WEBVTT 00:00:00.000 --> 00:00:02.919 Picture a rusted out industrial site in western 00:00:02.919 --> 00:00:07.240 Pennsylvania, a relic, a defunct coal plant that's 00:00:07.240 --> 00:00:10.320 been sitting silent for years. But now you see 00:00:10.320 --> 00:00:12.539 construction crews all over it and they're not 00:00:12.539 --> 00:00:14.320 tearing it down. They're bringing it back. They're 00:00:14.320 --> 00:00:17.320 bringing it back to life, but not for coal. They're 00:00:17.320 --> 00:00:20.219 turning it into this massive gas plant. And here's 00:00:20.219 --> 00:00:21.899 the thing that just stops you in your tracks. 00:00:22.359 --> 00:00:27.030 All that power. Every single megawatt isn't for 00:00:27.030 --> 00:00:29.969 homes. It's not for a town. It's being resurrected 00:00:29.969 --> 00:00:33.130 to feed a single data center campus. Yeah. It 00:00:33.130 --> 00:00:35.149 is such a stark imagery. We think of this AI 00:00:35.149 --> 00:00:39.289 revolution as this clean digital thing that lives 00:00:39.289 --> 00:00:42.310 up in the cloud. Right. Ethereal. Exactly. But 00:00:42.310 --> 00:00:43.829 then you see that plant or you see the pipelines 00:00:43.829 --> 00:00:45.689 going down in Texas and you realize the cloud 00:00:45.689 --> 00:00:48.109 is actually heavy. It's hot and it's hungry. 00:00:48.329 --> 00:00:51.210 And right now it's demanding we dig up the industrial 00:00:51.210 --> 00:00:54.000 pass to fuel our future. Welcome back to the 00:00:54.000 --> 00:00:55.920 Deep Dive. I'm really glad you're here. Today 00:00:55.920 --> 00:00:58.259 we're going to step back and just look at that 00:00:58.259 --> 00:01:02.079 friction. The friction between our digital progress 00:01:02.079 --> 00:01:06.319 and, well, physical reality. We've got a stack 00:01:06.319 --> 00:01:09.340 of sources here that paint a very complex picture. 00:01:09.480 --> 00:01:11.400 It's not just about the code anymore. It's about 00:01:11.400 --> 00:01:13.459 physics. Right. And we're going to navigate this 00:01:13.459 --> 00:01:16.159 in four parts. First, that gas boom. There's 00:01:16.159 --> 00:01:18.299 a massive environmental cost to all this compute 00:01:18.299 --> 00:01:21.920 and the numbers are, frankly, staggering. Then 00:01:21.920 --> 00:01:23.599 we'll shift to the business landscape, which 00:01:23.599 --> 00:01:27.060 is just chaotic right now. We're talking corporate 00:01:27.060 --> 00:01:30.480 espionage, a literal space race. Actual servers 00:01:30.480 --> 00:01:32.519 in orbit. Servers in orbit. We'll get there. 00:01:32.620 --> 00:01:34.939 Then we'll look at the toolkit new software that's 00:01:34.939 --> 00:01:37.680 changing how we work, plus a pretty serious security 00:01:37.680 --> 00:01:41.260 warning. And finally, a big technical breakthrough 00:01:41.260 --> 00:01:44.680 in how robots actually see the world. OK, let's 00:01:44.680 --> 00:01:46.579 unpack this. We started with that plant in Pennsylvania. 00:01:46.680 --> 00:01:48.900 That's just one data point from a much bigger 00:01:48.900 --> 00:01:51.159 report we have here from Global Energy Monitor. 00:01:51.379 --> 00:01:52.980 Yeah, but the core tension here is just simple 00:01:52.980 --> 00:01:55.040 physics. Yeah. AI models are getting exponentially 00:01:55.040 --> 00:01:58.060 larger. And as they get bigger, their appetite 00:01:58.060 --> 00:02:01.260 for power just grows with them. You're talking 00:02:01.260 --> 00:02:04.099 about data centers with thousands of GPUs. These 00:02:04.099 --> 00:02:06.040 aren't like your laptop that goes to sleep. No. 00:02:06.140 --> 00:02:08.599 These things need always -on power. They cannot 00:02:08.599 --> 00:02:11.759 afford any downtime. But hold on a second. I 00:02:11.759 --> 00:02:14.699 look at the sustainability reports from Microsoft, 00:02:15.219 --> 00:02:17.900 Google, Amazon. They all have these really aggressive 00:02:17.900 --> 00:02:21.900 net zero pledges for 2030. If they're pivoting 00:02:21.900 --> 00:02:25.400 to gas, doesn't that just kill those goals? It 00:02:25.400 --> 00:02:28.259 certainly puts them in serious jeopardy. But 00:02:28.259 --> 00:02:29.919 here's the engineering reality they're up against. 00:02:30.520 --> 00:02:32.699 Renewables are intermittent. Right. The sun doesn't 00:02:32.699 --> 00:02:35.080 always shine. The wind doesn't always blow. To 00:02:35.080 --> 00:02:37.840 run a GPU cluster at 100%, you need what's called 00:02:37.840 --> 00:02:40.699 baseload power, consistent flatline energy. And 00:02:40.699 --> 00:02:42.419 batteries. I mean, we always hear that battery 00:02:42.419 --> 00:02:44.060 storage is the answer to that. Batteries are 00:02:44.060 --> 00:02:45.759 great for smoothing out a few hours, but right 00:02:45.759 --> 00:02:49.039 now, utility -scale battery storage, it just 00:02:49.039 --> 00:02:52.159 hasn't scaled enough or gotten cheap enough to 00:02:52.159 --> 00:02:54.319 bridge those long gaps for something this big. 00:02:54.400 --> 00:02:56.860 We're talking gigawatts. So the industry is pivoting. 00:02:57.039 --> 00:02:59.159 They're pivoting hard back to natural gas because 00:02:59.159 --> 00:03:01.599 it provides that instant, reliable baseload. 00:03:01.780 --> 00:03:04.620 The data from Global Energy Monitor shows the 00:03:04.620 --> 00:03:07.259 U .S. is leading what they call a massive gas 00:03:07.259 --> 00:03:10.199 boom, specifically for AI's energy problem. I 00:03:10.199 --> 00:03:11.719 was looking at the emission projections in the 00:03:11.719 --> 00:03:14.719 report. They are difficult to wrap your head 00:03:14.719 --> 00:03:16.960 around. They're huge. If all the proposed U .S. 00:03:16.960 --> 00:03:19.080 projects go forward, we're looking at potential 00:03:19.080 --> 00:03:23.409 emissions of 12 .1 billion tons of CO2. Can we 00:03:23.409 --> 00:03:26.629 put that in context? 12 .1 billion tons. To put 00:03:26.629 --> 00:03:29.650 that in context, that is roughly double the current 00:03:29.650 --> 00:03:32.389 annual emissions of the entire United States. 00:03:32.569 --> 00:03:34.729 Double, just from the infrastructure to support 00:03:34.729 --> 00:03:37.129 this one tech sector. That's the projection if 00:03:37.129 --> 00:03:39.150 the full build -out happens. And if you look 00:03:39.150 --> 00:03:42.490 globally, it could hit 53 .2 billion tons. It's 00:03:42.490 --> 00:03:44.310 a complete reshaping of the energy landscape. 00:03:44.569 --> 00:03:47.069 In the U .S., electricity demand could surge 00:03:47.069 --> 00:03:50.009 60 % by 2050, and the main driver they cite is 00:03:50.009 --> 00:03:52.389 AI. And this isn't happening evenly across the 00:03:52.389 --> 00:03:54.050 country, is it? It seems really concentrated. 00:03:54.250 --> 00:03:57.270 Texas is the epicenter. As of 2025, they had 00:03:57.270 --> 00:04:00.490 57 .9 gigawatts of gas projects underway just 00:04:00.490 --> 00:04:03.349 in Texas. But it's elsewhere, too. Meta is building 00:04:03.349 --> 00:04:07.930 a $1 .5 billion data center in El Paso. And that 00:04:07.930 --> 00:04:11.539 facility, gas -powered. It feels like we're training 00:04:11.539 --> 00:04:14.300 the grid to depend on fossil fuels again at the 00:04:14.300 --> 00:04:16.579 exact moment we were supposed to be weaning ourselves 00:04:16.579 --> 00:04:19.139 off them. It's a pragmatic choice, I guess, but 00:04:19.139 --> 00:04:21.579 a controversial one. They need the juice. There's 00:04:21.579 --> 00:04:24.420 a political angle in here, too. The report mentions 00:04:24.420 --> 00:04:26.620 comments from Donald Trump about this. That's 00:04:26.620 --> 00:04:28.720 right. Trump has promised that companies like 00:04:28.720 --> 00:04:31.740 Microsoft will, and I'm quoting, pay their own 00:04:31.740 --> 00:04:34.540 way for this stuff. So no public funding. Right. 00:04:34.620 --> 00:04:37.279 The idea is that tech giants shouldn't rely on 00:04:37.279 --> 00:04:39.939 rate payers to build out these huge power draws. 00:04:40.389 --> 00:04:42.730 But the report does note that the actual details 00:04:42.730 --> 00:04:46.050 on how you'd enforce that are pretty scarce right 00:04:46.050 --> 00:04:48.329 now. So we're trying to invent the future AI 00:04:48.329 --> 00:04:51.350 autonomous agents. But to get there, we are doubling 00:04:51.350 --> 00:04:53.610 down on energy sources from the 20th century. 00:04:53.769 --> 00:04:55.649 We're burning the past to build the future. Yeah. 00:04:55.709 --> 00:04:58.850 So I have to ask, does the utility of AI justify 00:04:58.850 --> 00:05:02.370 doubling our carbon output? Immediate gains versus 00:05:02.370 --> 00:05:05.370 long -term climate debt. Long -term climate debt. 00:05:05.529 --> 00:05:08.589 That is a heavy bill. It really complicates the 00:05:08.589 --> 00:05:13.209 story of AI as this clean future. But speaking 00:05:13.209 --> 00:05:16.009 of complications, the business side is just as 00:05:16.009 --> 00:05:18.230 messy. Let's look at the corporate battlefield. 00:05:18.550 --> 00:05:20.670 It is absolutely chaotic. You're seeing this 00:05:20.670 --> 00:05:24.990 mix of incredible profit, fierce rivalry, and 00:05:24.990 --> 00:05:28.769 some like straight up spy novel stuff. Okay, 00:05:28.790 --> 00:05:31.610 let's start there. The spy novel stuff. This 00:05:31.610 --> 00:05:34.850 story about the Google engineer. Yeah, this is 00:05:34.850 --> 00:05:37.569 just wild. A former engineer at Google was convicted 00:05:37.569 --> 00:05:40.850 for stealing over 2 ,000 files, AI trade secrets. 00:05:41.350 --> 00:05:43.290 But it wasn't just, you know, code snippets. 00:05:43.290 --> 00:05:45.990 He was stealing model weights, architecture diagrams. 00:05:46.290 --> 00:05:48.250 And he was pitching investors while he was still 00:05:48.250 --> 00:05:50.490 there. While he was still employed. He was downloading 00:05:50.490 --> 00:05:52.649 files, pitching investors in China to build a 00:05:52.649 --> 00:05:55.370 rival startup, and then he tried to flee. It 00:05:55.370 --> 00:05:57.470 just highlights how valuable this IP is right 00:05:57.470 --> 00:05:59.759 now. Governments are treating these model weights 00:05:59.759 --> 00:06:02.100 like state secrets. Because if you have the weights, 00:06:02.240 --> 00:06:04.079 you don't need to spend the $100 million to train 00:06:04.079 --> 00:06:05.939 the model. You can just run it. Exactly. You 00:06:05.939 --> 00:06:08.160 skip the whole R &D cost and go right to deployment. 00:06:08.379 --> 00:06:10.819 And speaking of money, we saw Microsoft's earnings. 00:06:10.879 --> 00:06:13.839 The numbers are just eye -watering. Microsoft 00:06:13.839 --> 00:06:17.439 pulled in $7 .6 billion in profit from OpenAI 00:06:17.439 --> 00:06:19.500 in a single quarter. A single quarter. Yeah. 00:06:19.579 --> 00:06:22.730 But here's the stat that really got me. Forty 00:06:22.730 --> 00:06:24.589 five percent of their six hundred twenty five 00:06:24.589 --> 00:06:27.490 billion in total deals came just from that partnership. 00:06:27.750 --> 00:06:29.889 OK, that explains why they're so intertwined. 00:06:29.930 --> 00:06:32.050 This isn't a side project for them. It's a massive 00:06:32.050 --> 00:06:34.389 pillar of their whole valuation. Right. And the 00:06:34.389 --> 00:06:36.949 market is shifting all over. We see QAI raising 00:06:36.949 --> 00:06:39.129 one point five billion and now they're teaming 00:06:39.129 --> 00:06:41.730 up with Apple. That's a huge strategic move. 00:06:41.870 --> 00:06:44.569 Yeah. But while some are rising, others are pivoting. 00:06:45.259 --> 00:06:47.839 OpenAI is actually retiring models next month, 00:06:47.879 --> 00:06:51.019 including GPT -4 .0. I saw that, and the reason 00:06:51.019 --> 00:06:53.439 was that barely anyone uses it. Which is funny 00:06:53.439 --> 00:06:55.959 because I know for a fact some power users absolutely 00:06:55.959 --> 00:06:59.139 loved that specific iteration. But in this field, 00:06:59.240 --> 00:07:02.240 if a model isn't hitting critical mass, the maintenance 00:07:02.240 --> 00:07:05.459 cost, the inference cost, it's just too high. 00:07:05.540 --> 00:07:07.439 They cut it loose. Okay, so we talked about the 00:07:07.439 --> 00:07:09.759 energy limits on Earth. Maybe that explains this 00:07:09.759 --> 00:07:11.560 next story, which feels like it's pulled from 00:07:11.560 --> 00:07:14.199 science fiction. China announced plans to launch 00:07:14.199 --> 00:07:16.639 AI data centers into space. This is where it 00:07:16.639 --> 00:07:18.500 gets really interesting. China's laid out a roadmap 00:07:18.500 --> 00:07:22.240 to have this going by 2030. It's being framed 00:07:22.240 --> 00:07:26.060 as a direct competition, Beijing versus Elon 00:07:26.060 --> 00:07:30.279 Musk's SpaceX. The goal is to process AI data 00:07:30.279 --> 00:07:32.800 off Earth. I'm assuming the logic is cooling, 00:07:32.939 --> 00:07:35.060 right? I mean, space is cold. You solve the heat 00:07:35.060 --> 00:07:37.459 problem instantly. Well, that's actually a common 00:07:37.459 --> 00:07:40.360 misconception. Is it? Yeah. I mean, space is 00:07:40.360 --> 00:07:43.319 cold, sure, but it's a vacuum. On Earth, we cool 00:07:43.319 --> 00:07:45.379 things with air or water. That's convection. 00:07:45.759 --> 00:07:48.480 The air carries heat away. In a vacuum, there's 00:07:48.480 --> 00:07:50.879 no air. So there's nowhere for the heat to go. 00:07:52.079 --> 00:07:53.819 Thermodynamically, getting rid of heat in space 00:07:53.819 --> 00:07:55.980 is an absolute nightmare. So you can't just open 00:07:55.980 --> 00:07:58.019 a window. Exactly. You would literally roast 00:07:58.019 --> 00:07:59.939 your servers. You have to use these massive radiators 00:07:59.939 --> 00:08:02.819 to radiate the heat away as infrared light. So 00:08:02.819 --> 00:08:04.560 the engineering challenge isn't just launching 00:08:04.560 --> 00:08:06.680 them. It's keeping them from melting down without 00:08:06.680 --> 00:08:10.199 an atmosphere. Whoa. Imagine scaling that. managing 00:08:10.199 --> 00:08:12.699 a billion queries in orbit while trying to radiate 00:08:12.699 --> 00:08:15.000 heat into the void. It's the ultimate high ground. 00:08:15.500 --> 00:08:18.240 But the physics are brutal. Still, they pull 00:08:18.240 --> 00:08:20.660 it off. They bypass all the land and energy constraints 00:08:20.660 --> 00:08:22.660 we just talked about on Earth. So looking at 00:08:22.660 --> 00:08:25.180 this whole dynamic, is the rush for dominance 00:08:25.180 --> 00:08:29.139 compromising security? Speed and scale are outpacing 00:08:29.139 --> 00:08:31.180 safety protocols. All right, let's bring it back 00:08:31.180 --> 00:08:33.620 down to Earth, to our laptops. The newsletter 00:08:33.620 --> 00:08:36.399 highlighted a toolkit section, and I see two 00:08:36.399 --> 00:08:39.629 categories here. tools for bureaucracy, and tools 00:08:39.629 --> 00:08:42.830 that manipulate perception. But first, that security 00:08:42.830 --> 00:08:46.389 warning. Claudebot. Right, Claudebot. So this 00:08:46.389 --> 00:08:49.070 is an open source AI agent. It looks super powerful 00:08:49.070 --> 00:08:51.710 because it can execute code on your local machine 00:08:51.710 --> 00:08:54.529 to solve tasks for you. I have to admit, I still 00:08:54.529 --> 00:08:56.889 wrestle with that urge to try new code like this 00:08:56.889 --> 00:08:59.190 on my main machine. You see the demo, you get 00:08:59.190 --> 00:09:01.190 excited, you just want to install it. I totally 00:09:01.190 --> 00:09:04.269 get it. The curiosity takes over. But the source 00:09:04.269 --> 00:09:07.629 material is very, very clear here. This is experimental 00:09:07.629 --> 00:09:09.690 code. Do not install this on your main laptop. 00:09:09.889 --> 00:09:12.230 If there's a vulnerability or if the agent just 00:09:12.230 --> 00:09:15.269 hallucinates a command, it has access to your 00:09:15.269 --> 00:09:18.450 entire file system. So use a burner laptop, basically. 00:09:18.730 --> 00:09:22.210 Or use the alternative they mentioned, Cloudflare's 00:09:22.210 --> 00:09:25.029 Moldworker. It's pretty much the same capability, 00:09:25.230 --> 00:09:27.830 but it's all cloud -hosted. It's safer because 00:09:27.830 --> 00:09:29.850 it's sandboxed away from your personal data. 00:09:30.509 --> 00:09:32.929 Okay, let's look at what I'm calling the bureaucracy 00:09:32.929 --> 00:09:36.210 killers, Pandata and StoryCV. These are fascinating 00:09:36.210 --> 00:09:38.710 because they signal the end of the junior analyst 00:09:38.710 --> 00:09:43.809 grind. Pandata takes messy input like CSVs, PDFs, 00:09:43.809 --> 00:09:46.230 even photos of whiteboards, and it turns them 00:09:46.230 --> 00:09:50.029 into these McKinsey -level reports. It structures 00:09:50.029 --> 00:09:53.340 the data, writes the narrative for you. And StoryCV 00:09:53.340 --> 00:09:55.059 does the same thing for resumes. Right. It just 00:09:55.059 --> 00:09:57.179 interviews you. You talk to it about your work 00:09:57.179 --> 00:10:00.159 history, and it writes the resume. It removes 00:10:00.159 --> 00:10:02.519 that writer's block. But think about the implication. 00:10:02.940 --> 00:10:05.539 We're automating the entry -level tasks that 00:10:05.539 --> 00:10:07.320 used to teach people how to think strategically. 00:10:07.639 --> 00:10:10.279 That's a really deep point. If the AI does the 00:10:10.279 --> 00:10:12.519 grunt work, how do juniors ever learn the nuance? 00:10:13.389 --> 00:10:15.769 OK, then there's the second category, the prompting 00:10:15.769 --> 00:10:18.330 company. This sounds like reverse SEO. That is 00:10:18.330 --> 00:10:20.610 the perfect way to describe it. In the old days, 00:10:20.669 --> 00:10:22.210 you'd optimize your website so Google search 00:10:22.210 --> 00:10:24.769 would find you. The prompting company finds what 00:10:24.769 --> 00:10:27.389 questions people are asking chatbots, chat GPT, 00:10:27.429 --> 00:10:31.250 Claude, and it optimizes your content. So those 00:10:31.250 --> 00:10:35.190 AIs will cite your product as the answer. That 00:10:35.190 --> 00:10:38.639 is incredibly clever. And also. A little dystopian. 00:10:38.659 --> 00:10:41.120 We're optimizing our content so machines will 00:10:41.120 --> 00:10:43.419 read it and then recommend us to humans. It's 00:10:43.419 --> 00:10:45.600 the new marketing frontier. If the training data 00:10:45.600 --> 00:10:47.879 doesn't know you exist, you're invisible. And 00:10:47.879 --> 00:10:50.200 then highlight GPT. Simple browser extension. 00:10:50.299 --> 00:10:52.320 You just highlight text, get an explanation. 00:10:52.759 --> 00:10:55.820 It just reduces the friction of learning. So 00:10:55.820 --> 00:10:58.539 looking at all these tools together, are we optimizing 00:10:58.539 --> 00:11:01.279 for algorithms or humans? We're tailoring our 00:11:01.279 --> 00:11:04.019 reality for machine consumption. Okay. For our 00:11:04.019 --> 00:11:06.360 final segment, we're going to look at how robots 00:11:06.360 --> 00:11:09.340 see the world. Or maybe how they struggle to 00:11:09.340 --> 00:11:11.179 see it. Yeah, this is about a breakthrough from 00:11:11.179 --> 00:11:14.120 China's ant group called Lingbot Depth. The problem 00:11:14.120 --> 00:11:16.460 they identify is broken depth. Can you explain 00:11:16.460 --> 00:11:18.720 what that is? Sure. When a robot looks at the 00:11:18.720 --> 00:11:22.360 world with, say, a LiDAR or a camera, the data 00:11:22.360 --> 00:11:24.700 isn't perfect. You have surfaces like glass or 00:11:24.700 --> 00:11:27.120 shiny metal or just bad lighting, and they create 00:11:27.120 --> 00:11:29.399 these holes in the data. So the robot literally 00:11:29.399 --> 00:11:31.759 thinks there's a hole in the table. Or it just 00:11:31.759 --> 00:11:33.460 can't tell how far away the cup is. Right. It 00:11:33.460 --> 00:11:36.789 makes them clumsy. Lingbot Depth is a model designed 00:11:36.789 --> 00:11:39.909 to fix this. It takes that broken depth map and 00:11:39.909 --> 00:11:42.690 the color image, and it basically reconstructs 00:11:42.690 --> 00:11:45.230 the missing 3D information. It fills in the blanks. 00:11:45.330 --> 00:11:47.870 But the way they trained it is what's so brilliant. 00:11:47.929 --> 00:11:50.970 They call it masked training. How does that work? 00:11:51.230 --> 00:11:53.389 Okay, think of it like learning to read a sentence 00:11:53.389 --> 00:11:55.710 where someone has taken a black marker and crossed 00:11:55.710 --> 00:11:58.409 out every third word. Okay. At first, it's just 00:11:58.409 --> 00:12:01.799 nonsense. But if you practice enough... You start 00:12:01.799 --> 00:12:04.320 to understand the context. You know that after 00:12:04.320 --> 00:12:07.039 the cat sat on the word, the next word is probably 00:12:07.039 --> 00:12:09.639 Matt. You don't even need to see the word to 00:12:09.639 --> 00:12:12.639 know it's there. So they blinded the AI on purpose 00:12:12.639 --> 00:12:15.200 during training. Exactly. They forced it to learn 00:12:15.200 --> 00:12:18.340 the context of the physical world. So now when 00:12:18.340 --> 00:12:21.320 it sees broken data in the real world, it has 00:12:21.320 --> 00:12:23.700 the intuition to fill it in accurately. And the 00:12:23.700 --> 00:12:26.960 result is a robot that can handle like transparent 00:12:26.960 --> 00:12:30.480 objects and messy environments. It creates a 00:12:30.480 --> 00:12:34.440 really tight 2D to 3D understanding. And this 00:12:34.440 --> 00:12:38.480 is crucial for what they call embodied AI, getting 00:12:38.480 --> 00:12:40.740 robots out of the lab and into the real world. 00:12:40.879 --> 00:12:43.340 So the question here, does this bridge the gap 00:12:43.340 --> 00:12:46.259 between software and the physical world? Yes, 00:12:46.259 --> 00:12:48.659 it gives machines intuition for physical space. 00:12:49.080 --> 00:12:50.600 We're going to take a quick beat. We'll be right 00:12:50.600 --> 00:12:53.799 back. And we're back. So let's try to tie all 00:12:53.799 --> 00:12:56.059 this together. We started with a coal plant being 00:12:56.059 --> 00:12:58.480 resurrected as a gas plant. We went to space 00:12:58.480 --> 00:13:01.120 with servers that need massive radiators. We 00:13:01.120 --> 00:13:03.700 looked at tools that write our resumes and robots 00:13:03.700 --> 00:13:05.960 that learn to see by reading between the lines. 00:13:06.200 --> 00:13:08.759 There is such a clear through line here. We are 00:13:08.759 --> 00:13:11.539 witnessing a massive physical build out just 00:13:11.539 --> 00:13:14.340 to support a digital revolution. We love to focus 00:13:14.340 --> 00:13:16.279 on the software, the magic of the chat bot, right? 00:13:16.440 --> 00:13:19.740 But the reality is steel and concrete and gas 00:13:19.740 --> 00:13:22.039 and thermodynamics. It feels like a moment of 00:13:22.039 --> 00:13:24.370 friction. We want the magic of Lingbot. through 00:13:24.370 --> 00:13:26.730 broken pixels. We want the convenience of Pandata. 00:13:26.909 --> 00:13:30.230 But the cost is being paid in old world fossil 00:13:30.230 --> 00:13:32.990 fuels and really intense geopolitical rivalry. 00:13:33.330 --> 00:13:35.970 Exactly. We're building tools that are effectively 00:13:35.970 --> 00:13:39.230 giving machines intuition and creativity. But 00:13:39.230 --> 00:13:42.059 to power them? We're reverting to these very 00:13:42.059 --> 00:13:45.519 traditional, heavy industrial methods. The digital 00:13:45.519 --> 00:13:48.480 future has a very, very heavy physical footprint. 00:13:48.679 --> 00:13:50.279 Before we go, I want to leave you with a thought 00:13:50.279 --> 00:13:52.320 on that prompting company tool. It really stuck 00:13:52.320 --> 00:13:56.019 with me. If we start creating content primarily 00:13:56.019 --> 00:13:59.220 to please AI chatbots, so they'll recommend us, 00:13:59.399 --> 00:14:03.080 are we entering an era where humans no longer 00:14:03.080 --> 00:14:06.000 write for humans, but for the training data of 00:14:06.000 --> 00:14:08.450 the future? Are we just feeding the machine that 00:14:08.450 --> 00:14:10.710 feeds us? That is a thought that will keep me 00:14:10.710 --> 00:14:13.029 up tonight. Something to mull over. If you enjoyed 00:14:13.029 --> 00:14:14.370 this deep dive, make sure you're subscribed. 00:14:14.450 --> 00:14:16.690 We've got some great topics coming up. Thanks 00:14:16.690 --> 00:14:18.090 for listening, everyone. See you in the next 00:14:18.090 --> 00:14:18.529 deep dive.