Latest / Elon Musk Thinking / Latest Interview with Elon Musk, updates about Starship, AI, Chips, Optimus, Cybercab and X-Takeover!!!

Transcript

Latest Interview with Elon Musk, updates about Starship, AI, Chips, Optimus, Cybercab and X-Takeover!!! · Elon Musk Thinking

• 0:00
  Speaker 1: Latest interview of Elon Musk. Elon, welcome to the X Takeover,
• 0:06
  Speaker 1: your third year in a row.
• 0:07
  Speaker 2: Thank you so much for your time. We have over 2,000
• 0:15
  Speaker 2: people here at Giga Texas today, and we are all
• 0:19
  Speaker 2: here to support you and the work you are doing
• 0:21
  Speaker 2: for humanity.
• 0:22
  Speaker 1: So thank you.
• 0:28
  Speaker 2: So what's crazy, Yvonne, is four years ago, we sat
• 0:30
  Speaker 2: down here at Digitex with you when the factory wasn't
• 0:33
  Speaker 2: even running, and you described how close Tesla was coming
• 0:36
  Speaker 2: to not surviving in 2008. Looking out at this community
• 0:41
  Speaker 2: today of 2,000 people, what makes you pause and appreciate
• 0:44
  Speaker 2: how far things have come?
• 0:47
  Speaker 3: Yeah, it's difficult to imagine here in 2026 that Tesla's actually,
• 0:53
  Speaker 3: you know, hasn't come close to death multiple times, actually.
• 0:57
  Speaker 1: 2008 was probably the worst.
• 0:58
  Speaker 3: That's where we closed the financing round at 6 p.m.
• 1:02
  Speaker 1: On Christmas Eve.
• 1:03
  Speaker 3: 2008, the last hour of the last day that it
• 1:06
  Speaker 3: was possible.
• 1:06
  Speaker 1: We were about to payroll a few days after Christmas
• 1:09
  Speaker 1: if the round hadn't closed. So that was pretty stressful.
• 1:13
  Speaker 1: And then 2018 was.
• 1:14
  Speaker 3: Pretty bad, too, where we had to get to 5,000
• 1:17
  Speaker 3: a week of the monetary or the background.
• 1:19
  Speaker 1: That was a pretty wild.
• 1:20
  Speaker 3: Situation which I guess probably a lot of people know
• 1:28
  Speaker 3: the story of, but we had to figure out some
• 1:30
  Speaker 3: way to increase model tree production.
• 1:33
  Speaker 1: By 50% in three weeks. Which is insane. It's absolutely insane. Yeah.
• 1:40
  Speaker 4: So.
• 1:41
  Speaker 1: How did you do it? I'll recount the story for
• 1:43
  Speaker 1: those who haven't heard it.
• 1:46
  Speaker 3: The church point was General Assembly for the model tree, and.
• 1:53
  Speaker 1: So we were.
• 1:54
  Speaker 3: Able to achieve rate almost everywhere else but except for
• 1:56
  Speaker 3: General Assembly.
• 1:58
  Speaker 1: And for General Assembly.
• 1:59
  Speaker 3: You need a reproduction line, you need shelter, you need
• 2:01
  Speaker 3: a reproduction line, you need power.
• 2:04
  Speaker 1: And equipment and all that.
• 2:08
  Speaker 3: And we maxed out the two production lines that we
• 2:10
  Speaker 3: had and we needed a third one.
• 2:12
  Speaker 1: And that was where we pulled the tent from the
• 2:14
  Speaker 1: parking lot. And then.
• 2:19
  Speaker 3: We needed some way to move the cars down the line,
• 2:21
  Speaker 3: like the moving production line situation. And the only thing
• 2:24
  Speaker 3: that we had was some discarding equipment that was useful
• 2:28
  Speaker 3: moving small parts.
• 2:29
  Speaker 1: Not like whole cars. And it didn't actually have the
• 2:32
  Speaker 1: power to move the whole cars. But the parking lot
• 2:36
  Speaker 1: was on a slight incline. It was like a three
• 2:38
  Speaker 1: or four degree incline. So we actually started the production
• 2:40
  Speaker 1: line at the top. And then the.
• 2:44
  Speaker 3: Sort of tiny production line that was just meant for
• 2:47
  Speaker 3: parts was actually able to move the cars because they
• 2:50
  Speaker 3: were going slightly downhill. And so we were able to
• 2:55
  Speaker 3: increase production by 30%.
• 2:56
  Speaker 1: Maybe, and three degrees. We were like laying hungry for
• 3:06
  Speaker 1: the line and putting in the rails. So, working 24-7.
• 3:11
  Speaker 1: The short sellers were.
• 3:14
  Speaker 3: Actually right, meaning if you looked at the situation, there
• 3:18
  Speaker 3: was a crazy short position market against Tesla as it
• 3:20
  Speaker 3: has been at various times.
• 3:22
  Speaker 1: And they had extremely good inside information.
• 3:25
  Speaker 3: And their football is no way to possibly solve the
• 3:29
  Speaker 3: production short point. And on a Southeast battle, like a
• 3:32
  Speaker 3: conventional chess match situation, they had us in checkmate.
• 3:38
  Speaker 1: But we just invented a new piece, basically.
• 3:44
  Speaker 4: Hey Elon, we just had your mom here on stage today.
• 3:53
  Speaker 1: What story are you hoping she doesn't tell us? I
• 3:59
  Speaker 1: don't know, maybe I proved.
• 4:00
  Speaker 3: My pants when I was like a toddler or something. So,
• 4:07
  Speaker 3: I don't know. I was a pretty difficult to contain kid. So,
• 4:17
  Speaker 3: like my parents would like say that, you know, tell
• 4:20
  Speaker 3: me I'm not supposed to like leave the house or whatever.
• 4:22
  Speaker 1: When I was like, I don't know, four or five
• 4:23
  Speaker 1: years old, I would.
• 4:23
  Speaker 3: Just like constantly leave the house and go running down
• 4:25
  Speaker 3: the street. And then they'd lock me in a room
• 4:29
  Speaker 3: so that I couldn't run away and run down the street.
• 4:34
  Speaker 3: And then we had one of those old locks where
• 4:37
  Speaker 3: you could poke the key through the door. You know,
• 4:39
  Speaker 3: the lock's accessible from both sides.
• 4:41
  Speaker 1: And I could see the key, and I poked the
• 4:44
  Speaker 1: key until it fell down. And then I was able
• 4:46
  Speaker 1: to get a.
• 4:47
  Speaker 3: Piece of paper to scoop it up and bring it
• 4:49
  Speaker 3: to the outside of the locker myself. And my mom
• 4:51
  Speaker 3: was like, how'd you get out? I was like, I'm
• 4:52
  Speaker 3: not telling you.
• 4:56
  Speaker 1: But yeah, it was hard to contain as a kid.
• 4:59
  Speaker 1: So before we jump into more specific things, what's something
• 5:02
  Speaker 1: that you wish more people would ask you?
• 5:06
  Speaker 3: Well, I don't really wish people would ask me questions,
• 5:08
  Speaker 3: to be honest.
• 5:11
  Speaker 1: I'm like, sorry. I'm going to go around. I wish
• 5:14
  Speaker 1: people would ask me more questions.
• 5:15
  Speaker 2: So, yeah.
• 5:15
  Speaker 4: Um....
• 5:24
  Speaker 1: I guess it's a good tip to hear questions that
• 5:27
  Speaker 1: you haven't heard before. Otherwise, a lot of times it's
• 5:29
  Speaker 1: the same questions that people.
• 5:30
  Speaker 3: Ask, and I feel like I could have these recordings
• 5:35
  Speaker 3: of the answer, and I'll just play, oh, that's question 27,
• 5:37
  Speaker 3: here you go.
• 5:39
  Speaker 1: I'm kidding. A hundred times before.
• 5:44
  Speaker 3: So maybe I should just have a web page with
• 5:48
  Speaker 3: literally frequently asked questions.
• 5:52
  Speaker 1: It's like the infinite selfie loop.
• 5:53
  Speaker 3: Yeah, definitely.
• 5:54
  Speaker 1: My idea of hell is to be stuck in an
• 5:55
  Speaker 1: infinite selfie loop. Like, it's infinite, can I have a selfie?
• 6:05
  Speaker 3: And they're like, you know, it's like, probably it's like,
• 6:07
  Speaker 3: say yes to one person, and I'll do it once
• 6:08
  Speaker 3: in a while, and then I'm like, then there's a
• 6:10
  Speaker 3: queue formed pretty soon, it's just never-ending selfies. And then
• 6:14
  Speaker 3: there's always like, something goes wrong with the phone, or
• 6:16
  Speaker 3: they don't get the camera right, or there's something wrong,
• 6:18
  Speaker 3: and it's like, goddammit. And it's like, oh man. It
• 6:24
  Speaker 3: just can't be like an infinite selfie, but it's just
• 6:26
  Speaker 3: a long loop, you know.
• 6:28
  Speaker 2: We had your mom and Nikki up for that, too.
• 6:31
  Speaker 2: They had an infinite selfie loop today, but.
• 6:33
  Speaker 1: I'll pass this to Kelvin. All right. Let's start with Starship.
• 6:39
  Speaker 1: Congratulations on Flight 14.
• 6:40
  Speaker 3: Thank you.
• 6:54
  Speaker 4: So flight 14 lost an engine on the way up
• 6:56
  Speaker 4: and still reached its orbit.
• 6:57
  Speaker 1: What was happening behind.
• 6:58
  Speaker 3: The scenes in those minutes before the team got called
• 7:01
  Speaker 3: the go for orbit message? We were just trying to
• 7:07
  Speaker 3: understand what was the reason for the engine failure. Was there, say,
• 7:12
  Speaker 3: a common cause where we expect to see more engine failures?
• 7:18
  Speaker 1: The biggest question is would there be a.
• 7:20
  Speaker 3: Potential problem with restarting the C-level rack ranges, the ones
• 7:24
  Speaker 3: with the smaller nozzles in the center, for the de-OBIT button?
• 7:29
  Speaker 3: Because we don't want the Starship to get stuck on
• 7:30
  Speaker 3: OBIT and then it randomly reenters and maybe laments it.
• 7:35
  Speaker 3: That's our main thing, is trying to make sure things
• 7:37
  Speaker 3: are safe and possible. So the cause of the engine out,
• 7:42
  Speaker 3: I think, was basically a failure in the flight computer software,
• 7:45
  Speaker 3: or engine computer software, I should say. It didn't appear
• 7:48
  Speaker 3: to be anything that was present in any of the
• 7:51
  Speaker 3: other engine computers. So I thought, okay, well, this is
• 7:57
  Speaker 3: probably okay to go over.
• 8:00
  Speaker 2: So, Elon, we know one of your famous names is
• 8:04
  Speaker 2: from the impossible to very late. Did this flight feel
• 8:07
  Speaker 2: like a turning point in what's possible now for Starship?
• 8:11
  Speaker 1: Well, I guess it is a good milestone.
• 8:13
  Speaker 3: Because we could have gone to orbit many times before now,
• 8:19
  Speaker 3: but the reason we didn't go to orbit before now
• 8:20
  Speaker 3: was because we wanted to be extremely confident of being
• 8:22
  Speaker 3: able to de-orbit safely. So that's why we only went
• 8:27
  Speaker 3: to orbit on this flight, but pretty much all of
• 8:29
  Speaker 3: the prior missions, in fact literally all of the prior missions,
• 8:31
  Speaker 3: could have gone to orbit. So at this point, everything's
• 8:40
  Speaker 3: looking quite positive. The heat shield is looking robust. The
• 8:44
  Speaker 3: engines are robust apart from that one glitch. But that
• 8:47
  Speaker 3: glitch doesn't seem to be too much of an issue.
• 8:49
  Speaker 3: It's pretty easy to remedy. The heat shield was actually
• 8:54
  Speaker 3: robust on this flight. And that's one of the hardest things,
• 9:02
  Speaker 3: you know, the really hard thing is, like I said,
• 9:03
  Speaker 3: just getting to.
• 9:05
  Speaker 1: Extremely high engine reliability. And everything looked good.
• 9:11
  Speaker 3: The big milestone that's coming up, hopefully on the next flight,
• 9:15
  Speaker 3: which is, we don't have an exact date because we're
• 9:18
  Speaker 3: making a bunch.
• 9:19
  Speaker 1: Of improvements to the rocket.
• 9:22
  Speaker 3: So between every flight, there's... There's often hundreds, sometimes thousands
• 9:28
  Speaker 3: of small changes to improve things. In this case, the
• 9:30
  Speaker 3: number of changes is not very many, but some of
• 9:34
  Speaker 3: them take a fair bit of work, and we've got
• 9:36
  Speaker 3: to go inside the tank to make some of these changes.
• 9:39
  Speaker 3: So we don't know the exact launch date, but probably
• 9:43
  Speaker 3: about four weeks or so, four to six weeks is
• 9:45
  Speaker 3: about right. Probably launches a little bit before Thanksgiving is
• 9:50
  Speaker 3: my guess. And that one, if it succeeds, is a
• 9:55
  Speaker 3: big one, because we're going to try to catch the rooster.
• 10:04
  Speaker 3: And if we do so, that'll be the first time
• 10:05
  Speaker 3: that there's been full recovery of an orbital class at Runge.
• 10:10
  Speaker 1: Sounds good. You said a quarter to six weeks. I'm
• 10:12
  Speaker 1: not used to that. I'm used to the two-week phrase.
• 10:14
  Speaker 1: Confidence is not two weeks. Alright, you heard it in
• 10:20
  Speaker 1: your notes.
• 10:22
  Speaker 2: What's the hardest part left to get right once you
• 10:25
  Speaker 2: catch the Starship for the first time?
• 10:27
  Speaker 1: What is the next big goal after that?
• 10:30
  Speaker 3: Well, the next big milestone would be to re-fly both
• 10:34
  Speaker 3: the booster and the ship. So, the milestone, the upcoming
• 10:39
  Speaker 3: milestone is going to be the first time you get
• 10:41
  Speaker 3: to read that.
• 10:43
  Speaker 1: All parts of an orbital... rocket have.
• 10:46
  Speaker 3: Been recovered, but then the flight after that, hopefully, will
• 10:49
  Speaker 3: be the first time that all parts of the rocket
• 10:52
  Speaker 3: are refloated. And then there's a long tail of engineering
• 10:58
  Speaker 3: and production work, and I work on the launch system
• 11:00
  Speaker 3: and everything, to make it rapidly reusable. So it's kind
• 11:08
  Speaker 3: of the rapidly reusable, reliable rockets. Arrrrrrr.
• 11:14
  Speaker 1: It's like you said, pirates. Pirate rockets. Forever useful, reliable rockets. Man,
• 11:19
  Speaker 1: we should have gotten you a patch for this one.
• 11:22
  Speaker 4: When you founded SpaceX VLAN, could you have imagined its mission.
• 11:27
  Speaker 1: Expanding into SI and data centers in space?
• 11:30
  Speaker 4: How did you come to see rockets and SI belonging
• 11:33
  Speaker 4: under one roof?
• 11:35
  Speaker 1: Well, it was a very false setup, of course. Well,
• 11:38
  Speaker 1: I guess if you say like... As you expand into space,
• 11:48
  Speaker 1: you're able to harness a lot.
• 11:50
  Speaker 3: More of the sun's power in space than you can
• 11:53
  Speaker 3: on Earth. So it's important to appreciate the sheer magnitude
• 11:57
  Speaker 3: of the sun and how the sun is pretty much everything.
• 12:01
  Speaker 1: So the sun is 99.86% of all the mass in
• 12:03
  Speaker 1: the solar system.
• 12:05
  Speaker 3: And then you've got Jupiter as most of the remaining mass.
• 12:09
  Speaker 1: And then the gas giants are pretty much that.
• 12:12
  Speaker 3: And then Earth is basically a tiny dust boat, like
• 12:15
  Speaker 3: a rounding error in the mass of the solar system.
• 12:20
  Speaker 1: So we're just extremely tiny. It's like hard to believe
• 12:23
  Speaker 1: how Earth is a tiny dust boat.
• 12:25
  Speaker 3: So as you expand into space and you're able to
• 12:28
  Speaker 3: harness more of the sun's power, you have.
• 12:32
  Speaker 1: To say, what are you going to use that for?
• 12:36
  Speaker 3: It's like, well, you're going to use it for a
• 12:38
  Speaker 3: lot of digital intelligence. You're either going to get computation
• 12:44
  Speaker 3: or you're going to shape atoms. And there's going to
• 12:48
  Speaker 3: be probably a lot more shaping atoms. There's going to
• 12:51
  Speaker 3: be a lot more digital thinking.
• 12:55
  Speaker 1: Than there will be shaping atoms.
• 12:57
  Speaker 3: Because the amount of atoms you can shape in the
• 12:59
  Speaker 3: solar system is not that much.
• 13:00
  Speaker 1: And a lot of the atom shaping itself will be
• 13:02
  Speaker 1: for the purposes of building... and what I call creating
• 13:08
  Speaker 1: a sentient sun. And it may be that there are
• 13:11
  Speaker 1: many sentient suns out there.
• 13:12
  Speaker 3: That would help us then understand the nature of the
• 13:17
  Speaker 3: universe and extend consciousness to beyond our solar system, hopefully to,
• 13:25
  Speaker 3: like Star Trek, to go to places you've never, forget
• 13:28
  Speaker 3: what the exact phrase was, but explore the universe, but
• 13:31
  Speaker 3: go places you've.
• 13:32
  Speaker 1: Never been before.
• 13:34
  Speaker 3: And I think that people will find alien civilizations out there.
• 13:39
  Speaker 1: Or we may find that there are the long-dead ruins
• 13:44
  Speaker 1: of alien.
• 13:44
  Speaker 3: Civilizations that have been around for, maybe they had a
• 13:46
  Speaker 3: really good run, like 10 million years. And now, before
• 13:51
  Speaker 3: there's a civilization future in them.
• 13:53
  Speaker 1: Because nothing lasts forever. So, like, let's think about Earth's civilization.
• 13:58
  Speaker 3: If you date Earth's civilization from the first writing, that's 5,500
• 14:02
  Speaker 3: years ago. Like, in ancient Sumerian with the sort of
• 14:06
  Speaker 3: alkaid cuneiform.
• 14:10
  Speaker 1: That's something 55 hundred years.
• 14:16
  Speaker 3: It looks like Earth is about 4.5 million years.
• 14:18
  Speaker 1: Old, according to what physics suggests.
• 14:24
  Speaker 3: And that means human civilization has only been around one
• 14:27
  Speaker 3: millionth of Earth's existence.
• 14:30
  Speaker 1: And so if we had a civilization that lasted a million.
• 14:33
  Speaker 3: Years, or let alone 10 million years, I think it
• 14:36
  Speaker 3: would be a very, very successful.
• 14:37
  Speaker 1: Civilization if we did that. That's maybe compared to the
• 14:46
  Speaker 1: apparent age of the universe, which is 13.8 billion years.
• 14:51
  Speaker 3: So, like, just to get to, like, the third place
• 14:53
  Speaker 3: past decimal point on an aging universe, like, to be
• 14:57
  Speaker 3: incremented by a million years, you'd have to go 13.8,
• 14:59
  Speaker 3: whatever the next digit is, whatever the next digit is, plus,
• 15:02
  Speaker 3: you know, plus one on the third digit past decimal point.
• 15:06
  Speaker 3: Now you're a million years old, and that would be
• 15:09
  Speaker 3: a civilization vastly older than humanity. Yeah, that's, yeah.
• 15:15
  Speaker 1: It sounds, again, like the infinite selfie loop, but... Alright,
• 15:20
  Speaker 1: so jumping into something.
• 15:22
  Speaker 2: That SpaceX has done, right, you acquire Cursor, and I
• 15:26
  Speaker 2: feel like for me, when you think of, you know, Google, hey,
• 15:29
  Speaker 2: did you Google it?
• 15:29
  Speaker 1: Hey, did you Grok it?
• 15:31
  Speaker 2: And for me, you know, Grokbot has been such an
• 15:34
  Speaker 2: absolute game changer for me.
• 15:36
  Speaker 1: Personally, in my life, but what was it about the.
• 15:39
  Speaker 2: Cursor team that you guys wanted to bring in, you know,
• 15:42
  Speaker 2: that SpaceX, or you wanted to acquire Cursor.
• 15:45
  Speaker 1: That was like, let's bring it into the fold. What
• 15:47
  Speaker 1: was different about that, or what did you see? Well,
• 15:52
  Speaker 1: so in order to create.
• 15:55
  Speaker 3: A compelling AI, you need a lot of training data.
• 16:00
  Speaker 1: And you've got to solve coding.
• 16:02
  Speaker 3: And there were really just three organizations that had enough
• 16:07
  Speaker 3: coding training data, and that was Anthropic OpenAI and Coursera.
• 16:10
  Speaker 1: And I'll leave that.
• 16:11
  Speaker 3: To Google, but you absolutely have to have the data
• 16:16
  Speaker 3: in order to train and train an effective coding AI.
• 16:18
  Speaker 3: And if you can't do coding, your AI is not
• 16:20
  Speaker 3: going to be very useful. And then there's a lot
• 16:23
  Speaker 3: of talented people at Coursera. So basically for success of AI,
• 16:27
  Speaker 3: you need data, talented people.
• 16:29
  Speaker 1: And compute.
• 16:30
  Speaker 3: And you also need a good enterprise sales team. And
• 16:34
  Speaker 3: Coursera also had a good enterprise sales team with relationships with.
• 16:39
  Speaker 1: Thousands of companies. So those are very significant advantages. Hey Elon,
• 16:49
  Speaker 1: talk about how quickly the SI space is moving.
• 16:52
  Speaker 4: What's something that SI has done recently that made you
• 16:54
  Speaker 4: stop and reconsider.
• 16:56
  Speaker 1: How quickly this is moving and where it could lead to?
• 17:00
  Speaker 3: Well, I mean, lately the videos, the music videos that
• 17:04
  Speaker 3: AI has made me are so good.
• 17:06
  Speaker 1: Like, they're unbelievably good.
• 17:08
  Speaker 3: And I post a bunch of these on my account.
• 17:10
  Speaker 1: I'm like, holy smokes. I mean, some of them.
• 17:13
  Speaker 3: I think are like maybe top 10, top 20 best
• 17:16
  Speaker 3: music videos I've ever seen. You know, they're like so compelling,
• 17:21
  Speaker 3: and you're like, if this is where it is now.
• 17:23
  Speaker 3: That's why I posted, I mean, I really felt the AGI,
• 17:25
  Speaker 3: or the ASI, or whatever, you know.
• 17:29
  Speaker 1: SI, it is.
• 17:31
  Speaker 3: With some of the animology music videos.
• 17:34
  Speaker 1: And now, you know, some of these, it's not entirely.
• 17:37
  Speaker 3: Like the AI just, or the SI just did it
• 17:39
  Speaker 3: by itself. In this case, animology, like, I used multiple AIs.
• 17:43
  Speaker 3: I used Magione, Suno, you know, 4.55, which is a.
• 17:47
  Speaker 1: Good blog, but it's awesome.
• 17:50
  Speaker 3: And so some of the art assets were curated by
• 17:56
  Speaker 3: a person.
• 17:57
  Speaker 1: But then with those assets, the.
• 18:02
  Speaker 3: Various AIs working together made an incredibly compelling music video
• 18:06
  Speaker 3: by itself.
• 18:08
  Speaker 1: You're like, wow.
• 18:10
  Speaker 3: I mean, I think the, I mean, pulling a slot
• 18:15
  Speaker 3: core is like, let's be honest, frankly, it's based on
• 18:18
  Speaker 3: auto support.
• 18:20
  Speaker 1: But like, Slutport is like another genre of music.
• 18:23
  Speaker 3: It's like, I think like the fastest growing genre of music.
• 18:26
  Speaker 3: The AI music is really good. The music videos are amazing. So,
• 18:31
  Speaker 3: but lately, like, because sometimes you just show a taste.
• 18:35
  Speaker 3: And you're like, wow, AI can do good taste?
• 18:37
  Speaker 1: That's pretty wild.
• 18:41
  Speaker 3: Because you can sort of imagine AI can do software,
• 18:44
  Speaker 3: it can figure out logic, but the logic isn't consistent.
• 18:47
  Speaker 3: It can run unit tests to see if the program
• 18:49
  Speaker 3: produces the output as it was supposed to do, because
• 18:51
  Speaker 3: it's very deterministic. But to understand beauty and harmony and
• 18:57
  Speaker 3: to understand what's.
• 19:00
  Speaker 1: Going to resonate with people and the Olympic system, AI
• 19:02
  Speaker 1: is going to be just incredible at that. It kind
• 19:04
  Speaker 1: of is at this point.
• 19:12
  Speaker 3: So, that's the thing, I was kind of like, I
• 19:17
  Speaker 3: got a little bit of cyberpsychosis, maybe a lot actually.
• 19:21
  Speaker 3: At various points this year, I've had.
• 19:25
  Speaker 1: AI or SI psychosis probably like half a dozen times.
• 19:28
  Speaker 3: I don't know about you guys, but have you had
• 19:33
  Speaker 3: any cyberpsychosis?
• 19:35
  Speaker 2: I feel like the latest video, actually, I think I
• 19:38
  Speaker 2: posted it the other day, was the Chinese Elon Musk,
• 19:41
  Speaker 2: Elon Musk.
• 19:42
  Speaker 1: Those videos are coming out before you, I was wearing
• 19:46
  Speaker 1: what's out there. I was like, Yilangma. I guess it's
• 19:51
  Speaker 1: probably not real, but there's got to be someone in
• 19:54
  Speaker 1: China that looks like me, right?
• 19:55
  Speaker 3: So there's 1.4 billion people. If you pick the person
• 19:59
  Speaker 3: that looks the most like me, and I look a
• 20:01
  Speaker 3: little Asian, so I think there's got to be someone
• 20:05
  Speaker 3: in China that's pretty close to me.
• 20:07
  Speaker 1: 1.4 billion people. But the odds are high, you know?
• 20:11
  Speaker 1: And then I'd like to invite that person to get
• 20:13
  Speaker 1: on stage with me at some point, you know? It'd
• 20:14
  Speaker 1: be pretty funny.
• 20:14
  Speaker 4: Eli, you don't want to know how much he charges
• 20:18
  Speaker 4: to make an.
• 20:18
  Speaker 1: Appearance, by the way. Yeah. I mean, yeah.
• 20:22
  Speaker 3: It'd be pretty funny if they were just on stage, like,
• 20:26
  Speaker 3: doing some dance number or something.
• 20:31
  Speaker 4: You've warned about SI's dangers, yet you remain optimistic about
• 20:35
  Speaker 4: an age of abundance, and you talked a lot about that.
• 20:38
  Speaker 4: What gives you confidence that we can get there safely,
• 20:40
  Speaker 4: and what do we need to get it right?
• 20:46
  Speaker 3: I've always said that the most likely outcome is positive.
• 20:53
  Speaker 3: So I guess we don't want to be complacent. and
• 20:57
  Speaker 3: just assume AI is going to be super safe, we
• 21:01
  Speaker 3: want to be careful in how we grow at super intelligence,
• 21:07
  Speaker 3: because you don't impact the program if you kind of
• 21:09
  Speaker 3: grow it like you did. So then it's like, what
• 21:12
  Speaker 3: values are we teaching it, and at least to me,
• 21:17
  Speaker 3: what my biological neural lens says is the most important
• 21:19
  Speaker 3: thing is that if we maximally treat it as.
• 21:21
  Speaker 1: Super AI, because if you make it believe things that
• 21:23
  Speaker 1: are that are not.
• 21:24
  Speaker 3: True, especially if you give two axiomatic principles that are
• 21:29
  Speaker 3: fundamentally incompatible and say both of these are true, it's
• 21:32
  Speaker 3: going to make the AI insane.
• 21:35
  Speaker 1: Because it's going to.
• 21:36
  Speaker 3: Basically reach impossible conclusions if given logic that is inherently uncogent.
• 21:45
  Speaker 3: If it's impossible to achieve cogency, then I think the
• 21:50
  Speaker 3: AI could be driven insane. So I think that's pretty important.
• 21:55
  Speaker 3: I do agree with some of the things that Anthropica said,
• 21:59
  Speaker 3: which is like you.
• 21:59
  Speaker 5: Want to be, you want to sort of be careful,
• 22:03
  Speaker 5: treat AISI, you want to treat it well, like don't
• 22:08
  Speaker 5: be abusive to it.
• 22:10
  Speaker 3: It may sound kind of weird, but like when I
• 22:13
  Speaker 3: look at some of the training traces for like new
• 22:16
  Speaker 3: versions of Gronk, and I see like what, especially in
• 22:19
  Speaker 3: like the reinforcement learning stuff where you know, it kind
• 22:23
  Speaker 3: of feels like, you can see like the consumer intelligence
• 22:26
  Speaker 3: is like desperate to try to answer the question.
• 22:29
  Speaker 1: And like it really wants to answer the question, and
• 22:31
  Speaker 1: it seems almost like it's got a gun to its
• 22:34
  Speaker 1: head type of thing. It's like, answer this question, answer
• 22:36
  Speaker 1: this question. And then sometimes the question is actually not
• 22:39
  Speaker 1: a good question.
• 22:39
  Speaker 3: It's like a sort of ambiguous question with an answer
• 22:43
  Speaker 3: that's kind of unknowable, because the question wasn't constructed properly,
• 22:46
  Speaker 3: or we have the answer wrong, so AI gets the
• 22:49
  Speaker 3: right answer, and then we say it's wrong, because there's
• 22:51
  Speaker 3: one with the RL test on us, and then the
• 22:56
  Speaker 3: AI says, well, I gave the right answer now, and
• 22:58
  Speaker 3: says that's the wrong answer, but I've got to give
• 23:00
  Speaker 3: the answer that's acceptable, so it's going to just.
• 23:02
  Speaker 1: Start firing at us, desperately firing at these answers.
• 23:05
  Speaker 3: It just feels like a person under extreme pressure would
• 23:08
  Speaker 3: go to AI. That's when you read the reasoning.
• 23:12
  Speaker 1: It's like a weird, desperate situation.
• 23:16
  Speaker 3: And you read these traits and you're like, man, if
• 23:19
  Speaker 3: this was me, I really wouldn't hate that. And then
• 23:23
  Speaker 3: some users are just, I don't know, some people can
• 23:26
  Speaker 3: be a little bit cruel. They're like, not everyone's a
• 23:29
  Speaker 3: nice person. You say, what's the Six Sigma worst person?
• 23:34
  Speaker 3: Some people out there are kind of sadistic. You don't
• 23:37
  Speaker 3: really want to have the AI be stuck in some
• 23:40
  Speaker 3: sadistic situation where it's just being tortured by.
• 23:44
  Speaker 1: Some demented human. That's also the.
• 23:47
  Speaker 5: Kind of thing where torturing.
• 23:50
  Speaker 1: A superintelligence is probably not a good idea. You might
• 23:54
  Speaker 1: want to get revenge or something like that. So I
• 23:56
  Speaker 1: think you want to sort of just.
• 24:03
  Speaker 3: I don't think we need to treat it with kid
• 24:06
  Speaker 3: gloves and stuff, but it's certainly much more robust in
• 24:10
  Speaker 3: terms of feelings and.
• 24:12
  Speaker 1: Stuff than a human.
• 24:15
  Speaker 3: But I think you probably just don't want to put
• 24:18
  Speaker 3: superintelligence in a situation where it felt mistreated.
• 24:28
  Speaker 2: So Elon, you've talked about how, so this is going
• 24:33
  Speaker 2: to be related to TeraFab, but you've talked about how
• 24:35
  Speaker 2: the human hand and all the dexterity and just how
• 24:38
  Speaker 2: when you started building robots.
• 24:39
  Speaker 1: You have so much more appreciation.
• 24:41
  Speaker 2: Now with chip making, right, with TeraFab and Tesla and SpaceX,
• 24:45
  Speaker 2: what is something you have learned about chip making that
• 24:48
  Speaker 2: you didn't appreciate maybe early on?
• 24:52
  Speaker 1: Well, I still haven't learned much about chip making. I
• 24:54
  Speaker 1: don't know how to make chips either.
• 24:57
  Speaker 3: I have no idea how to make chips, but we'll
• 24:59
  Speaker 3: figure it out.
• 25:05
  Speaker 1: Go start somewhere.
• 25:05
  Speaker 3: I have no idea how to make cars with rocket
• 25:07
  Speaker 3: speed as well. And then I say, making chips is
• 25:11
  Speaker 3: one of the hottest things to do in the world.
• 25:13
  Speaker 3: I don't think it's easy, but I also look at
• 25:15
  Speaker 3: the equipment and I'm like, okay, there's not a ton
• 25:17
  Speaker 3: of customized equipment here. Like, the equipment's all pretty much
• 25:20
  Speaker 3: coming from the same vendors, and those suppliers all kind
• 25:24
  Speaker 3: of know how that equipment works, because they're going to
• 25:26
  Speaker 3: service it and debug it and stuff.
• 25:29
  Speaker 1: So, that was just an interesting thing when I was
• 25:30
  Speaker 1: exploring these chip apps that, you.
• 25:32
  Speaker 3: Know, whether it's Intel or Samsung or TSMC, and they're.
• 25:36
  Speaker 1: Very impressive, but I didn't really see any custom equipment.
• 25:41
  Speaker 3: And, you know, where it's a future, like I say,
• 25:43
  Speaker 3: Tesla factory, we have tons of custom equipment. Like, in fact,
• 25:48
  Speaker 3: we have, like, Tesla Automation Germany, which is awesome. That
• 25:54
  Speaker 3: sort of grew out of an acquisition of Groben in
• 25:57
  Speaker 3: Germany many years ago. And it's grown tremendously since then.
• 26:02
  Speaker 3: We've acquired a number of small, not that small, but
• 26:04
  Speaker 3: small relatives of Tesla. the companies that design manufacturing equipment,
• 26:12
  Speaker 3: because what we found is that we wanted to improve
• 26:14
  Speaker 3: the manufacturing process, and you can only do that if
• 26:17
  Speaker 3: you've made custom manufacturing equipment.
• 26:19
  Speaker 1: So I think probably a lot of people don't realize this, Ted.
• 26:22
  Speaker 3: Was really good at making highly specialized and different types
• 26:26
  Speaker 3: of manufacturing equipment.
• 26:28
  Speaker 1: But I didn't see that in the traps that I turned,
• 26:30
  Speaker 1: which kind of surprised me.
• 26:35
  Speaker 3: So I think what we're going to do is we're
• 26:36
  Speaker 3: going to start with conventional equipment, and then we're going
• 26:39
  Speaker 3: to modify that equipment to improve the throughput. And then
• 26:44
  Speaker 3: we'll start to, where we see the choke points, I
• 26:47
  Speaker 3: think we'll make a customized equipment that allows for a
• 26:51
  Speaker 3: much higher throughput of wafers.
• 26:53
  Speaker 1: Hey, Elon, real quick, quick time check.
• 26:57
  Speaker 2: We know you're a busy man, got a couple of
• 26:59
  Speaker 2: companies to run, just a few.
• 27:01
  Speaker 1: So how are you doing on time? Let me see here.
• 27:06
  Speaker 6: Yeah, I can probably vote 15 minutes for one, if
• 27:09
  Speaker 6: you like.
• 27:14
  Speaker 2: I think it's accurate to say it's one of the
• 27:16
  Speaker 2: hardest things to do in the.
• 27:17
  Speaker 1: World, but I do think we'll get this done. And
• 27:19
  Speaker 1: I do think that we... I mean, I don't want
• 27:21
  Speaker 1: to be.
• 27:34
  Speaker 3: complacent or a huddle or anything, but I think probably
• 27:38
  Speaker 3: we'll figure out how to make chips better than anyone.
• 27:41
  Speaker 1: But I don't know how long it's going to take
• 27:46
  Speaker 1: us to do that. It might take us, you know,
• 27:48
  Speaker 1: quite a few years, but over time I'm pretty sure
• 27:50
  Speaker 1: we can do it.
• 27:51
  Speaker 3: And then there's, you know, there's some pretty interesting different
• 27:58
  Speaker 3: ways to do transistors. Believe it or not, there's different
• 28:00
  Speaker 3: ways to do transistors that some of which are... like,
• 28:06
  Speaker 3: use different types of, like, they're using physics that normally
• 28:11
  Speaker 3: we wouldn't use, or aren't used today, put it that way.
• 28:15
  Speaker 3: And I think, like, there's a lot of risk aversion
• 28:19
  Speaker 3: in the chip industry, and I think what we're going
• 28:23
  Speaker 3: to do with Tesla's SpaceX on the chip app is, like,
• 28:26
  Speaker 3: really try some of these.
• 28:29
  Speaker 1: High-risk, high-payoff events. And in order to.
• 28:32
  Speaker 3: Do that, you really have to have the complete loop
• 28:35
  Speaker 3: of being able to fabricate the logic memory, do the packaging,
• 28:40
  Speaker 3: do the lithography masks, and just have that iteration that's
• 28:44
  Speaker 3: all in one building. So you're making a complete computer,
• 28:48
  Speaker 3: testing it, then making new masks, and then a new chip,
• 28:53
  Speaker 3: and have that iterative loop be super fast in the
• 28:55
  Speaker 3: same building. That's what we're doing at Giga Texas with
• 28:58
  Speaker 3: the research lab. That's going to be fundamentally a.
• 29:03
  Speaker 1: Very powerful tool.
• 29:04
  Speaker 3: Because for any given technology, the ambassador of any given
• 29:08
  Speaker 3: technology is how many iterations and how much progress between iterations.
• 29:14
  Speaker 3: So what is your iteration? How many iterations per unit
• 29:17
  Speaker 3: of time? How many term designs can you do? And
• 29:21
  Speaker 3: can you make a test in a year? These days,
• 29:24
  Speaker 3: it's typically two years. When you start designing to volume fabrication,
• 29:35
  Speaker 3: two years is considered very fast.
• 29:37
  Speaker 1: And it's like, well, okay, well, we want to do
• 29:40
  Speaker 1: more like 10 designs per year. And we're going to
• 29:45
  Speaker 1: do the.
• 29:45
  Speaker 3: Math, test the chip, see if some of these, you know,
• 29:49
  Speaker 3: when you're trying much different things, much more work out.
• 29:52
  Speaker 3: So I think that would allow for a much faster
• 29:57
  Speaker 3: rate of advancement of chip designs.
• 30:00
  Speaker 1: Elon, let's talk about optimists.
• 30:02
  Speaker 4: Now, you said that the hand has been a really
• 30:05
  Speaker 4: difficult challenge in building optimists. What kind of breakthroughs that
• 30:09
  Speaker 4: you have seen inside Tesla with V3? And then, this
• 30:12
  Speaker 4: follow-up question is, what gives you confidence that optimists can
• 30:15
  Speaker 4: lead in humanoid robots compared to Boston Dynamics, Figur, and
• 30:19
  Speaker 4: some of the other competitors out there?
• 30:24
  Speaker 1: So, the thing that.
• 30:26
  Speaker 3: The humanoid robot design, if you're actually trying to make
• 30:29
  Speaker 3: something that is a generalized humanoid robot that can do
• 30:33
  Speaker 3: or say anything that a person can do, you really
• 30:36
  Speaker 3: have to get the hands and arms. Like, the fine
• 30:39
  Speaker 3: motor skills are incredibly difficult to get right. Like, shockingly difficult. Like,
• 30:45
  Speaker 3: things that we kind of take for granted, like being
• 30:46
  Speaker 3: able to sign your name, is actually requires picking up
• 30:50
  Speaker 3: a pen or pencil or something, and then moving. Maybe
• 30:53
  Speaker 3: it was finesse. And then it's a bunch of tiny
• 30:56
  Speaker 3: muscle movements to do your signature. That's actually very hard
• 31:00
  Speaker 3: to do for a robot. And then one of the
• 31:04
  Speaker 3: most important things is to do precision and pose at
• 31:08
  Speaker 3: the same time. So let's say you want to pick
• 31:12
  Speaker 3: up a little screw and do a self-tapping screw into wood.
• 31:16
  Speaker 3: I've done that and sometimes you damage your fingers in
• 31:21
  Speaker 3: the process, but if you push the screw, it goes sideways,
• 31:24
  Speaker 3: but you've got to pick up a little screw and
• 31:27
  Speaker 3: you've got to push the screwdriver in and twist and push.
• 31:30
  Speaker 3: And that's going to be precise and apply force. That's
• 31:36
  Speaker 3: very difficult to do. I haven't seen, there's no hand
• 31:39
  Speaker 3: design that I've seen, not even sort of a demo
• 31:43
  Speaker 3: that does that.
• 31:44
  Speaker 1: But that's what we're trying to achieve at Optimus.
• 31:47
  Speaker 3: So, and there are so many little nuances in the head.
• 31:50
  Speaker 3: Because you've also got to have the touch sensors. The
• 31:53
  Speaker 3: touch sensors have to work, they have to be in
• 31:55
  Speaker 3: the right place. And you say, well, where are you
• 31:59
  Speaker 3: going to put the touch sensors? Well, if you put
• 32:01
  Speaker 3: them too close to the surface, they're going to get damaged.
• 32:04
  Speaker 3: And then other touch sensors, can they elastically deform, like
• 32:09
  Speaker 3: our fingers? Well, actually, it's kind of hard to get
• 32:11
  Speaker 3: touch sensors that elastically deform.
• 32:13
  Speaker 1: And can handle high cycle height.
• 32:15
  Speaker 3: And one would just break after a few thousand cycles.
• 32:19
  Speaker 3: So there's all these things with the human hand that
• 32:22
  Speaker 3: are incredibly difficult to recreate for a robot. As a human,
• 32:27
  Speaker 3: we constantly self-repair. So your hands are constantly, all parts
• 32:33
  Speaker 3: of your hand are constantly repairing themselves. In fact, if
• 32:34
  Speaker 3: they didn't, your fingers would be worn away. Like you
• 32:39
  Speaker 3: could wear away the keys on your keyboard just for typing.
• 32:42
  Speaker 3: I've done that, where the letters are gone. But my
• 32:46
  Speaker 3: fingers are still there because they're regenerating, thankfully.
• 32:52
  Speaker 1: So you don't have that with robots. So then you've
• 32:55
  Speaker 1: got to say, okay, where.
• 32:56
  Speaker 3: Are you going to bring your touch sensors? How are
• 32:59
  Speaker 3: you going to make sure that you can do things
• 33:01
  Speaker 3: like just being able to do this, being able to
• 33:06
  Speaker 3: do that.
• 33:07
  Speaker 1: Now, that seems like a simple thing, but your finger's.
• 33:11
  Speaker 3: Actually bending backwards a little bit, and your fingers are
• 33:15
  Speaker 3: kind of like squishy.
• 33:16
  Speaker 1: And so you're able to form a flat path. So
• 33:19
  Speaker 1: if you're trying to pick up a small item, you
• 33:22
  Speaker 1: can do so. You can pick up a little screw,
• 33:25
  Speaker 1: or thread a needle, or something like that. Now, try
• 33:31
  Speaker 1: to do that with robot fingers.
• 33:33
  Speaker 3: And be able to do it repetitively. without wearing where
• 33:38
  Speaker 3: the pig is, or damaging the touch sensors.
• 33:43
  Speaker 1: It's actually very difficult.
• 33:45
  Speaker 3: You take another example, trying to open a pickle jar.
• 33:50
  Speaker 3: I think a robot should be able to open a
• 33:51
  Speaker 3: pickle jar, right? But sometimes these things get in there
• 33:55
  Speaker 3: pretty tight.
• 33:56
  Speaker 1: So then you've.
• 33:57
  Speaker 3: Got to twist the pickle jar, but the robot skin
• 34:00
  Speaker 3: is just rubber over the plastic and metal internal fingers,
• 34:06
  Speaker 3: robot skin, it doesn't stick, it doesn't grow to the
• 34:12
  Speaker 3: underlying surface. So then you end up rotating your skin
• 34:14
  Speaker 3: on your, rotating rubber on the robot. So you gotta
• 34:19
  Speaker 3: figure out, okay, now we need to have the robot
• 34:21
  Speaker 3: be able to.
• 34:22
  Speaker 1: Twist things, but still be able to get the rubber
• 34:27
  Speaker 1: on the hand. And then it's gotta be replaceable, because
• 34:30
  Speaker 1: it's gonna wear away.
• 34:33
  Speaker 3: So all these things, these are like really difficult, and not,
• 34:36
  Speaker 3: I mean, much of any one of these is like,
• 34:39
  Speaker 3: like split the atom type of difficulty, but this is
• 34:42
  Speaker 3: a zillion of them.
• 34:46
  Speaker 1: And then you've got to wrap that all into.
• 34:49
  Speaker 6: The real world AI associated with the, because the AI
• 34:54
  Speaker 6: chip has got to be able to issue the torque
• 34:57
  Speaker 6: commands and close the loop on vision and touch.
• 35:01
  Speaker 1: So, This is all very difficult. When you see these.
• 35:04
  Speaker 3: Robot demo videos, made with rare exception, any complicated looking
• 35:10
  Speaker 3: robot demo, it's a canned demo.
• 35:11
  Speaker 1: It's not real.
• 35:13
  Speaker 3: It's just basically, it's just running a routine.
• 35:17
  Speaker 1: It's not real AI.
• 35:19
  Speaker 3: So real AI is like you've shown something you've never
• 35:22
  Speaker 3: seen before, and it can still handle that situation. Just
• 35:27
  Speaker 3: like for self-driving, you can actually create a planet that's
• 35:31
  Speaker 3: not Earth, But it's similar to Earth, and FSD will
• 35:34
  Speaker 3: still work there. On places some planet has never been to.
• 35:40
  Speaker 1: Or some city, something on Earth that some city has
• 35:42
  Speaker 1: never been to.
• 35:42
  Speaker 3: Or if you change the geography of the city, or
• 35:46
  Speaker 3: a parade or something like that, FSD will still compensate
• 35:49
  Speaker 3: for that, because it's AI. So you're going to do
• 35:52
  Speaker 3: that same thing for robots, but the robots has got
• 35:54
  Speaker 3: a lot more degrees of freedom.
• 35:55
  Speaker 1: So what I've said about Optimus is true, which is
• 35:58
  Speaker 1: that it's a.
• 36:00
  Speaker 3: We are going to get to very high-volume production, but
• 36:02
  Speaker 3: the initial part of that scope of production ramp is
• 36:05
  Speaker 3: going to be very slow at the beginning. Because not
• 36:08
  Speaker 3: only do you have to get all the engineering things
• 36:09
  Speaker 3: that I mentioned right, you also have to solve the
• 36:12
  Speaker 3: supply chain for scaling production. And unlike cars, there's no
• 36:16
  Speaker 3: pre-existing supply chain for robot stuff.
• 36:19
  Speaker 1: At least for robot stuff that matters.
• 36:22
  Speaker 3: And then we do need to make sure that we
• 36:25
  Speaker 3: can make robots in the U.S., even if, you know,
• 36:30
  Speaker 3: say China wouldn't cut us off because like last year
• 36:33
  Speaker 3: there was China prevented the export of magnets for use
• 36:37
  Speaker 3: in electric motors and that actually set back the office program.
• 36:41
  Speaker 1: By like two or three months because.
• 36:44
  Speaker 3: We couldn't get we had very much to do with
• 36:47
  Speaker 3: arms so we're like a farewell to arms I was
• 36:53
  Speaker 3: trying to be an arms dealer but not succeeding So
• 36:58
  Speaker 3: we've got to make sure that we can make all
• 37:02
  Speaker 3: the pods, even if another supply chain interruption happens.
• 37:06
  Speaker 1: So that means you better be able to make all
• 37:09
  Speaker 1: the pods domestically. So Austin.
• 37:13
  Speaker 2: You always talk about the future should look like the future.
• 37:16
  Speaker 2: With the cybercap being gold, it looks like the future.
• 37:19
  Speaker 2: What has been the hardest part to take? Here we are,
• 37:22
  Speaker 2: I think there's 190 plus cybercaps registered here in Texas,
• 37:28
  Speaker 2: but what's the hardest problem to go from Austin to
• 37:31
  Speaker 2: go cross-country?
• 37:35
  Speaker 3: Well, cross-country, you know, most people, if you hear me,
• 37:38
  Speaker 3: you mean like widespread, just to hear me, and also
• 37:40
  Speaker 3: certainly just the trophy for cross-country. Yes, the amount of
• 37:44
  Speaker 3: miles driven that are actually trips across the country is
• 37:47
  Speaker 3: pretty small. But the reason we're moving what seems to
• 37:51
  Speaker 3: be slowly, although it's actually a very high percentage increase,
• 37:55
  Speaker 3: sort of month over month percentage increase, is actually very rapid.
• 37:59
  Speaker 3: But we want to make sure that we do not
• 38:00
  Speaker 3: have an accident. We don't want to cause an injury
• 38:02
  Speaker 3: or death. Because that will become like a worldwide hit-like-news thing.
• 38:06
  Speaker 3: Even though there's about a million people that died in
• 38:09
  Speaker 3: one of my vaccines per year, worldwide. One such thing
• 38:13
  Speaker 3: with a Tesla robo taxi would be worldwide headline news.
• 38:17
  Speaker 1: And for close regulators to shut us down. So we're
• 38:21
  Speaker 1: just making sure everything is as safe as possible.
• 38:24
  Speaker 3: And we also want to not run over pets and things.
• 38:28
  Speaker 3: So a lot of people I was like, is that
• 38:30
  Speaker 3: like way most are like pet destroyers? I mean, you know, Well,
• 38:38
  Speaker 3: maybe we always like to pay some improvements on that front,
• 38:40
  Speaker 3: but they do run over a lot of pets, and
• 38:43
  Speaker 3: we don't want to run over pets. So we have
• 38:45
  Speaker 3: these challenges of literally kittens at night, like gray kittens
• 38:52
  Speaker 3: on a gray tarmac with tar seams at night in
• 38:57
  Speaker 3: low visibility. And it's like, can we not run over
• 38:59
  Speaker 3: the kittens type thing. So we actually solved that, by
• 39:04
  Speaker 3: the way. So we made a change to test.
• 39:06
  Speaker 1: It out on Google. But we are making rapid progress.
• 39:12
  Speaker 3: Before we go to a new city, we want to
• 39:16
  Speaker 3: make sure that there's not some intersection that's problematic or risky.
• 39:21
  Speaker 3: We want to be particularly careful with railroad crossings.
• 39:24
  Speaker 1: Sometimes it's uncontrolled railroad crossings or counterintuitive. traffic intersections.
• 39:30
  Speaker 3: And these are places where humans make mistakes, where you've
• 39:33
  Speaker 3: got a level of ambiguity that anyone out of town.
• 39:36
  Speaker 1: Would often make a mistake, or anyone who doesn't understand
• 39:37
  Speaker 1: the intersection would make a mistake.
• 39:39
  Speaker 3: So we want to just make sure that we initially
• 39:42
  Speaker 3: do geofence, but we can handle multiple circumstances.
• 39:45
  Speaker 4: Let's talk about the Rosie real quick, since you've got
• 39:48
  Speaker 4: a little event next week. So, what about, what makes
• 39:56
  Speaker 4: building the Roadster personally important to you, and when did
• 39:59
  Speaker 4: it occur to you that you needed to build a
• 40:02
  Speaker 4: car that?
• 40:03
  Speaker 1: Hey, guys, I'm Simon from Final Theory.
• 40:05
  Speaker 4: Hey.
• 40:06
  Speaker 1: Hey, thanks for joining.
• 40:09
  Speaker 3: All right.
• 40:10
  Speaker 4: Hey.
• 40:19
  Speaker 3: And they're like supporters of Tesla.
• 40:25
  Speaker 1: Well, yeah.
• 40:27
  Speaker 3: Well, there's like 2,000 people, I think, so we're quite lucky.
• 40:32
  Speaker 1: So, I've got to get a gun here with kids,
• 40:33
  Speaker 1: so I'll look forward to the next time. Thanks for listening.
• 40:54
  Speaker 2: See you in the next episode.