WEBVTT 00:00:00.000 --> 00:00:01.980 For years, we've been talking about AI living 00:00:01.980 --> 00:00:05.280 on our screens. You know, AI that generates text 00:00:05.280 --> 00:00:09.759 or creates images. But that whole era, it's fading. 00:00:10.179 --> 00:00:12.880 Fast. It really is. The game has totally changed. 00:00:13.060 --> 00:00:14.759 We're moving way beyond the sci -fi trailers 00:00:14.759 --> 00:00:17.339 now. AI has a body. It's seeing, it's moving, 00:00:17.399 --> 00:00:19.539 and it's reacting in the real world. And the 00:00:19.539 --> 00:00:22.339 stakes are just fundamentally different. A mistake 00:00:22.339 --> 00:00:26.039 on a screen is a typo. Maybe a bad recipe. But 00:00:26.039 --> 00:00:29.100 when physical AI makes a mistake in a factory, 00:00:29.469 --> 00:00:33.009 in a car it's immediate it's physical and it 00:00:33.009 --> 00:00:35.469 is incredibly costly it changes reality welcome 00:00:35.469 --> 00:00:38.170 to the deep dive today we are unpacking what 00:00:38.170 --> 00:00:40.070 our sources are calling the most important tech 00:00:40.070 --> 00:00:42.530 convergence happening right now the rise of physical 00:00:42.530 --> 00:00:45.640 ai and here's our argument The screen era of 00:00:45.640 --> 00:00:48.119 AI is, for all intents and purposes, over. Yeah, 00:00:48.179 --> 00:00:50.240 and this isn't some far -off 10 -year prediction. 00:00:50.380 --> 00:00:53.299 This is 2026. Physical AI is already running 00:00:53.299 --> 00:00:55.399 huge logistics operations. It's streamlining 00:00:55.399 --> 00:00:57.380 supply chains. It's controlling safety systems 00:00:57.380 --> 00:00:59.539 in our cars. Most people still think this is 00:00:59.539 --> 00:01:02.359 a decade away. Exactly. So our mission today 00:01:02.359 --> 00:01:04.340 is to really break down the four big shifts, 00:01:04.459 --> 00:01:06.599 the tech and the economic forces that are driving 00:01:06.599 --> 00:01:09.079 this massive change. We're moving from machines 00:01:09.079 --> 00:01:11.319 that just react to partners that can actually 00:01:11.319 --> 00:01:13.340 predict. So that means we're going to dive into 00:01:13.340 --> 00:01:15.560 predictive. math, this idea of collaborative 00:01:15.560 --> 00:01:18.480 robot learning, why hyper -specialized AI is 00:01:18.480 --> 00:01:21.819 winning, and this whole new data economy where 00:01:21.819 --> 00:01:24.760 robot performance itself becomes a tradable asset. 00:01:25.060 --> 00:01:28.359 Okay, let's get into it. First up, what exactly 00:01:28.359 --> 00:01:31.359 is physical AI and what's the economic engine 00:01:31.359 --> 00:01:33.780 driving this whole thing? So let's define our 00:01:33.780 --> 00:01:37.239 terms. What is physical AI? At its core, it's 00:01:37.239 --> 00:01:39.159 intelligence that's embedded right into a machine 00:01:39.159 --> 00:01:41.120 that has to deal with the real, messy, unpredictable 00:01:41.120 --> 00:01:43.859 world. We're talking systems with cameras, with 00:01:43.859 --> 00:01:46.340 sensors, with actuators that let them see and 00:01:46.340 --> 00:01:49.239 move and apply force. It's the absolute difference 00:01:49.239 --> 00:01:51.840 between thinking and doing. A digital AI, like 00:01:51.840 --> 00:01:53.920 a large language model, lives safely on a server. 00:01:54.019 --> 00:01:56.799 It processes text. Physical AI has to deal with 00:01:56.799 --> 00:01:59.560 motion, friction, physics. It has to survive 00:01:59.560 --> 00:02:02.359 in the real world. And that reality is the key 00:02:02.359 --> 00:02:05.200 differentiator. If a digital model hallucinates 00:02:05.200 --> 00:02:08.139 a fact, It's an annoyance. But if a self -driving 00:02:08.139 --> 00:02:10.780 car, a perfect example of physical AI, hallucinates 00:02:10.780 --> 00:02:13.680 a barrier that isn't there or worse, doesn't 00:02:13.680 --> 00:02:16.219 see one that is, the consequences are immediate 00:02:16.219 --> 00:02:19.400 and severe. It means that reliability and safety 00:02:19.400 --> 00:02:22.419 are just infinitely more important than creativity. 00:02:22.719 --> 00:02:24.639 And what's so fascinating is that this isn't 00:02:24.639 --> 00:02:27.360 just a robotics trend. It's really a chip war, 00:02:27.520 --> 00:02:31.539 a race to build the fastest, most reliable physical 00:02:31.539 --> 00:02:34.039 brain. Right. You see players like NVIDIA, ARM, 00:02:34.219 --> 00:02:36.689 all the big automakers. They're all battling 00:02:36.689 --> 00:02:38.669 to build the compute architecture for these things. 00:02:38.750 --> 00:02:40.770 And the money involved just confirms how high 00:02:40.770 --> 00:02:42.990 the stakes are. Oh, absolutely. Our sources are 00:02:42.990 --> 00:02:44.949 projecting that the automotive chip market alone, 00:02:45.050 --> 00:02:46.889 which is basically the prototype brain for all 00:02:46.889 --> 00:02:50.150 physical AI, is going to hit $123 billion by 00:02:50.150 --> 00:02:53.669 2032. That is an 85 % jump in less than a decade 00:02:53.669 --> 00:02:55.949 just for the brain. And that kind of growth, 00:02:55.990 --> 00:02:58.389 that level of investment, it points to one thing 00:02:58.389 --> 00:03:01.289 that is absolutely non -negotiable. Massive on 00:03:01.289 --> 00:03:03.629 -device compute. The thinking has to happen inside 00:03:03.629 --> 00:03:06.590 the machine. In real time, no waiting for the 00:03:06.590 --> 00:03:09.770 cloud. Think about it. For physical AI, latency 00:03:09.770 --> 00:03:13.770 is death. If a robot welding a car frame or a 00:03:13.770 --> 00:03:16.129 car on the highway has to send data to the cloud, 00:03:16.310 --> 00:03:19.650 wait for a decision and get it back, it's over. 00:03:19.750 --> 00:03:21.969 The moment has passed. The calculation has to 00:03:21.969 --> 00:03:24.830 be local. Instant. So with all this money and 00:03:24.830 --> 00:03:27.110 technical demand, what's the actual bottleneck 00:03:27.110 --> 00:03:29.449 right now? Are we still just waiting on better 00:03:29.449 --> 00:03:32.699 hardware? You know... Surprisingly, no. Hardware 00:03:32.699 --> 00:03:34.599 is still a challenge, but the real bottleneck 00:03:34.599 --> 00:03:37.120 has shifted. It's the software on the chip, that 00:03:37.120 --> 00:03:40.080 predictive math, and the power to run it in milliseconds. 00:03:40.240 --> 00:03:42.340 Okay, that's a perfect transition to our first 00:03:42.340 --> 00:03:45.259 major prediction. The next big leap forward is 00:03:45.259 --> 00:03:47.379 coming from math, not just from new hardware. 00:03:47.580 --> 00:03:50.439 That feels... A little counterintuitive. We always 00:03:50.439 --> 00:03:52.319 think we need more power. I know. It is a weird 00:03:52.319 --> 00:03:54.460 shift to think about. Most people expect, you 00:03:54.460 --> 00:03:56.960 know, better motors, stronger arms, faster sensors. 00:03:57.080 --> 00:03:59.620 But the real upgrade is in the math underneath 00:03:59.620 --> 00:04:02.039 it all. Current industrial robots, they're reactive. 00:04:02.219 --> 00:04:04.599 They just follow a very rigid script. And if 00:04:04.599 --> 00:04:06.599 anything unexpected happens, like a part is in 00:04:06.599 --> 00:04:08.599 the wrong place, they just stop. They enter a 00:04:08.599 --> 00:04:10.780 failure state and wait for a person to come fix 00:04:10.780 --> 00:04:14.259 it, which is so inefficient. But tomorrow's systems, 00:04:14.520 --> 00:04:16.699 the ones running this new predictive math, they 00:04:16.699 --> 00:04:19.160 work completely differently. They actually model 00:04:19.160 --> 00:04:22.079 the consequences before they move. They're simulating 00:04:22.079 --> 00:04:24.319 possibilities instead of just reacting to what 00:04:24.319 --> 00:04:26.439 the sensors tell them. So we're talking about 00:04:26.439 --> 00:04:29.220 concepts like dual numbers and jets. That sounds 00:04:29.220 --> 00:04:31.220 like some pretty heavy math. Can you break down 00:04:31.220 --> 00:04:34.240 what that actually lets a robot do? What's amazing 00:04:34.240 --> 00:04:36.379 here is that we're basically giving the robot 00:04:36.379 --> 00:04:40.459 calculus tools that let it see the future. think 00:04:40.459 --> 00:04:42.519 of it like a chess master seeing 10 moves ahead 00:04:42.519 --> 00:04:45.500 the robot is doing that with physics in real 00:04:45.500 --> 00:04:48.399 time it's calculating the future friction the 00:04:48.399 --> 00:04:51.180 inertia all of it before it even commits to a 00:04:51.180 --> 00:04:54.000 move so a robotic arm doesn't just grab a part 00:04:54.000 --> 00:04:56.740 it first simulates five different ways to grab 00:04:56.740 --> 00:04:59.870 it forecasts if a tiny change in angle will cause 00:04:59.870 --> 00:05:02.269 a problem later on, and then picks the best path. 00:05:02.569 --> 00:05:04.629 And it does all that in a fraction of a second. 00:05:04.769 --> 00:05:07.089 Yeah, this idea of modeling consequences is a 00:05:07.089 --> 00:05:09.470 huge, huge leap. And I'll be honest, I still 00:05:09.470 --> 00:05:11.310 wrestle with the subtle implications of state 00:05:11.310 --> 00:05:13.470 drift myself. That's a really important concept. 00:05:13.670 --> 00:05:16.610 What exactly is state drift here? Why is that 00:05:16.610 --> 00:05:19.029 predictive layer so necessary? It's when the 00:05:19.029 --> 00:05:21.790 world changes really slowly, so slowly the AI 00:05:21.790 --> 00:05:24.430 doesn't see it as a sudden error. Maybe the robot 00:05:24.430 --> 00:05:26.790 arm heats up over a long shift or the floor gets 00:05:26.790 --> 00:05:29.389 a little dusty. Its performance just degrades 00:05:29.389 --> 00:05:32.389 little by little until it fails. Predictive math 00:05:32.389 --> 00:05:34.970 can catch that slow slide before it becomes a 00:05:34.970 --> 00:05:37.709 catastrophe. So the result is this adaptive control 00:05:37.709 --> 00:05:41.550 that just feels intuitive. It leads to faster 00:05:41.550 --> 00:05:43.850 work, fewer mistakes. We're finally building 00:05:43.850 --> 00:05:46.089 machines that are smarter, not just stronger. 00:05:46.600 --> 00:05:49.660 And to work, it has to be incredibly fast. The 00:05:49.660 --> 00:05:52.139 machine has to anticipate the next 500 milliseconds 00:05:52.139 --> 00:05:55.180 of reality on device with zero help from the 00:05:55.180 --> 00:05:57.759 cloud. That need for speed brings us to prediction 00:05:57.759 --> 00:06:01.500 number two, the death of the solo robot. For 00:06:01.500 --> 00:06:04.139 decades, automation meant one robot in a cage 00:06:04.139 --> 00:06:06.660 following one program. And that era is over. 00:06:06.800 --> 00:06:09.079 It's just too slow, too inefficient, too rigid. 00:06:09.319 --> 00:06:11.620 The next wave is all about imitation learning. 00:06:12.060 --> 00:06:14.699 about cooperation so robots will literally watch 00:06:14.699 --> 00:06:17.839 humans or watch other robots copy what works 00:06:17.839 --> 00:06:19.899 refine it and then share that new skill across 00:06:19.899 --> 00:06:22.480 the whole fleet exactly true peer -to -peer learning 00:06:22.480 --> 00:06:24.560 is finally here we're getting away from that 00:06:24.560 --> 00:06:27.019 old model of programming each robot one by one 00:06:27.019 --> 00:06:29.180 and that changes everything on the factory floor 00:06:29.180 --> 00:06:32.220 especially setup time instead of spending days 00:06:32.220 --> 00:06:34.800 programming every single robot for a new task 00:06:34.800 --> 00:06:37.540 you just show one robot how to do it and the 00:06:37.540 --> 00:06:40.300 rest of the team just gets it instantly. That 00:06:40.300 --> 00:06:42.800 knowledge sharing makes setup five, maybe ten 00:06:42.800 --> 00:06:46.399 times faster. And the adaptability is just it's 00:06:46.399 --> 00:06:49.060 exponential. So if a part shows up on the conveyor 00:06:49.060 --> 00:06:51.740 belt at the wrong angle, the robots don't just 00:06:51.740 --> 00:06:54.620 freeze up. They can adapt together in real time 00:06:54.620 --> 00:06:56.920 because they can share the solution to the new 00:06:56.920 --> 00:07:00.079 problem. Yeah, companies like Universal Robots 00:07:00.079 --> 00:07:02.240 are already doing this with multi arm systems, 00:07:02.540 --> 00:07:05.629 but the price of entry is dropping fast. For 00:07:05.629 --> 00:07:08.189 this team learning to really work, you need the 00:07:08.189 --> 00:07:10.269 communication standards, the safety protocols, 00:07:10.550 --> 00:07:13.009 and the software to orchestrate it all. Those 00:07:13.009 --> 00:07:14.990 pieces are finally catching up to the hardware. 00:07:15.230 --> 00:07:17.310 So it's not really about the arms anymore. It's 00:07:17.310 --> 00:07:19.930 about how they talk to each other safely. Orchestration 00:07:19.930 --> 00:07:22.550 tools, communication standards, and safety rules 00:07:22.550 --> 00:07:25.250 are catching up to enable team learning. Okay, 00:07:25.310 --> 00:07:28.209 let's talk business. Prediction number three 00:07:28.209 --> 00:07:30.230 is a big one for anyone actually paying for this 00:07:30.230 --> 00:07:33.290 stuff. Vertical AI is crushing generic tools. 00:07:34.000 --> 00:07:36.160 For the last couple of years, the hype was all 00:07:36.160 --> 00:07:38.800 about general AI that could do anything. But 00:07:38.800 --> 00:07:41.120 the hard lesson learned in manufacturing in 2026 00:07:41.120 --> 00:07:44.199 is that a tool that tries to do everything usually 00:07:44.199 --> 00:07:46.600 does nothing well. If you're running a business, 00:07:46.720 --> 00:07:49.240 you need a system that does one job, but does 00:07:49.240 --> 00:07:52.660 it perfectly. And that's vertical AI, pre -trained, 00:07:52.660 --> 00:07:55.579 task -specific systems. We're talking AI welding, 00:07:55.759 --> 00:07:58.420 AI finishing, AI assembly. These aren't just 00:07:58.420 --> 00:08:00.279 concepts anymore. You can buy them off the shelf 00:08:00.279 --> 00:08:02.660 right now. This is such a big shift. I remember 00:08:02.660 --> 00:08:04.519 just five years ago, everyone was trying to build 00:08:04.519 --> 00:08:07.040 that one general purpose factory robot, and it 00:08:07.040 --> 00:08:09.759 was terrible at everything. This move to specialization, 00:08:09.920 --> 00:08:12.139 just do welding, just do pick and place. That's 00:08:12.139 --> 00:08:14.300 the industry growing up. Take Siemens' somatic 00:08:14.300 --> 00:08:17.420 robot pick AI. It's a perfect example. It uses 00:08:17.420 --> 00:08:19.899 deep learning tuned for one thing, picking up 00:08:19.899 --> 00:08:22.660 complex objects. It ships ready to go with hardware 00:08:22.660 --> 00:08:25.240 you already have. That eliminates weeks of custom 00:08:25.240 --> 00:08:27.680 programming. Welding is maybe the best use case. 00:08:27.860 --> 00:08:30.319 It requires these tiny constant adjustments. 00:08:30.660 --> 00:08:33.399 A human is great at it because seams can shift, 00:08:33.559 --> 00:08:36.539 temperatures change. Well, now AI vision can 00:08:36.539 --> 00:08:39.159 track the seam and machine learning adjusts the 00:08:39.159 --> 00:08:42.620 heat, the feed rate, all of it on the fly. You 00:08:42.620 --> 00:08:45.139 get perfect quality every time. The bigger impact 00:08:45.139 --> 00:08:48.019 here is that automation is becoming more like 00:08:48.019 --> 00:08:50.899 buying an appliance. It's not just for massive 00:08:50.899 --> 00:08:53.480 companies with huge R &D budgets anymore. This 00:08:53.480 --> 00:08:56.080 is democratizing efficiency for everyone. And 00:08:56.080 --> 00:08:57.419 while factors are leading the way, we have to 00:08:57.419 --> 00:09:00.100 look beyond them. So outside of logistics, which 00:09:00.100 --> 00:09:02.279 industry is going to feel the impact of vertical 00:09:02.279 --> 00:09:04.860 AI the most? I think it's clear that while logistics 00:09:04.860 --> 00:09:07.659 is already using it, retail is absolutely next. 00:09:08.179 --> 00:09:10.879 Stocking shelves, taking inventory, even basic 00:09:10.879 --> 00:09:13.399 customer tasks. Our fourth prediction really 00:09:13.399 --> 00:09:15.399 gets into the new economy being built around 00:09:15.399 --> 00:09:18.919 all this. Robot data is about to become a tradable 00:09:18.919 --> 00:09:21.620 asset. And this solves a massive bottleneck for 00:09:21.620 --> 00:09:24.299 AI development. Robots generate a ton of information. 00:09:24.580 --> 00:09:26.899 Sensor readings, error logs, vision data. It's 00:09:26.899 --> 00:09:29.539 incredibly valuable real -world data. But right 00:09:29.539 --> 00:09:32.179 now, it's all locked away inside each customer's 00:09:32.179 --> 00:09:34.980 facility. Which stalls improvement. AI developers 00:09:34.980 --> 00:09:38.220 need that high -quality real -world data to train 00:09:38.220 --> 00:09:41.240 the next generation of models. Not simulations. 00:09:41.440 --> 00:09:44.120 The best data is just sitting there, completely 00:09:44.120 --> 00:09:47.159 unused. The solution is these new secure data 00:09:47.159 --> 00:09:50.139 exchanges. They're opt -in systems that let companies 00:09:50.139 --> 00:09:52.539 share anonymized performance data safely with 00:09:52.539 --> 00:09:54.639 the developers who can actually use it. So we're 00:09:54.639 --> 00:09:57.139 talking about welding robots sharing de -identified 00:09:57.139 --> 00:10:00.860 data on scene quality. Or assembly robots sharing 00:10:00.860 --> 00:10:03.120 their error logs. This isn't just random noise. 00:10:03.320 --> 00:10:05.899 It's perfectly structured, high -quality fuel 00:10:05.899 --> 00:10:08.100 for training new models. And it creates this 00:10:08.100 --> 00:10:10.700 amazing virtuous cycle. The manufacturer gets 00:10:10.700 --> 00:10:12.600 a new revenue stream. The customer gets better 00:10:12.600 --> 00:10:15.519 AI tools trained on real -world conditions. And 00:10:15.519 --> 00:10:17.700 the developers get the data they need. Whoa. 00:10:18.490 --> 00:10:21.330 Just imagine scaling that. A continuous improvement 00:10:21.330 --> 00:10:23.730 loop across every factory and warehouse on the 00:10:23.730 --> 00:10:25.570 planet. The rate of optimization would just, 00:10:25.610 --> 00:10:27.830 it would become truly exponential. Exactly. It 00:10:27.830 --> 00:10:30.370 flips the script on data scarcity. The core idea 00:10:30.370 --> 00:10:33.710 is so simple, but so powerful. Every single robot 00:10:33.710 --> 00:10:36.190 that gets deployed makes every future robot smarter. 00:10:36.570 --> 00:10:38.429 But of course, the big question is always going 00:10:38.429 --> 00:10:41.169 to be about privacy and trade secrets. Can we 00:10:41.169 --> 00:10:43.669 be sure customer data will be protected in this 00:10:43.669 --> 00:10:46.539 new model? Data is being anonymized, privacy 00:10:46.539 --> 00:10:49.299 preserved, and shared safely with customer permission 00:10:49.299 --> 00:10:52.100 via opt -in systems. So let's put it all together. 00:10:52.200 --> 00:10:54.120 What does it mean when you combine these four 00:10:54.120 --> 00:10:56.779 trends? We have predictive math for anticipation, 00:10:57.200 --> 00:11:00.419 collaborative learning for smarter teams, vertical 00:11:00.419 --> 00:11:03.700 AI for specialization, and a data economy to 00:11:03.700 --> 00:11:06.320 fuel it all. It means convergence. And the automotive 00:11:06.320 --> 00:11:08.500 industry is the biggest driver of that convergence. 00:11:08.700 --> 00:11:11.059 It's colliding head -on with factory automation. 00:11:11.519 --> 00:11:14.669 The powerful tested systems being built for self 00:11:14.669 --> 00:11:17.210 -driving cars are migrating straight to the factory 00:11:17.210 --> 00:11:20.389 floor. The robot gets a pre -tested brain. You 00:11:20.389 --> 00:11:21.909 could see it in all the announcements from shows 00:11:21.909 --> 00:11:25.490 like CES 2026, Ford selling a hands -free system 00:11:25.490 --> 00:11:28.710 by 2028, Mercedes is debuting theirs this year, 00:11:28.769 --> 00:11:31.330 and Nvidia is supplying its architecture to major 00:11:31.330 --> 00:11:34.090 automakers like Geely. And here's why Nvidia 00:11:34.090 --> 00:11:36.250 is winning. They're taking the same architecture 00:11:36.250 --> 00:11:38.929 they used for training huge language models and 00:11:38.929 --> 00:11:40.850 just applying it to real -time physical control. 00:11:41.519 --> 00:11:48.320 Jensen Huang said, And you see the exact same 00:11:48.320 --> 00:11:51.179 thing with humanoids. It's now confirmed that 00:11:51.179 --> 00:11:54.379 humanoid robots from Google DeepMind, from Boston 00:11:54.379 --> 00:11:57.429 Dynamics, from Hyundai, They're hitting factory 00:11:57.429 --> 00:11:59.669 floors for actual production work in the next 00:11:59.669 --> 00:12:02.450 few months, not just demos. And that's only possible 00:12:02.450 --> 00:12:04.830 because these physical AI systems are finally 00:12:04.830 --> 00:12:07.570 robust enough to reason and act in those messy 00:12:07.570 --> 00:12:10.029 real -world spaces. They can handle that state 00:12:10.029 --> 00:12:12.129 drift we talked about. Which is all fueled by 00:12:12.129 --> 00:12:14.549 that explosion in compute. The central brain 00:12:14.549 --> 00:12:16.830 of the car, which is the prototype for the robot 00:12:16.830 --> 00:12:20.289 brain, is becoming, as one expert said, Quantum 00:12:20.289 --> 00:12:22.789 leaps bigger, hundreds of times as big. Look 00:12:22.789 --> 00:12:26.110 like separate trends, cars, robots, chips, all 00:12:26.110 --> 00:12:29.669 one converging system. Okay, let's bring this 00:12:29.669 --> 00:12:32.789 deep dive home. We've laid out the four big forces 00:12:32.789 --> 00:12:35.450 driving this shift to physical AI. You've got 00:12:35.450 --> 00:12:38.250 predictive math for anticipation, cooperative 00:12:38.250 --> 00:12:41.070 learning for smarter teams, specialized vertical 00:12:41.070 --> 00:12:44.490 AI for immediate value, and this new robot data 00:12:44.490 --> 00:12:46.870 economy fueling it all. Yeah, and this is not 00:12:46.870 --> 00:12:49.389 sci -fi in a lab. This is a fundamental restructuring 00:12:49.389 --> 00:12:51.450 of our economy and technology that's happening 00:12:51.450 --> 00:12:54.590 right now in 2026. We're moving away from machines 00:12:54.590 --> 00:12:57.070 that just follow orders toward cooperative partners 00:12:57.070 --> 00:12:59.830 that can learn and reason and even predict what's 00:12:59.830 --> 00:13:01.690 next. We do have to acknowledge the workforce 00:13:01.690 --> 00:13:03.789 implications here, which our sources were clear 00:13:03.789 --> 00:13:06.129 about. We're looking at a global drop of maybe 00:13:06.129 --> 00:13:08.330 10 to 15 percent in low -skill manufacturing 00:13:08.330 --> 00:13:11.230 and routine retail jobs over the next five to 00:13:11.230 --> 00:13:13.840 seven years. That's a huge transition, a massive 00:13:13.840 --> 00:13:17.279 one. But it's balanced by this skyrocketing demand 00:13:17.279 --> 00:13:20.200 for specialized engineering roles. The people 00:13:20.200 --> 00:13:22.639 who build, maintain, and train the vertical AI 00:13:22.639 --> 00:13:25.120 models we were just talking about. It's a huge 00:13:25.120 --> 00:13:28.799 and probably painful shift. It demands completely 00:13:28.799 --> 00:13:31.600 new skills. It definitely is. So here's a final 00:13:31.600 --> 00:13:34.500 thought. If physical AI is truly defined by its 00:13:34.500 --> 00:13:37.419 ability to use predictive math to survive in 00:13:37.419 --> 00:13:41.090 messy, real -world environments, What complex 00:13:41.090 --> 00:13:43.350 variable job that you currently think is safe 00:13:43.350 --> 00:13:45.669 from automation will be the very next one solved 00:13:45.669 --> 00:13:48.649 by a vertically trained predictive machine? Something 00:13:48.649 --> 00:13:50.269 to think about as you start seeing these systems 00:13:50.269 --> 00:13:52.090 pop up all around you. It's difficult thinking 00:13:52.090 --> 00:13:54.129 about how fast those goalposts are moving. We'll 00:13:54.129 --> 00:13:54.570 see you next time.