WEBVTT 1 00:00:00.760 --> 00:00:07.759 Which is a great segue. Optimists is I think going 2 00:00:07.839 --> 00:00:13.480 to be the greatest product in the history of humanity. 3 00:00:14.439 --> 00:00:17.679 What's the progress like and how much of your how 4 00:00:17.679 --> 00:00:20.760 many of your cycles are going specifically to Optimists. What's 5 00:00:20.800 --> 00:00:24.000 the timeline? I think you're on version three maybe four? 6 00:00:24.359 --> 00:00:25.120 Tell us everything? 7 00:00:27.559 --> 00:00:32.719 Well, yeah, everything will gonna take a long time. We've 8 00:00:32.759 --> 00:00:33.159 got time. 9 00:00:38.280 --> 00:00:44.359 We're finalizing the design of Optimist version three, and. 10 00:00:46.799 --> 00:00:49.200 That really is going to be a very re multiple robot. 11 00:00:51.960 --> 00:00:55.759 It will have the essentially the manual dexterity of a 12 00:00:55.840 --> 00:01:01.399 human It's meaning a very complex hand the an AI 13 00:01:01.560 --> 00:01:08.239 mind that can navigate and comprehend reality, and we made 14 00:01:08.280 --> 00:01:11.359 in very high volume. Those are the three things that 15 00:01:11.400 --> 00:01:15.480 I'm missing, Like if you see any other robotics company, 16 00:01:15.760 --> 00:01:19.079 they're missing those three things. Those are the three really 17 00:01:19.120 --> 00:01:23.079 hard things. 18 00:01:23.239 --> 00:01:27.239 And I spend Actually, at this point. 19 00:01:32.519 --> 00:01:35.560 It might be more of my mental cycles than anything 20 00:01:35.640 --> 00:01:40.280 anything else, any other single thing on Optimis. That's that 21 00:01:40.439 --> 00:01:46.079 that's solving pool real world AI all of the electro 22 00:01:46.239 --> 00:01:51.640 mechanical issues of Optimists, the supply chain and production challenges 23 00:01:51.640 --> 00:01:54.159 of it, because there is no supply chain that exists 24 00:01:54.200 --> 00:01:56.439 for humanoid robots, so it has to be we have 25 00:01:56.480 --> 00:02:00.120 to recreate it from scratch, and which requires doing a 26 00:02:00.120 --> 00:02:06.560 lot of vertical integration. None of the actuators in Optimism 27 00:02:06.719 --> 00:02:13.120 are available from an existing supply chain. But I think 28 00:02:13.159 --> 00:02:16.520 it is accurate to say that if successful, Optimus will 29 00:02:16.560 --> 00:02:20.759 be the biggest product ever and the. 30 00:02:20.680 --> 00:02:24.280 Cost of it at scale twenty thirty forty thousand dollars 31 00:02:24.319 --> 00:02:27.800 a robot. What do you think the first wave of 32 00:02:27.840 --> 00:02:32.360 them will cost and when will we be able to 33 00:02:32.560 --> 00:02:34.319 buy one to work on the ranch? 34 00:02:37.479 --> 00:02:41.120 I think that the Martinmall cost of production once you 35 00:02:41.240 --> 00:02:46.400 hit a million units per year is probably around the 36 00:02:46.560 --> 00:02:51.800 twenty thousand dollars range. It so of depends on how 37 00:02:51.879 --> 00:02:56.439 much you spend on the AI chip in the robot 38 00:03:00.080 --> 00:03:03.599 to achieve a lot of efficiencies in the actuators. There 39 00:03:03.639 --> 00:03:09.919 are twenty six actuators per arm, like twenty six like motors, 40 00:03:09.960 --> 00:03:11.960 gearboxes and par electronics. 41 00:03:16.719 --> 00:03:21.280 So, but the AI chip will be pretty expensive. 42 00:03:21.280 --> 00:03:23.199 But like that, that might be like five, five or 43 00:03:23.240 --> 00:03:26.319 six thousand dollars of the of the bill of materials, 44 00:03:26.319 --> 00:03:27.080 maybe more. 45 00:03:29.080 --> 00:03:33.879 And so but I think at volume. 46 00:03:33.960 --> 00:03:37.360 At a million units a year, the production cost is 47 00:03:37.400 --> 00:03:40.360 probably in the order of twenty thousand dollars maybe twenty 48 00:03:40.400 --> 00:03:44.120 five something like that, and it price will be as 49 00:03:44.159 --> 00:03:45.000 a function of demand. 50 00:03:45.319 --> 00:03:50.000 Lelon, can you maybe explain to everybody why the hand 51 00:03:50.120 --> 00:03:53.159 is so important to get right and why you know, 52 00:03:53.240 --> 00:03:56.439 the actuator design is so unique, and you know why 53 00:03:56.479 --> 00:03:59.120 it's so difficult, why nobody makes it, and why you 54 00:03:59.120 --> 00:04:01.039 have to start there almost to build the rest of 55 00:04:01.080 --> 00:04:02.840 the robot properly. 56 00:04:05.080 --> 00:04:09.080 It turns out human hands are incredibly that they've evolved 57 00:04:09.120 --> 00:04:13.120 to to be this incredibly sophisticated machine like that. Your 58 00:04:13.120 --> 00:04:19.079 hand is no actually a remarkable thing. It's look look 59 00:04:19.120 --> 00:04:23.480 closely at your hands and and think of all the 60 00:04:23.480 --> 00:04:25.439 things you can do with your hands, which is a lot. 61 00:04:28.480 --> 00:04:32.680 Just thinking about something. Your hands a very versatile instrument. 62 00:04:33.240 --> 00:04:34.920 Yeah, you can give them a high five. 63 00:04:36.639 --> 00:04:37.399 Very versatile. 64 00:04:39.720 --> 00:04:42.639 You know, you can swing a baseball bat, you can 65 00:04:43.240 --> 00:04:47.439 thread needles, you put thread in a needle, you can 66 00:04:48.160 --> 00:04:52.600 play the cannon with violin. You know, you could disassemble 67 00:04:52.680 --> 00:05:00.040 or assemble a car. The hands are incredibly versatile instruments. 68 00:05:00.120 --> 00:05:04.120 And most of the muscles of the hand are actually 69 00:05:04.279 --> 00:05:07.839 in the forearm, so your hand is kind of like 70 00:05:07.879 --> 00:05:10.360 a like a like it's like a puppet like it's 71 00:05:10.360 --> 00:05:11.040 mostly a puppet. 72 00:05:11.040 --> 00:05:11.399 The muscle. 73 00:05:11.399 --> 00:05:13.519 The muscles are coming from the forearm and they're pulling 74 00:05:13.519 --> 00:05:18.439 the tendons, which are you know, also human tendon desizing 75 00:05:18.879 --> 00:05:23.360 or human human tendent evolution is incredibly good. 76 00:05:23.560 --> 00:05:29.439 So you've got this web of tendons. You've got I 77 00:05:29.439 --> 00:05:30.399 think I think that. 78 00:05:30.600 --> 00:05:32.879 Human hand is something like depending on how you counted, 79 00:05:32.920 --> 00:05:35.639 twenty seven or twenty eight degrees of freedom per you know, 80 00:05:36.120 --> 00:05:42.639 in the hand, it's uh, it's amazing. So in order 81 00:05:42.639 --> 00:05:47.839 to create a robot that can be a generalized, uh, humanoid, 82 00:05:47.959 --> 00:05:52.000 you must solve the hand hands problem. 83 00:05:51.199 --> 00:05:55.879 And we had just got hands, knees, hands, and so is. 84 00:05:55.839 --> 00:05:58.800 It like when you were first building Tesla where the 85 00:05:58.839 --> 00:06:00.920 supply chain doesn't exist, So now you have to go 86 00:06:00.959 --> 00:06:03.240 out and find folks to work with and you know, 87 00:06:03.360 --> 00:06:05.759 build all this vertical integration, get support. 88 00:06:05.920 --> 00:06:06.160 Is it? 89 00:06:06.240 --> 00:06:08.600 Is it literally like it's just nowhere to be found 90 00:06:08.680 --> 00:06:10.519 and you're going to have to build all of this 91 00:06:10.560 --> 00:06:11.079 stuff up. 92 00:06:12.079 --> 00:06:15.040 Yes, we could not actually buy the actuators for any 93 00:06:15.040 --> 00:06:17.439 amount of money. They simply didn't exist, even though there 94 00:06:17.560 --> 00:06:20.639 are ten twenty thousand electric. 95 00:06:20.360 --> 00:06:21.360 Motors out there. 96 00:06:21.480 --> 00:06:27.160 Versus sizes and shapes, we've had to design every electric motor, 97 00:06:27.240 --> 00:06:31.759 gearbox and the controlling electronics from scratch, basically from physics 98 00:06:31.800 --> 00:06:32.759 post principles. 99 00:06:33.439 --> 00:06:35.800 The good news is you've got a lot of experience 100 00:06:36.040 --> 00:06:39.839 with factories over the last couple of decades. So how 101 00:06:39.959 --> 00:06:46.319 challenging is this versus cyber truck, Model YDEL, gigafactory, you 102 00:06:46.319 --> 00:06:50.279 know the Yeah, the Faberge egg known as the Model X. 103 00:06:50.680 --> 00:06:57.279 Yeah, right, it's harder than any any of those things. 104 00:06:57.480 --> 00:07:02.800 Okay, yeah much hart significantly am Yes. 105 00:07:02.839 --> 00:07:06.839 Well, there not hot starships harder. 106 00:07:08.439 --> 00:07:12.120 Okay, So somewhere between a Model X and a starship. 107 00:07:13.319 --> 00:07:16.560 Is it is the what's harder the hardware or the software? 108 00:07:19.680 --> 00:07:23.959 Right now we're struggling with the the final design of 109 00:07:23.959 --> 00:07:27.120 the hardware. Like I said, it's really primarily the hand, 110 00:07:27.600 --> 00:07:30.439 not to just dismiss the rest of the robot. Rest 111 00:07:30.439 --> 00:07:34.160 of it's also important, but that the hands are hands 112 00:07:34.160 --> 00:07:37.079 and inclusive of the forum, are a majority of the 113 00:07:37.079 --> 00:07:38.959 engineering difficulty of the entire robot. 114 00:07:39.519 --> 00:07:42.040 And then let's let's assume you get past the hardware challenges. 115 00:07:42.160 --> 00:07:44.920 How much do you sort of get for free based 116 00:07:44.959 --> 00:07:47.439 on all the progress that's happening with l ll ms. 117 00:07:47.959 --> 00:07:51.160 Will you know we'll consumers just be able to interact 118 00:07:51.160 --> 00:07:53.120 with this. Talk to the robot, ask you to do 119 00:07:53.240 --> 00:07:54.360 things and we'll understand. 120 00:07:54.759 --> 00:07:57.720 And so yeah, yeah, no problem. 121 00:07:58.120 --> 00:08:00.000 You're spending a lot of time with Annie. I know 122 00:08:00.040 --> 00:08:02.639 noticed online not that long. 123 00:08:03.800 --> 00:08:06.560 Maybe I went a little over the top promoting rock Imagine. 124 00:08:07.480 --> 00:08:13.399 Well, but in all seriousness, those characters and these robots 125 00:08:13.399 --> 00:08:16.199 that seems to be you know, like maybe. 126 00:08:15.920 --> 00:08:17.800 They you could get the embodiments of ani as. 127 00:08:18.600 --> 00:08:22.160 Yeah, why the human form factor? Elon, you could make 128 00:08:22.199 --> 00:08:25.439 something that's maybe better than a human, or maybe simpler 129 00:08:25.439 --> 00:08:27.680 than a human to do specific tasks, and maybe better 130 00:08:27.759 --> 00:08:29.600 than a human to do more things than a human 131 00:08:29.639 --> 00:08:32.039 can do. How do you decide to make it just 132 00:08:32.200 --> 00:08:32.720 like a human? 133 00:08:35.000 --> 00:08:36.600 Well, if you wanted to do all the things that 134 00:08:36.639 --> 00:08:38.200 a human can do, it turns out you need a 135 00:08:38.279 --> 00:08:43.120 humanoid robot. So if you wanted to do a subset, 136 00:08:44.000 --> 00:08:51.559 that's much easier. But it turns out humans evolved to 137 00:08:52.039 --> 00:08:59.960 the shape and capabilities that we have for good reasons. 138 00:09:00.440 --> 00:09:04.919 That actually is that there is like this value to 139 00:09:04.960 --> 00:09:09.799 having quiet you four fingers in the thumb, and even 140 00:09:09.799 --> 00:09:11.840 the pinky actually is quite useful. 141 00:09:13.519 --> 00:09:15.879 Tows are much more question walk but. 142 00:09:19.399 --> 00:09:20.480 Well also humans. 143 00:09:20.679 --> 00:09:23.399 Humans have designed the world as well, so we designed 144 00:09:23.440 --> 00:09:26.440 it for us. Exactly if we could make a humanoid robot, 145 00:09:26.480 --> 00:09:28.799 it'll be immediately backwards compatible with what we've. 146 00:09:28.600 --> 00:09:29.240 Built the world for. 147 00:09:30.519 --> 00:09:34.759 There's another there's another part of the robot. So there's 148 00:09:35.000 --> 00:09:37.559 the l l MS, there's the actuation in the hands, 149 00:09:38.399 --> 00:09:43.080 but also there's the the silicon that runs it, and 150 00:09:43.799 --> 00:09:48.000 there was you know dojo. I think you you posted 151 00:09:48.080 --> 00:09:50.720 on x A I five and AI six and it 152 00:09:50.799 --> 00:09:54.360 just seemed like you were incredibly excited about the direction 153 00:09:54.480 --> 00:09:57.320 in which the silicon layer was also going. Can you 154 00:09:57.480 --> 00:10:00.879 tell us about that and what that is? And what what? 155 00:10:00.600 --> 00:10:01.480 What are we? 156 00:10:01.720 --> 00:10:03.480 What are we building here? What is being built? Is 157 00:10:03.480 --> 00:10:06.679 it a complement to everything that exists in the world, 158 00:10:06.720 --> 00:10:08.399 Is it a potential long term competitor? 159 00:10:08.399 --> 00:10:08.879 What is it? 160 00:10:13.120 --> 00:10:13.600 Yeah, So. 161 00:10:15.639 --> 00:10:20.200 At Tesla, we basically had two different chip programs, one 162 00:10:20.320 --> 00:10:23.720 Dojo and one UH dojo on the training side, and 163 00:10:23.759 --> 00:10:28.320 then over call you know AI or it's just our 164 00:10:28.320 --> 00:10:33.440 inference jib that the AI Force is currently shipping in 165 00:10:33.440 --> 00:10:37.399 all vehicles and we're finalizing finalizing the design of AI five, 166 00:10:37.720 --> 00:10:42.240 which will be an immense jump from AI four by 167 00:10:42.279 --> 00:10:45.879 some metrics, the improvement in AI five will be forty 168 00:10:46.039 --> 00:10:47.639 times better. 169 00:10:47.360 --> 00:10:55.120 Than AI four times. And and this is because we 170 00:10:55.240 --> 00:10:57.480 work so closely at a. 171 00:10:57.039 --> 00:11:00.639 Very fine grade level on the AI software and the 172 00:11:00.679 --> 00:11:05.120 AI hardware, so we know exactly where the limiting factors are, 173 00:11:05.960 --> 00:11:09.840 and so effectively the AI hardware and software teams are 174 00:11:09.840 --> 00:11:10.799 co designing the chip. 175 00:11:12.960 --> 00:11:15.639 So forty x improvement in the silicon I think then, 176 00:11:15.799 --> 00:11:18.759 as it as everybody here in the audience experience is 177 00:11:18.799 --> 00:11:21.720 it is that just an almost like an order of 178 00:11:21.720 --> 00:11:25.320 magnitude increase in the quality of FSD and the safety 179 00:11:25.320 --> 00:11:27.759 that you experience as a Tesla driver, and then the 180 00:11:27.879 --> 00:11:30.120 quality of the robot. Like word, is it all manifest 181 00:11:30.159 --> 00:11:34.360 when you, you know, bring it up and actually get 182 00:11:34.360 --> 00:11:35.080 it into production. 183 00:11:36.360 --> 00:11:37.240 Yeah, it's be precise. 184 00:11:37.240 --> 00:11:39.960 The forty x is on if you said, like compared 185 00:11:40.000 --> 00:11:44.679 to the worst limitation on AI four, which is running 186 00:11:45.000 --> 00:11:49.039 the softmax operation. Yeah, we currently have to run softmax 187 00:11:49.240 --> 00:11:53.200 in around forty steps in emulation mode, whereas that will 188 00:11:53.600 --> 00:11:54.279 just be done in a. 189 00:11:54.279 --> 00:11:56.519 Few steps natively in AI five. 190 00:11:58.360 --> 00:12:04.480 A five trip will also be easily handled mixed precision models, 191 00:12:04.559 --> 00:12:07.720 so you don't have it'll dynamically handle mixed there's a 192 00:12:07.759 --> 00:12:10.720 bunch of sort of technical stuff that AIF I will 193 00:12:10.720 --> 00:12:15.720 do a lot better in terms of nominal sort of 194 00:12:15.960 --> 00:12:21.639 raw compute. It's it's eight times more compute, about nine 195 00:12:21.639 --> 00:12:27.720 times more memory, roughly five times more memory boundwidth. So 196 00:12:28.480 --> 00:12:31.519 but because we're addressing some core limitations in AI four, 197 00:12:31.679 --> 00:12:35.279 you multiply that by that that eight x compute improvement 198 00:12:35.360 --> 00:12:40.279 via another five x improvement because of optimization at at 199 00:12:40.279 --> 00:12:42.799 a very fine grain silicon level of things that are 200 00:12:42.799 --> 00:12:43.919 currently suboptimal in. 201 00:12:43.960 --> 00:12:46.279 AI four, that's where you get the forty x improvement 202 00:12:46.679 --> 00:12:48.440 you had. Keep going. 203 00:12:49.519 --> 00:12:53.639 So now that said, I am confident that the current 204 00:12:54.120 --> 00:12:57.080 chips AI A four chips that are in the cars 205 00:12:57.559 --> 00:13:02.480 will achieve self dry having safety that is at least 206 00:13:02.639 --> 00:13:05.960 two to three times that of human and and maybe 207 00:13:06.000 --> 00:13:10.159 even ten x. And the software that will be released 208 00:13:10.200 --> 00:13:13.799 for that is coming out over the next few months, 209 00:13:14.200 --> 00:13:18.639 so version fourteen will be the biggest upgrade until the 210 00:13:18.639 --> 00:13:24.879 software stance version twelve. We are increasing the parameter account 211 00:13:24.919 --> 00:13:30.559 by an order magnitude. The there's there's there's a lot 212 00:13:30.559 --> 00:13:38.759 of reinforcement learning that's been used. As we were there, 213 00:13:38.919 --> 00:13:41.120 There's like you can think of AI sort of as 214 00:13:41.159 --> 00:13:45.039 a way of compressing reality, and some of those compression 215 00:13:45.080 --> 00:13:51.039 steps we were too lossy, and we address the lostiness 216 00:13:51.039 --> 00:13:55.080 in the compression steps. So these are all software updates 217 00:13:55.080 --> 00:13:57.399 that will that will go out. So just over there updates. 218 00:13:58.080 --> 00:14:00.519 Your car has got to feel like it is sentient 219 00:14:01.080 --> 00:14:01.879 by the end of the year. 220 00:14:03.519 --> 00:14:07.320 It feels that way already, to be honest, I saw 221 00:14:07.360 --> 00:14:10.480 in the trades that you spent about seventeen billion dollars 222 00:14:10.519 --> 00:14:16.240 on some spectrum and that. Yeah, so some couch change 223 00:14:17.200 --> 00:14:22.679 to enable your satellites and the Starlink network to connect 224 00:14:22.679 --> 00:14:25.720 directly with phones. What will that look like in a 225 00:14:25.799 --> 00:14:28.360 year or two? Are we going to drop our Verizon 226 00:14:28.399 --> 00:14:35.279 account and just expand our Starlink account. We're kind of 227 00:14:35.320 --> 00:14:38.000 hoping because Verizon kind of sucks. 228 00:14:38.799 --> 00:14:40.919 How many of you on a Starlink phone? 229 00:14:41.000 --> 00:14:46.480 Who wants a Starlink phone? Is it technically possible? 230 00:14:46.559 --> 00:14:48.080 I know you can't see it, but it's everyone. 231 00:14:48.200 --> 00:14:52.759 Yeah, definitely A right cool. So this is a kind 232 00:14:52.759 --> 00:14:53.879 of a long term thing. 233 00:14:55.919 --> 00:15:04.600 It will allow SpaceX to deliver high band with connectivity 234 00:15:04.720 --> 00:15:10.440 directly from the satellites to the phones, but there are 235 00:15:10.960 --> 00:15:12.960 hardware changes that need to happen in the phone so 236 00:15:13.559 --> 00:15:14.120 since these. 237 00:15:14.000 --> 00:15:16.279 Frequencies aren't supported in current. 238 00:15:15.960 --> 00:15:21.559 Phones, the chipset has to be modified to add these frequencies, 239 00:15:22.519 --> 00:15:23.559 and that probably. 240 00:15:23.200 --> 00:15:24.200 Is a two year time frame. 241 00:15:24.279 --> 00:15:27.879 So the phones that are able to use the spectrum 242 00:15:27.919 --> 00:15:30.320 that was acquired probably start shipping. 243 00:15:30.039 --> 00:15:31.039 In around two years. 244 00:15:32.360 --> 00:15:36.240 And then we also need to build the satellites that 245 00:15:36.279 --> 00:15:39.759 are going to communicate on those frequencies. So in parallel 246 00:15:39.840 --> 00:15:43.440 building the satellites and working with the handset makers to 247 00:15:43.720 --> 00:15:47.720 add these frequencies to the phones, and then the satellites 248 00:15:47.799 --> 00:15:51.039 and the phones will then handshake very well to achieve 249 00:15:51.159 --> 00:15:53.480 high bound with connectivity. But the net effect is that 250 00:15:53.519 --> 00:15:57.519 you should be able to watch videos anywhere on your phone. 251 00:15:58.919 --> 00:16:00.759 Wow, and it's. 252 00:16:00.600 --> 00:16:01.320 Going to be crazy. 253 00:16:01.480 --> 00:16:04.120 What do these do? These frequencies? 254 00:16:04.120 --> 00:16:06.879 Would they work indoors inside buildings, you know, like like 255 00:16:06.960 --> 00:16:07.840 that phone currently does? 256 00:16:07.919 --> 00:16:10.200 Okay, and so. 257 00:16:10.159 --> 00:16:12.039 Will you be able to have basically like if you 258 00:16:12.159 --> 00:16:13.519 if you if you're in a building with it with 259 00:16:13.559 --> 00:16:15.600 it like a thick metal roof, then no, but. 260 00:16:16.600 --> 00:16:20.120 The same the same types of normal normal homes. 261 00:16:20.200 --> 00:16:23.000 Yes, yes, Elon, is your vision for this that instead 262 00:16:23.039 --> 00:16:25.039 of you know, having an AT and T account or 263 00:16:25.320 --> 00:16:27.960 and then roaming when you're in the UK or you're 264 00:16:28.000 --> 00:16:30.600 in India. It's just we could have one direct deal 265 00:16:30.600 --> 00:16:34.399 with starlink. It works all over the world eventually, not today, 266 00:16:34.559 --> 00:16:37.320 but at some point. Is that the end goal that 267 00:16:37.399 --> 00:16:40.759 basically we don't need a regional carrier, we have a 268 00:16:40.799 --> 00:16:42.159 global carrier and that would be you. 269 00:16:43.600 --> 00:16:45.000 Uh, that would be one of the options. 270 00:16:45.000 --> 00:16:46.440 To be clear, We're not going to put the other 271 00:16:46.480 --> 00:16:48.440 carriers out of business. They're still going to be around 272 00:16:48.519 --> 00:16:53.159 because they own a lot of spectrum. So, uh, there's 273 00:16:52.399 --> 00:16:55.399 a but. But yes, you'll you should be able to 274 00:16:55.480 --> 00:16:58.000 have a Starlink like you have like you have an 275 00:16:58.000 --> 00:17:00.559 A T and T or the mobile Bride and whatever 276 00:17:00.720 --> 00:17:03.559 you should you could have a you know account with 277 00:17:03.600 --> 00:17:09.279 stallink that works with your you know, starlink h antenna 278 00:17:09.599 --> 00:17:12.079 at home, uh, for your Wi Fi as well as 279 00:17:12.079 --> 00:17:16.079 on your phone. And yeah, it would be a comprehensive 280 00:17:16.079 --> 00:17:20.200 solution for high bandwidth at home and for high end 281 00:17:20.279 --> 00:17:21.039 with Director sell. 282 00:17:21.319 --> 00:17:26.119 Could you buy some carriers to have more spectrum, maybe 283 00:17:26.440 --> 00:17:27.480 five Hoverrizon. 284 00:17:27.200 --> 00:17:29.640 Not out of the question. I suppose that may happen. 285 00:17:30.599 --> 00:17:34.160 Let's talk about let's talk about Starship. You just had 286 00:17:34.160 --> 00:17:41.319 a really what appeared to be a phenomenal launch. How 287 00:17:41.359 --> 00:17:45.519 close is it to you know, being predictable and ready 288 00:17:45.519 --> 00:17:47.160 to go in a commercial setting. 289 00:17:49.880 --> 00:17:53.240 I think we'll recover the ship next year. 290 00:17:54.559 --> 00:18:00.119 We've got one more launch of the Stallink version two to. 291 00:18:02.000 --> 00:18:06.519 Stack, but there's only one one booster and ship left 292 00:18:06.720 --> 00:18:10.039 that's in the version two design. 293 00:18:10.839 --> 00:18:13.200 And then thereafter it's it's version three, which is a 294 00:18:13.240 --> 00:18:17.559 gigantic upgrade because that's got Raptor three and pretty much. 295 00:18:17.400 --> 00:18:19.440 Everything changes on the rocket with version three. 296 00:18:20.960 --> 00:18:24.960 So version three you might have some initial tieving pains 297 00:18:25.599 --> 00:18:31.400 because it's such a radical redesign, but it's it's capable 298 00:18:31.480 --> 00:18:35.240 of over one hundred tons to of it fully reusable, 299 00:18:36.480 --> 00:18:38.039 and I think it's I think I think. 300 00:18:39.359 --> 00:18:41.000 Unless we have unless we have some very. 301 00:18:41.000 --> 00:18:48.000 Major setbacks, SpaceX will demonstrate full reusability next year, catching 302 00:18:48.039 --> 00:18:51.559 both the booster and the ship and being able to 303 00:18:51.599 --> 00:18:55.079 deliver over one hundred tons to a useful orbit. 304 00:18:55.839 --> 00:18:58.200 What is the best rocket in the world do now 305 00:18:58.839 --> 00:19:01.440 in terms of page to space. 306 00:19:03.680 --> 00:19:07.759 Well, in terms of uh sort of commercial rockets, this 307 00:19:07.759 --> 00:19:14.720 this Falcon Heavy, which we'll do in with with with 308 00:19:15.200 --> 00:19:20.000 side booster reuse, we'll do about forty tons. 309 00:19:20.519 --> 00:19:21.799 So this is five times bigger. 310 00:19:21.880 --> 00:19:26.480 Yeah, well two and a half times bigger. But but 311 00:19:26.599 --> 00:19:30.400 starsha would be full reuse, full reusability, got it? Okay, Hey, 312 00:19:30.680 --> 00:19:31.920 everything comes back. 313 00:19:32.000 --> 00:19:37.640 Elan after after the explosion that happened with the failed launch, 314 00:19:39.119 --> 00:19:39.839 there was a lot. 315 00:19:39.640 --> 00:19:43.799 Of sorry which which failed, the more. 316 00:19:43.640 --> 00:19:45.680 Recent one, the more recent with the starship hat with 317 00:19:45.720 --> 00:19:48.799 the big boom, yeah, and the big boom, the and 318 00:19:48.799 --> 00:19:50.960 and there's a lot of There was a lot of 319 00:19:51.000 --> 00:19:54.079 proclamations that there's going to be environmental and f a 320 00:19:54.079 --> 00:19:57.240 A and all these other sorts. The recovery back to 321 00:19:57.839 --> 00:20:01.960 the launch pad again was incredible fast. How did you 322 00:20:02.079 --> 00:20:05.160 get back so fast? Not just technically and work wise, 323 00:20:05.200 --> 00:20:08.680 but just like regulatory clearance wise, because they said there 324 00:20:08.680 --> 00:20:11.039 were gonna be all these questions and reviews and so on. 325 00:20:11.400 --> 00:20:12.799 How did you guys manage that? 326 00:20:14.440 --> 00:20:16.079 Well, there were a lot of questions and reviews. We 327 00:20:16.160 --> 00:20:19.160 got through them all. And credit to the SpaceX team. 328 00:20:19.200 --> 00:20:24.920 They worked incredibly hard and they got the next sugar 329 00:20:24.960 --> 00:20:29.880 and booster tested and on the pad and plan and used. 330 00:20:29.920 --> 00:20:31.119 Credit to the SpaceX team. 331 00:20:31.200 --> 00:20:34.000 Very proud of them for doing doing such a job, 332 00:20:34.240 --> 00:20:39.720 a great job recovering, I mean, creating a fully reusable 333 00:20:39.759 --> 00:20:43.160 OpenAL rocket is one of the hottest engineering problems ever. 334 00:20:44.599 --> 00:20:49.920 Certainly you know, a candidate for most difficult interineering project ever. 335 00:20:50.720 --> 00:20:52.440 You know, it's on the podium at least. 336 00:20:54.759 --> 00:20:58.160 So it's a that that's been the goal of SpaceX 337 00:20:58.240 --> 00:21:01.759 from the beginning, from two thousand too, and here we. 338 00:21:01.799 --> 00:21:03.000 Are twenty three years later. 339 00:21:03.039 --> 00:21:09.519 So it's a long journey and with a super talent 340 00:21:09.559 --> 00:21:11.799 like by far, I think the most talented group of 341 00:21:11.960 --> 00:21:18.759 rocket engineers that's ever been assembled, and we're finally next year, 342 00:21:19.680 --> 00:21:20.200 I think. 343 00:21:20.039 --> 00:21:22.400 We'll be able to achieve full reusability. 344 00:21:23.480 --> 00:21:26.920 On what are the big technical blockers that you're focused 345 00:21:26.960 --> 00:21:30.039 on there between now and that full reusability? Are there 346 00:21:30.279 --> 00:21:35.119 some showstoppers where you're just kind of literally just obsessing 347 00:21:35.160 --> 00:21:37.519 over trying to figure out still or is it more 348 00:21:38.039 --> 00:21:40.880 about getting through a side of a laundry list of 349 00:21:40.920 --> 00:21:43.559 your learnings and just integrating it into the next launch. 350 00:21:46.440 --> 00:21:54.640 Well, well, for full reusability of the ship, there's still 351 00:21:54.720 --> 00:21:55.960 a lot of work that. 352 00:21:55.880 --> 00:21:58.960 Remains on the heat shield. So no one's ever made. 353 00:21:58.799 --> 00:22:01.680 A fully reusable opal heat shield like the Shuttle heat 354 00:22:01.680 --> 00:22:04.920 shield had to go through nine months of repair after 355 00:22:04.960 --> 00:22:09.839 every flight, right, so no one has ever made a. 356 00:22:09.759 --> 00:22:11.960 Fully reusable overal heat shield and is. 357 00:22:11.960 --> 00:22:15.240 That a material science problem or is that an engineering 358 00:22:15.240 --> 00:22:15.920 problem or both? 359 00:22:16.720 --> 00:22:20.359 Yeah, I mean it's a material science engineering problem. So 360 00:22:20.400 --> 00:22:25.720 it's but we really are uh looking at the fundamental. 361 00:22:25.359 --> 00:22:28.359 Physics here again physics post. 362 00:22:28.079 --> 00:22:30.720 Principles and trying to figure out how do we make 363 00:22:30.759 --> 00:22:38.440 something that is you know, it can withstand the heat, 364 00:22:38.720 --> 00:22:41.440 is very light, doesn't transmit. 365 00:22:40.960 --> 00:22:45.920 The heat to the primary primary structure. 366 00:22:47.799 --> 00:22:48.279 And. 367 00:22:50.960 --> 00:22:56.640 Integrity is holdile, stay on and they don't crack. Yeah. Uh. 368 00:22:56.720 --> 00:22:59.920 And then as you ascend, if you hit some rain, 369 00:23:00.759 --> 00:23:04.200 the tiles don't dissolve and rain. There's there's a lot 370 00:23:04.200 --> 00:23:07.960 of different issues, and then you really need to know 371 00:23:08.039 --> 00:23:12.440 that these tiles are working. You can't you know, go 372 00:23:12.559 --> 00:23:15.119 through this laborious inspection. So it really needs to be 373 00:23:15.480 --> 00:23:19.400 where you know, these these tens of thousands of tiles 374 00:23:19.519 --> 00:23:25.440 all work and don't need to be repurbished or checked 375 00:23:25.440 --> 00:23:26.039 one by one. 376 00:23:26.240 --> 00:23:27.519 That was the case with the shuttle. 377 00:23:28.319 --> 00:23:31.880 Can we maybe switch now, think, I mean, who who 378 00:23:31.880 --> 00:23:34.000 else were you talked about Tesla? Then you go to 379 00:23:34.000 --> 00:23:36.680 SpaceX Yeah, now I'd like to ask you some questions 380 00:23:36.720 --> 00:23:42.039 about rock and x AI. You want to just give 381 00:23:42.079 --> 00:23:43.759 us an update. I think you you kind of talked 382 00:23:43.759 --> 00:23:45.960 about where the next gen model is and you said 383 00:23:45.960 --> 00:23:49.039 something incredible. I still don't think people really understand it, 384 00:23:49.079 --> 00:23:50.920 which is, you know, there's going to be a next 385 00:23:50.960 --> 00:23:54.720 training run where you expect, you know, not to start 386 00:23:55.359 --> 00:23:59.160 from the you know, common web and common crawl, where 387 00:23:59.319 --> 00:24:03.359 you expected an enormous amount of synthetic data. Just tell 388 00:24:03.440 --> 00:24:06.559 us about how the evolution of GROC is going and 389 00:24:06.720 --> 00:24:09.079 this innovation and why it's so important. 390 00:24:12.359 --> 00:24:15.160 Yeah, So we're running a lot of using a lot 391 00:24:15.200 --> 00:24:20.920 of inference, compute and reasoning to look at all of 392 00:24:20.960 --> 00:24:24.920 the source data, which is really the corpus of human knowledge, 393 00:24:25.200 --> 00:24:29.799 and then thinking about each piece of information and then 394 00:24:30.079 --> 00:24:37.079 adding what adding what's missing, and correcting, correcting mistakes and 395 00:24:37.599 --> 00:24:41.160 removing falsehoods from the from that training data. So it's 396 00:24:41.200 --> 00:24:44.319 it's if you take data Wikipedia as an example, but 397 00:24:44.359 --> 00:24:50.759 this really applies to the books, PDFs, the websites, every 398 00:24:50.759 --> 00:24:59.480 form of information. The GROC is using heavy amounts of 399 00:24:59.480 --> 00:25:02.920 inference can put to say to look at, as an example, 400 00:25:03.000 --> 00:25:08.000 a Wikipedia page and say, what is true postially true 401 00:25:08.319 --> 00:25:09.200