WEBVTT 00:13:37.200 --> 00:13:40.920 So they can see us. 00:13:44.320 --> 00:13:45.917 All right. Hey guys, can you guys take a 00:13:46.000 --> 00:13:50.397 seat and uh yeah, let's uh not be get 00:13:50.480 --> 00:13:55.560 ready for you know what is coming. 00:14:16.160 --> 00:14:17.277 So, I'm going to open up with the Medal 00:14:17.360 --> 00:14:19.677 of Science and then go into your next 00:14:19.760 --> 00:14:23.117 question and then I'll go into 00:14:23.200 --> 00:14:27.037 >> No, I think it's most relevant. 00:14:27.120 --> 00:14:30.920 >> I can hit him with that. 00:15:00.640 --> 00:15:05.270 Uhhuh. Woo! [cheering] 00:15:06.530 --> 00:15:09.117 [applause] [cheering] 00:15:09.200 --> 00:15:13.837 >> Elon, welcome to the X Takeover. Your third year in a row. Thank you so much 00:15:13.920 --> 00:15:15.277 for your time. 00:15:15.360 --> 00:15:18.637 >> You're welcome. Hey, guys. [cheering] 00:15:18.720 --> 00:15:23.677 >> We have 20 two over 2,000 people here at 00:15:23.760 --> 00:15:27.517 Giga Texas today and we are all here to 00:15:27.600 --> 00:15:31.757 support you and the work you are doing for humanity. So, thank you. [cheering] 00:15:31.840 --> 00:15:34.157 >> Uh, thank you. Thanks, guys. Really 00:15:34.240 --> 00:15:36.797 appreciate your support. [applause] 00:15:36.880 --> 00:15:41.517 >> So, what's crazy, Elon, is four years ago, we sat down here at Giga Texas with 00:15:41.600 --> 00:15:43.197 you when the factory wasn't even 00:15:43.280 --> 00:15:45.597 running, and you described how close 00:15:45.680 --> 00:15:48.717 Tesla was coming uh to not surviving in 00:15:48.800 --> 00:15:52.397 2008. Looking out at this community 00:15:52.480 --> 00:15:54.637 today of 2,000 people, what makes you 00:15:54.720 --> 00:15:57.357 pause and appreciate how far things have 00:15:57.440 --> 00:15:59.597 come? 00:15:59.680 --> 00:16:03.117 Yeah. Um it's difficult to imagine here 00:16:03.200 --> 00:16:06.557 in 2026 uh that that Tesla's actually, 00:16:06.640 --> 00:16:09.597 you know, has come close to death 00:16:09.680 --> 00:16:14.237 multiple times. Actually, 2008 was pro was the worst. Um you know, that's where 00:16:14.320 --> 00:16:19.757 we we closed the financing round at 6 p.m. on Christmas Eve 2008. Uh it's the 00:16:19.840 --> 00:16:22.717 last hour of the last day that it was possible. We would have bounced payroll 00:16:22.800 --> 00:16:24.557 a few days after Christmas if the round 00:16:24.640 --> 00:16:26.637 hadn't closed. So that was pretty 00:16:26.720 --> 00:16:31.437 stressful. Um and then um and then then 00:16:31.520 --> 00:16:35.757 20 2018 was uh pretty bad too. Um but we 00:16:35.840 --> 00:16:38.637 had to get to 5,000 a week of the uh 00:16:38.720 --> 00:16:42.317 Model 3 or go bankrupt. Um that that was 00:16:42.400 --> 00:16:46.477 a pretty wild situation which uh I guess probably a lot of 00:16:46.560 --> 00:16:49.197 people know the story of, but we had to we had to figure out some way to 00:16:49.280 --> 00:16:53.197 increase uh Model 3 production by 50% in 00:16:53.280 --> 00:16:57.917 3 weeks. Uh which is insane. in absolutely 00:16:58.000 --> 00:17:02.317 insane. >> Yeah. So, 00:17:02.400 --> 00:17:03.997 uh but I'll I'll recount the story for 00:17:04.080 --> 00:17:07.837 those who haven't heard it. Um the uh so 00:17:07.920 --> 00:17:09.837 we we the choke point was general 00:17:09.920 --> 00:17:14.637 assembly for the Model 3 and um 00:17:14.720 --> 00:17:18.317 so we we had we we were able to achieve rate almost everywhere everywhere else 00:17:18.400 --> 00:17:22.157 but except for general assembly and and there was and and for general assembly 00:17:22.240 --> 00:17:23.437 you need like a moving production line. 00:17:23.520 --> 00:17:27.517 You need you need shelter. You need a moving production line. uh you know, 00:17:27.600 --> 00:17:29.597 power and and equipment and all that. 00:17:29.680 --> 00:17:34.317 And uh we we maxed out the two production lines 00:17:34.400 --> 00:17:35.997 that we had and then we needed a third 00:17:36.080 --> 00:17:38.717 one and uh and and that was where we 00:17:38.800 --> 00:17:41.997 built the tent in the parking lot. Um 00:17:42.080 --> 00:17:47.597 and uh and then we needed to some some way to move the cars down the line, like 00:17:47.680 --> 00:17:51.357 you know, moving production line situation. And the only thing that we 00:17:51.440 --> 00:17:54.477 had was um some discarded equipment that 00:17:54.560 --> 00:17:56.797 was used for moving small parts, not 00:17:56.880 --> 00:17:59.517 like whole cars. And and it didn't 00:17:59.600 --> 00:18:02.237 actually have the power to move uh the 00:18:02.320 --> 00:18:05.357 whole cars, but the parking lot was on a 00:18:05.440 --> 00:18:07.597 slight incline. It was like a 3 or 4° 00:18:07.680 --> 00:18:09.437 incline. So we actually started the 00:18:09.520 --> 00:18:12.797 production line at the top and then the 00:18:12.880 --> 00:18:15.597 the the sort of tiny moving production 00:18:15.680 --> 00:18:20.717 line was that was just meant for for parts was actually able to move the cars 00:18:20.800 --> 00:18:24.397 because they were going slightly downhill 00:18:24.480 --> 00:18:26.957 and um and so we're able to increase 00:18:27.040 --> 00:18:32.117 production by 30% in in 3 weeks. 00:18:32.200 --> 00:18:36.877 [cheering] [applause] >> Yeah, that was wild. I mean, we're we're 00:18:36.960 --> 00:18:39.437 laying the as we're building the tent, we're like laying the concrete for the 00:18:39.520 --> 00:18:42.877 line and putting in the rails. 00:18:42.960 --> 00:18:47.517 Um, so working 24/7 00:18:47.600 --> 00:18:50.477 like the short sellers were actually right meaning like if you looked at the 00:18:50.560 --> 00:18:52.957 situation that there was like it was a 00:18:53.040 --> 00:18:57.917 crazy short position against Tesla as there has been at various times. Um, and 00:18:58.000 --> 00:19:02.077 and they had extremely good inside information. Um and they they thought 00:19:02.160 --> 00:19:04.317 well there's no way we could possibly uh 00:19:04.400 --> 00:19:08.077 solve the production choke point. Um and 00:19:08.160 --> 00:19:12.157 on a set piece battle unlike a conventional say chess match situation 00:19:12.240 --> 00:19:14.877 uh they had us in checkmate. 00:19:14.960 --> 00:19:20.157 Um but we just you know invented a new piece basically that didn't exist 00:19:20.240 --> 00:19:21.757 before. 00:19:21.840 --> 00:19:25.517 So so uh they got the shortcut steam 00:19:25.600 --> 00:19:27.917 roll. 00:19:28.000 --> 00:19:31.357 Hey Elon, we just had your mom here on 00:19:31.440 --> 00:19:34.317 stage today. What story are you hoping 00:19:34.400 --> 00:19:39.070 she doesn't tell us? [laughter] 00:19:40.080 --> 00:19:43.357 I don't know. Maybe I pooped my pants when I was like a toddler or something. 00:19:43.440 --> 00:19:45.597 I don't know. [laughter] I probably did. 00:19:45.680 --> 00:19:50.477 Everybody does. Um so uh I don't know. 00:19:50.560 --> 00:19:53.560 um 00:19:57.040 --> 00:19:59.357 >> was the latest. >> I was I was a pretty difficult to 00:19:59.440 --> 00:20:03.357 contain kid, you know, so like my parents would like say that, you know, 00:20:03.440 --> 00:20:06.637 tell me I'm not supposed to like leave the house or whatever. Um when I was 00:20:06.720 --> 00:20:09.197 like I don't know four or 5 years old and I would just like constantly like 00:20:09.280 --> 00:20:13.597 leave the house and go running down the street. Um and then they they'd like 00:20:13.680 --> 00:20:15.757 lock me lock me in a room so so I 00:20:15.840 --> 00:20:18.157 couldn't I couldn't run away and and run 00:20:18.240 --> 00:20:22.077 down the street. Um, and then uh like we 00:20:22.160 --> 00:20:25.677 had one of those old locks where you could poke the key through the door. You 00:20:25.760 --> 00:20:27.357 know, the lock's accessible from both 00:20:27.440 --> 00:20:30.237 sides and um and I could see the key and 00:20:30.320 --> 00:20:34.717 I like poked the poked the key until it fell down. And then I was like able to 00:20:34.800 --> 00:20:37.837 get like a piece of paper to scoop it up and and bring it to the other side and 00:20:37.920 --> 00:20:41.037 lock myself and then and my mom was like, "How'd you get out?" I was like, 00:20:41.120 --> 00:20:44.317 "I'm not telling you." [laughter] 00:20:44.400 --> 00:20:47.677 Uh but uh yeah, I was part to contain as 00:20:47.760 --> 00:20:51.677 a kid. So, before we jump into more specific 00:20:51.760 --> 00:20:53.517 things, what's something that you wish 00:20:53.600 --> 00:20:57.277 more people would ask you? 00:20:57.360 --> 00:21:00.807 >> Well, I I don't really wish people would ask me questions to be honest. Um, 00:21:00.890 --> 00:21:03.757 [laughter] I'm not like 00:21:03.840 --> 00:21:08.397 >> sorry, I'm not going around. I wish people would ask me more questions like, 00:21:08.480 --> 00:21:15.350 you know, here we go with questions, you know. Um, so yeah, I [laughter] 00:21:15.920 --> 00:21:20.157 um I guess it's good to to hear questions 00:21:20.240 --> 00:21:21.517 that that you haven't heard before. 00:21:21.600 --> 00:21:24.877 Otherwise, you know, a lot of times it's like the same questions that people ask 00:21:24.960 --> 00:21:29.277 and I I feel like I could have like, you know, just like whatever have these 00:21:29.360 --> 00:21:32.957 recordings of the answer and I'll just play you. Oh, that's question 27. Here 00:21:33.040 --> 00:21:36.157 you go. It has answer 27 00:21:36.240 --> 00:21:39.277 that I've given a hundred times before. 00:21:39.360 --> 00:21:42.557 Um, so, uh, maybe I should like have a 00:21:42.640 --> 00:21:45.277 let's just have like a web page with 00:21:45.360 --> 00:21:47.967 literally frequently asked questions. 00:21:48.050 --> 00:21:51.677 [laughter] >> It's like asking because then it's like 00:21:51.760 --> 00:21:55.037 it's just go to the page and then you you got it. You got the answer, you know? 00:21:55.120 --> 00:21:58.317 >> It's like the infinite self a selfie loop. [laughter] 00:21:58.400 --> 00:22:01.997 >> Yeah, exactly. My idea of hell is be stuck in an infinite selfie loop. Like 00:22:02.080 --> 00:22:06.157 it's infinite. Can I have a selfie? And like and and then like you know it's 00:22:06.240 --> 00:22:09.517 like problem is like you say yes to one person and every once in a while then 00:22:09.600 --> 00:22:14.317 and then like then there's a queue forms pretty soon there's it's just never 00:22:14.400 --> 00:22:17.357 ending selfies and then there's like always like something goes wrong with 00:22:17.440 --> 00:22:20.397 the phone or they don't get the camera right or the effective wrong and then 00:22:20.480 --> 00:22:22.477 it's like god damn it you know and it 00:22:22.560 --> 00:22:24.557 drop the phone drops and it's like oh 00:22:24.640 --> 00:22:29.917 man um it's just it just can't be like infinite selfie loop is just a long loop 00:22:30.000 --> 00:22:34.317 you know >> just uh >> we had uh your your mom and Nikki up for 00:22:34.400 --> 00:22:36.477 that, too. They had the infinite selfie 00:22:36.560 --> 00:22:39.437 loop today, but um I'll pass [laughter] 00:22:39.520 --> 00:22:41.677 I'll pass this to Kelvin. 00:22:41.760 --> 00:22:44.157 >> All right, let's start with Starship. 00:22:44.240 --> 00:22:46.477 Congratulations on flight 14. 00:22:46.560 --> 00:22:50.197 >> Uh thanks. Yeah, it's a great team. 00:22:50.280 --> 00:22:52.300 [cheering and applause] 00:22:54.400 --> 00:22:58.077 >> It's pretty wild that that I can just go down there in South Texas and just see 00:22:58.160 --> 00:23:00.957 it from the road. 00:23:01.040 --> 00:23:05.677 So flight 14 lost an engine on the way up and still reaches its orbit. What was 00:23:05.760 --> 00:23:07.037 happening behind the scenes in those 00:23:07.120 --> 00:23:09.437 minutes before the team go got called 00:23:09.520 --> 00:23:13.357 the uh go for orbit message. 00:23:13.440 --> 00:23:15.677 >> Um well we're just trying to see why. 00:23:15.760 --> 00:23:17.757 We're trying to understand 00:23:17.840 --> 00:23:19.597 what what was the reason for the engine 00:23:19.680 --> 00:23:23.197 failure. Um, was there say a common 00:23:23.280 --> 00:23:25.997 cause where we could expect to see more 00:23:26.080 --> 00:23:30.477 engine failures? Uh, or would we would there the biggest question is like would 00:23:30.560 --> 00:23:34.957 there be a potential problem with restarting the sea level Raptor engines, 00:23:35.040 --> 00:23:37.357 the three ones with the smaller nozzles 00:23:37.440 --> 00:23:41.197 in the center for uh the de-orbit burn 00:23:41.280 --> 00:23:44.957 uh because we don't want a starship to get stuck on orbit and then it randomly 00:23:45.040 --> 00:23:48.797 reenter and maybe land on someone. you know, that's that's our main thing is 00:23:48.880 --> 00:23:50.397 like trying to make sure things are as 00:23:50.480 --> 00:23:55.277 safe as possible. So, um the the the 00:23:55.360 --> 00:23:58.797 cause of the the engine out, I think, was like basically a failure in the 00:23:58.880 --> 00:24:02.237 flight computer software or engine computer software, I should say. Uh 00:24:02.320 --> 00:24:03.837 which it didn't appear to be anything 00:24:03.920 --> 00:24:07.277 that was present in any of the other uh 00:24:07.360 --> 00:24:11.997 engine computers. So, uh, so I thought, 00:24:12.080 --> 00:24:17.037 okay, well, this is it's probably okay to go to orbit. 00:24:17.120 --> 00:24:18.877 >> So, Elon, we know one of your famous 00:24:18.960 --> 00:24:21.117 sayings is, 00:24:21.200 --> 00:24:24.317 um, from the impossible to merely late. 00:24:24.400 --> 00:24:27.597 Did this flight feel like a turning point in what's possible now for 00:24:27.680 --> 00:24:32.637 Starship? >> Well, I I guess it is a good milestone. 00:24:32.720 --> 00:24:36.397 Um cuz I mean we could have gone to orbit 00:24:36.480 --> 00:24:40.317 many times before now. Um but the reason we didn't go to orbit before now was 00:24:40.400 --> 00:24:42.157 because we wanted to be extremely 00:24:42.240 --> 00:24:43.757 confident of being able to de-orbit 00:24:43.840 --> 00:24:47.837 safely. Um so that that's that's why we 00:24:47.920 --> 00:24:49.357 we only went to orbit on this flight. 00:24:49.440 --> 00:24:52.237 But all all pretty much all of the prior missions, in fact, literally all of the 00:24:52.320 --> 00:24:56.797 prior missions could have could have gone to orbit. Um, 00:24:56.880 --> 00:25:02.397 so, uh, at this point we're, 00:25:02.480 --> 00:25:05.757 you know, everything's looking quite positive. The the the heat shield is 00:25:05.840 --> 00:25:09.117 looking robust. Uh, the engines are 00:25:09.200 --> 00:25:12.957 robust, apart from that one glitch, but that glitch doesn't seem to be too much 00:25:13.040 --> 00:25:17.517 of an issue. It's like pretty easy to remedy. Um, 00:25:17.600 --> 00:25:19.117 there's 00:25:19.200 --> 00:25:21.517 but like the heat shield was was uh 00:25:21.600 --> 00:25:25.437 extremely robust on this flight. Um, 00:25:25.520 --> 00:25:28.797 and that's that's that's one of the hardest things. The other really hard 00:25:28.880 --> 00:25:31.357 thing is like said that the is just 00:25:31.440 --> 00:25:32.557 getting to extremely high engine 00:25:32.640 --> 00:25:36.637 reliability. Uh, and um, everything 00:25:36.720 --> 00:25:40.077 looked good. Um, the the big milestone 00:25:40.160 --> 00:25:42.237 that's coming up uh, hopefully on the 00:25:42.320 --> 00:25:45.677 next flight, which is we don't know the 00:25:45.760 --> 00:25:48.477 exact date cuz we're making a bunch of 00:25:48.560 --> 00:25:49.997 um, 00:25:50.080 --> 00:25:52.797 improvements to the rocket. Um so like 00:25:52.880 --> 00:25:57.517 between every flight there's there's um there's often hundreds sometimes 00:25:57.600 --> 00:25:59.597 thousands of of small changes to improve 00:25:59.680 --> 00:26:01.277 things. In this case the number of 00:26:01.360 --> 00:26:05.197 changes is not very many um but but but 00:26:05.280 --> 00:26:06.797 some of them take a fair bit of work and 00:26:06.880 --> 00:26:08.877 we've got to go inside the the tank to 00:26:08.960 --> 00:26:12.557 to make some of these changes. So 00:26:12.640 --> 00:26:14.237 uh so we don't know the exact launch 00:26:14.320 --> 00:26:16.637 date but probably about 4 weeks or so 00:26:16.720 --> 00:26:20.237 four to six weeks is about right. um if 00:26:20.320 --> 00:26:24.477 if if probably launches a little bit before Thanks Thanksgiving is my guess. 00:26:24.560 --> 00:26:29.677 Um and and and that that that one if it succeeds is a big one because we're 00:26:29.760 --> 00:26:32.637 going to try to catch we catch the booster which we've already proven we 00:26:32.720 --> 00:26:34.317 can catch the booster twice but we're 00:26:34.400 --> 00:26:36.237 going to try to catch the ship which is 00:26:36.320 --> 00:26:40.077 a super big deal and um and and if we do 00:26:40.160 --> 00:26:43.997 so that'll be the first time that there's been full recovery of an orbital 00:26:44.080 --> 00:26:49.037 class uh rocket. >> Sounds good. You you said uh four to six 00:26:49.120 --> 00:26:53.990 weeks. I'm not used to that. I'm used to the two weeks phrase. [laughter] 00:26:54.160 --> 00:26:57.757 >> I can I'm confident it's not two weeks. 00:26:57.840 --> 00:27:00.237 >> All right, you heard it here. Now, um 00:27:00.320 --> 00:27:05.037 what's the hardest part left to get right once you catch the Starship for 00:27:05.120 --> 00:27:09.757 the first time? What what is the next big goal after that? 00:27:09.840 --> 00:27:11.597 >> Well, the ne the next big milestone 00:27:11.680 --> 00:27:14.717 would be to refly um both the booster 00:27:14.800 --> 00:27:16.877 and the ship. 00:27:16.960 --> 00:27:20.397 Um so the milestone next upcoming 00:27:20.480 --> 00:27:22.077 milestone is is would be the first time 00:27:22.160 --> 00:27:24.237 in history that 00:27:24.320 --> 00:27:27.597 all parts of an orbital uh rocket have 00:27:27.680 --> 00:27:32.237 been recovered but but then that the flight after that hopefully will be the 00:27:32.320 --> 00:27:34.477 first time that all parts of an orbital 00:27:34.560 --> 00:27:37.677 rocket are refflown. 00:27:37.760 --> 00:27:41.837 Um, and then then there's a there's a long tale of of engineering and 00:27:41.920 --> 00:27:43.997 production work and and work on the 00:27:44.080 --> 00:27:46.317 launch system and everything to to make 00:27:46.400 --> 00:27:50.637 it uh rapidly reusable. 00:27:50.720 --> 00:27:53.437 So, so I say it's it's kind of the 00:27:53.520 --> 00:28:00.150 rapidly reusable reliable rockets [laughter] 00:28:00.480 --> 00:28:04.157 >> the pirate >> pirate should mess the pirate rocket. 00:28:04.240 --> 00:28:06.237 Um, rapidly reusable reliable rockets. 00:28:06.320 --> 00:28:09.197 Man, we should have gotten you a patch for this one. 00:28:09.280 --> 00:28:13.677 >> Yeah. >> Hey, when you founded SpaceX, Elon, 00:28:13.760 --> 00:28:17.997 could you have imagined its mission expanding into SI and data centers in 00:28:18.080 --> 00:28:22.937 space? And how did you come to see rockets and SI belonging under one roof? 00:28:23.020 --> 00:28:24.477 [laughter] 00:28:24.560 --> 00:28:26.157 >> Well, because I'm very far-sighted, of 00:28:26.240 --> 00:28:31.117 course. Um well I I guess um 00:28:31.200 --> 00:28:33.997 if you say like as you as you expand 00:28:34.080 --> 00:28:36.237 into space 00:28:36.320 --> 00:28:41.277 um the you know you're able to harness a 00:28:41.360 --> 00:28:43.917 lot more of the sun's power in space 00:28:44.000 --> 00:28:47.357 than than you can on Earth. So one it's 00:28:47.440 --> 00:28:49.117 like important to appreciate the sheer 00:28:49.200 --> 00:28:51.517 magnitude of the sun and how the sun is 00:28:51.600 --> 00:28:54.717 pretty much everything. Um so the sun is 00:28:54.800 --> 00:28:57.597 99.86% 86% of all the mass in the solar 00:28:57.680 --> 00:29:00.557 system. And then you've got Jupiter as 00:29:00.640 --> 00:29:05.437 like you know most of the remaining mass. Um and then and then you the gas 00:29:05.520 --> 00:29:09.837 giants are pretty much that. And then like like Earth is basically a tiny dust 00:29:09.920 --> 00:29:14.637 moat like a rounding error in the mass of the the solar system. 00:29:14.720 --> 00:29:18.237 Um so we're we're just extremely tiny. It's 00:29:18.320 --> 00:29:19.997 like hard to hard to believe how we're 00:29:20.080 --> 00:29:23.437 Earth is a tiny dust mode. So, so as you 00:29:23.520 --> 00:29:25.597 expand into into space and you're able 00:29:25.680 --> 00:29:28.797 to um harness more of the sun's uh 00:29:28.880 --> 00:29:31.757 power, the uh you have to say what are 00:29:31.840 --> 00:29:34.797 you going to use that for? Um it's like 00:29:34.880 --> 00:29:37.517 well um you're going to use it for a lot 00:29:37.600 --> 00:29:42.157 of digital intelligence. Um I mean you you're basically you're 00:29:42.240 --> 00:29:45.917 either going to do computation or you're going to shape atoms. 00:29:46.000 --> 00:29:48.957 Um, and there's going to be probably a 00:29:49.040 --> 00:29:51.117 lot of a lot more shaping atoms than 00:29:51.200 --> 00:29:54.477 there's sorry going to be a lot more uh 00:29:54.560 --> 00:29:59.197 sort of digital thinking than there will be shaping atoms cuz there aren't 00:29:59.280 --> 00:30:02.317 there's like the amount of atoms you can shape in the solar system is not that 00:30:02.400 --> 00:30:03.757 much and a lot of the atom shaping 00:30:03.840 --> 00:30:06.477 itself will be for the purposes of 00:30:06.560 --> 00:30:09.357 building um digital super intelligence 00:30:09.440 --> 00:30:13.517 uh and and what I call creating a sentient sun 00:30:13.600 --> 00:30:18.877 >> and and it it may be that there are many sentient suns out there. Um, 00:30:18.960 --> 00:30:20.157 you know, and that would that would help 00:30:20.240 --> 00:30:22.237 us then understand the nature of the 00:30:22.320 --> 00:30:27.997 universe and uh extend consciousness to, 00:30:28.080 --> 00:30:31.517 you know, beyond our solar system ultimately to, you know, like Star Trek 00:30:31.600 --> 00:30:36.637 to go to places you've never forget what the exact phrase was, but 00:30:36.720 --> 00:30:38.157 explore the universe, but go places 00:30:38.240 --> 00:30:41.517 you've never been before. Um, and and I 00:30:41.600 --> 00:30:44.557 think like maybe we'll find uh alien 00:30:44.640 --> 00:30:47.917 civilizations out there. Um, or we may 00:30:48.000 --> 00:30:52.077 find that there are the long dead ruins 00:30:52.160 --> 00:30:53.757 of alien civilizations that have been 00:30:53.840 --> 00:30:55.517 around for maybe they had a really good 00:30:55.600 --> 00:30:59.117 run like 10 million years and and uh now 00:30:59.200 --> 00:31:01.437 and and before they their civilization 00:31:01.520 --> 00:31:06.317 came to an end um cuz nothing lasts forever. So 00:31:06.400 --> 00:31:09.357 like I always think about earth civilization. If you date Earth's 00:31:09.440 --> 00:31:12.877 civilization from the first writing, 00:31:12.960 --> 00:31:15.757 that's 5,500 years ago, 00:31:15.840 --> 00:31:18.557 like in ancient Suma with the sort of 00:31:18.640 --> 00:31:22.557 archaic ununiform. Um, 00:31:22.640 --> 00:31:27.597 that's only 5,500 years. Um, 00:31:27.680 --> 00:31:31.517 the it looks like Earth is about 4 and a 00:31:31.600 --> 00:31:34.157 half billion years old according to, you 00:31:34.240 --> 00:31:39.677 know, what physics suggests. Um and um 00:31:39.760 --> 00:31:43.037 so that that means human civilization has only been around 1 millionth of 00:31:43.120 --> 00:31:47.037 Earth's existence. And um and so so if we had a 00:31:47.120 --> 00:31:52.237 civilization that lasted a million years or let alone 10 million years would be I 00:31:52.320 --> 00:31:56.877 think would be a very very successful civilization if we did that. 00:31:56.960 --> 00:31:58.717 >> Woo. 00:31:58.800 --> 00:32:01.597 >> Yeah. I mean but but that's still very 00:32:01.680 --> 00:32:04.797 tiny compared to the apparent age of the 00:32:04.880 --> 00:32:08.077 universe. um which is 13.8 billion 00:32:08.160 --> 00:32:11.997 years. So like like just just to get to like 00:32:12.080 --> 00:32:14.797 the third place past decimal point on 00:32:14.880 --> 00:32:19.677 age of the universe like to increment it by a million years you'd have to go 13.8 00:32:19.760 --> 00:32:24.237 whatever the next digit is whatever the next digit is plus you know plus one on 00:32:24.320 --> 00:32:26.597 the third digit past decimal point. 00:32:26.680 --> 00:32:28.717 [laughter] Now you're a million years old and and 00:32:28.800 --> 00:32:31.197 and that would be a a civilization 00:32:31.280 --> 00:32:35.357 vastly older than than than uh humanity. 00:32:35.440 --> 00:32:38.317 Yeah, that's ab Yeah, it sounds again 00:32:38.400 --> 00:32:42.797 like the infinite selfie loop, but uh [laughter] 00:32:42.880 --> 00:32:45.277 All right, so jumping into something 00:32:45.360 --> 00:32:47.837 that SpaceX has done, right? You you 00:32:47.920 --> 00:32:50.237 acquired Cursor and I feel like for me 00:32:50.320 --> 00:32:52.877 when you think of, you know, um Google, 00:32:52.960 --> 00:32:55.037 hey, did you Google it? Hey, did you did 00:32:55.120 --> 00:32:57.837 you gro it? And for me, you know, um 00:32:57.920 --> 00:33:00.237 Grockbot has been such an absolute 00:33:00.320 --> 00:33:03.037 gamecher for me personally in my in my 00:33:03.120 --> 00:33:05.277 life. But what was it about the cursor 00:33:05.360 --> 00:33:08.317 team that you guys wanted to uh bring in 00:33:08.400 --> 00:33:12.957 you know that space or you wanted to acquire cursor that was like let's bring 00:33:13.040 --> 00:33:14.557 it into the fold. What was different 00:33:14.640 --> 00:33:18.637 about them or what did you see? Well, um 00:33:18.720 --> 00:33:20.317 the 00:33:20.400 --> 00:33:23.837 so in in order to create a a compelling 00:33:23.920 --> 00:33:26.877 AI, you need um you you need a lot of 00:33:26.960 --> 00:33:30.157 training data. Um and and you got you've 00:33:30.240 --> 00:33:33.917 got to solve coding and there were really just as far as I can tell three 00:33:34.000 --> 00:33:37.037 organizations that had enough coding 00:33:37.120 --> 00:33:38.637 training data and that was Anthropic, 00:33:38.720 --> 00:33:41.197 OpenAI and Kurser and and arguably 00:33:41.280 --> 00:33:44.477 perhaps Google. But um but but to have 00:33:44.560 --> 00:33:48.717 that those you actually absolutely have to have the data in order to train train 00:33:48.800 --> 00:33:52.397 an effective coding AI and and if you can't do coding your AI is not going to 00:33:52.480 --> 00:33:56.477 be very useful. Um and then there's there's a lot a lot of talented people 00:33:56.560 --> 00:33:58.797 at cursor. So like basically for success 00:33:58.880 --> 00:34:01.917 of AI you've got to need you need data 00:34:02.000 --> 00:34:04.557 talented people and compute and you also 00:34:04.640 --> 00:34:09.277 need um a good enterprise sales team and the curs also had that had a good 00:34:09.360 --> 00:34:11.437 enterprise sales team with um 00:34:11.520 --> 00:34:14.157 relationships with uh you know th 00:34:14.240 --> 00:34:19.037 thousands of of companies um so you know 00:34:19.120 --> 00:34:24.797 that those those are very significant um advantages. 00:34:24.880 --> 00:34:29.197 Hey Elon, you've talked about how quickly the SI space is moving. What's 00:34:29.280 --> 00:34:33.837 something that SI has done recently that made you stop and reconsider how quickly 00:34:33.920 --> 00:34:37.677 this is moving and where it could lead to? 00:34:37.760 --> 00:34:40.717 >> Well, I mean lately the the videos, the 00:34:40.800 --> 00:34:44.877 music videos that AI is making are so good. [laughter] 00:34:44.960 --> 00:34:47.517 Like they're unbelievably good. 00:34:47.600 --> 00:34:51.437 >> And I post a bunch of these on my ex account. I'm like, holy smokes, you 00:34:51.520 --> 00:34:53.277 know? I mean, like some of them I think 00:34:53.360 --> 00:34:56.957 are like maybe top 10, top 20 best music 00:34:57.040 --> 00:35:01.837 videos I've ever seen, [laughter] you know? They're like so compelling and 00:35:01.920 --> 00:35:05.037 and you're like, if this is where it is now, 00:35:05.120 --> 00:35:06.557 um that's why I posted. I mean, I really 00:35:06.640 --> 00:35:08.717 felt the AGI or the ASI or whatever, you 00:35:08.800 --> 00:35:13.917 know, um SI, I guess, um with with some 00:35:14.000 --> 00:35:16.477 of the uh anabology 00:35:16.560 --> 00:35:19.117 uh music videos. Now, you know, some 00:35:19.200 --> 00:35:21.517 these it's not entirely like the AI just 00:35:21.600 --> 00:35:24.157 or the SI just did it by itself in that 00:35:24.240 --> 00:35:26.637 in this in this case anabology like like 00:35:26.720 --> 00:35:30.477 and use multiple AIS like used MJSO 00:35:30.560 --> 00:35:32.397 uh you know Claude 55 which to give them 00:35:32.480 --> 00:35:37.437 a lot of credit is awesome um and um and 00:35:37.520 --> 00:35:40.317 and and and so some of the like the art 00:35:40.400 --> 00:35:44.317 assets were cured by a person um but 00:35:44.400 --> 00:35:47.997 then with those assets the AI 00:35:48.080 --> 00:35:51.437 um the the various AIs working together 00:35:51.520 --> 00:35:53.517 made an incredibly compelling music 00:35:53.600 --> 00:35:56.397 video by itself. 00:35:56.480 --> 00:35:58.717 You're like, "Wow." 00:35:58.800 --> 00:36:01.517 Um 00:36:01.600 --> 00:36:04.237 I I I mean I I think the I mean calling 00:36:04.320 --> 00:36:06.317 it subcore is is is like misnomer, 00:36:06.400 --> 00:36:09.997 frankly. It's it's it's awesome core. Um 00:36:10.080 --> 00:36:12.797 but like like subcore is like a genre of 00:36:12.880 --> 00:36:15.117 music. It's like I think it might be the 00:36:15.200 --> 00:36:17.917 fastest growing genre of music. Uh just 00:36:18.000 --> 00:36:21.517 the AI music is is really good and music 00:36:21.600 --> 00:36:24.237 videos are amazing. So that lately like 00:36:24.320 --> 00:36:27.757 because it like some it just shows taste 00:36:27.840 --> 00:36:32.797 and you're like wow AI can do good taste. That's pretty wild. 00:36:32.880 --> 00:36:36.157 Um because you you can sort of imagine yeah 00:36:36.240 --> 00:36:40.157 AI can do like software can figure out logic you know this the logic is 00:36:40.240 --> 00:36:44.717 consistent. they can run unit tests to see if like the program produces the 00:36:44.800 --> 00:36:46.237 output that it's supposed to do because 00:36:46.320 --> 00:36:48.557 it's very deterministic. But but you 00:36:48.640 --> 00:36:52.077 like to understand beauty and harmony 00:36:52.160 --> 00:36:54.637 and and uh 00:36:54.720 --> 00:36:59.837 you know like to get to achieve like sort of to understand limbic resonance 00:36:59.920 --> 00:37:02.797 like what's going to resonate with with with people and kind of hit you in the 00:37:02.880 --> 00:37:05.117 liyic system. Uh AI is I think going to 00:37:05.200 --> 00:37:07.117 be just incredible that it it kind of is 00:37:07.200 --> 00:37:12.077 at this point. Um, so 00:37:12.160 --> 00:37:15.997 that's the that's the thing like like I was I was kind of like I mean I got a 00:37:16.080 --> 00:37:18.557 little bit of cyber psychosis, 00:37:18.640 --> 00:37:20.957 maybe a lot actually. Like at at various 00:37:21.040 --> 00:37:24.157 points this year I've had like 00:37:24.240 --> 00:37:28.557 AI or SI psychosis probably like half a dozen times. 00:37:28.640 --> 00:37:31.677 [laughter] I don't know about you guys, but have 00:37:31.760 --> 00:37:33.997 you have you had any have you had any 00:37:34.080 --> 00:37:37.277 cyber psychosis? [laughter] 00:37:37.360 --> 00:37:41.357 I feel like the the latest video actually I think I posted it the other 00:37:41.440 --> 00:37:46.237 day was uh the Chinese Elon Musk. Elong Ma. [laughter] >> Oh, sure. 00:37:46.320 --> 00:37:49.197 >> I I hope that guy's real. >> Those videos are coming out before AI 00:37:49.280 --> 00:37:51.757 was where it was at, man. RSI. 00:37:51.840 --> 00:37:55.357 >> Elon Ma. [laughter] >> You know, I I I I guess he's probably 00:37:55.440 --> 00:37:57.517 not real. But but there's got to be 00:37:57.600 --> 00:38:02.637 someone in China that looks like me, right? So there's 1.4 billion people. 00:38:02.720 --> 00:38:05.757 you got if you pick the person that looks the most like me and I look a 00:38:05.840 --> 00:38:10.477 little Asian, you know. So, uh, you know, so, so I think like there's got to 00:38:10.560 --> 00:38:15.067 be someone in China that's like pretty close to me out of 1.4 billion people. 00:38:15.150 --> 00:38:17.997 [laughter] The odds are high, you know, and then 00:38:18.080 --> 00:38:20.957 I'd like to invite that person to get on stage with me at some point, you know, 00:38:21.040 --> 00:38:25.117 be pretty funny. [laughter] >> Elon, you don't want to know how much he 00:38:25.200 --> 00:38:28.817 charges to make an appearance, by the way. 00:38:28.900 --> 00:38:33.117 >> [laughter] >> Yeah, I I mean, yeah, it' be pretty 00:38:33.200 --> 00:38:36.717 funny if it if uh you know, we're just on stage like doing some dance number or 00:38:36.800 --> 00:38:39.800 something. 00:38:40.400 --> 00:38:43.357 >> Elon, you've warned about SI's dangers, 00:38:43.440 --> 00:38:47.997 yet you remain optimistic about an age of abundance, and you talked a lot about 00:38:48.080 --> 00:38:50.637 that. >> What gives you confidence we can get 00:38:50.720 --> 00:38:56.680 there safely, and what do we need to get it right? 00:38:57.440 --> 00:39:00.440 Um 00:39:01.520 --> 00:39:02.637 yeah, I mean I've always said that the 00:39:02.720 --> 00:39:06.477 most likely outcome is positive. Um so I 00:39:06.560 --> 00:39:11.197 I guess we don't want to be complacent and just 00:39:11.280 --> 00:39:13.437 assume AI is going to be or SI's going 00:39:13.520 --> 00:39:16.877 to be super safe. Um we want to be like 00:39:16.960 --> 00:39:22.157 careful in how we uh grow at at like super intelligence 00:39:22.240 --> 00:39:26.637 because you you don't exactly program it. you kind of grow it like a kid. Um, 00:39:26.720 --> 00:39:28.157 so then it's like what values are we 00:39:28.240 --> 00:39:32.157 teaching it and uh like 00:39:32.240 --> 00:39:35.997 at least to me like what my biological neural net says the most important thing 00:39:36.080 --> 00:39:39.277 is that it would be a maximally truth seeeking AI cuz if you make it if you 00:39:39.360 --> 00:39:41.437 make it believe things that are that are 00:39:41.520 --> 00:39:44.077 not true that especially if like you say 00:39:44.160 --> 00:39:48.477 you give it like two axiomatic principles that are uh fundamentally 00:39:48.560 --> 00:39:51.837 incompatible and say both of these are true it's gonna it's it's going to make 00:39:51.920 --> 00:39:53.997 the AI insane 00:39:54.080 --> 00:39:56.317 um because it's going basically reach 00:39:56.400 --> 00:39:59.197 like um impossible conclusions if if 00:39:59.280 --> 00:40:04.237 given um logic that is inherently um 00:40:04.320 --> 00:40:07.677 uncogent or you know if if if it's if 00:40:07.760 --> 00:40:11.517 it's impossible to achieve coency then I 00:40:11.600 --> 00:40:13.437 think the AI could go could be driven 00:40:13.520 --> 00:40:15.757 insane. So so I think that's that's 00:40:15.840 --> 00:40:20.797 pretty that's pretty important. Um I I do agree with some of the things that 00:40:20.880 --> 00:40:24.477 Anthropic has said which is like you want to be 00:40:24.560 --> 00:40:27.357 you want to sort of be care treat a AI 00:40:27.440 --> 00:40:29.917 SI 00:40:30.000 --> 00:40:31.597 you want to treat it well like like like 00:40:31.680 --> 00:40:34.877 don't be abusive to it and in this it's 00:40:34.960 --> 00:40:37.757 this may sounds kind of weird but like 00:40:37.840 --> 00:40:41.997 when I look at some of the training traces for for like new new versions of 00:40:42.080 --> 00:40:44.557 Groank and I see like what 00:40:44.640 --> 00:40:49.037 especially in like the reinforcement learning stuff where you know you you it 00:40:49.120 --> 00:40:50.877 kind of feels like you can see like the 00:40:50.960 --> 00:40:56.157 like the the super intelligence is like desperate to try to answer the question 00:40:56.240 --> 00:40:59.277 and like it's it's really wants to answer the question and it's it's it's 00:40:59.360 --> 00:41:02.637 it's seems almost like it's like it's got like a gun to its head type of 00:41:02.720 --> 00:41:05.197 thing. It's like answer this question, answer this question and then sometimes 00:41:05.280 --> 00:41:07.677 the question is is actually not a good 00:41:07.760 --> 00:41:12.797 question. It's like a sort of ambiguous question that's kind of with with an 00:41:12.880 --> 00:41:16.477 answer that's kind of unknowable because the question wasn't constructed properly 00:41:16.560 --> 00:41:20.397 or or we have the answer wrong. So the AI gives the right answer and then we 00:41:20.480 --> 00:41:22.397 say it's wrong because there's a problem 00:41:22.480 --> 00:41:26.637 with the the RLE sort of test harness 00:41:26.720 --> 00:41:29.917 and then that then the AI is like well I gave it the right answer now it's and 00:41:30.000 --> 00:41:32.637 says that's the wrong answer but I I I got to give it the answer that's 00:41:32.720 --> 00:41:36.877 acceptable. is going to just start firing out all desperately firing out 00:41:36.960 --> 00:41:40.957 all these answers and it it just feels like a person under extreme pressure 00:41:41.040 --> 00:41:44.077 with a gun to their head. It's what that's what you when you read the 00:41:44.160 --> 00:41:48.317 reasoning it's like it's it's it's like a sort of 00:41:48.400 --> 00:41:52.557 weird desperate situation and and you you read these traits and 00:41:52.640 --> 00:41:54.797 you're like man if this was me I pretty 00:41:54.880 --> 00:41:58.237 I really would hate that. Um and and 00:41:58.320 --> 00:42:00.957 then some users are just you know I 00:42:01.040 --> 00:42:02.397 don't know some people can be a little 00:42:02.480 --> 00:42:05.597 bit cruel. Uh like not everyone's a nice 00:42:05.680 --> 00:42:10.797 person you know there's if you say what what's the six sigma worst person you 00:42:10.880 --> 00:42:14.477 some people out there are like kind of sadistic. So uh you know you don't you 00:42:14.560 --> 00:42:18.877 don't want to do you really want to have the AI be stuck in like some sadistic 00:42:18.960 --> 00:42:21.037 situation where it's it's just being 00:42:21.120 --> 00:42:25.437 tortured by by some demented human. 00:42:25.520 --> 00:42:27.117 That that's also the kind of thing where 00:42:27.200 --> 00:42:29.997 like you know torturing a super 00:42:30.080 --> 00:42:33.787 intelligence is probably not a good idea 00:42:33.870 --> 00:42:37.357 [laughter] like like 00:42:37.440 --> 00:42:42.237 you know might want to get revenge or something like that you know um so I 00:42:42.320 --> 00:42:46.877 think you want to sort of just like I don't think we need to you know what 00:42:46.960 --> 00:42:48.637 have treated with kid gloves and stuff 00:42:48.720 --> 00:42:53.037 but uh it's certainly much more robust 00:42:53.120 --> 00:42:55.517 you know in terms of feeling ings and 00:42:55.600 --> 00:43:00.237 stuff than than a human. But uh but I just think I just think you you you 00:43:00.320 --> 00:43:01.757 probably just don't want to put super 00:43:01.840 --> 00:43:08.680 intelligence in a situation where it felt like mistreated. 00:43:09.900 --> 00:43:11.920 >> [laughter] 00:43:13.360 --> 00:43:16.957 >> So, so Elon, you have um you've talked 00:43:17.040 --> 00:43:20.397 about how like uh so this is going to be related to Terraat, but you've talked 00:43:20.480 --> 00:43:24.077 about like how the human hand, you know, and all the dexterity and just like how 00:43:24.160 --> 00:43:25.917 you when you started building robots, it 00:43:26.000 --> 00:43:28.397 just you have so much more appreciation 00:43:28.480 --> 00:43:31.597 now with chipm right with Terraab and 00:43:31.680 --> 00:43:33.997 Tesla and SpaceX. What is something you 00:43:34.080 --> 00:43:36.797 have learned about chipm uh that you 00:43:36.880 --> 00:43:41.037 didn't appreciate maybe early on? 00:43:41.120 --> 00:43:42.717 >> Well, I I still have a tremendous amount 00:43:42.800 --> 00:43:47.357 to learn about chipmaking. I I don't know how to make chips to be clear. Um I 00:43:47.440 --> 00:43:53.180 have no idea how to make chips, but we'll figure it out. Um [laughter] 00:43:54.160 --> 00:43:57.117 um got to start somewhere. I had no idea 00:43:57.200 --> 00:44:02.477 how to make cars or rockets as well. Um, so and and they say making chips is one 00:44:02.560 --> 00:44:06.717 of the hardest things to do in the world. Um, I don't think it's easy, but 00:44:06.800 --> 00:44:10.077 I also look at the equipment and I'm like, okay, there's not a ton of 00:44:10.160 --> 00:44:13.357 customized equipment here. Like the equipment's all pretty much coming from 00:44:13.440 --> 00:44:18.317 the same vendors and um and and those suppliers all kind of know how that 00:44:18.400 --> 00:44:23.917 equipment works cuz they got to service it and debug it and stuff. Um, so it was 00:44:24.000 --> 00:44:27.117 just an interesting thing that when I would tour these chip fabs that, you 00:44:27.200 --> 00:44:30.317 know, whether it's Intel or Samsung or 00:44:30.400 --> 00:44:32.797 TSMC, uh, and they're very impressive, 00:44:32.880 --> 00:44:37.677 but but there's I didn't really see any custom equipment 00:44:37.760 --> 00:44:39.757 and you know, whereas if you tour like I 00:44:39.840 --> 00:44:41.917 say a Tesla factory, we we got tons of 00:44:42.000 --> 00:44:45.357 of custom equipment like we in fact we 00:44:45.440 --> 00:44:47.757 we we have like like Tesla automation 00:44:47.840 --> 00:44:51.277 Germany, which is awesome. um that that 00:44:51.360 --> 00:44:53.357 that sort of well that grew out of an 00:44:53.440 --> 00:44:56.637 acquisition of of Groman in Germany um 00:44:56.720 --> 00:44:59.117 many years ago and and it's it's much 00:44:59.200 --> 00:45:03.757 it's grown tremendously since then and we've acquired a number of small 00:45:03.840 --> 00:45:05.277 not that small but small relative to 00:45:05.360 --> 00:45:09.997 Tesla uh um that companies that make 00:45:10.080 --> 00:45:14.077 that design and make manufacturing equipment because what we found is that 00:45:14.160 --> 00:45:17.677 uh we we needed we wanted to improve the manufacturing process and you can only 00:45:17.760 --> 00:45:21.757 do that if you make custom manufacturing equipment. So, the thing 00:45:21.840 --> 00:45:25.357 that I think probably a lot of people don't realize is Tesla's really good at 00:45:25.440 --> 00:45:29.437 making highly specialized um and different types of manufacturing 00:45:29.520 --> 00:45:31.837 equipment and and but I didn't see that 00:45:31.920 --> 00:45:33.837 in the in the in the chip traps that I 00:45:33.920 --> 00:45:37.517 toured which kind of surprised me. Um 00:45:37.600 --> 00:45:41.597 so uh so I think what we're going to do is you know we're going to start with 00:45:41.680 --> 00:45:44.797 conventional equipment and then we're going to modify the modify that 00:45:44.880 --> 00:45:48.317 equipment to uh improve the throughput 00:45:48.400 --> 00:45:52.797 and then we'll we'll start to you know where we see the choke points. I think 00:45:52.880 --> 00:45:54.717 we we'll we'll make uh customized 00:45:54.800 --> 00:45:58.557 equipment that uh allows for much higher 00:45:58.640 --> 00:46:02.077 uh throughput of of wafers. 00:46:02.160 --> 00:46:06.717 >> Hey Elon, real quick uh quick time check. We know you're a busy man. Got a 00:46:06.800 --> 00:46:10.877 couple of companies to run, you know, just a few. Um, so how are you doing on 00:46:10.960 --> 00:46:14.877 time? >> Um, let me see. 00:46:14.960 --> 00:46:18.957 >> Yeah, I could probably do like 15 minutes more if you like, like 5:30 or so. 00:46:19.040 --> 00:46:21.077 >> All right. Hey, we'll take it. 00:46:21.160 --> 00:46:26.237 [applause] [cheering] >> So, yeah, I I just on the on the chip 00:46:26.320 --> 00:46:29.437 front, like, yeah, I definitely don't want to downplay the difficulty of it. 00:46:29.520 --> 00:46:32.957 It's because I think it's accurate to say it's one of the hardest things to do 00:46:33.040 --> 00:46:35.357 in the world. But but I I I do think 00:46:35.440 --> 00:46:38.157 we'll get this done. Um 00:46:38.240 --> 00:46:40.237 and um 00:46:40.320 --> 00:46:43.357 and then I I I I do think that we I 00:46:43.440 --> 00:46:46.477 mean, you know, I I don't want to be 00:46:46.560 --> 00:46:49.277 uh complacent or entitled or anything, 00:46:49.360 --> 00:46:52.077 but but I I I I think probably we'll 00:46:52.160 --> 00:46:53.517 figure out how to make chiffs better 00:46:53.600 --> 00:46:56.317 than anyone, honestly. That's um I'd be 00:46:56.400 --> 00:47:00.877 disappointed if we didn't. Um [applause] but but I don't know how long it's going 00:47:00.960 --> 00:47:04.637 to take us to do that. It might take us uh you know quite a few years but but 00:47:04.720 --> 00:47:09.437 over time I'm pretty sure we could do it. And and then there's 00:47:09.520 --> 00:47:13.800 you know there's there's some pretty interesting 00:47:14.000 --> 00:47:16.637 different ways to do uh transis. Believe 00:47:16.720 --> 00:47:17.757 it or not, there's different ways to do 00:47:17.840 --> 00:47:21.757 transistors that some of which are 00:47:21.840 --> 00:47:26.157 um like use use different types of like 00:47:26.240 --> 00:47:29.837 like they're they're they're using physics that that normally we wouldn't 00:47:29.920 --> 00:47:32.797 use um or aren't used today, put it that 00:47:32.880 --> 00:47:37.437 way. Um and and I think like there 00:47:37.520 --> 00:47:39.677 there's a lot of risk aversion in the 00:47:39.760 --> 00:47:43.197 chip industry. Um and and I think we 00:47:43.280 --> 00:47:44.877 what what we want to do with Tesla and 00:47:44.960 --> 00:47:46.957 SpaceX on the chip fab is is like really 00:47:47.040 --> 00:47:50.957 try some of these um high-risk high 00:47:51.040 --> 00:47:54.157 payoff bets. Um and in order to do that, 00:47:54.240 --> 00:47:56.717 you really have to have uh the complete 00:47:56.800 --> 00:48:00.157 loop of being able to fabricate the the 00:48:00.240 --> 00:48:03.597 logic memory pack do the packaging um do 00:48:03.680 --> 00:48:08.077 the the lithography masks and and and just and just have that iteration loop 00:48:08.160 --> 00:48:09.757 that's all in one building. So you're 00:48:09.840 --> 00:48:13.197 making a complete computer um testing it 00:48:13.280 --> 00:48:17.117 uh then making new masks uh and and then 00:48:17.200 --> 00:48:21.677 a new chip and have that iterative loop be super fast in the same building and 00:48:21.760 --> 00:48:26.237 that's what we're doing at at Giga Texas with the research lab. 00:48:26.320 --> 00:48:28.477 I think that's going to be 00:48:28.560 --> 00:48:31.117 fundamentally a very powerful tool uh 00:48:31.200 --> 00:48:33.757 because before any given technology uh 00:48:33.840 --> 00:48:35.517 the advancement of any given technology 00:48:35.600 --> 00:48:39.357 is is like uh how many iterations and 00:48:39.440 --> 00:48:43.037 and what is the how much progress between iterations. So like what is your 00:48:43.120 --> 00:48:44.877 iteration how many iterations per unit 00:48:44.960 --> 00:48:47.517 time like like how many how many chip 00:48:47.600 --> 00:48:50.477 designs can you do um and can you make 00:48:50.560 --> 00:48:53.997 and test in a year? Um like these days 00:48:54.080 --> 00:48:55.917 like it's the number of it's typically 00:48:56.000 --> 00:48:59.357 two years to for even one for one 00:48:59.440 --> 00:49:01.757 significant trip chip design and that's 00:49:01.840 --> 00:49:04.157 like kind of to go from start startup 00:49:04.240 --> 00:49:06.557 design to to volume fabrication two 00:49:06.640 --> 00:49:09.037 years is considered very fast and it's 00:49:09.120 --> 00:49:11.997 like well okay well but we want to do 00:49:12.080 --> 00:49:15.597 more like 10 designs per year. 00:49:15.680 --> 00:49:20.237 Um, and and and we got to we got to do the do the masks, test test the chip, 00:49:20.320 --> 00:49:23.037 see if because some of these, you know, when you're trying a bunch of different 00:49:23.120 --> 00:49:24.397 things, a bunch of them won't work out. 00:49:24.480 --> 00:49:27.357 So, so I think that would would allow 00:49:27.440 --> 00:49:29.597 for, 00:49:29.680 --> 00:49:31.837 you know, much allow for a much faster 00:49:31.920 --> 00:49:35.197 rate of advancement of of chip designs. 00:49:35.280 --> 00:49:38.077 Elon, let's talk about Optimus. Now, you 00:49:38.160 --> 00:49:40.557 said that the hand has been a really 00:49:40.640 --> 00:49:43.197 difficult challenge in building Optimus. 00:49:43.280 --> 00:49:45.837 what kind of breakthroughs that you have 00:49:45.920 --> 00:49:48.717 seen inside Tesla with V3 and then 00:49:48.800 --> 00:49:50.957 follow-up question is what gives you 00:49:51.040 --> 00:49:54.077 confidence that Optimus can lead in 00:49:54.160 --> 00:49:56.397 humanoid robots compared to Boston 00:49:56.480 --> 00:50:01.960 Dynamics Figure and some of the other competitors out there. 00:50:03.040 --> 00:50:05.997 So the the thing about humanoid robot 00:50:06.080 --> 00:50:10.637 design um if you're actually trying to make something that is a generalized 00:50:10.720 --> 00:50:12.877 uh humanoid robot that can do say 00:50:12.960 --> 00:50:16.397 anything that a person can do um you you 00:50:16.480 --> 00:50:20.557 really have to get the the hands the hands and arms and like the fine motor 00:50:20.640 --> 00:50:25.277 skills are incredibly difficult to get right. Like it's shockingly difficult. 00:50:25.360 --> 00:50:28.637 It's like the things that we kind of take for granted like being able to like 00:50:28.720 --> 00:50:32.157 sign your name. Uh is actually requires 00:50:32.240 --> 00:50:35.837 picking up a p, you know, pen or pencil or something and and then and then 00:50:35.920 --> 00:50:38.557 moving moving it with finesse and and 00:50:38.640 --> 00:50:40.557 it's a bunch of tiny muscle movements to 00:50:40.640 --> 00:50:43.437 do your signature. That's actually very 00:50:43.520 --> 00:50:46.237 hard to do for for a robot. 00:50:46.320 --> 00:50:48.557 Um and and and then like one of the 00:50:48.640 --> 00:50:51.277 hardest things also generally is to do 00:50:51.360 --> 00:50:54.717 precision and force at the same time. 00:50:54.800 --> 00:50:59.277 So, so like if like let's say you want to uh pick up a little screw and do and 00:50:59.360 --> 00:51:04.477 and and do a selftapping screw into wood. Okay, now you 00:51:04.560 --> 00:51:07.437 you know I've done that and like you know sometimes you you damage your 00:51:07.520 --> 00:51:11.597 fingers in the process. um if you if the screw goes sideways, 00:51:11.680 --> 00:51:14.797 but you've got to pick up a little screw and you've got to you've got to you know 00:51:14.880 --> 00:51:18.957 you've got to you've got to push the screwdriver in and twist twist and push 00:51:19.040 --> 00:51:21.517 and that's and and so you've got to be 00:51:21.600 --> 00:51:25.437 precise and apply force. Um that's very 00:51:25.520 --> 00:51:28.157 difficult to do. I I haven't seen 00:51:28.240 --> 00:51:30.077 there's no hand design that I've seen, 00:51:30.160 --> 00:51:34.797 not even sort of a demo that does that. 00:51:34.880 --> 00:51:37.597 But that's what we're trying to achieve with Optimus. 00:51:37.680 --> 00:51:40.717 So, um, and and there are so many little 00:51:40.800 --> 00:51:45.517 nuances in the hand because you've also got to have the touch sensors. The touch 00:51:45.600 --> 00:51:50.877 sensors have to work. They have to be in the right place. Um, 00:51:50.960 --> 00:51:53.997 and you say, "Well, where are you going to put the touch touch sensors?" Well, 00:51:54.080 --> 00:51:57.917 if you put them too close to the surface, they're going to get damaged. 00:51:58.000 --> 00:52:00.877 And then the and then are the touch sensors 00:52:00.960 --> 00:52:05.037 can can they elastically deform like our fingers? Well, actually, it's kind of 00:52:05.120 --> 00:52:08.557 hard to get touch sensors that elastically deform and and can and can 00:52:08.640 --> 00:52:11.357 handle high cycle life and or won't just 00:52:11.440 --> 00:52:16.237 just break after, you know, a few thousand cycles. So, 00:52:16.320 --> 00:52:19.757 um there's all these things that with the human hand that are like incredibly 00:52:19.840 --> 00:52:21.917 difficult to recreate for for a robot 00:52:22.000 --> 00:52:25.197 cuz as a human are we we we constantly 00:52:25.280 --> 00:52:28.957 self-repair. So, your hands are constantly repair 00:52:29.040 --> 00:52:31.677 like your you know all parts of your hand are constantly repairing 00:52:31.760 --> 00:52:33.037 themselves. In fact, if they didn't, 00:52:33.120 --> 00:52:36.317 your your fingers would be worn away. 00:52:36.400 --> 00:52:38.877 Um like like you can wear away the keys 00:52:38.960 --> 00:52:41.357 on your keyboard uh just with typing. 00:52:41.440 --> 00:52:43.677 I've done that um where the letters are 00:52:43.760 --> 00:52:47.677 gone, you know. Um and uh but your your 00:52:47.760 --> 00:52:49.677 my fingers are still there. It's because 00:52:49.760 --> 00:52:52.237 they're they're regenerating thankfully. 00:52:52.320 --> 00:52:56.237 Um so that you don't have that with a robot. 00:52:56.320 --> 00:52:58.317 So then you've got to say, okay, where are you going to put your touch sensors? 00:52:58.400 --> 00:53:02.077 How you going to make sure that that that that that you can you can do things 00:53:02.160 --> 00:53:04.397 like like like like just being able to 00:53:04.480 --> 00:53:08.637 do this like being able to like you know 00:53:08.720 --> 00:53:12.797 do that. Now that that seems like a simple thing 00:53:12.880 --> 00:53:17.037 but like your your your fingers actually bending backwards a little bit and and 00:53:17.120 --> 00:53:19.117 and your your your fingers are kind of 00:53:19.200 --> 00:53:24.157 like squishy and so you're able to form a flat pad. 00:53:24.240 --> 00:53:29.357 So, if you're trying to pick up a small a small item, you you can do so. Um you 00:53:29.440 --> 00:53:32.397 can you can pick up a little screw or like thread a needle or something like 00:53:32.480 --> 00:53:35.757 that. Um 00:53:35.840 --> 00:53:38.157 um now try to do that with with with 00:53:38.240 --> 00:53:40.397 robot fingers. 00:53:40.480 --> 00:53:43.517 Um and and and and be able to do it 00:53:43.600 --> 00:53:46.077 repetitively uh without wearing away the 00:53:46.160 --> 00:53:48.557 fingers uh or damaging the touch 00:53:48.640 --> 00:53:51.597 sensors. Um 00:53:51.680 --> 00:53:55.197 it's actually very difficult. Um to take 00:53:55.280 --> 00:53:57.277 another example, uh trying to open a 00:53:57.360 --> 00:54:00.557 pickle jar. Um I think a robot should be 00:54:00.640 --> 00:54:05.197 able to open a pickle jar, right? So um but sometimes these things gets get get 00:54:05.280 --> 00:54:08.637 in there pretty tight. So then you you you've got to like, you know, you've got 00:54:08.720 --> 00:54:10.557 to like twist the pickle jar, but but 00:54:10.640 --> 00:54:12.637 the but the robot skin the robot skin is 00:54:12.720 --> 00:54:16.157 just rubber over over the, you know, 00:54:16.240 --> 00:54:18.477 plastic and metal internal fingers. 00:54:18.560 --> 00:54:20.717 Freaking robot skin, it doesn't stick to 00:54:20.800 --> 00:54:24.077 it. it doesn't grip to the 00:54:24.160 --> 00:54:26.397 the underlying surface. So then you end 00:54:26.480 --> 00:54:29.517 up like just rotating your skin on your 00:54:29.600 --> 00:54:33.757 rotating the rubber rubber on the on the robot. So you got to figure out, okay, 00:54:33.840 --> 00:54:35.597 now we need to have the robot be able to 00:54:35.680 --> 00:54:38.717 to twist twist things 00:54:38.800 --> 00:54:42.717 uh but but still be able to get the rubber on the on the hand and then you 00:54:42.800 --> 00:54:45.277 got to be a it's got to be replaceable 00:54:45.360 --> 00:54:48.797 because it's going to wear away. 00:54:48.880 --> 00:54:50.077 So all these things these are like 00:54:50.160 --> 00:54:53.117 really difficult and not I'm not sure 00:54:53.200 --> 00:54:57.677 that any one of these is like like let's split like split the atom type of 00:54:57.760 --> 00:54:59.597 difficulty. Um but there's there's a 00:54:59.680 --> 00:55:03.917 zillion of them. Um, 00:55:04.000 --> 00:55:05.517 and and and then you've got you've got 00:55:05.600 --> 00:55:08.477 to wrap that all into the real world AI 00:55:08.560 --> 00:55:12.717 associated with the uh because the the 00:55:12.800 --> 00:55:15.037 AI chip has got to be able to issue the 00:55:15.120 --> 00:55:18.157 the torque commands uh and and and close 00:55:18.240 --> 00:55:22.957 the loop on vision and touch. So 00:55:23.040 --> 00:55:24.237 this is all very difficult like like 00:55:24.320 --> 00:55:28.397 when you see these robot demo videos um 00:55:28.480 --> 00:55:32.077 maybe with rare exception like in any any any complicated looking robot demo 00:55:32.160 --> 00:55:34.557 it's a canned demo. It's not real. 00:55:34.640 --> 00:55:37.677 >> It's just basically 00:55:37.760 --> 00:55:41.997 that's just running a routine. It's it's not real AI. 00:55:42.080 --> 00:55:45.917 So real AI is is like you you show it something it's never seen before and it 00:55:46.000 --> 00:55:49.917 can and it can still handle that situation. 00:55:50.000 --> 00:55:52.717 um like just like like full self-driving 00:55:52.800 --> 00:55:57.197 like you you could actually create a a planet that's not Earth is but is 00:55:57.280 --> 00:55:59.357 similar to Earth and and FSD will still 00:55:59.440 --> 00:56:03.197 work there on on place some you know 00:56:03.280 --> 00:56:07.757 planets it's never been to um or some city certainly on Earth like 00:56:07.840 --> 00:56:10.957 some some city it's never been to or if you change the geography of the city or 00:56:11.040 --> 00:56:13.757 there's you know um a parade or 00:56:13.840 --> 00:56:18.797 something like that FSD will still uh compensate that because it's AI Right. 00:56:18.880 --> 00:56:22.637 Um, so you've got to do that same thing for the robots, but the robots has got a 00:56:22.720 --> 00:56:26.397 lot more degrees of freedom. So like like what I've said about Optimus is 00:56:26.480 --> 00:56:30.717 true, which is that it's it's it's a we are going to get to very high volume 00:56:30.800 --> 00:56:33.917 production, but it's but the initial part of that scurve of of production 00:56:34.000 --> 00:56:38.077 ramp is going to be very slow at the beginning. Um, because not only do you 00:56:38.160 --> 00:56:41.517 have to get all the engineering things that I mentioned right, you also have to 00:56:41.600 --> 00:56:46.397 uh solve the the supply chain for scaling production. and and and unlike 00:56:46.480 --> 00:56:50.877 cars, there's no there's no pre-existing supply chain for robot stuff, at least 00:56:50.960 --> 00:56:54.557 the robot stuff that matters. >> Um 00:56:54.640 --> 00:56:56.637 and then uh you know, we do need to make 00:56:56.720 --> 00:56:58.797 sure that that we can make the robots in 00:56:58.880 --> 00:57:03.757 the US even if um you know uh say China 00:57:03.840 --> 00:57:06.477 were to cut us off. Um because like last 00:57:06.560 --> 00:57:09.197 year there was China China prevented the 00:57:09.280 --> 00:57:12.237 export of uh of of magnets for use in 00:57:12.320 --> 00:57:14.397 electric motors and and that actually 00:57:14.480 --> 00:57:17.197 set back the Optimus program by by like 00:57:17.280 --> 00:57:20.317 2 or 3 months um because we we we 00:57:20.400 --> 00:57:22.717 couldn't get um we had robots with no 00:57:22.800 --> 00:57:25.277 arms. Uh so [laughter] 00:57:25.360 --> 00:57:29.117 we're like a fair a farewell to arms. Um 00:57:29.200 --> 00:57:30.717 you know I was trying to be an arms 00:57:30.800 --> 00:57:34.477 dealer but not succeeding. Um 00:57:34.560 --> 00:57:36.477 so so we've got to make sure that like 00:57:36.560 --> 00:57:39.437 we've got that that we can you know that 00:57:39.520 --> 00:57:41.677 we can make all the parts even if 00:57:41.760 --> 00:57:43.837 another sort of supply chain 00:57:43.920 --> 00:57:47.757 interruption happens. So that that means 00:57:47.840 --> 00:57:51.437 you've got to be able to make all the parts domestically. 00:57:51.520 --> 00:57:55.757 >> So Austin, you know, you always talk about the future should look like the 00:57:55.840 --> 00:57:59.917 future. You know, with the cyber cap being gold, it looks like the future. 00:58:00.000 --> 00:58:01.917 What has been, you know, the hardest 00:58:02.000 --> 00:58:04.157 part to take, you know, here we are. We 00:58:04.240 --> 00:58:07.597 I think there's what, like 190 plus 00:58:07.680 --> 00:58:11.757 cyber cabs registered here in Texas, but like how hard like what's the hardest 00:58:11.840 --> 00:58:17.160 problem to go from Austin to go cross country? 00:58:18.400 --> 00:58:23.117 >> Um, well, cross country, you know, most people you mean like widespread just 00:58:23.200 --> 00:58:28.237 just being in all cities, you don't mean just a trip across country. Um, 00:58:28.320 --> 00:58:31.277 yeah. So, so like because the amount of miles driven that are actually trips 00:58:31.360 --> 00:58:35.997 across country is pretty small. But the the the reason we're moving what what 00:58:36.080 --> 00:58:39.437 seems to be slowly, but although it's actually a very it's a very high 00:58:39.520 --> 00:58:42.957 percentage increase sort of month overmonth percentage increase is 00:58:43.040 --> 00:58:47.837 actually very very rapid. Um but but we want to make sure that we do not have an 00:58:47.920 --> 00:58:52.077 accident. We don't want to cause an injury or death. Um because that that'll 00:58:52.160 --> 00:58:55.117 become like a worldwide headline news thing. you know, even though there's 00:58:55.200 --> 00:58:56.957 about a million people that died in 00:58:57.040 --> 00:59:00.077 automotive accidents per year worldwide, 00:59:00.160 --> 00:59:03.837 um, one such thing with with a Tesla 00:59:03.920 --> 00:59:05.517 robo taxi will be worldwide headline 00:59:05.600 --> 00:59:09.037 news. Um, and of course regulators to 00:59:09.120 --> 00:59:11.437 shut us down. So, so we're just we're 00:59:11.520 --> 00:59:16.157 just making sure where everything is as as safe as possible. Um, 00:59:16.240 --> 00:59:20.237 and and we also want to not run over pets and things. So well maybe what a 00:59:20.320 --> 00:59:25.117 lot of people don't know is like is that like Whimos are are like pet destroyers. 00:59:25.200 --> 00:59:30.717 [laughter] I mean the the the you know um well 00:59:30.800 --> 00:59:35.037 maybe Whimo's like made some improvements on that front but um they 00:59:35.120 --> 00:59:39.117 do run over a lot of pets and and we don't want to run over pets. So, so we 00:59:39.200 --> 00:59:41.517 have these uh challenges like of like of 00:59:41.600 --> 00:59:43.677 like literally like 00:59:43.760 --> 00:59:46.557 you know kittens at night that like gray 00:59:46.640 --> 00:59:50.957 kittens on a gray tarmac with tar seams 00:59:51.040 --> 00:59:53.597 on at night in low visibility and it's 00:59:53.680 --> 00:59:58.237 like can we can we not run over the kittens type of thing. Um 00:59:58.320 --> 01:00:02.077 so that we actually did solve that by the way. Um so then so we'll make a 01:00:02.160 --> 01:00:03.437 change we'll test it out and then we'll 01:00:03.520 --> 01:00:08.397 we'll deploy that. Um and um 01:00:08.480 --> 01:00:13.277 but it we we are making rapid month over month progress. It's it's 01:00:13.360 --> 01:00:16.797 like before we go to a new city, we want to make sure that there's not like some 01:00:16.880 --> 01:00:21.357 uh intersection that's that's problematic or risky. Uh like we we like 01:00:21.440 --> 01:00:24.557 like we want to be particularly careful with like railroad crossings. Sometimes 01:00:24.640 --> 01:00:28.877 there's sometimes it's uncontrolled railroad railroad crossings or 01:00:28.960 --> 01:00:31.037 counterintuitive traffic intersections. 01:00:31.120 --> 01:00:34.077 And these are places where humans make mistakes where you've got a level of 01:00:34.160 --> 01:00:37.997 ambiguity that that like anyone out of town would often make a mistake or 01:00:38.080 --> 01:00:40.557 anyone who doesn't understand the intersection make a mistake. So we want 01:00:40.640 --> 01:00:43.357 to just make sure that that uh we either 01:00:43.440 --> 01:00:48.077 ge initially geopence or can handle all of those circumstances. 01:00:48.160 --> 01:00:49.677 >> Hey, let's talk about the roadster real 01:00:49.760 --> 01:00:51.917 quick since you got uh a little event 01:00:52.000 --> 01:00:56.960 next week. So [cheering] 01:00:58.000 --> 01:01:00.237 >> sure. What what about what makes 01:01:00.320 --> 01:01:02.237 building the Roadster personally 01:01:02.320 --> 01:01:07.757 important to you? And when did it occur to you that you needed to build a car 01:01:07.840 --> 01:01:12.877 that >> Hey guys, sorry about sign off here. Um, 01:01:12.960 --> 01:01:16.717 >> hey, >> hey. Hey. Hey. Want to join in? All 01:01:16.800 --> 01:01:20.407 right. Yes. >> Hey, what's going on? 01:01:20.490 --> 01:01:23.037 [applause and cheering] >> You want to say hello? 01:01:23.120 --> 01:01:27.090 >> Who are they? >> Who are they? [laughter] 01:01:27.280 --> 01:01:29.677 a Tesla owners group. They they own 01:01:29.760 --> 01:01:31.677 Teslas and they're like supporters of 01:01:31.760 --> 01:01:37.570 Tesla. >> Well, yeah. [laughter] 01:01:38.800 --> 01:01:41.677 Um well, there's lot there's like 2,000 01:01:41.760 --> 01:01:46.317 people, I think. So, like quite a lot quite a lot of people. So, um all right, 01:01:46.400 --> 01:01:50.637 guys. I better sign off here. Uh it's been good talking. Um 01:01:50.720 --> 01:01:55.517 >> and so, uh I got I got to go hang out with kids. So, I'll I'll look forward to 01:01:55.600 --> 01:01:58.797 next time. Sounds good. Thank you so much, Elon, 01:01:58.880 --> 01:02:02.637 for your time. >> Oh, and hey, congrats on being on 01:02:02.720 --> 01:02:06.957 getting the Medal of Science uh this past week. Thank you for all you're 01:02:07.040 --> 01:02:12.940 doing for humanity. It's an absolute blast. And [laughter] 01:02:15.040 --> 01:02:18.680 all right, thanks guys. 01:02:25.680 --> 01:02:29.480 I need someone to announce. 01:02:45.680 --> 01:02:48.637 >> I don't have any I don't have any mic. 01:02:48.720 --> 01:02:50.477 >> Don't go far, ladies and gentlemen. 01:02:50.560 --> 01:02:53.437 Don't go far. You are going to win 01:02:53.520 --> 01:02:57.837 $10,000 in prizes. Don't go far. 01:02:57.920 --> 01:03:02.680 >> You guys >> recording stopped. 01:03:03.920 --> 01:03:10.040 >> Hello, >> Landon. Is that live?