WEBVTT 00:00:00.000 --> 00:00:03.459 Imagine a world where the software on your screen 00:00:03.459 --> 00:00:06.639 isn't just mimicking conversation, but it's actually 00:00:06.639 --> 00:00:09.480 thinking. I mean, really contemplating a problem. 00:00:10.419 --> 00:00:12.580 Elon Musk recently dropped a statistic that, 00:00:12.599 --> 00:00:15.339 frankly, it just stops you in your tracks. He 00:00:15.339 --> 00:00:17.859 believes there is a 10 % probability, one in 00:00:17.859 --> 00:00:21.219 10, that his new model, Grok 5, will be as smart 00:00:21.219 --> 00:00:23.420 as a human being. And just to give you a sense 00:00:23.420 --> 00:00:26.989 of the gravity there, that single claim... was 00:00:26.989 --> 00:00:29.850 reportedly enough to trigger an emergency strategy 00:00:29.850 --> 00:00:32.789 meeting at OpenAI. Wow. They didn't just shrug 00:00:32.789 --> 00:00:35.350 it off as marketing hype. They hit the accelerator. 00:00:35.710 --> 00:00:37.570 Welcome back to the Deep Dive. Today, we are 00:00:37.570 --> 00:00:42.590 unpacking XAI, Grok 5, and this whole race to 00:00:42.590 --> 00:00:44.789 artificial general intelligence. It feels like 00:00:44.789 --> 00:00:46.770 everyone, there's a new model, a new claim, a 00:00:46.770 --> 00:00:49.109 new benchmark. But looking at the source material 00:00:49.109 --> 00:00:51.600 we have today, this one. It feels different. 00:00:51.859 --> 00:00:53.899 It does. It feels less like a software update 00:00:53.899 --> 00:00:57.320 and more like a philosophical shift in how we 00:00:57.320 --> 00:00:59.600 approach intelligence. It absolutely is. And 00:00:59.600 --> 00:01:02.619 today, we are going to map out exactly why. We 00:01:02.619 --> 00:01:04.319 have a lot of ground to cover. OK, so what's 00:01:04.319 --> 00:01:07.159 the plan? First, we're going to talk about Colossus, 00:01:07.540 --> 00:01:09.980 this massive hardware scaling happening right 00:01:09.980 --> 00:01:13.019 now in Memphis, Tennessee. Right. We'll break 00:01:13.019 --> 00:01:16.239 down the three specific ways Grok 5 is fundamentally 00:01:16.239 --> 00:01:19.079 different from Chat, GPT, and Gemini. We'll look 00:01:19.079 --> 00:01:22.560 at the architecture. the six trillion parameter 00:01:22.560 --> 00:01:26.000 structure aiming for AGI, and then we'll get 00:01:26.000 --> 00:01:29.760 practical coding markets, image generation. And 00:01:29.760 --> 00:01:31.459 of course, the part that always feels a bit like 00:01:31.459 --> 00:01:33.280 science fiction, the integration with Tesla and 00:01:33.280 --> 00:01:35.159 the optimist robots. Putting the brain inside 00:01:35.159 --> 00:01:37.099 the body. That's the end game. So let's start 00:01:37.099 --> 00:01:39.459 with the timeline, because in the world of Musk 00:01:39.459 --> 00:01:42.200 companies, timelines can be. Well, let's call 00:01:42.200 --> 00:01:44.780 them aspirational. For sure. But we have a confirmed 00:01:44.780 --> 00:01:48.260 window now. We do. We do. MUST confirmed in November 00:01:48.260 --> 00:01:52.019 2025 that the target release for Grok 5 is the 00:01:52.019 --> 00:01:54.700 first quarter of 2026. So we're looking at January, 00:01:54.739 --> 00:01:57.239 February, March. OK. And what's interesting is 00:01:57.239 --> 00:02:00.000 that despite the history of delays with rockets 00:02:00.000 --> 00:02:03.640 or cars, XAI seems incredibly confident about 00:02:03.640 --> 00:02:06.540 this one. So why the confidence? Usually when 00:02:06.540 --> 00:02:09.060 you get this close to a release of this magnitude, 00:02:09.460 --> 00:02:11.939 you start hearing about delays, but the sources 00:02:11.939 --> 00:02:14.120 suggest the hardware is already humming. Because 00:02:14.120 --> 00:02:16.599 it's not just code anymore, it's physics. This 00:02:16.599 --> 00:02:18.340 is where we have to talk about Colossus. Right, 00:02:18.379 --> 00:02:20.800 the supercomputer. It's the name of the supercomputer, 00:02:21.099 --> 00:02:24.159 XAI, built in Memphis. And when I say supercomputer, 00:02:24.300 --> 00:02:26.460 I don't mean a server room with blinking lights. 00:02:26.539 --> 00:02:29.340 I mean a facility that consumes as much electricity 00:02:29.340 --> 00:02:32.000 as a small city. A small city, just to train 00:02:32.000 --> 00:02:35.389 an AI. It's staggering. To get Grok 5 to that 00:02:35.389 --> 00:02:38.789 level of intelligence, they are moving from using 00:02:38.789 --> 00:02:42.770 100 ,000 GPUs, which is already a mind -bending 00:02:42.770 --> 00:02:46.050 amount of compute, to trying to connect 1 million 00:02:46.050 --> 00:02:48.389 GPUs. OK, so let's unpack that scale. I think 00:02:48.389 --> 00:02:51.250 for most people, 1 million GPUs is just a number. 00:02:51.770 --> 00:02:54.069 What does that actually represent in terms of 00:02:54.069 --> 00:02:57.009 engineering? Think of a GPU like a tiny brain 00:02:57.009 --> 00:03:00.680 or a neuron. Your home computer usually has one. 00:03:01.219 --> 00:03:03.020 Colossus is trying to link a million of them 00:03:03.020 --> 00:03:06.740 together to a single cohesive system. It is arguably 00:03:06.740 --> 00:03:09.139 one of the biggest engineering projects in history. 00:03:09.879 --> 00:03:12.939 It's like trying to wire up a synthetic neocortex. 00:03:13.039 --> 00:03:14.900 And from what I understand, the challenge isn't 00:03:14.900 --> 00:03:16.939 just plugging them in. It's the latency, right? 00:03:17.099 --> 00:03:19.740 The speed of communication between them. That 00:03:19.740 --> 00:03:22.099 is the bottleneck. It's a networking nightmare. 00:03:22.680 --> 00:03:25.240 If you have a brain where the neurons take five 00:03:25.240 --> 00:03:27.219 seconds to talk to each other, you don't have 00:03:27.219 --> 00:03:29.819 intelligence. You have a disconnected mess. You 00:03:29.819 --> 00:03:31.460 have to get these chips to talk to each other 00:03:31.460 --> 00:03:34.180 instantly, almost at the speed of light. That's 00:03:34.180 --> 00:03:37.319 why we have the delay until 2026. It's not about 00:03:37.319 --> 00:03:39.539 the software learning. It's about physically 00:03:39.539 --> 00:03:42.039 building the brain big enough to hold the thoughts. 00:03:42.280 --> 00:03:44.680 It reminds me of that paper, The Bitter Lesson 00:03:44.680 --> 00:03:47.500 by Rich Sutton. The idea that we keep trying 00:03:47.500 --> 00:03:49.780 to be clever with programming, but eventually 00:03:49.780 --> 00:03:52.180 brute force scaling, just adding more compute, 00:03:52.460 --> 00:03:56.180 always wins. That's the bet XAI is making. They 00:03:56.180 --> 00:03:58.599 are betting the farm on the idea that if you 00:03:58.599 --> 00:04:01.039 build a big enough computer, intelligence just 00:04:01.039 --> 00:04:04.340 emerges. It's not magic, it's scale. So why do 00:04:04.340 --> 00:04:07.280 we actually need a million ships linked together? 00:04:07.360 --> 00:04:09.460 To create an intelligence smarter than anything 00:04:09.460 --> 00:04:11.840 previously seen. That leads us perfectly into 00:04:11.840 --> 00:04:14.699 the software itself. Because once you build this 00:04:14.699 --> 00:04:18.180 massive brain, what makes it different from what 00:04:18.180 --> 00:04:21.300 we already have? If I'm sitting there with ChatGPT 00:04:21.300 --> 00:04:23.819 open in one tab and Grok 5 in another, what's 00:04:23.819 --> 00:04:26.240 the tangible difference? There are a few core 00:04:26.240 --> 00:04:28.899 differentiators, but the biggest one is time. 00:04:29.730 --> 00:04:33.509 Most AI models, like Chat, GPT, or Gemini, operate 00:04:33.509 --> 00:04:36.529 like a library book. A library book? How so? 00:04:36.610 --> 00:04:38.649 Well, think about it. If you pick up a book printed 00:04:38.649 --> 00:04:41.649 in 2023, it has zero idea what happened yesterday. 00:04:41.689 --> 00:04:44.170 Right. It's frozen. It's frozen in time. Those 00:04:44.170 --> 00:04:46.750 models rely on training data that has a cutoff 00:04:46.750 --> 00:04:48.910 date. Now, they are getting better at searching 00:04:48.910 --> 00:04:51.550 Google to patch that hole, but it's often slow. 00:04:51.720 --> 00:04:54.160 clunky, or a bit messy. I have noticed that you 00:04:54.160 --> 00:04:56.019 ask about a news event from this morning and 00:04:56.019 --> 00:04:58.420 it sometimes hallucinates or just says, I don't 00:04:58.420 --> 00:05:01.199 have that information. Precisely. Grok 5 is different 00:05:01.199 --> 00:05:04.060 because it lives inside the X platform. It has 00:05:04.060 --> 00:05:06.160 a direct pipeline to the global conversation. 00:05:06.300 --> 00:05:09.019 So it's reading posts in real time. Instantly. 00:05:09.339 --> 00:05:12.199 When news breaks, an earthquake, a stock market 00:05:12.199 --> 00:05:14.879 crash, a political scandal, people posted on 00:05:14.879 --> 00:05:18.889 X instantly. Grok reads that. It doesn't wait 00:05:18.889 --> 00:05:21.250 for a news article to be written and edited and 00:05:21.250 --> 00:05:23.810 published and then indexed by Google. It sees 00:05:23.810 --> 00:05:26.649 the raw data as it happens. That is a massive 00:05:26.649 --> 00:05:28.649 advantage for anyone who needs to be current. 00:05:28.790 --> 00:05:30.990 It's the difference between a historian and a 00:05:30.990 --> 00:05:33.689 news anchor. And it's drinking from the fire 00:05:33.689 --> 00:05:35.759 hose. It has to filter the noise, obviously, 00:05:36.279 --> 00:05:37.759 but the signal is there before it's anywhere 00:05:37.759 --> 00:05:39.339 else. But there's another difference that people 00:05:39.339 --> 00:05:41.500 talk about a lot, and that's the personality, 00:05:41.819 --> 00:05:44.519 the uncensored aspect. Root -seeking mode, yeah. 00:05:44.600 --> 00:05:47.259 Yeah. And I have to admit, I sometimes feel this 00:05:47.259 --> 00:05:49.779 frustration. I feel like I wrestle with other 00:05:49.779 --> 00:05:51.839 AIs when I ask a slightly sensitive question. 00:05:51.920 --> 00:05:55.160 You know the response. As an AI lineage model, 00:05:55.399 --> 00:05:58.579 I cannot provide an opinion. It feels very corporate. 00:05:58.639 --> 00:06:02.370 It feels very safe, and that is by design. OpenAI 00:06:02.370 --> 00:06:04.810 and Google have a massive enterprise reputation 00:06:04.810 --> 00:06:06.610 to protect. They have shareholders. They have 00:06:06.610 --> 00:06:08.769 corporate clients. They optimize for safety. 00:06:09.230 --> 00:06:12.529 XAI is taking a different approach. Croc 5 is 00:06:12.529 --> 00:06:15.110 designed to be less strict. It treats the user 00:06:15.110 --> 00:06:18.370 like an adult. So fun mode isn't just a gimmick. 00:06:18.629 --> 00:06:20.829 No, it's a philosophy. Yeah. It allows the model 00:06:20.829 --> 00:06:23.949 to be sarcastic, to make jokes, but more importantly, 00:06:24.370 --> 00:06:27.029 to give answers that might be considered controversial 00:06:27.029 --> 00:06:30.089 or sensitive without immediately shutting down. 00:06:30.199 --> 00:06:33.399 It's trying to provide the raw truth. Or at least 00:06:33.399 --> 00:06:35.879 the available information rather than a sanitized 00:06:35.879 --> 00:06:38.720 version of it. So is being uncensored actually 00:06:38.720 --> 00:06:41.680 a utility though, or is it just a gimmick? It's 00:06:41.680 --> 00:06:44.500 essential for users wanting raw truth over corporate 00:06:44.500 --> 00:06:47.100 safety. I can see that. If you are a researcher 00:06:47.100 --> 00:06:50.079 or a journalist, you don't want a tool that filters 00:06:50.079 --> 00:06:52.800 reality. You want the data, even if it's ugly. 00:06:53.019 --> 00:06:54.560 Exactly. You want to know what the sentiment 00:06:54.560 --> 00:06:57.199 actually is, not what the AI thinks the sentiment 00:06:57.199 --> 00:06:59.100 should be. That makes sense. It's a tool for 00:06:59.100 --> 00:07:01.879 deep thinking, not just, you know, drafting HR 00:07:01.879 --> 00:07:04.379 emails. And speaking of deep thinking, we have 00:07:04.379 --> 00:07:08.220 to talk about the Holy Grail, AGI, artificial 00:07:08.220 --> 00:07:10.680 general intelligence. This is the heavy terminology. 00:07:10.819 --> 00:07:13.199 Let's define it quickly. So narrow AI is good 00:07:13.199 --> 00:07:15.920 at one thing, playing chess, recognizing faces. 00:07:16.439 --> 00:07:19.000 AGI is good at everything. It learns like a human. 00:07:19.100 --> 00:07:23.160 Right. And musks. 10 % probability claim is about 00:07:23.160 --> 00:07:26.500 this specific threshold. Can't Grok 5 look at 00:07:26.500 --> 00:07:28.819 a problem it has never seen before, no training 00:07:28.819 --> 00:07:31.500 data on it, and just figure out the solution 00:07:31.500 --> 00:07:34.240 using logic? And how do they plan to get there? 00:07:34.439 --> 00:07:36.920 We talked about the chips, but what is the strategy? 00:07:37.240 --> 00:07:39.970 The strategy is brute force scaling. Imagine 00:07:39.970 --> 00:07:41.730 you're trying to build a library that contains 00:07:41.730 --> 00:07:43.930 the answer to every question. You just keep adding 00:07:43.930 --> 00:07:46.790 floors. You just keep adding floors. In AI, we 00:07:46.790 --> 00:07:48.889 measure this in parameters. And parameters are 00:07:48.889 --> 00:07:51.670 like the synapses in the brain, right? The connections. 00:07:51.910 --> 00:07:54.089 Yeah, exactly. To give you a sense of scale, 00:07:54.550 --> 00:07:57.990 GPT -4 is estimated to have around 1 .7 trillion 00:07:57.990 --> 00:08:01.459 parameters. 1 .7 trillion. Grok 5 is expected 00:08:01.459 --> 00:08:04.160 to have around 6 trillion parameter. Oh, that 00:08:04.160 --> 00:08:06.560 is a massive jump. That's not just a linear increase. 00:08:06.759 --> 00:08:09.360 That's exponential complexity. It is. And the 00:08:09.360 --> 00:08:12.579 theory the BetX AI is making is that by making 00:08:12.579 --> 00:08:14.660 the brain significantly bigger and feeding it 00:08:14.660 --> 00:08:17.259 more information, intelligence will naturally 00:08:17.259 --> 00:08:19.360 emerge. So they're not programming it to be smart. 00:08:19.759 --> 00:08:22.300 No, they're building a structure complex enough 00:08:22.300 --> 00:08:24.899 for smartness to happen. But does making the 00:08:24.899 --> 00:08:28.000 model bigger actually create understanding? XAI 00:08:28.000 --> 00:08:30.439 believes intelligence naturally emerges from 00:08:30.439 --> 00:08:32.679 massive scale. And they are focusing heavily 00:08:32.679 --> 00:08:36.759 on reasoning versus just autocomplete. What's 00:08:36.759 --> 00:08:40.779 the distinction there? Most current AIs are essentially 00:08:40.779 --> 00:08:43.059 guessing the next word in a sentence. They're 00:08:43.059 --> 00:08:45.679 hyper -advanced autocomplete. Right. Grok 5 is 00:08:45.679 --> 00:08:49.519 being trained to stop think, and check its logic. 00:08:50.039 --> 00:08:52.259 If you ask it a math problem, it doesn't just 00:08:52.259 --> 00:08:54.019 guess the answer. It works through the steps. 00:08:54.559 --> 00:08:57.440 If you ask for code, it checks the logic of that 00:08:57.440 --> 00:08:59.159 code before giving it to you. It's the difference 00:08:59.159 --> 00:09:02.080 between memorizing a textbook and actually understanding 00:09:02.080 --> 00:09:05.139 the physics. Precisely. It's simulating an internal 00:09:05.139 --> 00:09:07.019 monologue. It's what researchers call a chain 00:09:07.019 --> 00:09:09.259 of thought. And if they pull it off at six trillion 00:09:09.259 --> 00:09:12.059 parameters, that internal monologue gets very 00:09:12.059 --> 00:09:15.019 sophisticated. Okay, so we have this massive, 00:09:15.259 --> 00:09:17.740 sarcastic, truth -seeking brain being built in 00:09:17.740 --> 00:09:20.480 Memphis, but let's bring it down to Earth. For 00:09:20.480 --> 00:09:22.259 the person listening who isn't trying to solve 00:09:22.259 --> 00:09:25.039 the theory of relativity, how do I actually use 00:09:25.039 --> 00:09:27.960 this? It comes down to a few very powerful use 00:09:27.960 --> 00:09:30.799 cases. The first is handling large documents. 00:09:31.360 --> 00:09:33.879 Grok 5 is going to have a massive context window, 00:09:34.299 --> 00:09:37.940 likely around 256 ,000 tokens. Which translates 00:09:37.940 --> 00:09:41.440 to what in the real world? book. A very thick 00:09:41.440 --> 00:09:44.120 book. Yeah. Or a massive stack of legal contracts. 00:09:44.419 --> 00:09:47.299 You could upload a 300 page PDF and say, find 00:09:47.299 --> 00:09:49.580 every mention of financial risk in this contract 00:09:49.580 --> 00:09:51.779 and tell me if any of them contradict each other. 00:09:51.980 --> 00:09:54.639 That's huge because right now with some models 00:09:54.639 --> 00:09:56.860 you have to chop the document up into pieces 00:09:56.860 --> 00:09:59.240 or you hit a limit. Right. Grok 5 can hold the 00:09:59.240 --> 00:10:02.120 whole thing in memory at once. It sees the big 00:10:02.120 --> 00:10:04.399 picture. It understands the narrative arc of 00:10:04.399 --> 00:10:07.009 the data. not just isolated sentences. And what 00:10:07.009 --> 00:10:08.529 about the market research side? You mentioned 00:10:08.529 --> 00:10:10.929 the X integration earlier. This is a game changer 00:10:10.929 --> 00:10:13.610 for businesses. Imagine asking, what is the sentiment 00:10:13.610 --> 00:10:16.730 on X right now regarding the Apple launch? Instead 00:10:16.730 --> 00:10:18.529 of getting a summary of articles from yesterday, 00:10:19.070 --> 00:10:21.830 Grok 5 scans the live posts from the last hour. 00:10:22.250 --> 00:10:24.429 Wow. It can tell you people are complaining about 00:10:24.429 --> 00:10:27.389 the battery life or everyone loves the new camera. 00:10:27.669 --> 00:10:30.210 It's real time market intelligence. I can see 00:10:30.210 --> 00:10:32.730 that being addictive for investors. You're getting 00:10:32.730 --> 00:10:34.990 the signal before the market moves. Absolutely. 00:10:35.470 --> 00:10:37.590 And then there's coding. If you're a developer, 00:10:37.789 --> 00:10:41.230 Grok 5 acts like a senior engineer. Because of 00:10:41.230 --> 00:10:43.889 that large memory, it can look at your entire 00:10:43.889 --> 00:10:46.509 project structure, not just one script. So you 00:10:46.509 --> 00:10:48.409 could ask it to rewrite something for speed. 00:10:48.950 --> 00:10:51.009 Exactly. You could say, rewrite this Python script 00:10:51.009 --> 00:10:53.610 for speed. And it understands the context of 00:10:53.610 --> 00:10:56.679 the whole application. And finally, images. I 00:10:56.679 --> 00:11:00.019 saw a note about Grok Imagine. Yes. The struggle 00:11:00.019 --> 00:11:03.080 with AI images has always been text. You ask 00:11:03.080 --> 00:11:05.460 for a picture of a shop with a sign, and the 00:11:05.460 --> 00:11:08.120 sign usually has alien hieroglyphics on it. Yeah, 00:11:08.299 --> 00:11:10.559 just jumbled letters. Grok 5 understands the 00:11:10.559 --> 00:11:13.059 world better, so it can actually spell correctly 00:11:13.059 --> 00:11:15.970 inside the image. You can ask for a futuristic 00:11:15.970 --> 00:11:18.950 Tokyo Cafe with a neon sign that says Grok Cafe, 00:11:19.350 --> 00:11:22.169 and it will actually spell Grok Cafe. Which sounds 00:11:22.169 --> 00:11:24.570 simple, but is technically very difficult. It 00:11:24.570 --> 00:11:26.769 implies the model understands what letters are. 00:11:27.129 --> 00:11:29.629 Conceptually, yes. Not just visually. So which 00:11:29.629 --> 00:11:31.870 user actually benefits the most from this tool? 00:11:32.250 --> 00:11:34.330 Investors and researchers needing real -time 00:11:34.330 --> 00:11:37.129 data and logic. Before we move to the final segment, 00:11:37.169 --> 00:11:39.149 I want to take a brief moment to thank our partners 00:11:39.149 --> 00:11:42.480 who make this deep dive possible. Okay, we're 00:11:42.480 --> 00:11:45.019 back. We've covered the brain. Now I want to 00:11:45.019 --> 00:11:47.659 talk about the body. Because Elon Musk isn't 00:11:47.659 --> 00:11:49.620 just building a chatbot to live in your browser. 00:11:50.259 --> 00:11:53.159 He's building a mind that can move. This is where 00:11:53.159 --> 00:11:55.019 we connect to the bigger picture of the Musk 00:11:55.019 --> 00:11:58.580 ecosystem. The first physical container for Grok 00:11:58.580 --> 00:12:03.340 5 is the car. The Tesla! Yes. Right now, voice 00:12:03.340 --> 00:12:07.539 commands in cars are... let's be honest, pretty 00:12:07.539 --> 00:12:11.289 basic. Call mom. Navigate home. I didn't understand 00:12:11.289 --> 00:12:14.250 that. Playing music by Cher. Exactly. It's frustrating. 00:12:14.710 --> 00:12:17.289 With Grok 5 integrated, the car becomes a smart 00:12:17.289 --> 00:12:19.250 companion. So what does that look like? You could 00:12:19.250 --> 00:12:20.769 be driving through a town you've never visited 00:12:20.769 --> 00:12:22.490 and ask, Grok, tell me a story about the history 00:12:22.490 --> 00:12:24.590 of this town. Or find me a restaurant with a 00:12:24.590 --> 00:12:26.870 4 .5 star rating that is open right now and actually 00:12:26.870 --> 00:12:29.929 has parking. It turns the drive into a conversation. 00:12:30.429 --> 00:12:32.509 But it's more than just a concierge, isn't it? 00:12:32.730 --> 00:12:35.409 It's about the car understanding the world. It 00:12:35.409 --> 00:12:38.929 is. But the even wilder application is Optimus, 00:12:39.289 --> 00:12:41.610 the humanoid robot. This is the part that really 00:12:41.610 --> 00:12:44.049 captures the imagination. We've seen the videos 00:12:44.049 --> 00:12:46.190 of the robots folding shirts or walking around, 00:12:46.610 --> 00:12:49.970 but they still look a bit... robotic. It's stiff, 00:12:50.450 --> 00:12:53.330 because currently they are dumb in the cognitive 00:12:53.330 --> 00:12:55.750 sense. They follow programming. They're executing 00:12:55.750 --> 00:12:58.929 a script. Got it. But if you put Grok 5's brain 00:12:58.929 --> 00:13:02.049 into Optimus, the robot can understand abstract 00:13:02.049 --> 00:13:05.039 concepts. You could say, go to the kitchen and 00:13:05.039 --> 00:13:07.460 make a sandwich. And the robot has to understand 00:13:07.460 --> 00:13:09.519 what a kitchen is, what a sandwich is, where 00:13:09.519 --> 00:13:12.299 the bread is kept. It has to understand the concept 00:13:12.299 --> 00:13:15.120 of a knife. Not just the coordinates of a tool, 00:13:15.539 --> 00:13:18.100 but what a knife is for. It figures that out 00:13:18.100 --> 00:13:21.100 by thinking, using that massive training data. 00:13:21.279 --> 00:13:23.659 It provides the common sense that robots have 00:13:23.659 --> 00:13:26.120 always lacked. Exactly. And where does it get 00:13:26.120 --> 00:13:27.899 that common sense? It gets it from the cars, 00:13:27.980 --> 00:13:30.220 right? From the fleet. That's the closed loop. 00:13:30.330 --> 00:13:33.169 This is the secret weapon XAI has. They have 00:13:33.169 --> 00:13:35.370 millions of Teslas driving around with cameras, 00:13:35.710 --> 00:13:38.190 recording video of physics, of human behavior, 00:13:38.409 --> 00:13:41.370 of how the world moves. That data trains the 00:13:41.370 --> 00:13:44.070 model. The car teaches the robot how to see. 00:13:44.350 --> 00:13:46.769 It's incredible potential. But I have to play 00:13:46.769 --> 00:13:49.570 the skeptic for a second. We talked about colossus 00:13:49.570 --> 00:13:52.210 consuming the power of a small city. Yes. If 00:13:52.210 --> 00:13:54.929 we have millions of these robots and cars all 00:13:54.929 --> 00:13:58.440 accessing this super intelligence. The energy 00:13:58.440 --> 00:14:01.879 demand must be astronomical. Is a massive challenge. 00:14:02.179 --> 00:14:04.340 This is the dark side of the moon here. The more 00:14:04.340 --> 00:14:06.340 intelligent the model, the more energy it consumes. 00:14:07.059 --> 00:14:09.860 Colossus in Memphis is already straining the 00:14:09.860 --> 00:14:12.080 grid. So what's the solution? Musk is betting 00:14:12.080 --> 00:14:15.799 on solar and batteries, Tesla energy, to solve 00:14:15.799 --> 00:14:19.149 this. But it's a race. Can we generate green 00:14:19.149 --> 00:14:22.149 energy faster than these AI brains consume it? 00:14:22.289 --> 00:14:25.509 So is this just software or something physically 00:14:25.509 --> 00:14:28.490 tangible? It is the future mind for cars and 00:14:28.490 --> 00:14:30.909 humanoid robots. It really reframes how you look 00:14:30.909 --> 00:14:32.710 at a Tesla driving down the street. It's just 00:14:32.710 --> 00:14:35.049 waiting for its brain upgrade. It is. It's hardware 00:14:35.049 --> 00:14:37.350 waiting for the soul. So let's bring this all 00:14:37.350 --> 00:14:38.929 home. We've covered a huge amount of ground. 00:14:39.129 --> 00:14:41.330 We have these three giants in the arena now. 00:14:41.350 --> 00:14:43.419 We do. And it's important to see where they sit. 00:14:43.639 --> 00:14:46.019 You have OpenAI with GPT -5 coming, which is 00:14:46.019 --> 00:14:48.659 the enterprise standard. It's safe. It's corporate. 00:14:48.860 --> 00:14:51.919 It's reliable. The suit and tie. Exactly. Then 00:14:51.919 --> 00:14:54.059 you have Google with Gemini. That's the ecosystem 00:14:54.059 --> 00:14:56.720 play. It's integrated into your Docs, your Gmail, 00:14:56.799 --> 00:14:59.039 your Android phone. It's all about convenience. 00:14:59.200 --> 00:15:03.580 And then XAI with Grok. The truth seeker. It's 00:15:03.580 --> 00:15:06.980 about raw power, real time access to the world 00:15:06.980 --> 00:15:10.179 via X, and deep reasoning without the safety 00:15:10.179 --> 00:15:12.700 wheels. Regardless of which one wins, or if they 00:15:12.700 --> 00:15:14.480 all win in different ways, what does this mean 00:15:14.480 --> 00:15:17.259 for us? It forces the whole industry to evolve. 00:15:17.480 --> 00:15:20.480 Even if Grok 5 doesn't hit that 10 % chance of 00:15:20.480 --> 00:15:23.919 being human level smart, just the attempt pushes 00:15:23.919 --> 00:15:26.539 OpenAI and Google to be better. It accelerates 00:15:26.539 --> 00:15:28.779 everything. So here's my challenge to you, the 00:15:28.779 --> 00:15:31.580 listener. We are expecting this release in Q1 00:15:31.580 --> 00:15:34.879 2026. When Grok 5 drops, don't just read the 00:15:34.879 --> 00:15:37.779 headlines. Test it. Absolutely. Compare it. Ask 00:15:37.779 --> 00:15:40.340 ChatGPT a question, then ask Grok the same question. 00:15:40.759 --> 00:15:42.600 See which reasoning style fits your brain better. 00:15:42.960 --> 00:15:45.200 Are you looking for the safe summary or the raw 00:15:45.200 --> 00:15:47.399 data? It's going to be a fascinating year. The 00:15:47.399 --> 00:15:49.360 car you drive and the robot you might one day 00:15:49.360 --> 00:15:51.559 own are just waiting for this mind to wake up. 00:15:51.620 --> 00:15:53.539 And when it does, we'll be here to talk about 00:15:53.539 --> 00:15:55.379 it. I can't wait. Thanks for diving in with us. 00:15:55.460 --> 00:15:56.279 We'll catch you in the next one.