1 00:00:00,120 --> 00:00:02,319 Speaker 1: Imagine a world, and I'm not talking about, you know, 2 00:00:02,359 --> 00:00:05,440 fifty years from now, I mean maybe five or less 3 00:00:05,559 --> 00:00:08,759 even or less. A world where we are launching more 4 00:00:08,839 --> 00:00:13,960 computing power into space every single year then exists on 5 00:00:14,000 --> 00:00:16,920 the entire planet Earth today. And I don't mean a 6 00:00:16,920 --> 00:00:20,199 few satellites for GPS. I'm not talking about getting Wi 7 00:00:20,239 --> 00:00:21,320 Fi in your flight, right. 8 00:00:21,399 --> 00:00:24,320 Speaker 2: This isn't Starling two point zero. This is something else entirely. 9 00:00:24,519 --> 00:00:32,359 Speaker 1: I'm talking about massive orbiting server farms, gigawatt class supercomputers 10 00:00:32,399 --> 00:00:35,560 just floating up there in the void, and they make 11 00:00:35,840 --> 00:00:38,520 everything we have down here look like a pocket calculator. 12 00:00:38,520 --> 00:00:40,679 Speaker 2: And the reason we're doing it, the reason, yeah. 13 00:00:40,479 --> 00:00:43,079 Speaker 1: Because Earth simply ran out of electricity. 14 00:00:43,159 --> 00:00:45,159 Speaker 2: It really does sound like the opening crawl of a 15 00:00:45,200 --> 00:00:46,439 sci fi movie, doesn't it. 16 00:00:46,439 --> 00:00:49,039 Speaker 1: It does. It sounds like the backstory for like Blade 17 00:00:49,079 --> 00:00:49,600 Runner or. 18 00:00:49,520 --> 00:00:52,000 Speaker 2: Something where a game like Mass Effect, you know, humanity 19 00:00:52,000 --> 00:00:54,560 expanded to the star is driven by an insatiable need 20 00:00:54,600 --> 00:00:57,000 for energy. But the wild thing is when you actually 21 00:00:57,600 --> 00:01:01,520 sit down and look at the physics, the raw ormodynamics. 22 00:01:00,799 --> 00:01:02,799 Speaker 1: Of it all, and the economics. 23 00:01:02,320 --> 00:01:06,040 Speaker 2: And the economics exactly, it stops being science fiction. It's well, 24 00:01:06,079 --> 00:01:09,480 it's a roadmap, a very specific, calculated and honestly maybe 25 00:01:09,560 --> 00:01:11,799 a necessary one if you just follow the trend lines. 26 00:01:12,159 --> 00:01:14,680 Speaker 1: Welcome back to thrilling Threads. I am so hyped to 27 00:01:14,680 --> 00:01:15,319 get into. 28 00:01:15,079 --> 00:01:16,840 Speaker 2: This one too. This one's a mind bendor. 29 00:01:17,439 --> 00:01:20,840 Speaker 1: Today we are pulling on a thread that starts with 30 00:01:21,200 --> 00:01:24,840 a very real, very gritty energy crisis right here on 31 00:01:24,879 --> 00:01:28,599 the ground, and it ends with well ends with the 32 00:01:28,680 --> 00:01:29,719 robots building other. 33 00:01:29,680 --> 00:01:32,680 Speaker 2: Robots and something that gets called the infinite money glitch. 34 00:01:32,840 --> 00:01:35,359 Speaker 1: The infinite money glitch. Yeah, we are definitely getting to that. 35 00:01:35,519 --> 00:01:40,239 Speaker 2: So we're basing this whole discussion on a really fascinating, sprawling, 36 00:01:40,359 --> 00:01:44,159 three hour conversation. It was between Elon Musk, Patrick Collison, 37 00:01:44,239 --> 00:01:47,760 Stripe co founder, right, and the podcaster Dwarkish Patel, And 38 00:01:47,840 --> 00:01:50,239 it wasn't your typical tech interview. It felt more like 39 00:01:50,280 --> 00:01:52,120 an intellectual jam session. 40 00:01:51,920 --> 00:01:56,200 Speaker 1: Totally three people who are you know, deeply deeply embedded 41 00:01:56,239 --> 00:01:59,159 in the engineering and economic reality of where this is 42 00:01:59,200 --> 00:02:02,439 all going. And you know, usually when we cover these 43 00:02:02,480 --> 00:02:04,799 AI topics, we end up talking about the software, right, 44 00:02:04,840 --> 00:02:05,840 the models. 45 00:02:05,760 --> 00:02:08,080 Speaker 2: M hmm, GPT, this Gemini that. 46 00:02:08,240 --> 00:02:12,479 Speaker 1: Yeah, parameters, tokens, context windows, We get stuck the weeds 47 00:02:12,520 --> 00:02:15,400 of is the chatbot funny or can it write a 48 00:02:15,439 --> 00:02:16,080 good email? 49 00:02:16,439 --> 00:02:19,000 Speaker 2: But this conversation was different. It wasn't about the code. 50 00:02:19,039 --> 00:02:21,479 It was about the physics. It was about the metal, 51 00:02:21,599 --> 00:02:25,759 the copper, the electricity, the hard stuff, the hard stuff exactly. 52 00:02:25,840 --> 00:02:28,159 It's about the hard constraints, the things you just can't 53 00:02:28,199 --> 00:02:31,479 code your way out of. You can optimize an algorithm, sure, 54 00:02:31,599 --> 00:02:33,639 you can make it twice as efficient, but you. 55 00:02:33,599 --> 00:02:36,479 Speaker 1: Can't optimize your way past the laws of thermodynamic You can't. 56 00:02:36,960 --> 00:02:40,080 Speaker 2: Eventually you hit the wall of physics. You need electrons 57 00:02:40,120 --> 00:02:42,800 to flip the switches, and you need atoms to build 58 00:02:42,800 --> 00:02:45,960 the switches. And that right now is where the real 59 00:02:46,039 --> 00:02:46,719 bottleneck is. 60 00:02:46,919 --> 00:02:49,319 Speaker 1: Okay, So let's unpack this, this idea of the great 61 00:02:49,400 --> 00:02:52,280 migration of AI to orbit. I want to start there 62 00:02:52,319 --> 00:02:55,400 because it is by far the most visually insane part 63 00:02:55,439 --> 00:02:56,000 of this whole thing. 64 00:02:56,159 --> 00:02:56,719 Speaker 2: It's the hook. 65 00:02:56,919 --> 00:03:00,280 Speaker 1: It's the hook. Musk dropped a timeline that, honestly, it 66 00:03:00,360 --> 00:03:03,199 made me rewind. I thought I misheard him. He said 67 00:03:03,199 --> 00:03:06,680 that within thirty to thirty six months from the time 68 00:03:06,719 --> 00:03:07,919 of that recording. 69 00:03:07,560 --> 00:03:09,319 Speaker 2: So a two and a half to three year window. 70 00:03:09,680 --> 00:03:13,840 Speaker 1: Yeah, space becomes the most economically compelling place to put 71 00:03:13,919 --> 00:03:14,879 new AI. 72 00:03:14,599 --> 00:03:16,639 Speaker 2: Compute, and just to put a pin in that for 73 00:03:16,680 --> 00:03:20,319 everyone listening. That conversation took place a while back. So 74 00:03:20,400 --> 00:03:22,520 we're sitting here in early twenty twenty six. 75 00:03:22,960 --> 00:03:23,879 Speaker 1: We're in the window. 76 00:03:23,960 --> 00:03:26,080 Speaker 2: We are right inside that window he was talking about. 77 00:03:26,080 --> 00:03:29,280 We're supposedly standing on the precipice of that exact tipping 78 00:03:29,319 --> 00:03:30,680 point right now. 79 00:03:30,719 --> 00:03:32,319 Speaker 1: But I have to push back on this. My brain 80 00:03:32,439 --> 00:03:36,039 just rebels against this idea because it feels completely counterintuitive. 81 00:03:36,120 --> 00:03:39,879 How so, well, space is expensive, right, We've spent our 82 00:03:40,039 --> 00:03:42,800 entire lives hearing about the cost per kilogram to get 83 00:03:42,840 --> 00:03:45,560 something into orbit. It used to be tens of thousands 84 00:03:45,599 --> 00:03:46,759 of dollars per kila. 85 00:03:46,599 --> 00:03:48,759 Speaker 2: The price of a small car to launch a milk jug. 86 00:03:49,159 --> 00:03:52,840 Speaker 1: Exactly. Yeah, So why on Earth, and pardon the pun, 87 00:03:53,240 --> 00:03:56,919 would it possibly be cheaper to build a massive complex 88 00:03:57,039 --> 00:03:59,719 data center in a hard vacuum than it would be 89 00:03:59,800 --> 00:04:02,120 in I don't know, an empty field in Nevada. 90 00:04:02,159 --> 00:04:04,439 Speaker 2: It's a great question, and it's the right question. To 91 00:04:04,439 --> 00:04:06,319 get the answer, you have to stop thinking about the 92 00:04:06,400 --> 00:04:09,199 launch cost and start thinking about the operational cost of 93 00:04:09,199 --> 00:04:12,479 a data center over its entire lifetime. If you're running 94 00:04:12,520 --> 00:04:16,079 a massive compute cluster and something truly huge training a 95 00:04:16,120 --> 00:04:18,639 model that's ten times bigger than anything we have today. 96 00:04:19,720 --> 00:04:22,519 What are your two biggest line items on the bill 97 00:04:22,720 --> 00:04:23,480 day after day. 98 00:04:23,800 --> 00:04:26,160 Speaker 1: Well, the hardware itself is number one. I assume the 99 00:04:26,199 --> 00:04:28,199 capital cost of buying all those GPUs. 100 00:04:28,319 --> 00:04:30,879 Speaker 2: Okay, after that, yeah, the operating expenses. 101 00:04:30,920 --> 00:04:32,639 Speaker 1: It's the electricity bill. It's gotta be. 102 00:04:32,759 --> 00:04:35,759 Speaker 2: Electricity is huge. But there's a second one that people 103 00:04:35,800 --> 00:04:39,240 always forget, and it's tied directly to the electricity. It's cooling. 104 00:04:39,439 --> 00:04:41,680 Speaker 1: Right, of course, all that electricity turns. 105 00:04:41,519 --> 00:04:45,360 Speaker 2: Into heat, massive amounts of heat. Data centers are basically 106 00:04:45,480 --> 00:04:49,279 just giant, very inefficient heaters that happen to do math 107 00:04:49,360 --> 00:04:52,600 as a side effect. You pump a megawatt of electricity in, 108 00:04:52,680 --> 00:04:54,680 a megawat of heat comes out. You have to get 109 00:04:54,759 --> 00:04:57,079 rid of that heat or the silicon melts. 110 00:04:56,879 --> 00:05:00,000 Speaker 1: And your multi billion dollar investment turns into a puddle. 111 00:05:00,120 --> 00:05:04,720 Speaker 2: Precisely on Earth. That means giant air conditioning units, huge 112 00:05:04,759 --> 00:05:08,040 water cooling towers, environmental impact studies. 113 00:05:07,680 --> 00:05:11,439 Speaker 1: And a second massive energy bill just to run the coolers. 114 00:05:11,240 --> 00:05:13,800 Speaker 2: To paying a double tax. You pay to power the 115 00:05:13,879 --> 00:05:16,199 chip and then you pay again to cool the chip 116 00:05:16,240 --> 00:05:17,800 down from the power you just fed it. 117 00:05:17,920 --> 00:05:21,319 Speaker 1: Okay, so that's the problem on Earth. How is space better? 118 00:05:21,600 --> 00:05:25,399 Speaker 2: Let's look at the physics of space. First, power on Earth, 119 00:05:25,720 --> 00:05:30,079 solar power is it's good, it's great actually, but it's limited. 120 00:05:30,160 --> 00:05:32,120 The Sun goes down, the Sun goes down half the time. 121 00:05:32,639 --> 00:05:35,920 You also have an atmosphere scattering the light which reduces 122 00:05:35,920 --> 00:05:39,079 the intensity. You have clouds that can drop production to 123 00:05:39,160 --> 00:05:42,040 almost zero. So to run a two hundred and forty 124 00:05:42,079 --> 00:05:44,199 seven data center on solar on Earth, you. 125 00:05:44,199 --> 00:05:47,240 Speaker 1: Need batteries, a mountain of batteries. 126 00:05:46,800 --> 00:05:51,319 Speaker 2: A mountain of very expensive complex batteries to store power 127 00:05:51,360 --> 00:05:55,040 for the night. That basically triples your infrastructure cost and 128 00:05:55,120 --> 00:05:58,000 your points of failure. But in orbit, in orbit, especially 129 00:05:58,000 --> 00:06:00,519 if you pick the right one, like a sun synchronic orbit, 130 00:06:00,560 --> 00:06:03,120 the Sun never sets. It's just always there. Wow, you 131 00:06:03,160 --> 00:06:06,639 get direct unfiltered solar radiation. It's about thirteen hundred and 132 00:06:06,680 --> 00:06:11,199 sixty watts per square meter, constantly pure raw fusion energy 133 00:06:11,240 --> 00:06:14,279 from the Sun, uninterrupted, no batteries needed, So you just. 134 00:06:14,279 --> 00:06:16,240 Speaker 1: Plug the panels directly into the computers. 135 00:06:16,519 --> 00:06:19,360 Speaker 2: That's the idea. So the power is effectively free and 136 00:06:19,399 --> 00:06:22,680 it's constant that's a huge economic wind right there. 137 00:06:22,800 --> 00:06:26,319 Speaker 1: Okay, that's one half of the double tax, the power. 138 00:06:26,480 --> 00:06:29,040 What about the other half, the cooling? You said that 139 00:06:29,120 --> 00:06:31,519 was the other huge cost. And this is where I 140 00:06:31,560 --> 00:06:35,680 get confused. Space is a vacuum. I remember from high 141 00:06:35,680 --> 00:06:39,399 school physics. A vacuum is a thermos. It's an insulator. 142 00:06:40,079 --> 00:06:43,319 It keeps hot things hot because there's no air to 143 00:06:43,360 --> 00:06:44,279 conduct the heat away. 144 00:06:44,639 --> 00:06:47,360 Speaker 2: That's the common misconception, and it's a good one. It's 145 00:06:47,360 --> 00:06:49,040 where the engineering gets really clever. 146 00:06:49,199 --> 00:06:51,160 Speaker 1: So how do you cool a supercomputer if there's no 147 00:06:51,240 --> 00:06:52,480 air to blow over it with a fan. 148 00:06:52,720 --> 00:06:55,600 Speaker 2: You're right, you can't use convection cooling, no fans. But 149 00:06:55,680 --> 00:06:59,680 space is also really really cold. The background temperature if 150 00:06:59,720 --> 00:07:02,560 you're in the shadow is close to absolute zero. So 151 00:07:02,680 --> 00:07:05,160 instead of convection, you use radiative cooling. 152 00:07:05,240 --> 00:07:07,720 Speaker 1: Okay, hold on, break that down for me. Radiative cooling. 153 00:07:07,920 --> 00:07:09,759 That sounds like star trek technobabble. 154 00:07:09,839 --> 00:07:12,399 Speaker 2: Think about how the Sun heats the Earth. There's no 155 00:07:12,480 --> 00:07:14,680 air between us and the Sun, right, it's a vacuum. 156 00:07:15,120 --> 00:07:18,959 The heat gets here through radiation light infrared. Well, the 157 00:07:19,000 --> 00:07:24,360 reverse works too, Any hot object emits infrared radiation, so 158 00:07:24,399 --> 00:07:28,199 you attach these giant radiator panels to your servers, big 159 00:07:28,319 --> 00:07:31,600 flat surfaces. And because the background temperature of space is 160 00:07:31,600 --> 00:07:34,920 so low, the heat literally radiates off the panel as 161 00:07:34,959 --> 00:07:37,199 infrared light straight into the void. 162 00:07:37,439 --> 00:07:40,519 Speaker 1: So the air conditioning isn't a machine. It's it's just 163 00:07:40,839 --> 00:07:42,160 a shape, a big piece. 164 00:07:42,000 --> 00:07:45,439 Speaker 2: Of metal, exactly. It's passive. You don't need compressors that break, 165 00:07:45,519 --> 00:07:48,040 you don't need water pumps that clog, you don't need 166 00:07:48,079 --> 00:07:50,879 free on that leaks. You just need surface area. Once 167 00:07:50,920 --> 00:07:53,800 you build the structure, the cooling is effectively free and 168 00:07:53,959 --> 00:07:55,000 maintenance free forever. 169 00:07:55,160 --> 00:07:58,199 Speaker 1: Wow. Okay, let's tell you this up. Then you've got free, 170 00:07:58,319 --> 00:08:01,920 unlimited constant power from this on check, and you have passive, 171 00:08:02,040 --> 00:08:04,879 maintenance free cooling from the void of space. Check. That 172 00:08:05,000 --> 00:08:08,040 sounds economically incredible. But there was a third factor. Musk 173 00:08:08,120 --> 00:08:10,240 mentioned the Nimby factor. 174 00:08:10,360 --> 00:08:14,720 Speaker 2: Ah. Yes, not in my backyard. This is quietly becoming 175 00:08:14,759 --> 00:08:17,720 the single biggest hurdle for building AI on Earth. 176 00:08:18,360 --> 00:08:20,160 Speaker 1: I can absolutely imagine. I mean, try to get a 177 00:08:20,160 --> 00:08:22,560 permit to build a new gigawatt scale power plant, let 178 00:08:22,600 --> 00:08:24,519 alone a data center next to a nice suburb in 179 00:08:24,600 --> 00:08:25,680 Virginia today. 180 00:08:25,560 --> 00:08:29,279 Speaker 2: It's impossible. The noise from the cooling fans alone. It's 181 00:08:29,319 --> 00:08:33,320 a constant low frequency industrial drone that drives people crazy 182 00:08:33,360 --> 00:08:34,159 for miles, and. 183 00:08:34,120 --> 00:08:35,879 Speaker 1: The power draw must be insane. 184 00:08:36,120 --> 00:08:40,200 Speaker 2: It crashes the local grid. Residents get brownouts. The water 185 00:08:40,320 --> 00:08:44,120 usage for the cooling towers drains local aquifers. So you 186 00:08:44,200 --> 00:08:46,960 get sued, you get protested, you get blocked by local 187 00:08:47,000 --> 00:08:51,320 councils for years. And in space, in space, your backyard 188 00:08:51,360 --> 00:08:55,240 is a geostationary orbit thirty five thousand kilometers up. There's 189 00:08:55,360 --> 00:08:56,960 nobody to complain about the hum. 190 00:08:57,000 --> 00:08:59,759 Speaker 1: Not in my low Earth orbit. We need a new 191 00:08:59,759 --> 00:09:01,799 app acronym nil EO. 192 00:09:02,200 --> 00:09:04,759 Speaker 2: We'll work on the branding. Yeah, but this all leads 193 00:09:04,759 --> 00:09:07,639 to the really mind blowing part. The scale, the sheer 194 00:09:07,679 --> 00:09:10,360 magnitude of what Musk is proposing is what really sets 195 00:09:10,360 --> 00:09:11,799 the context for this whole discussion. 196 00:09:11,879 --> 00:09:14,360 Speaker 1: Yes, let's drill into those numbers. Yeah, because they are 197 00:09:14,519 --> 00:09:17,679 absolutely staggering. They sound made up until you actually stop 198 00:09:17,720 --> 00:09:18,399 and think about it. 199 00:09:18,480 --> 00:09:20,559 Speaker 2: So let's set the stage for you listening. Think about 200 00:09:20,600 --> 00:09:23,720 Earth's current capacity. If you take every single data center 201 00:09:23,759 --> 00:09:26,840 amazon Aws, Google Cloud, Microsoft Azures. 202 00:09:26,600 --> 00:09:30,240 Speaker 1: All the crypto mining rigs in Texas, every university supercomputer, 203 00:09:30,440 --> 00:09:32,519 the servers, running Wall Street everything. 204 00:09:32,600 --> 00:09:35,080 Speaker 2: The sum total of human computation built over the last 205 00:09:35,080 --> 00:09:39,120 forty years, that adds up to roughly fifty to sixty 206 00:09:39,159 --> 00:09:40,720 gigawatts of power consumption. 207 00:09:40,879 --> 00:09:44,399 Speaker 1: Okay, so sixty gigawytes. That's our baseline. That's humanity status quo, 208 00:09:44,720 --> 00:09:47,159 that's the engine of the entire modern world. 209 00:09:47,279 --> 00:09:51,120 Speaker 2: Right. Must prediction was that within five years we would 210 00:09:51,159 --> 00:09:54,919 be launching hundreds of gigawatts of capacity into space. 211 00:09:55,159 --> 00:09:58,879 Speaker 1: See that per year, per year. That breaks my brain 212 00:09:58,919 --> 00:10:01,200 a little bit. That's not great growth. That's not More's law. 213 00:10:01,240 --> 00:10:04,039 That's a phase change. It's like going from water to steam. 214 00:10:04,240 --> 00:10:04,559 Speaker 2: It is. 215 00:10:05,000 --> 00:10:09,559 Speaker 1: He's suggesting we'll be launching several internets worth of compute annually. 216 00:10:09,679 --> 00:10:13,320 Speaker 2: Yes. The implication is that the demand for compute for 217 00:10:13,440 --> 00:10:17,440 raw intelligence is so bottomless, so insatiable, that we will 218 00:10:17,480 --> 00:10:19,759 literally have to leave the planet to find enough energy 219 00:10:19,759 --> 00:10:20,559 to satisfy it. 220 00:10:20,639 --> 00:10:23,440 Speaker 1: And he suggested Earth itself just caps out, that we 221 00:10:23,519 --> 00:10:24,159 hit a hard wall. 222 00:10:24,200 --> 00:10:26,080 Speaker 2: He put a number on it. He thinks we hit 223 00:10:26,080 --> 00:10:29,559 a limit around one tarrawat of AI capacity on the surface. 224 00:10:29,240 --> 00:10:33,279 Speaker 1: A tarowat a trillion watts. Why why that limit? We 225 00:10:33,279 --> 00:10:36,320 have nuclear power, we have other energy sources. Why can't 226 00:10:36,320 --> 00:10:37,840 we just build more down here? 227 00:10:38,200 --> 00:10:42,519 Speaker 2: It becomes a logistical and environmental nightmare. A tarowat of 228 00:10:42,600 --> 00:10:45,759 stable two hundred and forty seven power, you need to 229 00:10:45,759 --> 00:10:50,559 build hundreds of new nuclear plants. That takes decades of permitting, construction, 230 00:10:50,720 --> 00:10:51,639 political battles. 231 00:10:51,679 --> 00:10:53,919 Speaker 1: Okay, so nuclear is slow. What about gas? 232 00:10:54,639 --> 00:10:58,159 Speaker 2: Then you have pipeline constraints, carbon capture issues, and the 233 00:10:58,159 --> 00:11:01,879 grid itself, the physical copper wires strung across the country. 234 00:11:01,919 --> 00:11:03,080 They can't handle that load. 235 00:11:03,240 --> 00:11:04,759 Speaker 1: So the wires literally. 236 00:11:04,399 --> 00:11:08,320 Speaker 2: Melt effectively, yes, or the transformers at the substations just 237 00:11:08,399 --> 00:11:11,720 blow up. The terrestrial grid is a fragile beast. It 238 00:11:11,759 --> 00:11:14,360 was built over one hundred years for a distributed load. 239 00:11:14,679 --> 00:11:18,000 It was never designed for these hyper localized gigawatt scale 240 00:11:18,000 --> 00:11:19,960 loads just popping up overnight in the desert. 241 00:11:20,120 --> 00:11:21,480 Speaker 1: But space is a green field. 242 00:11:21,720 --> 00:11:24,600 Speaker 2: Space is a clean slate. You aren't plugging into an old, 243 00:11:24,679 --> 00:11:27,279 creaky grid. You're building a brand new one from scratch, 244 00:11:27,679 --> 00:11:29,240 optimized purely for AI. 245 00:11:29,600 --> 00:11:31,960 Speaker 1: And this, of course, of where SpaceX comes in. This 246 00:11:32,039 --> 00:11:35,720 whole plan doesn't work without cheap, reliable, heavy lift launch. 247 00:11:36,039 --> 00:11:38,559 Speaker 2: He's building the pickax and the shovel for the gold Rush. 248 00:11:38,679 --> 00:11:42,039 Speaker 1: He explicitly made that comparison, right. He sees SpaceX as 249 00:11:42,120 --> 00:11:43,799 the new Union Pacific Railroad. 250 00:11:44,120 --> 00:11:46,759 Speaker 2: Yeah, in the eighteen hundreds, if you wanted to get 251 00:11:46,759 --> 00:11:49,120 in on the gold Rush, you needed the railroad to 252 00:11:49,120 --> 00:11:51,679 get your equipment to California. Today, if you want to 253 00:11:51,720 --> 00:11:53,759 compete in the AI gold Rush, you're going to need 254 00:11:53,799 --> 00:11:55,879 Starship to get your GPUs to orbit. 255 00:11:56,039 --> 00:11:59,000 Speaker 1: It's the ultimate vertical integration. It's Tesla for the energy 256 00:11:59,000 --> 00:12:02,039 and robots, space X for the logistics, x AI for 257 00:12:02,080 --> 00:12:02,840 the intelligence. 258 00:12:02,919 --> 00:12:03,720 Speaker 2: It's the full stack. 259 00:12:03,840 --> 00:12:06,559 Speaker 1: Okay, But before we get to this, you know that's 260 00:12:06,600 --> 00:12:09,799 incredible future of orbital server farms, we have to survive 261 00:12:09,879 --> 00:12:12,320 the mad Max energy scramble happening right now. 262 00:12:12,559 --> 00:12:13,639 Speaker 2: Right, We're not there yet. 263 00:12:13,759 --> 00:12:16,919 Speaker 1: This brings us to the next thread, the bottleneck. 264 00:12:16,559 --> 00:12:19,679 Speaker 2: Because we can't launch that stuff. Today's starship is still 265 00:12:19,759 --> 00:12:22,519 ramping up, so we have this gap to bridge and 266 00:12:22,639 --> 00:12:25,639 Musk laid out what he called a timeline of constraints, 267 00:12:25,960 --> 00:12:28,600 which is really important for anyone trying to understand the space. 268 00:12:28,679 --> 00:12:31,120 Speaker 1: He broke it down by time horizons. He said, the 269 00:12:31,120 --> 00:12:33,639 problem actually changes depending on how far out you look. 270 00:12:33,720 --> 00:12:36,159 Speaker 2: Exactly, he said, on a one year horizon, So like 271 00:12:36,279 --> 00:12:40,759 right now, immediately, the crisis is not chips, which. 272 00:12:40,559 --> 00:12:43,120 Speaker 1: Is funny because all you hear about is the chip shortage. 273 00:12:43,279 --> 00:12:46,759 Speaker 2: Everyone is screaming about a chip shortage. But Musk says, no, 274 00:12:47,679 --> 00:12:50,639 we have the silicon, the H one hundreds, B one hundreds. 275 00:12:51,679 --> 00:12:54,080 They are rolling off the assembly line at t SMC. 276 00:12:54,480 --> 00:12:55,120 They exist. 277 00:12:55,320 --> 00:12:57,720 Speaker 1: So what's the problem If we have the brains, why 278 00:12:57,759 --> 00:12:58,519 aren't they thinking? 279 00:12:58,679 --> 00:13:02,720 Speaker 2: The problem is voltage? It's electricity, but it's not the generation. 280 00:13:02,879 --> 00:13:08,519 It's the plumbing. Yeah, specifically the step down transformers. No, 281 00:13:08,639 --> 00:13:12,320 what you have these massive high voltage lines running near 282 00:13:12,320 --> 00:13:15,639 a data center, carrying hundreds of thousands of volts, but 283 00:13:15,720 --> 00:13:20,600 a server rag needs drastically lower stable voltage to do 284 00:13:20,639 --> 00:13:24,440 that conversion. You need these huge bus sized transformers at 285 00:13:24,480 --> 00:13:25,039 a substation. 286 00:13:25,320 --> 00:13:27,240 Speaker 1: Let me guess you can't just buy those on Amazon Prime. 287 00:13:27,320 --> 00:13:30,440 Speaker 2: Now, those specific utility grade transformers are on a two 288 00:13:30,480 --> 00:13:33,679 year back order. The global supply chain for heavy electrical 289 00:13:33,759 --> 00:13:35,720 equipment is completely jammed. 290 00:13:35,759 --> 00:13:38,240 Speaker 1: So you could build a brand new billion dollar data center, 291 00:13:38,279 --> 00:13:40,440 you could fill it with the latest, most powerful AI 292 00:13:40,519 --> 00:13:42,639 chips in the world, and it just sits there. 293 00:13:42,720 --> 00:13:44,639 Speaker 2: It just sits there because You literally can't plug it 294 00:13:44,679 --> 00:13:45,200 into the wall. 295 00:13:45,360 --> 00:13:46,200 Speaker 1: That's insane. 296 00:13:46,279 --> 00:13:48,759 Speaker 2: And this leads to this concept he talked about called 297 00:13:49,039 --> 00:13:52,360 dark silicon. It sounds ominous, doesn't it. 298 00:13:52,360 --> 00:13:55,039 Speaker 1: It sounds like a villain in a Marvel movie. I 299 00:13:55,080 --> 00:13:57,200 am Dark Silicon destroyer of grids. 300 00:13:57,480 --> 00:14:00,559 Speaker 2: It's more of an economic villain. Dark silic is the 301 00:14:00,639 --> 00:14:04,360 term for chips that have been manufactured, purchased for tens 302 00:14:04,399 --> 00:14:07,799 of thousands of dollars each, shipped across the ocean, installed 303 00:14:07,799 --> 00:14:10,919 in a server rack, and then they just sit there dark, 304 00:14:11,720 --> 00:14:13,240 unpowered because. 305 00:14:13,080 --> 00:14:16,080 Speaker 1: The local utility company called and said, sorry, we can't 306 00:14:16,120 --> 00:14:17,399 give you the juice for two years. 307 00:14:17,559 --> 00:14:18,399 Speaker 2: That's exactly it. 308 00:14:18,799 --> 00:14:21,960 Speaker 1: Wow, that is a wild level of inefficiency. You have 309 00:14:22,440 --> 00:14:25,519 billions of dollars of cutting edge capital equipment just gathering dust. 310 00:14:26,399 --> 00:14:29,000 Musk's quote was people are going to have real trouble 311 00:14:29,039 --> 00:14:29,879 turning on the chips. 312 00:14:30,399 --> 00:14:33,279 Speaker 2: And this is where it gets strategic. This reality, this 313 00:14:33,399 --> 00:14:36,840 bottleneck creates a fascinating shift and who is likely to win? 314 00:14:37,200 --> 00:14:39,840 Speaker 1: Right? He was very candid about his own company, XAI. 315 00:14:39,919 --> 00:14:41,440 He didn't come out and say, oh, our research is 316 00:14:41,440 --> 00:14:44,440 a smarter than Google's, or our code is magic, which. 317 00:14:44,200 --> 00:14:46,519 Speaker 2: Is what every other tech CEO does. It's always about 318 00:14:46,559 --> 00:14:49,799 their secret sauce algorithm, their superior model architecture. 319 00:14:50,240 --> 00:14:55,720 Speaker 1: Right. But Musk's argument is that in AI today, ideas 320 00:14:56,159 --> 00:14:59,039 diffuse way too quickly for that to be a real advantage. 321 00:14:59,039 --> 00:14:59,960 Speaker 2: What do you mean by diffuse. 322 00:15:00,240 --> 00:15:03,960 Speaker 1: Well, if a researcher, a DeepMind discovers some brilliant new 323 00:15:04,039 --> 00:15:07,639 transformer architecture or a new way to do reinforcement learning, 324 00:15:08,440 --> 00:15:11,919 that idea leaks. It's in a research paper on Arskill 325 00:15:12,000 --> 00:15:14,399 in a week, it's in a discord server that night, 326 00:15:14,720 --> 00:15:16,000 it's on GitHub in a month. 327 00:15:16,200 --> 00:15:18,720 Speaker 2: The half life of a trade secret in AI software 328 00:15:18,799 --> 00:15:20,240 is incredibly short, so. 329 00:15:20,159 --> 00:15:22,360 Speaker 1: There's no real software mote anymore. You can't just be 330 00:15:22,399 --> 00:15:24,320 the smartest coder in the room and expect to win. 331 00:15:24,360 --> 00:15:26,720 Speaker 2: Not for long. The community is too open, too fast. 332 00:15:26,960 --> 00:15:29,759 But you know, it doesn't leak a power plant. You 333 00:15:29,799 --> 00:15:33,399 can't just copy paste a gigawatt substation. You can't download 334 00:15:33,399 --> 00:15:36,519 a cooling solution for one hundred thousand GPU cluster. Those 335 00:15:36,519 --> 00:15:38,279 are physical industrial modes. 336 00:15:38,360 --> 00:15:40,639 Speaker 1: So the winner of the AI race isn't the person 337 00:15:40,679 --> 00:15:44,399 with the best algorithm, is the person with the best electrician. 338 00:15:44,159 --> 00:15:47,519 Speaker 2: In reductionist sense, yes, the winner is the one who 339 00:15:47,559 --> 00:15:50,159 can actually plug in the most machines without blowing up 340 00:15:50,200 --> 00:15:53,639 the grid. And Musk believes that because Tesla and SpaceX 341 00:15:53,679 --> 00:15:58,120 are fundamentally hardware and energy companies, XAI has a built 342 00:15:58,159 --> 00:15:59,720 in advantage. 343 00:15:59,240 --> 00:16:02,879 Speaker 1: And advantage that a pure software lab like open AI 344 00:16:03,080 --> 00:16:04,759 or Anthropic doesn't have. 345 00:16:05,120 --> 00:16:09,039 Speaker 2: Exactly, they're playing a video game about code. He's playing Factorio. 346 00:16:09,679 --> 00:16:11,799 Speaker 1: I love that analogy for anyone who doesn't know. Factorio 347 00:16:11,919 --> 00:16:14,279 is this game where you just build automated supply chains. 348 00:16:14,320 --> 00:16:16,679 You don't win by being a hero. You win by 349 00:16:16,679 --> 00:16:19,919 building better conveyor belts, better power grids, better logistics. 350 00:16:20,200 --> 00:16:22,600 Speaker 2: And that is the reality of scaling AI right now. 351 00:16:22,679 --> 00:16:25,480 It is an industrial engineering problem far more than it 352 00:16:25,559 --> 00:16:29,240 is a computer science problem. The code is almost secondary 353 00:16:29,279 --> 00:16:31,360 to the calorie intake of the machine. 354 00:16:31,519 --> 00:16:34,679 Speaker 1: Okay, so we're building this massive industrial machine, we're fighting 355 00:16:34,679 --> 00:16:37,320 for power on Earth, dealing with dark silicon, and eventually 356 00:16:37,320 --> 00:16:40,559 we're moving the whole circus into space. But what is 357 00:16:40,600 --> 00:16:42,679 all this compute for? What are we actually trying to 358 00:16:42,679 --> 00:16:44,000 create with all these watts? 359 00:16:44,519 --> 00:16:47,200 Speaker 2: That brings us to the next really big concept, the 360 00:16:47,240 --> 00:16:48,000 digital human. 361 00:16:48,360 --> 00:16:51,000 Speaker 1: Now, I want to be really clear here for everyone listening. 362 00:16:51,519 --> 00:16:55,440 When Musk says digital human, he is not talking about 363 00:16:55,519 --> 00:16:58,480 a cartoon avatar in the metaverse with no legs. Yeah, 364 00:16:58,759 --> 00:17:01,720 he's not talking about the annoying customer service chatbot that 365 00:17:01,799 --> 00:17:03,360 can only reset your password. 366 00:17:03,519 --> 00:17:05,759 Speaker 2: No, this is something much more profound. He's talking about 367 00:17:05,759 --> 00:17:09,920 a human emulator, a software entity that can do anything 368 00:17:09,960 --> 00:17:11,079 a human being can do. 369 00:17:11,440 --> 00:17:14,720 Speaker 1: Sitting at a computer anything. Think about your job. If 370 00:17:14,720 --> 00:17:17,960 you work in an office, think about a generic knowledge worker. 371 00:17:18,200 --> 00:17:18,680 Speaker 2: What do you do. 372 00:17:18,759 --> 00:17:20,519 Speaker 1: You sit at a desk, you look at a screen, 373 00:17:20,720 --> 00:17:22,519 you type on a keyboard, you move a. 374 00:17:22,480 --> 00:17:25,759 Speaker 2: Mouse, You read emails, you write documents, You design things 375 00:17:25,759 --> 00:17:28,359 in pigma, you code in vs code, you browse the web, 376 00:17:28,440 --> 00:17:31,279 you process information, and you output information. 377 00:17:31,839 --> 00:17:34,319 Speaker 1: Musk's argument is that if you can truly solve that, 378 00:17:35,000 --> 00:17:37,640 if you can build an AI that can fully emulate 379 00:17:37,680 --> 00:17:41,839 that entire process, including the reasoning, the creativity, the tool use. 380 00:17:41,880 --> 00:17:45,240 Speaker 2: You unlock the single biggest economic shift in human history. 381 00:17:45,279 --> 00:17:47,039 He just kept saying trillions of dollars. 382 00:17:47,400 --> 00:17:49,880 Speaker 1: And to prove this he used this theory that I'm 383 00:17:49,920 --> 00:17:54,480 calling the digital output theory. He basically broke down the 384 00:17:54,519 --> 00:17:58,359 world's biggest, most valuable companies to their atomic level, or 385 00:17:58,440 --> 00:17:59,759 rather their bit level. 386 00:18:00,000 --> 00:18:02,759 Speaker 2: The very first principle's way of looking at where value 387 00:18:02,759 --> 00:18:05,240 actually comes from. He took in Nvidia as the first 388 00:18:05,240 --> 00:18:08,960 example multi trillion dollar company. We think of them as 389 00:18:08,960 --> 00:18:09,759 a hardware company. 390 00:18:09,799 --> 00:18:12,240 Speaker 1: Right, Well, of course they mean chips. 391 00:18:12,359 --> 00:18:14,880 Speaker 2: Do they do they physically make anything? 392 00:18:15,000 --> 00:18:18,839 Speaker 1: Well? No, I know that TSMC, the foundry in Taiwan, 393 00:18:19,000 --> 00:18:21,480 makes the actual physical silicon chips exactly. 394 00:18:21,519 --> 00:18:24,880 Speaker 2: So what does in Nvidia, the company in Santa Clara, California, 395 00:18:24,960 --> 00:18:27,200 actually produce what comes out of their headquarters? 396 00:18:27,240 --> 00:18:30,039 Speaker 1: I guess blueprints, schematics. 397 00:18:29,400 --> 00:18:32,720 Speaker 2: For the chips, design files, a bitstream, a sequence of 398 00:18:32,720 --> 00:18:36,160 ones and ViOS that describes a chip architecture. Musk's line was, 399 00:18:36,359 --> 00:18:39,599 in Nvidia's entire output is ftpen files to Taiwan. 400 00:18:40,160 --> 00:18:43,319 Speaker 1: That is a hilariously reductionist, but technically true way of 401 00:18:43,319 --> 00:18:46,400 putting it. Here is your two trillion dollar company. It's 402 00:18:46,440 --> 00:18:48,039 a file transfer, but it's true. 403 00:18:48,079 --> 00:18:53,000 Speaker 2: And then he did Apple an iPhone is made of atoms, right, glass, aluminum, silicon. 404 00:18:53,640 --> 00:18:56,559 But Apple doesn't mind the aluminum or screw the glass. 405 00:18:56,200 --> 00:18:58,559 Speaker 1: On Fox con does that in China, right. 406 00:18:58,720 --> 00:19:01,880 Speaker 2: Apple's contribution. The thing that comes out of Kupertino is 407 00:19:01,960 --> 00:19:05,519 purely the design files for the hardware and the software 408 00:19:05,519 --> 00:19:08,160 CID for iOS, it's pure bits. 409 00:19:08,319 --> 00:19:11,720 Speaker 1: And then of course Google, Meta, Microsoft. 410 00:19:11,279 --> 00:19:14,799 Speaker 2: They're just pure bit factories. Their entire product is digital 411 00:19:15,039 --> 00:19:19,200 search results, social media, feeds, spreadsheets. They digest bits and 412 00:19:19,240 --> 00:19:20,319 they excrete bits. 413 00:19:20,640 --> 00:19:24,480 Speaker 1: So the fundamental insight here is the most valuable and 414 00:19:24,559 --> 00:19:29,759 powerful companies in the world produce almost nothing but digital output. Therefore, 415 00:19:30,079 --> 00:19:33,119 if you create a digital human that can generate that 416 00:19:33,200 --> 00:19:34,559 same digital output. 417 00:19:34,480 --> 00:19:36,720 Speaker 2: Then you can automate the core function of the entire 418 00:19:36,720 --> 00:19:39,640 global economy. If you have an AI that can design 419 00:19:39,640 --> 00:19:41,759 a chip as well as an Nvidia engineer, or write 420 00:19:41,759 --> 00:19:44,319 code as well as a Google engineer, or design a 421 00:19:44,400 --> 00:19:47,319 user interface as well as an Apple designer, you don't 422 00:19:47,359 --> 00:19:48,279 just cut costs. 423 00:19:48,359 --> 00:19:50,480 Speaker 1: You create a company that can scale infinitely. 424 00:19:50,559 --> 00:19:52,000 Speaker 2: That's the key. You can't just go out and hire 425 00:19:52,039 --> 00:19:54,960 one hundred thousand senior electrical engineers overnight. They don't exist. 426 00:19:55,079 --> 00:19:56,680 There aren't enough PhDs in the world. 427 00:19:56,880 --> 00:20:00,279 Speaker 1: But you can spin up one hundred thousand instants is 428 00:20:00,319 --> 00:20:03,640 of a digital human AI on a server farm overnight, 429 00:20:03,880 --> 00:20:05,680 assuming you have the compute in the power. 430 00:20:05,440 --> 00:20:09,640 Speaker 2: Exactly the barrier to entry for creating a mega corporation. 431 00:20:09,920 --> 00:20:12,920 A new and Video or a new Google drops from 432 00:20:12,960 --> 00:20:16,559 needing tens of thousands of elite human minds to just 433 00:20:17,079 --> 00:20:18,839 the cost of electricity, you could have. 434 00:20:18,759 --> 00:20:21,880 Speaker 1: A one person company that employs a million digital workers. 435 00:20:21,880 --> 00:20:23,880 It completely changes the nature of the firm. 436 00:20:23,960 --> 00:20:28,960 Speaker 2: It's both exhilarating and terrifying. It essentially decouples economic output 437 00:20:28,960 --> 00:20:31,480 from human population. We used to need more people to 438 00:20:31,519 --> 00:20:34,000 make more stuff. Now you just need more GPUs. 439 00:20:34,119 --> 00:20:36,839 Speaker 1: And the timeline he put on this on solving this 440 00:20:36,960 --> 00:20:38,279 digital human benchmark. 441 00:20:38,440 --> 00:20:42,200 Speaker 2: He was talking late twenty twenty five, early twenty twenty six. 442 00:20:42,279 --> 00:20:45,160 Speaker 1: Which again for us sitting here in early twenty twenty six, 443 00:20:45,279 --> 00:20:47,319 is right now. And I mean, if you look at 444 00:20:47,319 --> 00:20:49,960 the rise of AI agents AI that can browse the web, 445 00:20:50,039 --> 00:20:53,039 use a mouse, write and execute its own code. We're 446 00:20:53,079 --> 00:20:54,799 seeing the first prototypes of this today. 447 00:20:54,880 --> 00:20:57,400 Speaker 2: We are we're definitely scratching the surface. But of course 448 00:20:57,519 --> 00:20:59,759 Musk doesn't stop with the digital world. He doesn't stop 449 00:20:59,759 --> 00:21:03,000 at this because once you have the brain, the digital human, 450 00:21:03,480 --> 00:21:06,480 the next logical step is to give it a body, and. 451 00:21:06,400 --> 00:21:08,079 Speaker 1: This is where we get to the infinite money. 452 00:21:08,079 --> 00:21:11,079 Speaker 2: Glitch optimists the robot, the humanoid robot. 453 00:21:11,200 --> 00:21:14,319 Speaker 1: He calls the Optimist robot the infinite money glitch. Yeah, 454 00:21:14,319 --> 00:21:16,160 and I love that phrase because it sounds like a 455 00:21:16,200 --> 00:21:18,160 cheat code in the video game, like you're just getting 456 00:21:18,160 --> 00:21:21,960 free resources. But he means it in a very strict 457 00:21:22,119 --> 00:21:23,079 economic sense. 458 00:21:23,160 --> 00:21:25,799 Speaker 2: He does, and the math behind it, as he explained 459 00:21:25,799 --> 00:21:29,559 it is fascinating. He describes it as the product of 460 00:21:29,680 --> 00:21:33,039 three different exponentials, all multiplying each other. 461 00:21:33,279 --> 00:21:36,240 Speaker 1: Okay, let's break those down. What are the three exponential 462 00:21:36,279 --> 00:21:38,440 curves that need to happen to unlock this glitch? 463 00:21:38,559 --> 00:21:42,240 Speaker 2: Okay, so first, you have the exponential increase in digital intelligence. 464 00:21:42,640 --> 00:21:46,440 The brains the AI models are getting smarter every month. 465 00:21:46,480 --> 00:21:47,519 That's the digital human we. 466 00:21:47,519 --> 00:21:49,400 Speaker 1: Just talked about. You got it, Brain's getting smarter. 467 00:21:49,720 --> 00:21:54,480 Speaker 2: Second, you have the exponential increase in chip capability, specifically 468 00:21:54,640 --> 00:21:58,200 the inference chips, the hardware that actually runs inside the 469 00:21:58,279 --> 00:22:01,960 robot's head. They're getting more powerful and more energy efficient 470 00:22:02,039 --> 00:22:03,680 at an incredible rate. 471 00:22:03,680 --> 00:22:05,839 Speaker 1: Which means you can run a really smart brain on 472 00:22:05,880 --> 00:22:08,440 a small battery pack. You don't need a robot that 473 00:22:08,480 --> 00:22:11,279 has to be plugged into a server rack to think exactly. 474 00:22:11,279 --> 00:22:14,599 Speaker 2: It needs to be untothered. And then the third exponential 475 00:22:14,759 --> 00:22:18,279 is the increase in electro mechanical dexterity, so the body, 476 00:22:18,400 --> 00:22:21,559 the body, the actuators, the motors, the hands, the sensors, 477 00:22:21,839 --> 00:22:24,799 The ability of the robot to move and grasp and 478 00:22:24,920 --> 00:22:28,559 balance and manipulate the physical world is also improving on 479 00:22:28,599 --> 00:22:29,880 a very steep curve. 480 00:22:30,119 --> 00:22:32,519 Speaker 1: So you have a smarter brain running on a more 481 00:22:32,519 --> 00:22:35,319 efficient chip inside a more capable body. And when you 482 00:22:35,359 --> 00:22:38,240 multiply those three accelerating curves together. 483 00:22:38,079 --> 00:22:41,440 Speaker 2: You get a supernova, a productivity explosion, because at a 484 00:22:41,480 --> 00:22:44,240 certain point the robot becomes capable enough to walk onto 485 00:22:44,240 --> 00:22:47,119 a factory floor and manufacture another robot. 486 00:22:47,279 --> 00:22:48,559 Speaker 1: The feedback loop. 487 00:22:48,480 --> 00:22:52,599 Speaker 2: The recursive loop, robots building robots. This is the concept 488 00:22:52,599 --> 00:22:55,880 people sometimes call the final invention. Think about why anything 489 00:22:55,880 --> 00:22:59,240 physical is expensive today, A car, a house, a phone. 490 00:22:59,519 --> 00:23:00,559 Why does it cost. 491 00:23:00,359 --> 00:23:04,279 Speaker 1: So much labor and materials? But mostly labor right all 492 00:23:04,319 --> 00:23:07,000 the way down the supply chain, the miners, the truck drivers, 493 00:23:07,000 --> 00:23:09,759 the factory workers, the builders, the inspectors. 494 00:23:09,599 --> 00:23:13,000 Speaker 2: Exactly, and even the materials are expensive because of the 495 00:23:13,079 --> 00:23:16,119 human labor required to extract and refine them. But if 496 00:23:16,160 --> 00:23:18,480 you have a machine that can dig the ore, refine 497 00:23:18,480 --> 00:23:21,319 the metal, manufacture the parts, and then assemble a perfect 498 00:23:21,359 --> 00:23:25,359 copy of itself, the cost of labor effectively drops to zero. 499 00:23:25,759 --> 00:23:27,960 Speaker 1: It just becomes the cost of the raw materials and 500 00:23:28,000 --> 00:23:28,519 the energy. 501 00:23:28,599 --> 00:23:31,319 Speaker 2: And if we remember segment one, we're planning on getting 502 00:23:31,359 --> 00:23:33,000 that energy from space for cheap. 503 00:23:33,480 --> 00:23:37,160 Speaker 1: This is the glitch in economics. Labor is always the 504 00:23:37,240 --> 00:23:41,440 ultimate constraint. You can't just spawn more trained human welders 505 00:23:41,480 --> 00:23:44,880 or engineers. It takes twenty years to raise and educate one. 506 00:23:44,960 --> 00:23:47,680 Speaker 2: But you can spawn more robots. If the cost of 507 00:23:47,720 --> 00:23:50,599 one robot is say ten thousand dollars in materials and energy, 508 00:23:50,640 --> 00:23:52,799 and that one robot can build ten more robots in. 509 00:23:52,720 --> 00:23:54,960 Speaker 1: Its lifetime, then the cost of the next generation of 510 00:23:55,039 --> 00:23:57,720 robots plummets, and the next and the next. It's a 511 00:23:57,759 --> 00:24:00,880 deflationary spiral for the cost of everything physical it is. 512 00:24:01,000 --> 00:24:05,480 Speaker 2: It's a post scarcity utopia in theory. Suddenly housing is 513 00:24:05,559 --> 00:24:08,079 cheap because robots can build houses two hundred four seven. 514 00:24:08,559 --> 00:24:12,079 Food is cheap because robots can farm. All physical goods 515 00:24:12,079 --> 00:24:14,759 become essentially as free as the energy required to make them. 516 00:24:15,000 --> 00:24:17,000 It's the star Trek replicator economy. 517 00:24:17,240 --> 00:24:20,359 Speaker 1: That's the optimistic view. That's the age of abundance. But 518 00:24:20,480 --> 00:24:24,440 it immediately raises a massive and kind of scary question, 519 00:24:24,599 --> 00:24:27,480 which is, if the robots are building the robots, and 520 00:24:27,480 --> 00:24:31,559 the robots are running the entire physical economy, who's in charge, 521 00:24:31,720 --> 00:24:34,039 who owns a fleet, who holds the off switch? 522 00:24:34,200 --> 00:24:37,880 Speaker 2: And that brings us to the final and honestly the 523 00:24:37,920 --> 00:24:40,559 most controversial thread of this entire tapestry. 524 00:24:40,680 --> 00:24:41,119 Speaker 1: The villain. 525 00:24:41,359 --> 00:24:44,000 Speaker 2: Yes, this was the big curve ball that I think 526 00:24:44,039 --> 00:24:45,880 surprised a lot of people who are listening to that 527 00:24:45,960 --> 00:24:47,559 original conversation. 528 00:24:47,319 --> 00:24:50,599 Speaker 1: Absolutely, because we've all been trained by decades of movies 529 00:24:50,640 --> 00:24:53,039 and books to fear the evil corporation. 530 00:24:53,240 --> 00:24:58,759 Speaker 2: Of course, Tyroll Corporation in Blade Runner Cyberdyne Systems, and Terminator. 531 00:24:58,359 --> 00:25:02,000 Speaker 1: Whalen Utani in Alien. The narrative is always the same, 532 00:25:02,079 --> 00:25:05,880 some greedy corporation in his blind pursuit of profit, creates 533 00:25:05,880 --> 00:25:07,680 a monster that ends up destroying us. 534 00:25:07,680 --> 00:25:10,960 Speaker 2: All The story is that the profit motive overrides safety 535 00:25:11,000 --> 00:25:14,480 and common sense. That's the standard trope. We assume the 536 00:25:14,480 --> 00:25:16,799 CEO is the bad guy because he cares more about 537 00:25:16,880 --> 00:25:19,440 next quarter share price than he does about the survival 538 00:25:19,480 --> 00:25:20,119 of humanity. 539 00:25:20,480 --> 00:25:23,799 Speaker 1: But Musk flips this completely on its head. He says, 540 00:25:23,799 --> 00:25:27,559 the entity you should actually fear is not the corporation. 541 00:25:27,440 --> 00:25:28,519 Speaker 2: He says, it's the government. 542 00:25:28,720 --> 00:25:32,000 Speaker 1: It was a really stark statement, and he defined government 543 00:25:32,039 --> 00:25:34,480 in a way that I think, you know, political scientists 544 00:25:34,480 --> 00:25:37,240 would recognize, but that a lot of regular people might 545 00:25:37,359 --> 00:25:38,240 find jarring. 546 00:25:38,680 --> 00:25:42,480 Speaker 2: He called government just the biggest corporation with a monopoly 547 00:25:42,519 --> 00:25:43,160 on violence. 548 00:25:43,240 --> 00:25:47,039 Speaker 1: Oo monopoly on violence. That is a heavy phrase. It 549 00:25:47,079 --> 00:25:48,200 sounds almost anarchistic. 550 00:25:48,400 --> 00:25:51,480 Speaker 2: It does, but it's actually the standard sociological definition of 551 00:25:51,480 --> 00:25:54,279 the state. It goes back to the sociologist Max Weber 552 00:25:54,319 --> 00:25:57,519 in the early twentieth century. The state is defined as 553 00:25:57,559 --> 00:26:00,599 the only entity within a given territory that has the 554 00:26:00,680 --> 00:26:03,440 legitimate right to use physical force against its. 555 00:26:03,240 --> 00:26:05,920 Speaker 1: People, to arrest you, to imprison you, to wage war. 556 00:26:06,119 --> 00:26:09,440 Speaker 2: Right and Musk's argument is that this unique power makes 557 00:26:09,480 --> 00:26:12,559 them infinitely more dangerous than any company could ever be. 558 00:26:13,079 --> 00:26:16,160 How So, well, he argues that corporations, even the most 559 00:26:16,160 --> 00:26:19,359 greedy and powerful ones, are ultimately constrained. They have to 560 00:26:19,400 --> 00:26:22,160 obey laws, They have to serve customers, or they go 561 00:26:22,200 --> 00:26:22,839 out of business. 562 00:26:23,079 --> 00:26:25,640 Speaker 1: If a car company makes a car that explodes, they 563 00:26:25,640 --> 00:26:28,799 get sued into oblivion, their customers abandon them, they go bankrupt. 564 00:26:28,920 --> 00:26:30,640 The market punishes them exactly. 565 00:26:30,680 --> 00:26:34,880 Speaker 2: The CEO goes to jail. But governments, they don't have 566 00:26:34,960 --> 00:26:38,359 those same constraints. They can literally rewrite the laws to 567 00:26:38,480 --> 00:26:41,680 make their actions legal. They can print money to avoid 568 00:26:41,680 --> 00:26:44,400 going bankrupt. They don't really have competition in the same way. 569 00:26:44,519 --> 00:26:48,759 Speaker 1: So his argument is that corporations actually have a better 570 00:26:49,119 --> 00:26:52,759 or at least more functional morality than governments because they 571 00:26:52,759 --> 00:26:55,480 are held in check by the market and the legal system. 572 00:26:55,680 --> 00:26:58,400 Speaker 2: That's his take. So his ultimate fear isn't that a 573 00:26:58,519 --> 00:27:03,079 CEO maximized profits and accidentally turns the world into paper clips. 574 00:27:03,240 --> 00:27:05,359 Speaker 1: His fear is that a government gets control of this 575 00:27:05,480 --> 00:27:09,000 technology and uses it to intentionally suppress the population. 576 00:27:09,240 --> 00:27:13,160 Speaker 2: It's the ultimate authoritarian nightmare. Imagine that infinite money glitch 577 00:27:13,200 --> 00:27:16,319 we just talked about an army of millions of Optimus robots. 578 00:27:16,480 --> 00:27:18,680 Now imagine the government controls that army. 579 00:27:18,839 --> 00:27:21,680 Speaker 1: You don't need hewing soldiers anymore. You don't need human 580 00:27:21,720 --> 00:27:24,279 police who might have a conscience, who might hesitate to 581 00:27:24,319 --> 00:27:27,240 follow an unjust order or fire on their own neighbors. 582 00:27:27,599 --> 00:27:33,759 Speaker 2: You have a fleet of perfectly obedient, unthinking, physically superior enforcers. 583 00:27:34,240 --> 00:27:38,240 It removes the human check on tyranny throughout history, most 584 00:27:38,279 --> 00:27:41,440 revolutions and coups happen because the army eventually refuses to 585 00:27:41,440 --> 00:27:43,640 shoot the people they switch sides. 586 00:27:43,759 --> 00:27:46,880 Speaker 1: A robot will never switch sides. It'll just execute the code. 587 00:27:46,960 --> 00:27:50,279 Speaker 2: It just executes the code. This completely flips the standard 588 00:27:50,279 --> 00:27:54,680 cyberpunk narrative. Usually it's the plucky rebels fighting against the 589 00:27:54,720 --> 00:27:55,599 evil corporation. 590 00:27:56,200 --> 00:27:59,240 Speaker 1: Musk is saying, no, the corporation might be the only 591 00:27:59,240 --> 00:28:03,720 thing standing between you and a totalitarian superstate armed with AI. 592 00:28:04,039 --> 00:28:06,799 Speaker 2: It's a very libertarian viewpoint, obviously, and we should definitely 593 00:28:06,839 --> 00:28:08,920 discuss the validity of it. I mean, is he right 594 00:28:09,039 --> 00:28:11,680 or is this just a billionaire who doesn't like being 595 00:28:11,720 --> 00:28:14,160 regulated trying to make an argument for less oversight. 596 00:28:14,359 --> 00:28:17,279 Speaker 1: That's the cynical take, isn't it. Don't regulate me. I'm 597 00:28:17,319 --> 00:28:19,400 the good guy. The government is a real bad guy. 598 00:28:19,920 --> 00:28:22,119 It's a very convenient argument for him to make when 599 00:28:22,160 --> 00:28:25,519 he has regulators breathing down his neck about self driving 600 00:28:25,559 --> 00:28:26,519 cars and rockets. 601 00:28:26,880 --> 00:28:29,480 Speaker 2: It is convenient, But we also have to be intellectually 602 00:28:29,480 --> 00:28:32,000 honest and look at the scoreboard of history in the 603 00:28:32,000 --> 00:28:36,759 twentieth century. Who caused more death than suffering corporations or governments? 604 00:28:37,000 --> 00:28:40,880 Speaker 1: I mean, just looking at the raw numbers, it's not 605 00:28:40,920 --> 00:28:46,960 even a contest. It's governments, world wars, genocides, state sponsored famines, purges. 606 00:28:47,480 --> 00:28:50,759 The Goolegs Ford Motor Company didn't do that exactly. 607 00:28:51,000 --> 00:28:54,640 Speaker 2: So while the Evil Corp is a powerful and scary story, 608 00:28:55,000 --> 00:28:58,200 the evil State has a much longer and bloodier historical 609 00:28:58,240 --> 00:29:02,359 track record and giving entity the power of superintelligence and 610 00:29:02,400 --> 00:29:06,079 an infinite army of robots, that is a risk profile 611 00:29:06,079 --> 00:29:07,799 that we have to take very, very seriously. 612 00:29:07,920 --> 00:29:10,559 Speaker 1: It really complicates the whole AI safety debate, doesn't it, 613 00:29:10,599 --> 00:29:13,599 Because the default position for so many people is the 614 00:29:13,640 --> 00:29:16,400 government must save us from AI. Pass a law, stop 615 00:29:16,440 --> 00:29:17,359 the labs. 616 00:29:17,039 --> 00:29:18,839 Speaker 2: Right, regulate it before it's too late. 617 00:29:19,119 --> 00:29:21,160 Speaker 1: Musk is sort of the whispering from the side. But 618 00:29:21,200 --> 00:29:22,839 who is going to save you from the government with 619 00:29:22,880 --> 00:29:23,279 the AI? 620 00:29:23,720 --> 00:29:27,920 Speaker 2: Precisely, it's a classic who watches the Watchmen problem. If 621 00:29:27,960 --> 00:29:31,200 you centralize all of this unprecedented power in the hands 622 00:29:31,200 --> 00:29:33,799 of the state in order to protect us, you are 623 00:29:33,839 --> 00:29:38,200 creating a single point of failure that is historically overwhelmingly 624 00:29:38,279 --> 00:29:39,200 prone to abuse. 625 00:29:39,400 --> 00:29:42,799 Speaker 1: Man, Okay, let's try to zoom out and put a 626 00:29:42,799 --> 00:29:45,880 bow on this, because we have covered a wild amount 627 00:29:45,920 --> 00:29:48,599 of ground in this thread, and the implications are just huge. 628 00:29:48,680 --> 00:29:50,599 It's a lot to process, for sure, So let's just 629 00:29:50,640 --> 00:29:53,200 recap the roadmap we just walked through. We started with 630 00:29:53,240 --> 00:29:55,799 the lights flickering on Earth. We hit the one tarawatte limit. 631 00:29:56,079 --> 00:29:57,119 We simply ran out. 632 00:29:56,960 --> 00:29:58,839 Speaker 2: Of juice, the terrestrial bottleneck. 633 00:29:59,079 --> 00:30:02,039 Speaker 1: So we looked up. We saw the constant sun and 634 00:30:02,079 --> 00:30:05,680 the perfect cold vacuum of space, and we decided to 635 00:30:05,759 --> 00:30:09,480 launch the entire Internet into orbit to access those hundreds 636 00:30:09,480 --> 00:30:10,759 of gigawatts of clean. 637 00:30:10,640 --> 00:30:12,039 Speaker 2: Power, the Great Migration. 638 00:30:12,319 --> 00:30:15,039 Speaker 1: Then with that massive new brain power, we created the 639 00:30:15,079 --> 00:30:17,960 digital humans. We automated the bits. We took the core 640 00:30:18,039 --> 00:30:20,960 output of companies like Apple and Google and Nvidia, and 641 00:30:21,000 --> 00:30:22,160 we automated. 642 00:30:21,640 --> 00:30:25,200 Speaker 2: It, unlocking trillions and trillions of dollars of economic value. 643 00:30:25,319 --> 00:30:28,640 Speaker 1: And then we gave those digital brains physical bodies. We 644 00:30:28,720 --> 00:30:32,240 built the Optimus robot. We unlocked the infinite money glitch. 645 00:30:32,480 --> 00:30:36,799 We started the recursive feedback loop of physical production, the end. 646 00:30:36,640 --> 00:30:40,200 Speaker 2: Of scarcity, the deflation of the entire physical world. 647 00:30:40,640 --> 00:30:44,359 Speaker 1: And finally, after all that, we find ourselves standing at 648 00:30:44,359 --> 00:30:48,279 a crossroads, looking at this godlike power, an army of 649 00:30:48,440 --> 00:30:52,640 tireless robots and infinite intelligence, and we have to decide 650 00:30:52,799 --> 00:30:53,680 who gets to hold the. 651 00:30:53,640 --> 00:30:56,319 Speaker 2: Leash, the CEO who wants to maximize profit, or the 652 00:30:56,319 --> 00:30:58,200 state that holds the monopoly on violence. 653 00:30:58,400 --> 00:31:00,440 Speaker 1: It's quite the journey, and I think the most striking 654 00:31:00,440 --> 00:31:03,559 thing about it all is the compression of time. This 655 00:31:03,640 --> 00:31:05,799 isn't a science fiction story, said in twenty two hundred, 656 00:31:05,880 --> 00:31:06,319 not at all. 657 00:31:06,440 --> 00:31:08,440 Speaker 2: The roadmap we just walked through, the one laid out 658 00:31:08,440 --> 00:31:12,000 in that conversation. It's largely focused on a five year horizon. 659 00:31:12,200 --> 00:31:15,279 Speaker 1: That for me is the real thrilling threads takeaway. The 660 00:31:15,319 --> 00:31:18,640 future isn't happening linearly anymore. It's not one step after another. 661 00:31:18,680 --> 00:31:21,680 It's happening exponentially. We are on the vertical part of 662 00:31:21,680 --> 00:31:22,440 that curve. 663 00:31:22,759 --> 00:31:26,519 Speaker 2: We are, and exponentials feel flat for a very long time. 664 00:31:26,519 --> 00:31:28,440 Nothing is happening, nothing is happening, and then all of 665 00:31:28,440 --> 00:31:31,880 a sudden they feel completely vertical. We are in the 666 00:31:32,000 --> 00:31:33,680 knee of that curve right now. 667 00:31:34,039 --> 00:31:36,759 Speaker 1: So that brings me to the big question, the provocative 668 00:31:36,759 --> 00:31:39,039 thought I want to leave you and everyone listening with. 669 00:31:39,319 --> 00:31:39,799 Speaker 2: Let's hear it. 670 00:31:40,240 --> 00:31:43,400 Speaker 1: Musk makes the claim that the government is the biggest 671 00:31:43,440 --> 00:31:47,720 threat to AI safety, not corporations. When you look at history, 672 00:31:47,960 --> 00:31:50,039 and when you look at the raw, absolute power of 673 00:31:50,079 --> 00:31:54,200 this technology, the ability to watch everyone, enforce anything and 674 00:31:54,279 --> 00:31:57,799 build everything. Do you trust the monopoly on violence more 675 00:31:58,279 --> 00:32:01,519 or the maximizers of profit to hold the keys to 676 00:32:01,559 --> 00:32:02,160 that kingdom? 677 00:32:02,319 --> 00:32:04,400 Speaker 2: That is the question. It's a terrible choice, isn't it? 678 00:32:04,559 --> 00:32:07,519 Or is there a third option we're missing entirely readymon 679 00:32:07,920 --> 00:32:11,519 Is there a way to decentralize this power, to distribute 680 00:32:11,519 --> 00:32:14,160 it so that neither the CEO nor the general gets 681 00:32:14,200 --> 00:32:15,759 to be the sole master of it. 682 00:32:15,759 --> 00:32:19,400 Speaker 1: A decentralized AI future. Yeah. That that is a thread 683 00:32:19,440 --> 00:32:20,079 for another day. 684 00:32:20,200 --> 00:32:21,680 Speaker 2: It might be the only way we get through this. 685 00:32:22,119 --> 00:32:26,119 Speaker 1: Listeners, we really want to know what you think. Drop 686 00:32:26,160 --> 00:32:29,559 a comment. Who do you trust less? Are you Team CORP, 687 00:32:29,720 --> 00:32:32,960 Team GOV? Or are you hoping for Team something else? 688 00:32:33,599 --> 00:32:36,319 Speaker 2: Let us know I am genuinely curious to see those answers. 689 00:32:36,400 --> 00:32:38,960 Speaker 1: Me too. Thanks for reveling this with us. We'll catch 690 00:32:39,039 --> 00:32:39,640 the next thread.