WEBVTT 00:00:00.000 --> 00:00:05.179 We are building an $850 billion future. But the 00:00:05.179 --> 00:00:09.660 very tools powering it might be quietly erasing 00:00:09.660 --> 00:00:11.900 our ability to think. Yeah, that's the ultimate 00:00:11.900 --> 00:00:14.679 paradox of leverage, really. Welcome to today's 00:00:14.679 --> 00:00:19.050 deep dive to sex silence. You know, we often 00:00:19.050 --> 00:00:21.230 think of technology as something outside of us, 00:00:21.309 --> 00:00:23.929 like a tool in our hand. But today we're looking 00:00:23.929 --> 00:00:26.230 at what happens when the tool starts rewiring 00:00:26.230 --> 00:00:28.789 the hand that holds it. Exactly. So our mission 00:00:28.789 --> 00:00:31.789 today is to explore three structural shifts happening 00:00:31.789 --> 00:00:34.070 right now. And they are all deeply connected. 00:00:34.210 --> 00:00:36.829 Right. First, we're unraveling a tangled corporate 00:00:36.829 --> 00:00:39.920 web. OpenAI is maneuvering toward an immense 00:00:39.920 --> 00:00:42.479 IPO, and the financials are raising fundamental 00:00:42.479 --> 00:00:45.399 questions about incentives. Yeah. Second, we're 00:00:45.399 --> 00:00:47.640 tracking the physical and digital reality of 00:00:47.640 --> 00:00:51.359 AI's current speed, sprinting robots, a quiet 00:00:51.359 --> 00:00:53.939 escalating war over memory chips. A hardware 00:00:53.939 --> 00:00:56.899 layer. Right. And finally, we'll dissect a fascinating 00:00:56.899 --> 00:01:00.140 new joint study from MIT. It measures the hidden 00:01:00.140 --> 00:01:02.280 cognitive tax of using our favorite productivity 00:01:02.280 --> 00:01:04.459 tools. Let's start at the macro level, you know, 00:01:04.459 --> 00:01:06.900 the flow of capital. Right. We have to start 00:01:06.900 --> 00:01:12.099 with a staggering number, $852 billion. Beat. 00:01:12.260 --> 00:01:15.299 That is the reported internal valuation for OpenAI 00:01:15.299 --> 00:01:18.540 right now. Wow. Yeah. They are preparing for 00:01:18.540 --> 00:01:20.439 an initial public offering in the fourth quarter 00:01:20.439 --> 00:01:24.120 of 2026. It's a valuation that dwarfs most legacy 00:01:24.120 --> 00:01:27.390 automakers and banks combined. Yeah. But getting 00:01:27.390 --> 00:01:29.829 to that IPO requires navigating what insiders 00:01:29.829 --> 00:01:32.450 actually call the Sam problem. The Sam problem. 00:01:32.629 --> 00:01:34.510 Right. The tension between corporate treasury 00:01:34.510 --> 00:01:37.069 and personal ambition is reaching an expensive 00:01:37.069 --> 00:01:39.689 inflection point. He occupies a bizarre space 00:01:39.689 --> 00:01:42.890 in the tech ecosystem. He directs this near trillion 00:01:42.890 --> 00:01:46.549 dollar entity. Yet he famously holds zero official 00:01:46.549 --> 00:01:48.849 equity in the company itself. Which is highly 00:01:48.849 --> 00:01:50.870 unusual for a founder. I mean, he frequently 00:01:50.870 --> 00:01:54.430 leans on this fact. He frames himself as a disinterested 00:01:54.430 --> 00:01:57.379 director. Someone whose motivations are purely 00:01:57.379 --> 00:01:59.620 tethered to the overarching mission of building 00:01:59.620 --> 00:02:02.459 artificial general intelligence. But recent financial 00:02:02.459 --> 00:02:04.680 disclosures paint a much more complicated reality. 00:02:04.920 --> 00:02:07.079 The line between his personal venture portfolio 00:02:07.079 --> 00:02:10.060 and OpenAI's corporate needs is heavily blurred. 00:02:10.259 --> 00:02:13.120 We are looking at a situation where the company's 00:02:13.120 --> 00:02:16.020 survival depends on resources that he personally 00:02:16.020 --> 00:02:18.599 backs. The underlying mechanism here is compute. 00:02:19.599 --> 00:02:24.280 OpenAI is locked in a brutal scaling war. To 00:02:24.280 --> 00:02:26.400 build smarter models, they need exponentially 00:02:26.400 --> 00:02:29.139 more computing power. Right. So they're planning 00:02:29.139 --> 00:02:31.900 an infrastructure project called Stargate. It's 00:02:31.900 --> 00:02:34.740 a proposed data center that will require gigawatts 00:02:34.740 --> 00:02:37.500 of continuous reliable power. You can't just 00:02:37.500 --> 00:02:39.520 plug something like that into the existing municipal 00:02:39.520 --> 00:02:42.740 grid. It would brown out entire cities. Exactly. 00:02:42.740 --> 00:02:45.439 So they need an independent, colossal energy 00:02:45.439 --> 00:02:47.439 source. And that leads us to a company called 00:02:47.439 --> 00:02:51.530 Helium. Helion is a nuclear fusion startup. They 00:02:51.530 --> 00:02:53.669 are trying to build the holy grail of clean, 00:02:53.909 --> 00:02:56.530 limitless energy. Here's the twist. Yeah. He 00:02:56.530 --> 00:02:59.310 is the largest individual shareholder in Helion. 00:02:59.370 --> 00:03:01.629 He is. And according to recent reports, he heavily 00:03:01.629 --> 00:03:05.610 pushed OpenAI to make a $500 million direct investment 00:03:05.610 --> 00:03:08.469 into Helion. Wow. Yeah. He wanted the AI company 00:03:08.469 --> 00:03:11.289 to fund the fusion company directly. That's where 00:03:11.289 --> 00:03:14.250 the internal friction ignited. Open AI employees 00:03:14.250 --> 00:03:16.430 reportedly blocked that direct capital injection. 00:03:16.610 --> 00:03:19.569 Right. They saw the optics. Yeah. It looked remarkably 00:03:19.569 --> 00:03:22.710 like a massive transfer of corporate wealth directly 00:03:22.710 --> 00:03:26.189 into the CEO's personal asset column. The internal 00:03:26.189 --> 00:03:28.409 governance mechanisms actually worked in this 00:03:28.409 --> 00:03:30.270 instance. They flagged the conflict of interest. 00:03:30.389 --> 00:03:34.479 But the energy problem didn't vanish. OpenAI 00:03:34.479 --> 00:03:37.979 still desperately needs that future power. And 00:03:37.979 --> 00:03:40.439 Helion still needs the capital to build the reactors. 00:03:40.680 --> 00:03:43.259 It's like the mayor of a city using public funds 00:03:43.259 --> 00:03:46.620 to build a dam. And it just happens to perfectly 00:03:46.620 --> 00:03:49.280 irrigate his own private farm downriver. That 00:03:49.280 --> 00:03:51.759 is a great way to look at it. But he might argue 00:03:51.759 --> 00:03:54.060 something completely different. Like what? Well, 00:03:54.099 --> 00:03:56.199 he might say that without that dam, the entire 00:03:56.199 --> 00:03:58.900 city dies of thirst anyway. The tension isn't 00:03:58.900 --> 00:04:01.500 just about whether he profits. Right. It's whether 00:04:01.500 --> 00:04:04.439 we care more about enriching his farm or ensuring 00:04:04.439 --> 00:04:06.879 the city's survival. So they had to find a structural 00:04:06.879 --> 00:04:09.360 compromise, a way to fund the farm without writing 00:04:09.360 --> 00:04:11.680 a direct check. They landed on a power purchase 00:04:11.680 --> 00:04:15.219 agreement. Yes. OpenAI officially agreed to buy 00:04:15.219 --> 00:04:19.560 12 .5 % of Helion's future energy output. Let's 00:04:19.560 --> 00:04:21.019 make sure we're totally clear on the mechanics 00:04:21.019 --> 00:04:23.959 here. What exactly is a power purchase agreement? 00:04:24.410 --> 00:04:27.310 A long -term contract to buy electricity before 00:04:27.310 --> 00:04:30.870 the power plant exists. So by signing that contract, 00:04:31.230 --> 00:04:34.300 OpenAI guarantees a buyer for the product. That 00:04:34.300 --> 00:04:36.459 solves the energy bottleneck for the Stargate 00:04:36.459 --> 00:04:38.500 data center. Right. And it dramatically lowers 00:04:38.500 --> 00:04:41.720 the risk for helium. Exactly. When a near trillion 00:04:41.720 --> 00:04:44.620 dollar giant promises to buy your experimental 00:04:44.620 --> 00:04:48.279 energy, your valuation skyrockets. It does. That 00:04:48.279 --> 00:04:50.399 commercial validation is potentially worth billions 00:04:50.399 --> 00:04:53.600 to his personal net worth. Even without a direct 00:04:53.600 --> 00:04:56.100 500 million dollar investment, the corporate 00:04:56.100 --> 00:04:58.459 leverage heavily serves the personal portfolio. 00:04:58.939 --> 00:05:01.339 It's a masterclass in indirect value capture. 00:05:02.079 --> 00:05:05.160 I mean, he doesn't need equity in OpenAI if OpenAI's 00:05:05.160 --> 00:05:07.660 operational needs guarantee the success of his 00:05:07.660 --> 00:05:10.060 outside investments. So a quick question on that. 00:05:10.139 --> 00:05:12.720 Does this energy deal fundamentally blur the 00:05:12.720 --> 00:05:15.339 line between a software company and a personal 00:05:15.339 --> 00:05:18.980 hard tech slush fund? It practically erases the 00:05:18.980 --> 00:05:21.449 line. When a company's infrastructure requirements 00:05:21.449 --> 00:05:24.110 are so vast that they dictate national energy 00:05:24.110 --> 00:05:27.029 policy, well, the CEO's personal investments 00:05:27.029 --> 00:05:30.189 in that infrastructure become inextricably linked 00:05:30.189 --> 00:05:32.889 to the company's core mission. So it solves their 00:05:32.889 --> 00:05:35.769 energy crisis, but heavily enriches his personal 00:05:35.769 --> 00:05:38.790 portfolio. Yes. The corporate treasury isn't 00:05:38.790 --> 00:05:41.089 funding him directly, but the corporate demand 00:05:41.089 --> 00:05:44.279 is. You know, you can't... Pour gigawatts of 00:05:44.279 --> 00:05:46.779 fusion energy into a data center without an endpoint. 00:05:47.060 --> 00:05:49.959 That power is flowing directly into the physical 00:05:49.959 --> 00:05:53.060 and digital architecture of tomorrow. Let's look 00:05:53.060 --> 00:05:55.279 at what that power is actually buying us today. 00:05:55.540 --> 00:05:57.800 The translation of energy into capability is 00:05:57.800 --> 00:05:59.620 happening at an alarming rate. We're seeing it 00:05:59.620 --> 00:06:03.439 in the physical world first. Beat. Whoa. Imagine 00:06:03.439 --> 00:06:05.759 a humanoid robot running a half marathon faster 00:06:05.759 --> 00:06:08.279 than any human. It's a surreal visual. It really 00:06:08.279 --> 00:06:11.060 is. An all robot race in China just recorded 00:06:11.060 --> 00:06:13.920 that exact milestone. We're talking about bipedal 00:06:13.920 --> 00:06:16.660 machines maintaining a sub five minute mile pace 00:06:16.660 --> 00:06:19.680 over distance. Just 12 months ago, these same 00:06:19.680 --> 00:06:21.620 prototypes were literally tripping over their 00:06:21.620 --> 00:06:23.600 own feet. Yeah, they couldn't navigate a curb, 00:06:23.660 --> 00:06:26.399 let alone a race course. Yeah. We are witnessing 00:06:26.399 --> 00:06:29.360 the collapse of more of X paradox. Wheels X paradox. 00:06:29.560 --> 00:06:32.439 Yeah. For decades, it was incredibly easy to 00:06:32.439 --> 00:06:34.920 make a computer play Grandmaster Chess, but it 00:06:34.920 --> 00:06:36.959 was nearly impossible to make a robot walk up 00:06:36.959 --> 00:06:39.279 a flight of stairs. Reasoning was cheap, but 00:06:39.279 --> 00:06:41.720 physical mobility was terrible expensive. Exactly. 00:06:41.920 --> 00:06:45.279 Now, the models have internalized physical physics. 00:06:45.480 --> 00:06:47.740 They aren't just calculating, they are moving. 00:06:48.240 --> 00:06:50.800 But as fast as the hardware is running, the software 00:06:50.800 --> 00:06:53.680 layer is compounding even faster. Look at the 00:06:53.680 --> 00:06:57.100 coding ecosystem. Coding is the ultimate bottleneck 00:06:57.100 --> 00:07:00.519 for technological expansion. If you can automate 00:07:00.519 --> 00:07:02.980 the creation of software, you automate the creation 00:07:02.980 --> 00:07:05.519 of everything else. Right. Like a company called 00:07:05.519 --> 00:07:09.240 Cursor just raised $2 billion. Their valuation 00:07:09.240 --> 00:07:11.779 is now sitting at an astonishing $50 billion. 00:07:12.439 --> 00:07:15.639 Cursor is an AI -native code editor. It doesn't 00:07:15.639 --> 00:07:18.110 just autocomplete your typing. It anticipates 00:07:18.110 --> 00:07:21.750 entire architectural structures. Wow. Yeah. Investors 00:07:21.750 --> 00:07:23.870 are pouring billions into it because they realize 00:07:23.870 --> 00:07:26.910 human keystrokes are the slowest part of software 00:07:26.910 --> 00:07:29.709 development. And OpenAI is pivoting aggressively 00:07:29.709 --> 00:07:33.170 to capture this exact space. They just announced 00:07:33.170 --> 00:07:37.379 a major internal reorganization. Kevin Weil is 00:07:37.379 --> 00:07:39.480 leaving the company. He was a massive product 00:07:39.480 --> 00:07:42.060 leader. He essentially shaped the modern Instagram 00:07:42.060 --> 00:07:44.920 experience. His departure signals a distinct 00:07:44.920 --> 00:07:48.480 philosophical shift. OpenAI is moving away from 00:07:48.480 --> 00:07:51.540 consumer -facing glossy products. Right. They 00:07:51.540 --> 00:07:54.259 are doubling down on hardcore developer platforms. 00:07:54.560 --> 00:07:57.439 Which explains their other major move. They are 00:07:57.439 --> 00:08:00.180 folding an internal project called Prism directly 00:08:00.180 --> 00:08:02.810 into Codex. For those tracking the architecture, 00:08:03.069 --> 00:08:05.870 Codex is an AI system that translates plain English 00:08:05.870 --> 00:08:08.589 instructions into functional software code. By 00:08:08.589 --> 00:08:11.529 merging Prism into Codex, they aren't just shuffling 00:08:11.529 --> 00:08:13.389 engineering teams. What are they doing? They 00:08:13.389 --> 00:08:15.230 are betting that coding isn't merely a feature 00:08:15.230 --> 00:08:17.649 of artificial intelligence. It is the foundational 00:08:17.649 --> 00:08:20.589 language that will allow AI agents to eventually 00:08:20.589 --> 00:08:24.509 build and deploy other AI agents. But software 00:08:24.509 --> 00:08:26.550 is essentially just organized thought. It still 00:08:26.550 --> 00:08:28.629 requires a physical brain to process it. And 00:08:28.629 --> 00:08:30.910 that brings us to the silicon layer. Right, the 00:08:30.910 --> 00:08:34.019 hardware. NVIDIA has held an absolute monopoly 00:08:34.019 --> 00:08:36.759 on the chips required to train these models. 00:08:37.039 --> 00:08:40.200 The geopolitical stakes around silicon cannot 00:08:40.200 --> 00:08:44.559 be overstated. NVIDIA's CEO recently delivered 00:08:44.559 --> 00:08:47.220 a very stark public warning. Oh yeah. He stated 00:08:47.220 --> 00:08:49.220 clearly that it would be a horrible outcome for 00:08:49.220 --> 00:08:52.759 America's tech ecosystem if China secures dominance 00:08:52.759 --> 00:08:54.740 in the next generation of chip manufacturing. 00:08:55.080 --> 00:08:57.879 The war isn't just international, though. The 00:08:57.879 --> 00:08:59.580 domestic battle is heating up significantly. 00:09:00.360 --> 00:09:02.620 Google is reportedly partnering with Marvell 00:09:02.620 --> 00:09:05.980 to manufacture two entirely new classes of AI 00:09:05.980 --> 00:09:09.379 chips. They are directly targeting Nvidia's Moat. 00:09:09.870 --> 00:09:11.870 and they're doing it by addressing a very specific 00:09:11.870 --> 00:09:14.529 engineering bottleneck. What's that? Well, one 00:09:14.529 --> 00:09:16.610 of these new chips includes a dedicated memory 00:09:16.610 --> 00:09:19.250 processing unit attached directly to their TPUs. 00:09:19.330 --> 00:09:21.590 Let's clarify the terminology for everyone listening. 00:09:21.769 --> 00:09:24.269 What exactly is a TPU? Custom computer chips 00:09:24.269 --> 00:09:26.269 designed specifically to accelerate artificial 00:09:26.269 --> 00:09:28.789 intelligence calculations. Got it. So why is 00:09:28.789 --> 00:09:31.230 a memory processing unit so critical here? Because 00:09:31.230 --> 00:09:33.690 the math isn't the problem anymore. Moving the 00:09:33.690 --> 00:09:36.970 data is the problem. Right now, it takes more 00:09:36.970 --> 00:09:39.519 electrical energy... to move the data from the 00:09:39.519 --> 00:09:42.220 memory bank into the processor than it does to 00:09:42.220 --> 00:09:45.080 actually perform the calculation. It's a massive 00:09:45.080 --> 00:09:47.539 traffic jam. So Google is building the memory 00:09:47.539 --> 00:09:50.480 directly into the processor to eliminate the 00:09:50.480 --> 00:09:52.940 commute. Exactly. It fundamentally changes the 00:09:52.940 --> 00:09:55.720 energy efficiency of training a model. The hardware 00:09:55.720 --> 00:09:58.379 is accelerating. The software is absorbing billions 00:09:58.379 --> 00:10:01.580 in capital. Robots are physically sprinting past 00:10:01.580 --> 00:10:05.200 us. Beat. But then we look at the human workforce. 00:10:05.980 --> 00:10:08.720 The actual economic reality feels incredibly 00:10:08.720 --> 00:10:11.379 disconnected from this momentum. OpenAI's chief 00:10:11.379 --> 00:10:13.779 economist recently mapped out the trajectory 00:10:13.779 --> 00:10:17.899 for over 900 distinct job categories. The data 00:10:17.899 --> 00:10:20.720 is sobering. They calculate that 18 % of all 00:10:20.720 --> 00:10:23.820 jobs face a severe near -term risk of total AI 00:10:23.820 --> 00:10:26.600 automation. And another 24 % of jobs will likely 00:10:26.600 --> 00:10:29.500 shrink in scope. They won't disappear, but the 00:10:29.500 --> 00:10:31.720 human footprint required to do them will contract 00:10:31.720 --> 00:10:34.860 significantly. And only 12 % of map jobs are 00:10:34.860 --> 00:10:37.299 projected to actually grow. Right. We are looking 00:10:37.299 --> 00:10:39.460 at a structural hollowing out of the traditional 00:10:39.460 --> 00:10:42.740 labor market. Yet we are faced with a profound 00:10:42.740 --> 00:10:46.379 contradiction here. With all this immense compounding 00:10:46.379 --> 00:10:49.799 power, thousands of enterprise CEOs are reporting 00:10:49.799 --> 00:10:51.919 a totally different reality on the ground. They 00:10:51.919 --> 00:10:53.940 are buying the enterprise licenses. They are 00:10:53.940 --> 00:10:56.679 deploying the tools. But they report that AI 00:10:56.679 --> 00:10:59.139 hasn't meaningfully improved their actual corporate 00:10:59.139 --> 00:11:02.169 productivity. It feels exactly like the technology 00:11:02.169 --> 00:11:05.330 paradox we saw in the late 1980s. Back then, 00:11:05.529 --> 00:11:08.210 corporations put a personal computer on every 00:11:08.210 --> 00:11:11.350 single desk in America. The processing power 00:11:11.350 --> 00:11:14.110 skyrocketed. But national economic productivity 00:11:14.110 --> 00:11:17.049 didn't budge for almost a decade. It's like giving 00:11:17.049 --> 00:11:19.909 a 19th century accounting firm a modern Excel 00:11:19.909 --> 00:11:22.490 spreadsheet, but they still insist on printing 00:11:22.490 --> 00:11:25.149 out every single digital cell and mailing it 00:11:25.149 --> 00:11:27.730 to the boss for a physical signature. That is 00:11:27.730 --> 00:11:30.309 so true. The bottleneck isn't the software anymore. 00:11:30.490 --> 00:11:32.970 It's the corporate muscle memory. So let me ask 00:11:32.970 --> 00:11:35.330 you this. If the tech is evolving this fast, 00:11:35.570 --> 00:11:39.090 why are CEOs stuck in a 1980s -style productivity 00:11:39.090 --> 00:11:42.370 paradox? Because technology scales exponentially. 00:11:42.789 --> 00:11:46.929 But human habits adapt linearly. Buying the software 00:11:46.929 --> 00:11:50.129 license is incredibly easy. But rewiring deeply 00:11:50.129 --> 00:11:52.889 embedded human behavior and dismantling legacy 00:11:52.889 --> 00:11:56.090 workflows takes years of painful cultural friction. 00:11:56.960 --> 00:11:59.120 Right. We have the powerful tools, but we haven't 00:11:59.120 --> 00:12:01.460 reinvented the daily workflows yet. Exactly. 00:12:01.620 --> 00:12:03.960 We're trying to use quantum tools to speed up 00:12:03.960 --> 00:12:06.659 industrial era processes. It doesn't work. That 00:12:06.659 --> 00:12:09.399 paradox perfectly tees up our final exploration 00:12:09.399 --> 00:12:12.480 today. We have these incredibly powerful instant 00:12:12.480 --> 00:12:14.539 answer machines. We know they are disrupting 00:12:14.539 --> 00:12:16.919 corporate structures, but we rarely ask what 00:12:16.919 --> 00:12:19.139 they are actually doing to our internal architecture, 00:12:19.480 --> 00:12:21.820 our brains. This is arguably the most critical 00:12:21.820 --> 00:12:24.720 question we face right now. A massive new joint 00:12:24.720 --> 00:12:28.009 study from MIT. Oxford, and Carnegie Mellon just 00:12:28.009 --> 00:12:30.769 released some deeply unsettling data. The researchers 00:12:30.769 --> 00:12:34.210 gathered 1 ,222 participants. They assigned them 00:12:34.210 --> 00:12:36.309 a series of complex mathematical and reading 00:12:36.309 --> 00:12:38.950 comprehension tasks. We all intimately know the 00:12:38.950 --> 00:12:41.330 magic of typing a complex problem into a prompt 00:12:41.330 --> 00:12:43.789 box and getting a brilliant answer in three seconds. 00:12:43.950 --> 00:12:46.480 It feels like cognitive super strength. It really 00:12:46.480 --> 00:12:48.659 does. But this paper proves that our favorite 00:12:48.659 --> 00:12:51.299 productivity hack carries the hidden biological 00:12:51.299 --> 00:12:54.820 tax. The setup was brilliant. They gave the users 00:12:54.820 --> 00:12:58.120 a highly advanced GPT -5 -based cognitive assistant. 00:12:58.340 --> 00:13:01.179 And initially, the results were incredible. Yeah. 00:13:01.240 --> 00:13:03.759 The users performed much faster. They showed 00:13:03.759 --> 00:13:06.179 significantly higher accuracy on their early 00:13:06.179 --> 00:13:08.600 problems. They looked like absolute geniuses. 00:13:08.600 --> 00:13:10.379 For the first 10 minutes, it seemed like the 00:13:10.379 --> 00:13:14.559 ultimate proof of human -machine symbiosis. But 00:13:14.559 --> 00:13:16.580 the researchers were measuring something else. 00:13:16.740 --> 00:13:18.960 They identified a phenomenon they call answer 00:13:18.960 --> 00:13:22.120 outsourcing. Answer outsourcing. Right. It acts 00:13:22.120 --> 00:13:25.200 as a cognitive crutch. After just 10 minutes 00:13:25.200 --> 00:13:28.200 of heavy AI assistance, the researchers suddenly 00:13:28.200 --> 00:13:31.080 removed the GPT -5 tool. And the participants' 00:13:31.200 --> 00:13:33.509 performance plummeted instantly. It wasn't a 00:13:33.509 --> 00:13:36.409 slow decline. It was a catastrophic drop in capability. 00:13:36.730 --> 00:13:39.009 Those exact same users who were flying through 00:13:39.009 --> 00:13:41.710 the test moments before suddenly solved fewer 00:13:41.710 --> 00:13:44.250 problems. They stalled out constantly. And they 00:13:44.250 --> 00:13:45.990 actually quit trying sooner than the control 00:13:45.990 --> 00:13:48.990 group. Yeah. The people who were working entirely 00:13:48.990 --> 00:13:53.110 alone, with no AI help from the very start, exhibited 00:13:53.110 --> 00:13:55.509 far more resilience. It demonstrates something 00:13:55.509 --> 00:13:57.929 profound about the mechanics of learning. AI 00:13:57.929 --> 00:14:01.230 removes the friction from our workflow. But it 00:14:01.230 --> 00:14:04.789 quietly erases our mental persistence. Persistence 00:14:04.789 --> 00:14:07.269 isn't just an attitude. It's a biological adaptation. 00:14:07.570 --> 00:14:10.009 When you struggle with a problem, your neurons 00:14:10.009 --> 00:14:13.259 build myelin. That insulation allows signals 00:14:13.259 --> 00:14:16.139 to travel faster next time. Friction is the actual 00:14:16.139 --> 00:14:18.320 fuel for building intelligence. I have to be 00:14:18.320 --> 00:14:21.759 honest. I still wrestle with prompt drift myself. 00:14:22.200 --> 00:14:24.320 It's incredibly common. For those who haven't 00:14:24.320 --> 00:14:26.700 experienced it, prompt drift is gradually losing 00:14:26.700 --> 00:14:28.740 your original train of thought while constantly 00:14:28.740 --> 00:14:32.299 tweaking AI inputs. Yeah. I find myself reaching 00:14:32.299 --> 00:14:34.940 for the AI before I've even let the actual problem 00:14:34.940 --> 00:14:37.620 settle in my own mind. I outsource the framing 00:14:37.620 --> 00:14:39.700 of the question, not just the answer. We all 00:14:39.700 --> 00:14:42.000 do it. The dopamine loop is deeply tempting. 00:14:42.240 --> 00:14:45.139 But by removing the agonizing friction of a hard 00:14:45.139 --> 00:14:48.019 problem, the AI stops the crucial mental training. 00:14:48.139 --> 00:14:51.639 You are outsourcing your own cognitive load -bearing 00:14:51.639 --> 00:14:54.620 capacity. That training is exactly what builds 00:14:54.620 --> 00:14:57.879 the habit of pushing through confusion. We need 00:14:57.879 --> 00:15:00.500 that confusion. The MIT study suggests we are 00:15:00.500 --> 00:15:02.779 undergoing a major shift in human capability. 00:15:03.320 --> 00:15:06.179 We are rapidly building a world of orchestrators. 00:15:06.519 --> 00:15:08.679 Orchestrators are people who know how to manage 00:15:08.679 --> 00:15:11.960 complex AI agents. They supervise systems like 00:15:11.960 --> 00:15:14.519 codecs. They direct the traffic. But in the rush 00:15:14.519 --> 00:15:17.519 to become orchestrators, we are losing the fundamental 00:15:17.519 --> 00:15:20.759 ability to execute. Right. We are literally forgetting 00:15:20.759 --> 00:15:23.139 how to do the actual granular work ourselves. 00:15:23.559 --> 00:15:27.259 So is shifting from an executor to an orchestrator 00:15:27.259 --> 00:15:30.480 really a bad thing for the future of work? It 00:15:30.480 --> 00:15:32.720 becomes catastrophic the moment the system produces 00:15:32.720 --> 00:15:36.000 an error. Orchestration requires a deep fundamental 00:15:36.000 --> 00:15:38.899 understanding of the underlying mechanics. If 00:15:38.899 --> 00:15:41.700 you lack execution skills, you cannot evaluate 00:15:41.700 --> 00:15:44.120 the quality or safety of what the AI just produced. 00:15:44.419 --> 00:15:46.980 You become entirely dependent on a black box 00:15:46.980 --> 00:15:48.980 you don't understand. Yeah, if you can't execute, 00:15:49.100 --> 00:15:50.740 you won't know if the AI's orchestration actually 00:15:50.740 --> 00:15:53.019 works. Exactly. You become a passenger holding 00:15:53.019 --> 00:15:55.139 a plastic steering wheel. But the researchers 00:15:55.139 --> 00:15:57.860 didn't just diagnose the problem. They actually 00:15:57.860 --> 00:16:01.200 found a very clear behavioral solution hidden 00:16:01.200 --> 00:16:03.960 in the data. Right. The cognitive damage was 00:16:03.960 --> 00:16:07.340 significantly mitigated for a specific subgroup 00:16:07.340 --> 00:16:09.940 of users. These were the participants who intentionally 00:16:09.940 --> 00:16:13.559 used the AI as a hint system rather than a direct 00:16:13.559 --> 00:16:16.600 answer machine. Yes. They employed what the researchers 00:16:16.600 --> 00:16:19.840 call the hint rule. If you want to maintain your 00:16:19.840 --> 00:16:22.120 neuroplasticity and your mental edge, you have 00:16:22.120 --> 00:16:24.720 to refuse the direct answer. Mid -roll sponsor 00:16:24.720 --> 00:16:28.500 read goes here. So if we pull all these threads 00:16:28.500 --> 00:16:31.460 together, a very clear, continuous pattern emerges 00:16:31.460 --> 00:16:33.580 today. We started by looking at the absolute 00:16:33.580 --> 00:16:36.679 macro scale of our economy. The $852 billion 00:16:36.679 --> 00:16:39.799 balancing act. The blurring of corporate treasury 00:16:39.799 --> 00:16:42.320 and personal ambitions secure gigawatts of nuclear 00:16:42.320 --> 00:16:45.240 fusion. Right. And we trace that immense energy 00:16:45.240 --> 00:16:47.919 demand down into the physical reality. The breakneck 00:16:47.919 --> 00:16:50.639 speed of humanoid robots breaking marathon records. 00:16:50.840 --> 00:16:53.519 The geopolitical and corporate wars over memory 00:16:53.519 --> 00:16:56.120 processing units. And then we drilled all the 00:16:56.120 --> 00:16:59.259 way down to the microscopic scale. Our own individual 00:16:59.259 --> 00:17:02.649 cognition. The way a three -second answer quietly 00:17:02.649 --> 00:17:05.450 erodes our biological capacity for mental persistence. 00:17:05.910 --> 00:17:08.529 What stands out to you is the connective tissue 00:17:08.529 --> 00:17:10.789 here. It's the concept of invisible taxation. 00:17:11.549 --> 00:17:14.130 Incredible technological leverage always comes 00:17:14.130 --> 00:17:16.589 with a hidden cost. We're just usually looking 00:17:16.589 --> 00:17:18.769 in the wrong place to see the bill. Well said. 00:17:18.930 --> 00:17:21.170 Whether it's a megacorporation blurring ethical 00:17:21.170 --> 00:17:23.630 boundaries to secure energy or an individual 00:17:23.630 --> 00:17:26.069 trading their own mental resilience for a faster 00:17:26.069 --> 00:17:29.029 email draft, the leverage is intensely real. 00:17:30.059 --> 00:17:32.579 But the tax is always paid eventually. We just 00:17:32.579 --> 00:17:34.420 have to decide if we are consciously willing 00:17:34.420 --> 00:17:37.140 to pay it. Awareness is the only defense. So 00:17:37.140 --> 00:17:39.460 what does this all mean for you? Today, we want 00:17:39.460 --> 00:17:41.619 to challenge you to actively practice the hint 00:17:41.619 --> 00:17:44.180 rule in your own life. The next time you face 00:17:44.180 --> 00:17:47.259 a difficult, frustrating problem at work, don't 00:17:47.259 --> 00:17:50.240 ask the AI to generate the final output. Ask 00:17:50.240 --> 00:17:53.069 it for a structural hint. Right. Force yourself 00:17:53.069 --> 00:17:55.789 to sit with the agonizing friction for at least 00:17:55.789 --> 00:17:58.069 five minutes before you type a single prompt. 00:17:58.230 --> 00:18:01.509 Build that persistence muscle back up. Maintain 00:18:01.509 --> 00:18:04.789 your execution skills. It might be the only way 00:18:04.789 --> 00:18:07.769 to truly keep your edge in a world of orchestrators. 00:18:08.359 --> 00:18:10.619 Thank you for taking the time to slowly, carefully 00:18:10.619 --> 00:18:12.740 learn with us today. It's been a great exploration. 00:18:13.059 --> 00:18:14.759 But before we go, I want to leave you with one 00:18:14.759 --> 00:18:17.839 final thought to mull over. We talked heavily 00:18:17.839 --> 00:18:21.039 today about losing our ability to execute, about 00:18:21.039 --> 00:18:23.759 the danger of just becoming orchestrators. But 00:18:23.759 --> 00:18:26.299 if AI development continues at this exponential 00:18:26.299 --> 00:18:28.960 curve, it will eventually automate both the execution 00:18:28.960 --> 00:18:31.559 and the orchestration. If machines eventually 00:18:31.559 --> 00:18:34.099 map the problems and solve them. Right. What 00:18:34.099 --> 00:18:36.740 happens to the innate human drive to overcome 00:18:36.740 --> 00:18:40.230 friction? T -Sex silence. Does the feeling of 00:18:40.230 --> 00:18:42.369 frustration eventually become a luxury good? 00:18:42.509 --> 00:18:44.150 Think about it. Out to your own music.