,539 It's about measurement. It's about the difference between a claim and a fact. 65 00:04:10,539 --> 00:04:13,951 Because here's the thing that made me want to do this this week. 66 00:04:14,359 --> 00:04:18,310 So this is July 2026. It's not one story. 67 00:04:18,310 --> 00:04:22,991 It's never one story, right? It's a collision of two stories. 68 00:04:22,991 --> 00:04:25,292 On one side, you've got the money. 69 00:04:25,292 --> 00:04:27,813 For incredibly large companies, 70 00:04:27,813 --> 00:04:31,273 Microsoft, Amazon, Google, Meta. 71 00:04:31,273 --> 00:04:36,995 They are projected to spend $670 billion, 72 00:04:36,995 --> 00:04:41,496 ba-ba-ba-ba-ba-billion, on AI infrastructure this year alone. 73 00:04:42,459 --> 00:04:45,150 Six hundred and seventy billion dollars. 74 00:04:45,150 --> 00:04:48,490 That's a sixty-two percent jump from last year. 75 00:04:48,490 --> 00:04:51,871 And they have burned through their free cash flow. 76 00:04:51,871 --> 00:04:56,742 They're actually issuing debt to keep building what they're building because they 77 00:04:56,742 --> 00:05:01,644 think the value is going to eventually outweigh the detriment and what it's costing 78 00:05:01,644 --> 00:05:07,295 them. Data center debt doubled to $182 billion last year. 79 00:05:07,295 --> 00:05:10,036 On the other side of this, you have Goldman Sachs. 80 00:05:10,750 --> 00:05:12,711 Not some, you know, skeptical blog. 81 00:05:12,711 --> 00:05:15,132 it's not an Alex Jones scenario, 82 00:05:15,132 --> 00:05:19,323 although the Onion has taken over ~ Alex Jones' old show, 83 00:05:19,323 --> 00:05:21,184 and it's hilarious, you have to check it out. 84 00:05:21,184 --> 00:05:23,125 So this is Goldman Sachs, right? 85 00:05:23,125 --> 00:05:27,387 If you haven't heard of them, crawl out from under the rock you've been hiding 86 00:05:27,387 --> 00:05:31,828 under. But their own senior economist said in March of this year, 87 00:05:31,828 --> 00:05:38,451 we will not find a meaningful relationship between productivity and AI adoption 88 00:05:38,791 --> 00:05:41,992 At the economy wide level. You didn't hear that wrong. 89 00:05:41,992 --> 00:05:46,896 No meaningful relationship after all of that money. 90 00:05:46,896 --> 00:05:52,991 Now the reason both the history and the moment matter here is that we have seen this 91 00:05:52,991 --> 00:05:55,563 exact movie before, more than once. 92 00:05:55,563 --> 00:05:59,626 And every single time the people telling you the revolution is here 93 00:05:59,847 --> 00:06:04,010 and it's measurable, they were the people getting paid for you to believe it. 94 00:06:04,010 --> 00:06:05,251 So we're gonna measure. 95 00:06:05,861 --> 00:06:08,972 That's the whole game today. We're gonna do the one thing that the 96 00:06:09,032 --> 00:06:11,603 the AI hype artists don't want you doing. 97 00:06:11,603 --> 00:06:14,874 We're gonna ask quietly, patiently, 98 00:06:14,874 --> 00:06:18,296 okay, show me where's the proof. 99 00:06:18,296 --> 00:06:21,457 And if the claims were true, we'd have the proof by now. 100 00:06:21,457 --> 00:06:25,678 But we don't. I want you to hold on to that phrase throughout this whole episode 101 00:06:25,678 --> 00:06:29,680 because it's gonna keep coming back and it's gonna feel different every time. 102 00:06:29,680 --> 00:06:33,021 If the claims were true, we'd have proof by now. 103 00:06:33,740 --> 00:06:36,252 Not a forecast, not a survey, not a feeling. 104 00:06:36,252 --> 00:06:41,535 Proof. So let's go back. Because to understand why this pattern keeps working, 105 00:06:41,535 --> 00:06:43,937 you have to understand that it is a pattern, 106 00:06:43,937 --> 00:06:49,530 and that has a shape. And the shape is older than most of the people building these 107 00:06:49,530 --> 00:06:51,151 models. Nineteen eighty seven, 108 00:06:51,151 --> 00:06:53,263 an economist named Robert Solaw, 109 00:06:53,263 --> 00:06:59,547 Nobel laureate, MIT, one of the most respected people in this entire field, 110 00:06:59,547 --> 00:07:01,348 wrote a short observation to 111 00:07:01,640 --> 00:07:03,121 In the New York Review of Books. 112 00:07:03,121 --> 00:07:08,565 And it becomes one of the most quoted lines in the history of economics. 113 00:07:08,565 --> 00:07:12,588 He said, You can see the computer age everywhere, 114 00:07:12,588 --> 00:07:15,870 except in productivity statistics. 115 00:07:15,870 --> 00:07:19,193 Sit with that for a second. 1987, 116 00:07:19,193 --> 00:07:21,794 I was three years old. By then, 117 00:07:21,794 --> 00:07:26,538 American business had spent well over a decade pouring money into computers. 118 00:07:26,538 --> 00:07:28,800 the nineteen seventies, the nineteen eighties, 119 00:07:28,800 --> 00:07:34,562 offices are filling up with these machines that basically generate heat. 120 00:07:34,562 --> 00:07:40,027 Everyone could see them, everyone believed they were changing everything. 121 00:07:40,027 --> 00:07:43,109 The revolution was really visible on every desk. 122 00:07:43,109 --> 00:07:45,161 But the productivity numbers flat. 123 00:07:45,161 --> 00:07:51,656 Nothing. You could not find the computer revolution in the one place that 124 00:07:51,656 --> 00:07:56,370 it was sold, that it was supposed to show up in the aggregate data. 125 00:07:56,370 --> 00:07:59,072 The bottom line for the whole economy. 126 00:07:59,072 --> 00:08:02,344 They actually called it the Sola Paradox, 127 00:08:02,344 --> 00:08:07,068 and I want to be really careful here because the hype artists around 128 00:08:07,068 --> 00:08:11,722 AI really love this story. They love it because they think it ends well for them. 129 00:08:12,580 --> 00:08:14,791 They say see, the gains were real, 130 00:08:14,791 --> 00:08:17,312 they just came late. Be patient, 131 00:08:17,312 --> 00:08:19,553 the gains from AI are coming too. 132 00:08:19,553 --> 00:08:22,834 And here's the thing that's half true. 133 00:08:22,834 --> 00:08:27,716 And half that's true is the half that actually damns them. 134 00:08:27,716 --> 00:08:31,717 Because yes, eventually the computer revolution did show up 135 00:08:31,717 --> 00:08:34,759 in the productivity numbers in the late 1990s. 136 00:08:34,759 --> 00:08:37,600 But I need you to listen to how it happened. 137 00:08:38,475 --> 00:08:43,358 It took roughly a decade longer than the boosters of the eighties promised. 138 00:08:43,358 --> 00:08:46,400 And it didn't happen because the machines got magic. 139 00:08:46,400 --> 00:08:50,203 It happened because organizations spent years, 140 00:08:50,203 --> 00:08:56,046 a decade, two decades restructuring how they actually 141 00:08:56,047 --> 00:09:00,289 did work around the machine. New processes, 142 00:09:00,289 --> 00:09:02,911 new training, new job descriptions, 143 00:09:02,911 --> 00:09:07,604 new management layers built specifically to extract value from the 144 00:09:08,043 --> 00:09:10,144 out of the technology that they had purchased. 145 00:09:10,144 --> 00:09:12,706 The technology alone really did nothing. 146 00:09:12,706 --> 00:09:16,809 The technology plus a decade of expensive, 147 00:09:16,809 --> 00:09:21,733 painful, human organizational change actually did something. 148 00:09:21,733 --> 00:09:24,705 That's the real solo paradox, right? 149 00:09:24,705 --> 00:09:27,957 It's not be patient, the magic is coming, 150 00:09:27,957 --> 00:09:32,901 it's the technology is the easy part and the productivity gain is the hard part 151 00:09:33,021 --> 00:09:36,304 and it takes 10 years longer than anyone selling it. 152 00:09:36,742 --> 00:09:38,812 is willing to admit. And importantly, 153 00:09:38,812 --> 00:09:43,244 the companies that actually got the productivity gain weren't the ones that bought 154 00:09:43,244 --> 00:09:48,106 the most machines. They were the ones that changed how people worked around 155 00:09:48,106 --> 00:09:50,536 the machines. The hardware was necessary, 156 00:09:50,536 --> 00:09:53,897 but not really sufficient. The human restructure, 157 00:09:53,897 --> 00:09:56,288 that was the variable that really mattered, 158 00:09:56,288 --> 00:09:59,569 and nobody selling hardware ever led with that, 159 00:09:59,569 --> 00:10:01,580 because frankly that's expensive. 160 00:10:01,580 --> 00:10:02,580 Now fast forward. 161 00:10:02,770 --> 00:10:06,032 The 1990s, we probably remember this. 162 00:10:06,032 --> 00:10:09,484 This is the dot-com boom. And ~ the claims. 163 00:10:09,484 --> 00:10:13,586 The internet was going to permanently accelerate GDP growth. 164 00:10:13,586 --> 00:10:20,330 Analysts predicted productivity gains of three to four percent every year forever. 165 00:10:20,330 --> 00:10:23,351 Some of them said the business cycle itself was over. 166 00:10:23,351 --> 00:10:26,293 Recessions, a thing of the past. 167 00:10:26,293 --> 00:10:28,654 We had transcended economics. We were 168 00:10:28,804 --> 00:10:31,356 We were post a post economic society. 169 00:10:31,356 --> 00:10:34,698 There was a famous business week cover, 170 00:10:34,698 --> 00:10:39,262 and there there were wired manuf manifesto, 171 00:10:39,262 --> 00:10:43,025 serious, credentialed people with a lot of letters behind their name, 172 00:10:43,025 --> 00:10:49,450 stood at podiums all over the world and said that the old rules no longer applied. 173 00:10:49,450 --> 00:10:55,394 That the the internet had changed the underlying physics of growth. 174 00:10:55,394 --> 00:10:57,195 The NASDAQ peaked in March of 175 00:10:57,195 --> 00:10:58,195 two thousand 176 00:10:58,346 --> 00:11:00,967 By October of two thousand two, 177 00:11:00,967 --> 00:11:03,307 it had lost seventy eight percent of its value. 178 00:11:03,307 --> 00:11:08,008 Five trillion dollars in market value were gone overnight. 179 00:11:08,008 --> 00:11:12,389 Vaporized. Not over years, over months. 180 00:11:12,389 --> 00:11:16,870 And the companies that disappeared weren't small bets. 181 00:11:16,870 --> 00:11:20,271 Pets dot com WebVans 182 00:11:21,233 --> 00:11:22,233 E toys. 183 00:11:21,926 --> 00:11:26,047 These were businesses that had raised hundreds of millions of dollars 184 00:11:26,407 --> 00:11:30,888 on the strength story, on the feeling that the internet changed everything. 185 00:11:30,888 --> 00:11:37,210 So the old rules about revenue and profit and unit economics didn't really apply 186 00:11:37,210 --> 00:11:42,001 to them anymore. And the productivity gains and the the the hypers promise 187 00:11:42,001 --> 00:11:45,092 was you know, they came eventually, 188 00:11:45,092 --> 00:11:49,423 sort of, but a decade later than it was actually advertised. 189 00:11:50,373 --> 00:11:52,534 And nowhere near three to four percent forever. 190 00:11:52,534 --> 00:11:55,316 The internet was real. It's still very real. 191 00:11:55,316 --> 00:11:57,838 You're listening to this podcast over the internet. 192 00:11:57,838 --> 00:12:00,560 The internet was transformative. 193 00:12:00,560 --> 00:12:05,683 I'm not denying that. The internet is is ~ an infrastructure that's 194 00:12:05,683 --> 00:12:10,626 now so incredibly foundational that we'd now kind of forget to notice it. 195 00:12:10,626 --> 00:12:16,470 And the people who told you exactly how much and exactly when were wrong 196 00:12:16,991 --> 00:12:19,162 by an order of magnitude and 197 00:12:19,523 --> 00:12:22,824 And a decade. The technology is real. 198 00:12:22,824 --> 00:12:27,496 It was real. But the timeline and the magnitude completely fabricated. 199 00:12:27,496 --> 00:12:31,718 Confly stated, completely fabricated, 200 00:12:31,718 --> 00:12:37,420 right? Same shape every time. So let me give you a third one because this one's 201 00:12:37,420 --> 00:12:41,782 good, and I mean that in the the darkest dystopian way possible. 202 00:12:41,782 --> 00:12:46,904 IBM's Watson. Between twenty eleven and twenty sixteen, 203 00:12:46,904 --> 00:12:47,984 IBM 204 00:12:48,238 --> 00:12:52,860 marketed Watson as this revolutionary AI that would transform medicine 205 00:12:52,860 --> 00:12:58,732 and law and and finance. They spent billions of dollars promoting Watson health. 206 00:12:58,732 --> 00:13:04,335 And here was the pitch. This thing can diagnose cancer better than your doctor, 207 00:13:04,335 --> 00:13:05,915 better than a human oncologist, 208 00:13:05,915 --> 00:13:08,576 the future of medicine running on a mainframe. 209 00:13:08,576 --> 00:13:14,428 Indy Anderson Cancer Center, one of the best arguably cancer hospitals on the planet 210 00:13:15,196 --> 00:13:20,417 Bought in all the way. They spent sixty two million dollars on 211 00:13:20,417 --> 00:13:26,069 a Watson oncology project. Sixty two million to a research hospital. 212 00:13:26,069 --> 00:13:30,330 That's not a rounding error. That's years of clinical trials. 213 00:13:30,330 --> 00:13:35,921 That's research positions and equipment and opportunity costs that don't just come 214 00:13:35,921 --> 00:13:39,831 back. They actually canceled it in twenty seventeen because it failed 215 00:13:39,831 --> 00:13:42,072 to produce usable results. 216 00:13:42,692 --> 00:13:47,055 sixty two million dollars for a system that couldn't do the one thing it 217 00:13:47,055 --> 00:13:51,096 was sold to do. It's like if you spent sixty two million dollars on 218 00:13:51,096 --> 00:13:54,592 a blender that didn't make smoothies. 219 00:13:54,592 --> 00:13:55,923 And the Watson Health Division, 220 00:13:55,923 --> 00:13:58,294 the flagship, you know, the future of medicine, 221 00:13:58,294 --> 00:13:59,544 the thing that was going to cure cancer. 222 00:13:59,544 --> 00:14:06,227 IBM actually quietly sold it off in 2022 for a fraction of what they they believed 223 00:14:06,227 --> 00:14:07,908 it to be worth and hyped it to be worth. 224 00:14:07,908 --> 00:14:09,469 There was not a press conference, 225 00:14:09,469 --> 00:14:11,390 there was no, you know, Mia Culpa, 226 00:14:11,390 --> 00:14:16,723 no accountability whatsoever. Just a quiet sale and a hope that you'd forgotten 227 00:14:16,723 --> 00:14:19,514 the diagnosed cancer better than doctors' part. 228 00:14:19,514 --> 00:14:21,725 The press releases from 2020. 229 00:14:22,185 --> 00:14:24,156 13 or actually still on the internet. 230 00:14:24,156 --> 00:14:26,817 ~ I pulled a couple of them and you can read them too. 231 00:14:26,817 --> 00:14:31,800 The gap between what was claimed and what was delivered is right there. 232 00:14:31,800 --> 00:14:36,643 Archived and it's big. And it's kind of sad. 233 00:14:36,643 --> 00:14:39,265 Did they know it might not work when they were selling it? 234 00:14:39,265 --> 00:14:42,457 I I can't I mean I honestly I can't tell you what was in their hearts. 235 00:14:42,457 --> 00:14:43,678 I can't tell you what was in their heads, 236 00:14:43,678 --> 00:14:46,969 but I know the money moved before the proof did, 237 00:14:46,969 --> 00:14:48,160 every single time. 238 00:14:48,750 --> 00:14:50,170 The money moves first. The proof, 239 00:14:50,170 --> 00:14:53,211 if it comes at all, comes years later. 240 00:14:53,211 --> 00:14:57,063 And by then, the people who made the money are gone, 241 00:14:57,063 --> 00:15:01,734 and the people who lost the sixty-two million dollars are writing it off as well, 242 00:15:01,734 --> 00:15:06,375 it's a lesson learned. It's just make sure not to throw good money after bad. 243 00:15:06,375 --> 00:15:07,815 One more, the self-driving car. 244 00:15:07,815 --> 00:15:11,756 Between 2016 and 2018, Waymo, 245 00:15:11,756 --> 00:15:13,837 Uber, Tesla, all of them repeat. 246 00:15:13,897 --> 00:15:17,419 Repeatedly telling you full autonomy was two years away. 247 00:15:17,419 --> 00:15:21,342 Two years. Always two years. Two weeks. 248 00:15:21,342 --> 00:15:24,984 Travis Kalimek, the CEO of Uber, 249 00:15:24,984 --> 00:15:31,449 predicted in twenty sixteen that self driving cars would replace every single Uber 250 00:15:31,449 --> 00:15:36,352 driver within a decade. Elon Musk said in twenty nineteen that Tesla would have 251 00:15:36,352 --> 00:15:39,734 a million robotaxis on the road by twenty twenty. 252 00:15:39,734 --> 00:15:42,286 A million. By twenty twenty. 253 00:15:42,286 --> 00:15:45,588 That decade is basically now. And as of twenty twenty six, 254 00:15:45,588 --> 00:15:50,751 full level five autonomy, the car that drives everywhere, 255 00:15:50,751 --> 00:15:53,433 anywhere, any conditions with no human, 256 00:15:53,433 --> 00:15:56,355 it's not commercially available at scale. 257 00:15:56,355 --> 00:15:59,437 It doesn't exist as a product you can buy. 258 00:15:59,437 --> 00:16:04,980 Now Waymo operates you know in a handful of highly geofenced cities. 259 00:16:04,980 --> 00:16:08,943 And it's pretty cool if you've ever had the opportunity to ride in one. 260 00:16:08,943 --> 00:16:10,184 And 261 00:16:10,460 --> 00:16:14,902 Outside of them, in the rain, in a construction zone, 262 00:16:14,902 --> 00:16:19,524 in a small town in rural Oklahoma with no HD map data, 263 00:16:19,524 --> 00:16:23,506 just doesn't work. Uber's self driving division actually got sold off 264 00:16:23,506 --> 00:16:27,958 to Aurora in twenty twenty after one of their test vehicles killed 265 00:16:27,958 --> 00:16:30,189 a woman in Arizona. You might remember that from the news. 266 00:16:30,189 --> 00:16:34,631 The technology moved forward, but the promises from twenty sixteen, 267 00:16:34,631 --> 00:16:36,632 they've yet to materialize. 268 00:16:36,632 --> 00:16:38,582 Billions of dollars on the promise, 269 00:16:38,582 --> 00:16:44,173 on two years away, and the investors who funded these promises just moved 270 00:16:44,173 --> 00:16:47,954 on to the next story, and the drivers who were told their jobs would 271 00:16:47,954 --> 00:16:49,585 be automated out of existence, 272 00:16:49,585 --> 00:16:54,376 they're still driving. Now with the added anxiety of having been told 273 00:16:54,376 --> 00:16:56,447 for a decade that they're just temporary, 274 00:16:56,447 --> 00:16:58,747 they're a contract, they're 1099. 275 00:16:58,747 --> 00:17:03,348 So here's the pattern. And let's name it clean before we walk into the present. 276 00:17:03,806 --> 00:17:07,038 Because you're about to see it wearing a completely brand new suit. 277 00:17:07,038 --> 00:17:11,722 The pattern is this a genuinely interesting technology shows up. 278 00:17:11,722 --> 00:17:16,145 The people positioned to profit from it make enormous, 279 00:17:16,145 --> 00:17:22,330 specific, confident claims about how much it will change the economy and how fast. 280 00:17:22,330 --> 00:17:25,393 Those claims drive a flood of money. 281 00:17:25,393 --> 00:17:28,635 And look, the money creates its own momentum. 282 00:17:28,635 --> 00:17:32,098 You have to invest because everyone else is investing because 283 00:17:32,581 --> 00:17:33,581 What if it's real? 284 00:17:33,452 --> 00:17:37,514 And the actual measured results when they finally arrive are smaller 285 00:17:37,875 --> 00:17:41,077 and slower and messier than anyone promised. 286 00:17:41,077 --> 00:17:43,559 Computers in nineteen eighty seven, 287 00:17:43,559 --> 00:17:48,252 the internet in two thousand, Watson in twenty seventeen, 288 00:17:48,252 --> 00:17:51,144 self driving cars. We're still waiting on all of it. 289 00:17:51,144 --> 00:17:53,445 Solo told us nineteen eighty seven, 290 00:17:53,445 --> 00:17:55,807 and you can see it everywhere, 291 00:17:55,807 --> 00:18:00,180 except the statistics. And now there's a machine that talks back. 292 00:18:00,180 --> 00:18:02,982 And the claims are bigger than they have ever been. 293 00:18:02,982 --> 00:18:06,304 So let me ask the question I I keep asking myself. 294 00:18:06,304 --> 00:18:09,066 Why does this still work? Why, 295 00:18:09,066 --> 00:18:12,098 after computers and the internet and Watson and the self-driving 296 00:18:12,098 --> 00:18:14,139 car that's always two years away, 297 00:18:14,139 --> 00:18:18,131 why does anyone still fall for this specific confident number thing? 298 00:18:18,131 --> 00:18:22,674 And the oddest answer is because falling for it is extremely profitable 299 00:18:23,355 --> 00:18:25,216 for the people telling you to fall for it. 300 00:18:25,216 --> 00:18:28,108 That's it. It's the whole engine. 301 00:18:28,108 --> 00:18:30,299 Let's follow the money because it's not subtle. 302 00:18:30,299 --> 00:18:32,620 So Jensen Wong, the CEO of NVIDIA. 303 00:18:32,620 --> 00:18:36,793 Now, if you don't know, if you're not involved in all of that, 304 00:18:36,793 --> 00:18:40,605 you are hiding under a rock. NVIDIA makes the chips that everybody needs 305 00:18:40,605 --> 00:18:44,457 to build these AI systems. And since ChatGPT launched, 306 00:18:44,457 --> 00:18:49,110 NVIDIA's net income has grown roughly 20 times over. 307 00:18:49,110 --> 00:18:54,653 That's a 20x return. Their market cap hit around $3 trillion. 308 00:18:56,255 --> 00:18:58,546 And Jensen Wong is, without exaggeration, 309 00:18:58,546 --> 00:19:02,647 the single biggest financial winner of the entire AI boom. 310 00:19:02,647 --> 00:19:06,709 He is the arms dealer in a gold rush, 311 00:19:06,709 --> 00:19:11,031 and he has been richly historically rewarded for it. 312 00:19:11,031 --> 00:19:14,802 And here's the thing that he said that you should be sitting with 313 00:19:14,802 --> 00:19:18,083 at night when you're waiting on your popcorn to finish. 314 00:19:18,083 --> 00:19:22,275 Wong called the CEOs who blame AI for layoffs lazy. 315 00:19:23,239 --> 00:19:26,972 Lazy. He said it doesn't even make sense that companies are already using 316 00:19:26,972 --> 00:19:29,435 AI enough to be replacing people. 317 00:19:29,435 --> 00:19:30,626 Play that back if you need to. 318 00:19:30,626 --> 00:19:36,378 The man selling the shovels is telling you the gold isn't really there yet. 319 00:19:36,378 --> 00:19:42,221 Meaning the person who profits the most from this AI race is saying the thing 320 00:19:42,455 --> 00:19:46,283 the AI race is justified by just hasn't happened. 321 00:19:46,283 --> 00:19:50,145 And he's still selling shovels as fast as he can make them. 322 00:19:50,145 --> 00:19:53,327 He's even got some pre-sold for the future. 323 00:19:53,327 --> 00:19:57,699 2026 to try to buy a chip is impossible because they're sold all the 324 00:19:57,699 --> 00:19:58,990 way out until 2028. 325 00:19:59,972 --> 00:20:05,425 That's why you see Anthropic and and Grok are renting out their servers 326 00:20:05,425 --> 00:20:09,947 and and you know putting leases on Google Gemini to make it worse for everybody. 327 00:20:09,947 --> 00:20:13,209 Is because they don't have the chips to to purchase. 328 00:20:13,209 --> 00:20:15,390 They have to rent them from other people. 329 00:20:15,390 --> 00:20:18,612 And look, Jensen Wong is not contradicting himself. 330 00:20:18,612 --> 00:20:23,435 He is being perfectly consistent from the perspective of someone whose business 331 00:20:23,435 --> 00:20:26,356 model is infrastructure, not the output. 332 00:20:26,356 --> 00:20:27,857 He makes money whether or not 333 00:20:28,283 --> 00:20:30,704 The productivity gain ever materializes. 334 00:20:30,704 --> 00:20:32,572 Hy makes money on the belief. 335 00:20:32,572 --> 00:20:35,776 And so he can afford to be honest about the belief's limits because 336 00:20:35,776 --> 00:20:39,842 his balance sheet doesn't depend on the belief being true. 337 00:20:39,842 --> 00:20:43,577 It depends on you buying the chips while you still believe. 338 00:20:43,577 --> 00:20:45,568 And that's not a contradiction he's embarrassed by. 339 00:20:45,568 --> 00:20:49,250 That's the business model. Sell the infrastructure for a revolution that 340 00:20:49,250 --> 00:20:52,833 has yet to be proven. And if anyone get hurt gets hurt buying into it, 341 00:20:52,833 --> 00:20:57,446 you just call them lazy. Now, the four big buyers are the hyperscalers, 342 00:20:57,446 --> 00:20:59,117 right? Microsoft, Amazon, Google, 343 00:20:59,117 --> 00:21:06,081 Meta. Six hundred and sixty-seven billion dollars in projected capex this year. 344 00:21:06,081 --> 00:21:07,081 And I told you. 345 00:21:07,381 --> 00:21:10,763 They've burned through their operating cash flow and started borrowing. 346 00:21:10,763 --> 00:21:15,306 Data center debt doubled to $182 billion last year. 347 00:21:15,306 --> 00:21:18,648 Why? Why keep spending when your own bank, 348 00:21:18,648 --> 00:21:23,371 Goldman Sachs, tells you there's no measurable economy-wide return? 349 00:21:23,371 --> 00:21:26,172 Goldman's James Cavello had the answer, 350 00:21:26,172 --> 00:21:30,145 and it's pretty brutal. He diagnosed the whole thing as FOMO. 351 00:21:30,145 --> 00:21:33,167 For those of you who don't keep up with internet slang, 352 00:21:33,167 --> 00:21:34,938 FOMO means the fear of missing out. 353 00:21:34,938 --> 00:21:40,742 He said, and I quote, FOMO has proven a stronger incentive than poor stock 354 00:21:40,742 --> 00:21:43,204 performance. Think about what that means. 355 00:21:43,204 --> 00:21:47,547 These are the four most sophisticated companies in the world. 356 00:21:47,547 --> 00:21:49,909 They have armies of economists, 357 00:21:49,909 --> 00:21:52,611 risk analysts, strategy teams, 358 00:21:52,611 --> 00:21:58,365 and the thing driving six hundred and sixty-seven billion dollars in spending. 359 00:21:59,167 --> 00:22:04,191 Is the same thing that makes a teenager buy a stock because their group chat won't 360 00:22:04,191 --> 00:22:05,892 shut up about it. What if that's real? 361 00:22:05,892 --> 00:22:08,113 What if the other guy figures it out first? 362 00:22:08,113 --> 00:22:09,795 I can't be the one who missed it. 363 00:22:09,795 --> 00:22:13,358 The CEO of a company with a trillion dollar market cap is, 364 00:22:13,358 --> 00:22:14,919 at the moment of this decision, 365 00:22:14,919 --> 00:22:19,022 operating on the same psychology as someone panic buying during 366 00:22:19,022 --> 00:22:22,044 the GameStop GME rut. 367 00:22:22,044 --> 00:22:25,097 That's not a strategy, right? That's panic. 368 00:22:25,815 --> 00:22:28,537 Dressed up as a keynote. And then the third leg of the stool, 369 00:22:28,537 --> 00:22:31,800 the consultants. Mmm, everybody's favorite. 370 00:22:31,800 --> 00:22:37,314 McKenzie Global Institute. McKenzie projected that AI could contribute four point 371 00:22:37,314 --> 00:22:40,807 four trillion dollars in productivity gains annually. 372 00:22:40,807 --> 00:22:43,529 Trillion, that's with a T, you didn't hear me wrong. 373 00:22:43,529 --> 00:22:44,630 Annually. 374 00:22:44,630 --> 00:22:50,351 They anticipated three to four percentage points of additional GDP growth every year 375 00:22:50,652 --> 00:22:55,713 from now to twenty forty. That report got cited in boardrooms everywhere. 376 00:22:55,713 --> 00:23:01,725 It got cited by executives who had not read it to justify decisions that they 377 00:23:01,725 --> 00:23:04,556 had already made. It was the permission slip. 378 00:23:04,556 --> 00:23:08,477 four point four trillion dollars per year. 379 00:23:08,477 --> 00:23:10,147 Hold that number. 380 00:23:10,147 --> 00:23:14,568 Next to what an actual sober Nobel adjacent economist said. 381 00:23:14,568 --> 00:23:19,229 Because in 2024, Darren Asamoglu, 382 00:23:19,229 --> 00:23:25,510 MIT Nobel laureate, published a paper called The Simple Macroeconomics of AI. 383 00:23:25,510 --> 00:23:30,772 In his estimate, AI would produce somewhere between one point one and 384 00:23:30,772 --> 00:23:34,433 one point six percent of GDP growth over ten years. 385 00:23:34,433 --> 00:23:37,324 Not per year, total. 386 00:23:37,870 --> 00:23:41,972 over ten years. McKinsey said three to four points per year. 387 00:23:41,972 --> 00:23:47,425 Asa Moblu said one to one and a half points over a decade. 388 00:23:47,425 --> 00:23:50,856 That's not a disagreement or a rounding error. 389 00:23:50,856 --> 00:23:54,938 That's two different worlds. McKinsey's number is roughly twenty 390 00:23:54,938 --> 00:23:59,301 to thirty times bigger than than Asimoblu, 391 00:23:59,301 --> 00:24:01,922 measured over the same horizon. 392 00:24:01,922 --> 00:24:04,994 One of those two estimates is going to look very, 393 00:24:04,994 --> 00:24:05,994 very wrong. 394 00:24:05,942 --> 00:24:09,665 10 years. And I'll give you one guess which one was produced by 395 00:24:09,665 --> 00:24:13,118 the firm getting paid to help companies make the spending decision. 396 00:24:13,118 --> 00:24:17,642 And who pays McKinsey? The companies making the AI decisions. 397 00:24:17,642 --> 00:24:23,648 That's who. Average monthly enterprise AI spend hit almost $63,000 398 00:24:24,008 --> 00:24:29,873 in 2024, projected to rise to over $85,000 in 2025, 399 00:24:29,873 --> 00:24:32,976 a 36% jump, driven in part by 400 00:24:33,503 --> 00:24:34,813 By consulting recommendations. 401 00:24:34,813 --> 00:24:37,804 Mackenzie helps produce the number that justifies the spending. 402 00:24:37,804 --> 00:24:40,625 And then gets paid to help you do the spending, 403 00:24:40,625 --> 00:24:46,127 and nobody, nobody, has a contract clause that says if the four point four trillion 404 00:24:46,127 --> 00:24:48,188 doesn't materialize, we give your fee back. 405 00:24:48,188 --> 00:24:50,989 The accountability runs in one direction. 406 00:24:50,989 --> 00:24:53,140 That direction is the invoice. 407 00:24:53,140 --> 00:24:56,993 Look, nobody ever goes back and checks Mackenzie's four point four trillion dollar 408 00:24:56,993 --> 00:24:59,475 number against reality. You make the projection, 409 00:24:59,475 --> 00:25:02,177 you cash the check, and by the time the number's proven wrong, 410 00:25:02,177 --> 00:25:05,320 the engagement is over and there's a new number for a new technology. 411 00:25:05,320 --> 00:25:09,073 The thing that holds this whole machine together makes me angry in 412 00:25:09,073 --> 00:25:11,635 a very specific way. So in March of this year, 413 00:25:11,635 --> 00:25:16,599 Goldman found that seventy seventy give or take percent of S 414 00:25:16,599 --> 00:25:20,772 P five hundred management teams talked about AI on their earnings calls. 415 00:25:20,772 --> 00:25:21,772 Seventy percent. 416 00:25:21,955 --> 00:25:24,747 That means practically everyone's talking about it, 417 00:25:24,747 --> 00:25:30,312 but only 10% quantified its impact on any specific use case. 418 00:25:30,312 --> 00:25:35,415 And only 1%, 1% quantified its impact on earnings. 419 00:25:35,415 --> 00:25:38,737 1% of people showed their work, 420 00:25:38,737 --> 00:25:41,760 right? Showed the math. 70% of people are talking about it. 421 00:25:41,760 --> 00:25:43,971 One percent can actually point to a number. 422 00:25:43,971 --> 00:25:48,723 And the census data. So fewer than than twenty percent of US establishments 423 00:25:48,723 --> 00:25:51,935 are even using AI for any business function at all. 424 00:25:51,935 --> 00:25:55,097 Fewer than one in five businesses is using it. 425 00:25:55,097 --> 00:26:00,510 But a major component of the biggest companies are are on stage talking about 426 00:26:00,510 --> 00:26:02,801 it like it's already rebuilt the world, 427 00:26:02,801 --> 00:26:04,842 like the restructuring is done, 428 00:26:04,842 --> 00:26:06,923 like the productivity is already in the bag. 429 00:26:06,923 --> 00:26:09,124 And that's the machine, right? 430 00:26:09,124 --> 00:26:10,124 That's that's the thing. 431 00:26:10,541 --> 00:26:13,053 It runs on talk, it runs on the number nobody checks, 432 00:26:13,053 --> 00:26:17,246 and the FOMO nobody admits, and the earnings call where you have to say 433 00:26:17,527 --> 00:26:20,029 the word AI enough times to keep the stock up, 434 00:26:20,029 --> 00:26:23,752 whether or not you can actually point to a single dollar it's earned. 435 00:26:23,752 --> 00:26:28,596 six hundred and sixty seven billion dollars and one percent can prove 436 00:26:28,596 --> 00:26:32,039 it earned anything. If the claims were true, 437 00:26:32,039 --> 00:26:33,230 we'd have proof by now. 438 00:26:33,230 --> 00:26:38,954 We don't. So let's talk about Clairna because this is where it starts getting to, 439 00:26:38,954 --> 00:26:42,066 you know, like real people and real things. 440 00:26:42,066 --> 00:26:49,011 Because the AI hype artists are showing Clarna to everybody because they're 441 00:26:49,011 --> 00:26:53,474 holding it up like a trophy. So Clairna's a Swedish fintech company, 442 00:26:53,474 --> 00:26:55,905 financial technologies, buy now, 443 00:26:55,905 --> 00:27:01,339 pay later. And they deployed an open AI powered customer service assistant, 444 00:27:01,339 --> 00:27:02,339 Chatchi. 445 00:27:02,612 --> 00:27:04,573 And then they went out and told the world about it. 446 00:27:04,573 --> 00:27:06,714 And the numbers were incredible. 447 00:27:06,714 --> 00:27:11,458 The AI handled 2.3 million conversations in its first month. 448 00:27:11,458 --> 00:27:17,822 Clarina said it did the work of 853 full time human agents. 449 00:27:17,822 --> 00:27:22,265 That means 853 people were essentially replaced by a chatbot. 450 00:27:22,265 --> 00:27:27,199 They said the resolution time dropped from 11 minutes to under 2. 451 00:27:28,129 --> 00:27:32,291 They said they saved sixty million dollars of twenty twenty five, 452 00:27:32,291 --> 00:27:37,472 and they posted a hundred and fifty two percent increase in revenue per employee. 453 00:27:37,472 --> 00:27:39,433 That was the case study globally. 454 00:27:39,433 --> 00:27:42,694 All of these AI hype artists, they pointed at Clarina and said, 455 00:27:42,694 --> 00:27:45,415 Hey look, here it is. The proof you keep asking for. 456 00:27:45,415 --> 00:27:47,496 Here's a real company, real money, 457 00:27:47,496 --> 00:27:50,317 real people replaced, real savings. 458 00:27:50,317 --> 00:27:52,028 Stop being a skeptic. And look, 459 00:27:52,028 --> 00:27:53,969 honestly, that's a really cool story. 460 00:27:53,969 --> 00:27:55,530 If you only hear the first half. 461 00:27:55,530 --> 00:27:58,701 Because the second half, which came in late 2025, 462 00:27:58,701 --> 00:28:01,613 Clarin's own CEO, Sebastian I don't know, 463 00:28:01,613 --> 00:28:05,364 Simiatikowski, admitted the company had, 464 00:28:05,364 --> 00:28:08,315 in his word, over pivoted on AI. 465 00:28:08,315 --> 00:28:10,796 Quality problems, the service got worse, 466 00:28:10,796 --> 00:28:13,017 customers were having bad experiences. 467 00:28:13,017 --> 00:28:16,539 Tickets that should have been closed in two minutes were getting mishandled 468 00:28:16,539 --> 00:28:19,210 in ninety seconds instead. Is it faster? 469 00:28:19,210 --> 00:28:22,271 Sure. Is it correct? No, it's not. 470 00:28:22,902 --> 00:28:25,433 And they had to bring the human agents back to Clara. 471 00:28:25,433 --> 00:28:28,534 Bring them back, like 853 people. 472 00:28:28,534 --> 00:28:33,706 The trophy case study for AI replacing humans became a case study 473 00:28:33,706 --> 00:28:37,558 in bringing the humans back because the AI started to falter. 474 00:28:37,558 --> 00:28:42,209 So, because I mean this is kind of like the meter study all over again, 475 00:28:42,209 --> 00:28:44,100 just at a scale of a whole company. 476 00:28:44,100 --> 00:28:46,521 So Clarina felt the productivity gain. 477 00:28:46,521 --> 00:28:48,392 They announced it, they put it in a press release. 478 00:28:48,914 --> 00:28:52,666 They believed it's hard enough to restructure their entire customer service 479 00:28:52,666 --> 00:28:57,259 operation around it. And then when the actual quality got measured 480 00:28:57,679 --> 00:29:00,701 by their own customers having actual bad experiences, 481 00:29:00,701 --> 00:29:05,863 reality decided to step in. This gain that they felt just evaporated, 482 00:29:05,863 --> 00:29:07,744 and the humans had to come back. 483 00:29:07,744 --> 00:29:13,187 The gap between what they believed and what was true cost real people their real 484 00:29:13,187 --> 00:29:16,089 jobs. Temporarily, maybe, but you tell me. 485 00:29:16,089 --> 00:29:19,450 You tell the customer service worker who got laid off in the trophy announcement 486 00:29:19,711 --> 00:29:22,312 and quietly rehired in the walk back. 487 00:29:22,312 --> 00:29:24,673 You tell them that that gap was harmless. 488 00:29:24,673 --> 00:29:29,645 You tell them the months of unemployment and uncertainty while a company figured 489 00:29:29,645 --> 00:29:34,597 out the AI wasn't actually doing the job that they built the AI for. 490 00:29:34,597 --> 00:29:39,599 Tell the you tell them that it was a ~ it's a minor inconvenience and a story with 491 00:29:39,599 --> 00:29:44,089 a happy ending. Fifty five percent of companies that cut jobs in AI 492 00:29:44,361 --> 00:29:47,982 In 2024 and in 2025, they now regret it. 493 00:29:47,982 --> 00:29:51,183 That's from HR Dive actually, 55% regret. 494 00:29:51,183 --> 00:29:57,305 More than half. They fired real people based on what felt like a game, 495 00:29:57,305 --> 00:29:59,405 and then they wish they hadn't. 496 00:29:59,405 --> 00:30:01,985 And the the regret doesn't rehire anybody. 497 00:30:01,985 --> 00:30:05,666 It's just an admission. It's a Mia culpa in the aggregate. 498 00:30:05,666 --> 00:30:08,377 But the feeling wasn't proof. 499 00:30:08,377 --> 00:30:11,359 Now let me tell you about a case where nobody gets to be rehired, 500 00:30:11,359 --> 00:30:13,591 where the stakes aren't a customer service ticket. 501 00:30:13,591 --> 00:30:18,464 The Centers for Medical Medicare and Medicaid Services, 502 00:30:18,464 --> 00:30:22,177 CMS, in January of this year launched a program called Wiser. 503 00:30:22,177 --> 00:30:26,650 And I I have to tell you the full name because the full name is kind 504 00:30:26,650 --> 00:30:31,903 of a confession. Wasteful and inappropriate service reduction. 505 00:30:31,903 --> 00:30:35,546 Reduction. That's the goal. Right there in the name. 506 00:30:35,963 --> 00:30:38,996 Not better medicine, not more accurate decisions, 507 00:30:38,996 --> 00:30:44,721 reduction. The metric baked into the program's identity is denying more things, 508 00:30:44,721 --> 00:30:46,102 not getting more things right. 509 00:30:46,102 --> 00:30:51,627 Wiser uses AI to evaluate and effectively to automate the denial 510 00:30:51,888 --> 00:30:55,270 of requests for medical care for people on Medicaid. 511 00:30:55,270 --> 00:31:02,147 Poor people, sick people, people asking for care they and their doctors believe they 512 00:31:02,147 --> 00:31:03,147 need. 513 00:31:02,437 --> 00:31:07,138 And what should make your skin crawl is that the program was deployed with 514 00:31:07,238 --> 00:31:11,259 no public transparency about how the AI made its decisions. 515 00:31:11,259 --> 00:31:15,721 No published error rate, no independent audit before launch, 516 00:31:15,721 --> 00:31:21,983 no clear, accessible path for a patient to appeal an algorithmic denial. 517 00:31:21,983 --> 00:31:26,004 Nothing. A machine is deciding whether a poor sick person gets care, 518 00:31:26,004 --> 00:31:30,575 and the public, the people who essentially pay for it. 519 00:31:30,575 --> 00:31:33,096 Is not allowed to know how well the machine works. 520 00:31:33,096 --> 00:31:35,217 We don't know how often it's wrong. 521 00:31:35,217 --> 00:31:37,537 We don't know what it optimizes for. 522 00:31:37,537 --> 00:31:41,879 We don't know whether it was validated against actual patient outcomes 523 00:31:42,439 --> 00:31:45,080 or just against prior authorization approval rates. 524 00:31:45,080 --> 00:31:47,360 We know the name of a program, 525 00:31:47,360 --> 00:31:49,540 and we know it denies things, that's about it. 526 00:31:49,540 --> 00:31:54,432 Did you see it? Do you see how the hype connects to this? 527 00:31:54,432 --> 00:31:55,432 The whole reason. 528 00:31:55,793 --> 00:31:59,294 A government agency can deploy a black box to deny medical care, 529 00:31:59,294 --> 00:32:02,835 is that we've spent three years being told these systems work, 530 00:32:02,835 --> 00:32:06,666 that they're smart, that they're better than the humans. 531 00:32:06,666 --> 00:32:10,356 Nobody demanded the error rate because the culture had already assumed 532 00:32:10,356 --> 00:32:13,517 the machine was competent. The hype did that. 533 00:32:13,517 --> 00:32:19,439 The hype built the permission structure for a machine to deny your grandmother's 534 00:32:19,439 --> 00:32:24,070 care without ever having to prove that it can tell a valid claim from an invalid. 535 00:32:24,070 --> 00:32:25,973 And the hypers primed the audience. 536 00:32:25,973 --> 00:32:31,240 And CMS walked through the door and the hypers opened it. 537 00:32:31,240 --> 00:32:33,604 The Electronic Frontier Foundation, 538 00:32:33,604 --> 00:32:35,597 the EFF. If you haven't heard of these guys, 539 00:32:35,597 --> 00:32:36,597 you should really check them out. 540 00:32:36,498 --> 00:32:37,590 They do some interesting work. 541 00:32:37,590 --> 00:32:41,832 They filed a FOIA request, Freedom of Information Act lawsuit, 542 00:32:41,832 --> 00:32:47,256 against CMS to try to get the records to find out how this thing actually decides. 543 00:32:47,256 --> 00:32:48,937 They just wanted to see the algorithm, 544 00:32:48,937 --> 00:32:50,217 they wanted to see the training data, 545 00:32:50,217 --> 00:32:52,289 the validation results, the error rates, 546 00:32:52,289 --> 00:32:56,851 the basic information you would want before trusting a system with that 547 00:32:56,851 --> 00:32:59,383 big and that consequential of a decision. 548 00:32:59,383 --> 00:33:02,145 And as of this month, July of twenty twenty six, 549 00:33:02,145 --> 00:33:03,936 the lawsuit is still pending, still. 550 00:33:04,439 --> 00:33:08,212 They're still fighting just to see the numbers and how it was trained 551 00:33:08,312 --> 00:33:10,714 and whether there are even numbers at all. 552 00:33:10,714 --> 00:33:12,576 And that's the through line made physical. 553 00:33:12,576 --> 00:33:14,377 The one percent problem. Remember, 554 00:33:14,377 --> 00:33:18,740 only one percent of companies can actually quantify AI's impact on earnings? 555 00:33:18,740 --> 00:33:22,083 Well, CMS can't. Well, maybe they won't. 556 00:33:22,083 --> 00:33:25,245 Quantify Weiser's impact on patients. 557 00:33:25,245 --> 00:33:28,408 Same darkness. Same refusal to measure. 558 00:33:28,408 --> 00:33:32,121 Except now the thing being measured isn't a stock price. 559 00:33:32,835 --> 00:33:36,137 It's whether a human being gets the care that keeps them alive. 560 00:33:36,137 --> 00:33:38,859 Remember the 16 developers from the cold open? 561 00:33:38,859 --> 00:33:44,863 Experts, paid to notice, who couldn't feel that a tool was slowing them down. 562 00:33:44,863 --> 00:33:47,885 The perception gap wasn't malice. 563 00:33:47,885 --> 00:33:51,988 It was just human. People feel the tool working, 564 00:33:51,988 --> 00:33:54,010 they feel the responses generating, 565 00:33:54,010 --> 00:33:59,313 they feel productive. And the gap between the feeling and the measurement was 19%. 566 00:33:59,313 --> 00:34:00,314 Now imagine. 567 00:34:00,818 --> 00:34:07,163 That same perception gap, that same confident wrongness sitting inside 568 00:34:07,163 --> 00:34:11,656 a government system that denies medical care to people who could 569 00:34:11,656 --> 00:34:13,627 not otherwise afford medical care. 570 00:34:13,627 --> 00:34:17,510 Nobody at CMS is going home feeling like they made wrong decisions. 571 00:34:17,510 --> 00:34:20,242 The machine says no, it feels efficient, 572 00:34:20,242 --> 00:34:22,834 it feels like fiscal responsibility, 573 00:34:22,834 --> 00:34:26,376 which feels like progress. And there's no meter study for. 574 00:34:27,052 --> 00:34:30,334 People wiser denies. There's no gold standard trial, 575 00:34:30,334 --> 00:34:35,256 no, you know, $150 an hour to sit down and measure whether the machine is right. 576 00:34:35,256 --> 00:34:39,177 There's just a denial and a sick person and a lawsuit still pending. 577 00:34:39,177 --> 00:34:42,018 If the claims were true, we'd have proof by now. 578 00:34:42,018 --> 00:34:45,990 They're denying care with a machine that they won't let anybody measure. 579 00:34:45,990 --> 00:34:49,021 So let me pull this all the way back and I will tell you what I think 580 00:34:49,021 --> 00:34:50,941 of this and what it actually is. 581 00:34:50,941 --> 00:34:53,422 Because I actually don't think it's about AI. 582 00:34:53,422 --> 00:34:56,213 AI is just the current, you know, 583 00:34:56,213 --> 00:34:58,143 costume, it's the flavor of the week. 584 00:34:58,143 --> 00:35:01,004 The pattern is the perception reality gap. 585 00:35:01,004 --> 00:35:04,125 And it runs in the exact same direction every single time. 586 00:35:04,125 --> 00:35:07,505 What people believe about the productivity gain is always, 587 00:35:07,505 --> 00:35:11,186 always bigger than what you can measure. 588 00:35:11,186 --> 00:35:14,017 And the industry, this is the the the 589 00:35:14,231 --> 00:35:18,452 important part. The industry systematically avoids funding 590 00:35:18,652 --> 00:35:21,073 the measurement that would close the gap. 591 00:35:21,073 --> 00:35:22,894 Think about the the meter study again. 592 00:35:22,894 --> 00:35:28,175 It's the first gold standard randomized controlled trial of AI 593 00:35:28,175 --> 00:35:30,616 on real developer work. The first one. 594 00:35:30,616 --> 00:35:32,977 And it came out in July of twenty twenty five. 595 00:35:32,977 --> 00:35:36,458 That's why if you're following all the the you know foundational models, 596 00:35:36,458 --> 00:35:40,319 it used models that are now deprecated and older. 597 00:35:40,319 --> 00:35:41,369 But 598 00:35:41,711 --> 00:35:44,362 ChatGPT launched in November of 2022. 599 00:35:44,362 --> 00:35:46,073 Before that, we didn't have anything like this. 600 00:35:46,073 --> 00:35:52,156 That's more than two and a half years of the biggest technology story on Earth. 601 00:35:52,156 --> 00:35:55,037 Hundreds of billions of dollars. 602 00:35:55,037 --> 00:35:59,339 Entire companies restructured before anyone ran 603 00:36:00,039 --> 00:36:05,662 one rigorously controlled experiment on whether the flagship use case actually makes 604 00:36:05,662 --> 00:36:08,603 people faster. Two and a half years. 605 00:36:08,603 --> 00:36:11,304 And the gap wasn't filled by the companies building the tools. 606 00:36:11,304 --> 00:36:14,095 GitHub's own study, the vendor study, 607 00:36:14,095 --> 00:36:16,605 claimed a 55% productivity boost. 608 00:36:16,605 --> 00:36:21,706 55%. Now that was based on self-reported feelings 609 00:36:21,706 --> 00:36:26,968 at narrow synthetic tasks that looked nothing like real development 610 00:36:26,968 --> 00:36:30,038 and production work. That number was cited everywhere, 611 00:36:30,038 --> 00:36:33,919 like in earnings calls and board presentations and layoff announcements. 612 00:36:33,919 --> 00:36:35,710 For years it was the number. 613 00:36:37,430 --> 00:36:39,391 The vendor's feeling based number, 614 00:36:39,391 --> 00:36:41,813 right? The one before the independent trial existed. 615 00:36:41,813 --> 00:36:43,694 And when they finally ran the independent trial, 616 00:36:43,694 --> 00:36:48,617 the answer was 19% slower. Why do you think it took so long to measure that? 617 00:36:48,617 --> 00:36:53,160 Why do you think the measurement that existed was a vendor survey 618 00:36:53,480 --> 00:36:58,683 and not a controlled trial? Because the measurement was the enemy in this scenario. 619 00:36:58,683 --> 00:37:00,194 The gap is the product. 620 00:37:01,026 --> 00:37:03,577 As long as everyone's operating on vibes, 621 00:37:03,577 --> 00:37:06,719 the perception always favors the sale. 622 00:37:06,719 --> 00:37:09,300 The moment you measure, the moment you run the controlled trial, 623 00:37:09,300 --> 00:37:11,982 the moment you you know FOIA the algorithm, 624 00:37:11,982 --> 00:37:16,874 that gap starts to close. And the closing always goes the same direction, 625 00:37:16,874 --> 00:37:23,628 down, toward reality, toward smaller and slower and messier than promised. 626 00:37:23,628 --> 00:37:26,750 Solov told us in 1987, 627 00:37:26,750 --> 00:37:29,231 everywhere except the statistics. 628 00:37:29,805 --> 00:37:31,986 And here's the human cost. And this is where it stops being 629 00:37:31,986 --> 00:37:35,168 an interesting economic story and starts being an injustice, 630 00:37:35,168 --> 00:37:39,271 which is what this show tries to point out for everyone. 631 00:37:39,271 --> 00:37:44,104 The Harvard Business Review surveyed over a thousand global executives, 632 00:37:44,104 --> 00:37:48,717 and the finding was devastating in how plainly they said it. 633 00:37:48,717 --> 00:37:54,040 Companies are cutting jobs because of AI's anticipated future potential. 634 00:37:54,040 --> 00:37:59,454 Not because AI has demonstrated it can do the displaced workers' job anticipated. 635 00:38:00,162 --> 00:38:04,995 Future potential. These are not words that describe a thing that's been proven. 636 00:38:04,995 --> 00:38:07,876 Those are words that describe a bet. 637 00:38:07,876 --> 00:38:11,008 And the people losing their jobs aren't losing them because the machine 638 00:38:11,008 --> 00:38:12,419 has proven it can do the work. 639 00:38:12,419 --> 00:38:17,181 They're losing their jobs because a board decided the machine might do it someday. 640 00:38:17,181 --> 00:38:21,203 And someday is cheap, and people are expensive. 641 00:38:21,203 --> 00:38:26,446 Nearly a hundred thousand jobs since late twenty twenty three were blamed on AI. 642 00:38:27,709 --> 00:38:29,820 fifty five thousand in twenty twenty five alone. 643 00:38:29,820 --> 00:38:35,743 By May of this year, over eighty seven thousand layoffs blamed on AI just this year. 644 00:38:35,743 --> 00:38:39,764 Tech sector unemployment climbed to five point eight percent. 645 00:38:39,764 --> 00:38:44,046 That's the highest tech unemployment has been since the dot com bust. 646 00:38:44,046 --> 00:38:46,087 The last time we did exactly this thing. 647 00:38:46,087 --> 00:38:50,889 The last time the feeling outran the reality and regular people paid 648 00:38:50,889 --> 00:38:53,240 the price while the infrastructure sellers cashed out. 649 00:38:53,801 --> 00:38:58,984 An anthropic. Anthropic is an AI company doing the research. 650 00:38:58,984 --> 00:39:03,818 They found that job finding rates for workers aged 22 to 651 00:39:03,818 --> 00:39:08,331 25 entering AI exposed jobs fell 652 00:39:08,331 --> 00:39:11,394 14% since ChatGPT launched. 653 00:39:11,394 --> 00:39:17,598 Young people trying to start a career locked out of the first rung of the ladder. 654 00:39:17,598 --> 00:39:20,751 Not because a machine is doing the job really well. 655 00:39:20,751 --> 00:39:24,134 But because a hiring manager has a feeling that a machine might do 656 00:39:24,134 --> 00:39:29,028 it really well eventually, the cost of the hype is landing on the on 657 00:39:29,028 --> 00:39:32,101 the people who can least afford to absorb it. 658 00:39:32,101 --> 00:39:34,593 Not the executives deciding to cut, 659 00:39:34,593 --> 00:39:36,645 not the investors funding the infrastructure, 660 00:39:36,645 --> 00:39:40,309 the twenty three year old who can't get their first interview. 661 00:39:40,309 --> 00:39:41,310 That's the injustice. 662 00:39:41,310 --> 00:39:43,491 The gains are theoretical, the layoffs they're real. 663 00:39:43,491 --> 00:39:49,254 The $4.4 trillion is a McKinsey slide in their latest, 664 00:39:49,254 --> 00:39:55,018 you know, keynote. The 87,000 people are human beings with rent 665 00:39:55,278 --> 00:40:00,681 and student loans and families who were counting on a job because they were promised 666 00:40:00,761 --> 00:40:04,944 that if they went into this field and got this degree and went into this debt, 667 00:40:04,944 --> 00:40:07,005 that there would be something they're waiting for them. 668 00:40:07,005 --> 00:40:10,868 We are transferring wealth and security away from workers toward NVIDIA 669 00:40:10,968 --> 00:40:12,910 and hyperscalers and consultants, 670 00:40:12,910 --> 00:40:16,293 and we're doing it on the strength of a of a vibe. 671 00:40:16,293 --> 00:40:21,497 A vibe that sixteen experts couldn't even trust with their own work. 672 00:40:21,497 --> 00:40:25,640 A feeling Clarina believed until their own customers proved it wrong. 673 00:40:25,640 --> 00:40:31,946 A feeling CMS won't even let us look at while it denies people health care. 674 00:40:31,946 --> 00:40:33,927 The perception reality gap isn't a quirk. 675 00:40:33,927 --> 00:40:37,010 It's not an accident. It's a mechanism for moving money. 676 00:40:37,010 --> 00:40:40,212 And money always moves in a very specific direction, 677 00:40:40,212 --> 00:40:42,454 we've said that before, away from people who work. 678 00:40:42,638 --> 00:40:46,098 Word people who own. And the people it moves money away from, 679 00:40:46,098 --> 00:40:48,279 they don't get to file a lawsuit and wait. 680 00:40:48,279 --> 00:40:50,780 They don't get a FOIA request, 681 00:40:50,780 --> 00:40:53,580 they just lose the job. And then they watch the earnings call where the 682 00:40:53,580 --> 00:40:57,882 CEO says the word AI 14 times and the stock goes up. 683 00:40:57,882 --> 00:41:01,303 So here's what I want you to watch more closely. 684 00:41:01,303 --> 00:41:03,443 Not the claims, the measurement. 685 00:41:03,443 --> 00:41:05,534 Or, in this case, the absence of it. 686 00:41:05,534 --> 00:41:09,258 Every time you hear a number about what AI is doing to productivity or jobs 687 00:41:09,258 --> 00:41:12,102 or the economy, I want you to ask one question. 688 00:41:12,102 --> 00:41:17,289 Who measured this? And can I see it now or am I going to see it later? 689 00:41:17,289 --> 00:41:21,425 If the answer is a company survey or a self reported feeling 690 00:41:21,425 --> 00:41:24,900 or a McKenzie projection about twenty forty, 691 00:41:24,900 --> 00:41:27,383 that's not proof. That's marketing. 692 00:41:27,383 --> 00:41:29,355 What I want you to carry out here is this too. 693 00:41:29,355 --> 00:41:33,679 Seventy percent of the biggest companies in America talk about AI on earnings costs. 694 00:41:33,679 --> 00:41:37,362 One percent can point to what it did to their earnings. 695 00:41:37,362 --> 00:41:42,266 Seventy talking, one numbing. Hold those numbers next to each other for 696 00:41:42,266 --> 00:41:45,149 the rest of this year, and you'll see the whole thing pretty clearly. 697 00:41:45,149 --> 00:41:50,651 And then Jensen Wong, the man who's made more money off of this boom than anyone 698 00:41:50,651 --> 00:41:56,272 alive. Even, yes, your beloved and beloved to be hated Elon Musk, 699 00:41:56,272 --> 00:42:01,764 whose net income grew 20 times over selling picks and shovels. 700 00:42:01,764 --> 00:42:04,334 He's the one telling you the gold isn't there. 701 00:42:04,334 --> 00:42:08,115 He called the CEOs blaming AI for layoffs lazy. 702 00:42:08,115 --> 00:42:11,156 He said it doesn't make sense that AI is replacing people already. 703 00:42:11,476 --> 00:42:15,959 The biggest beneficiary of the hype is on record saying the hype's central claim 704 00:42:16,320 --> 00:42:19,342 is false. And he said it while selling more shovels. 705 00:42:19,342 --> 00:42:23,485 That's not a slip. That's just like the entire logic naked. 706 00:42:23,485 --> 00:42:28,188 You can profit enormously from a revolution you don't even believe has come, 707 00:42:28,188 --> 00:42:32,712 as long as everyone else is too afraid to miss out to stop and measure. 708 00:42:32,712 --> 00:42:33,833 So two things. 709 00:42:33,833 --> 00:42:35,924 radical over a drink things. One, 710 00:42:35,924 --> 00:42:38,306 if you work somewhere that's making AI decisions, 711 00:42:38,306 --> 00:42:41,538 ask the right questions for measurement. 712 00:42:41,538 --> 00:42:43,790 Not like a vendor deck, the internal number. 713 00:42:43,790 --> 00:42:47,492 What did we measure and how? And then watch what happens. 714 00:42:47,492 --> 00:42:52,496 In a lot of places, a question is going to be the most disruptive thing that 715 00:42:52,496 --> 00:42:56,538 you do because making the Emperor's tailor put something in writing, 716 00:42:56,538 --> 00:42:58,611 the paper trail matters, you know, 717 00:42:58,611 --> 00:43:00,671 later. And two, if you're not if you're hiring, 718 00:43:00,927 --> 00:43:02,128 Or you know someone else who is, 719 00:43:02,128 --> 00:43:05,631 go and hire the twenty two to twenty five year olds getting locked 720 00:43:05,631 --> 00:43:07,252 out of this economy right now. 721 00:43:07,252 --> 00:43:12,226 The ones down fourteen percent on job finding because a machine might someday maybe 722 00:43:12,226 --> 00:43:14,297 be able to do their job adequately. 723 00:43:14,297 --> 00:43:18,581 That's the most ~ underpriced talent in the market. 724 00:43:18,581 --> 00:43:21,363 So everybody's too spooked by a feeling to grab it. 725 00:43:21,363 --> 00:43:25,026 Be the person who measures instead of a person who feels. 726 00:43:25,026 --> 00:43:27,318 If the claims were true, we'd have proof by now we don't. 727 00:43:28,238 --> 00:43:32,230 So stop accepting that vibe and start asking for the number. 728 00:43:32,230 --> 00:43:36,673 This has been the overlap. You can find us at FOF.foundation. 729 00:43:36,673 --> 00:43:40,015 We are available on Apple Podcasts and Spotify, 730 00:43:40,015 --> 00:43:43,657 and you know, wherever you like to get your podcasts. 731 00:43:43,657 --> 00:43:45,488 Will, I hope you're enjoying yourself. 732 00:43:45,488 --> 00:43:49,881 To the rest of you, go ask somebody to pull out a ruler and measure something. 733 00:43:49,881 --> 00:43:53,423 Take care of one another. Bye.