Latest / Elon Musk Thinking / Latest Interview of Elon Musk.

Transcript

Latest Interview of Elon Musk. · Elon Musk Thinking

• 0:00
  Speaker 1: Latest interview of Elon Musk. What's the reason to put
• 0:03
  Speaker 1: them in face? Well, the availability of energy is the issue.
• 0:10
  Speaker 1: So I mean, if you look at electrical output outside
• 0:15
  Speaker 1: of China, everywhere outside of China, it's more or less flat.
• 0:17
  Speaker 1: It's where you know, maybe slide increase, but for pretty
• 0:20
  Speaker 1: much flat. China has a rapid increase in electrical output.
• 0:24
  Speaker 1: But if you're putting data centers of work except China,
• 0:26
  Speaker 1: where you're gonna get your electricity, especially as you scale,
• 0:30
  Speaker 1: the output of chips is growing pretty much exponentially, but
• 0:34
  Speaker 1: the output of electricity is flat. So how are you
• 0:36
  Speaker 1: going to tell them what chips are? You know, magical
• 0:40
  Speaker 1: power sources, magical electricity ferries.
• 0:43
  Speaker 2: I mean your famous, your famous epic fan of solar
• 0:46
  Speaker 2: one terra a lot of solar power. So with the
• 0:48
  Speaker 2: twenty five percent computed factor, like four terrawats of solar panels,
• 0:51
  Speaker 2: it's like one percent of the.
• 0:52
  Speaker 1: Land area of the United States.
• 0:54
  Speaker 2: And that's like far in this You're in the singularity
• 0:56
  Speaker 2: when we've got one tarrat wat of data centers, right,
• 0:59
  Speaker 2: So what are you running out of?
• 1:01
  Speaker 1: Exactly? How far into the singularity are you? You tell me? Yeah,
• 1:04
  Speaker 1: exactly so I think I think we'll find we're in
• 1:06
  Speaker 1: the singularity and like, okay, let's look a long way
• 1:08
  Speaker 1: you go.
• 1:10
  Speaker 2: But is this like a is the plan to like
• 1:12
  Speaker 2: put it in the space after we will covered Nevada
• 1:14
  Speaker 2: and solar panels.
• 1:15
  Speaker 1: I think it's pretty hard to cover about sol panels.
• 1:17
  Speaker 1: You have to get like permits from Like the proach
• 1:19
  Speaker 1: for try.
• 1:20
  Speaker 2: Getting the pomots for that space is really it's really
• 1:23
  Speaker 2: regulatory play. It's like hard harder to build on land
• 1:26
  Speaker 2: than it is in space.
• 1:27
  Speaker 1: It's harder to scale on the ground than it is
• 1:31
  Speaker 1: to scale in space. But also the the you're going
• 1:36
  Speaker 1: to get about five times the effectiveness of sold panels
• 1:41
  Speaker 1: in space versus the ground, and you don't need batteries.
• 1:45
  Speaker 1: I almost wore my other shirt, which says it's always
• 1:47
  Speaker 1: sunny in space, which it is so because you don't
• 1:54
  Speaker 1: have a day night cycle or seasonality clouds, or or
• 2:00
  Speaker 1: an atmosphere in space. Because the atmosphere alone, we're still
• 2:06
  Speaker 1: about a thirty percent loss of energy. So you're can
• 2:12
  Speaker 1: any given soul panels can do about five times more
• 2:15
  Speaker 1: powering space than on the ground, and you avoid the
• 2:18
  Speaker 1: cost of having batteries to carry you through the night.
• 2:22
  Speaker 1: So it's actually much cheaper to do in space, and
• 2:26
  Speaker 1: my prediction is that it will be by far the
• 2:30
  Speaker 1: cheapest place to put AI will be space and thirty
• 2:34
  Speaker 1: six months or less, maybe thirty thursday months less than
• 2:37
  Speaker 1: thirty six months.
• 2:38
  Speaker 2: How do you service GPUs as they fail, which happens
• 2:40
  Speaker 2: quite often in training, Actually it.
• 2:45
  Speaker 1: Depends on how recent the GPUs are that arrived. I mean,
• 2:48
  Speaker 1: at this point we find our gps to be quite reliable.
• 2:52
  Speaker 1: There's infant mortality which you can obviously iron out on
• 2:54
  Speaker 1: the ground, so you can just run them on the
• 2:56
  Speaker 1: ground and confirm that you don't have infort mortality with
• 2:59
  Speaker 1: what the GPU. But once they once they start working,
• 3:02
  Speaker 1: their actual reliability and once they start working and you're
• 3:05
  Speaker 1: past the initial you know, debug cycle of video or whatever,
• 3:09
  Speaker 1: or whoever's making the trips could be Tesla Tesla AI
• 3:14
  Speaker 1: six trips or something like that, or could be you know,
• 3:17
  Speaker 1: TPUs or trainings or whatever the rivalry is. Actually they're
• 3:22
  Speaker 1: quite reliable past certain point. So I don't think I
• 3:28
  Speaker 1: don't think you need that. The servicing thing is an issue,
• 3:32
  Speaker 1: but you can mark my words, uh in thirty six months,
• 3:37
  Speaker 1: but probably closer to thirty months. The most economically compelling
• 3:41
  Speaker 1: place to put AI will be space, and then and
• 3:44
  Speaker 1: and and then it will get from it'll it'll then'
• 3:47
  Speaker 1: forget like ridiculously better to be in space. And then
• 3:51
  Speaker 1: the scaling. The only place you can really scale is space.
• 3:56
  Speaker 1: You know, once you start thinking in terms of what
• 3:58
  Speaker 1: percentage of the Sun's power are you, honestly you realize
• 4:01
  Speaker 1: you have to go to space. You can't scale very
• 4:05
  Speaker 1: very much on Earth, but by very much, to be clear,
• 4:08
  Speaker 1: you're talking like terro arts. Yeah, well all of the
• 4:13
  Speaker 1: United States currently uses only half a terrawork power on average, right,
• 4:18
  Speaker 1: so you know, if you say a terror work, that
• 4:20
  Speaker 1: would be twice as much electricity as the United States
• 4:23
  Speaker 1: currently consumes. So that's quite a lot. And can you
• 4:26
  Speaker 1: imagine building that many data centers, that many power plants.
• 4:31
  Speaker 1: It's like those who have lived in software land don't
• 4:34
  Speaker 1: realize that they're about to have a hard lesson in hardware.
• 4:40
  Speaker 1: That there's there's it's actually very difficult to build power plants.
• 4:45
  Speaker 1: And and then you don't need just need you need
• 4:47
  Speaker 1: power plants, you need all of the electrical equipment, need
• 4:49
  Speaker 1: that the electrical transformers to run the transformers, the transformers.
• 4:55
  Speaker 1: Now you told the industry is a very slow industry.
• 4:58
  Speaker 1: That they are they pretty much uh you know, the
• 5:02
  Speaker 1: impedis smash to the to the government, to the Public
• 5:04
  Speaker 1: Utility Commission. So they're the impedius smash like literally invigoraily.
• 5:10
  Speaker 1: So they're very slow because the history of their past
• 5:13
  Speaker 1: has been very slow. So trying to get them to
• 5:16
  Speaker 1: move fast is like, you know, like if you're trying
• 5:19
  Speaker 1: to do an interconnect agreement with you have you ever
• 5:20
  Speaker 1: tried to do an Internet interconnect agreement with a utility
• 5:23
  Speaker 1: at scale? Like I put a lot of power.
• 5:25
  Speaker 2: As a professional podcaster, I can say it and I'm not.
• 5:27
  Speaker 2: In fact, the need many more before that comes in
• 5:32
  Speaker 2: this year, they.
• 5:32
  Speaker 1: Have to do a study for a year. Okay, like
• 5:35
  Speaker 1: a year later they'll come back to you with their
• 5:37
  Speaker 1: interconnected study.
• 5:38
  Speaker 3: Can you tell this with your own behind the meter
• 5:41
  Speaker 3: power stuff.
• 5:43
  Speaker 1: You can build power plants. Yeah, that's what we did. Actually,
• 5:45
  Speaker 1: I fu clusses to it so for full classes too.
• 5:49
  Speaker 3: So yeah, while we're talking about the bridge, why not
• 5:50
  Speaker 3: just like build GPUs and power co located. That's what
• 5:53
  Speaker 3: we did, right, right, But I'm saying, why isn't this
• 5:55
  Speaker 3: generalized solution when you're talking about.
• 5:56
  Speaker 1: All the issues where you get the power plants from?
• 5:57
  Speaker 3: I'm saying, when you talk about the issues working working
• 5:59
  Speaker 3: with you two those use you can just build private
• 6:01
  Speaker 3: power plants with the with the data centers, right, But
• 6:04
  Speaker 3: it begs the question.
• 6:04
  Speaker 1: Of where do you get the power plants? Where do
• 6:06
  Speaker 1: where do you get the power plants from? I mean
• 6:08
  Speaker 1: the power plant makers.
• 6:10
  Speaker 3: As what you're saying, like, does the gas turbine backlog?
• 6:12
  Speaker 3: Basically yes, you can drill down to level further. It's
• 6:17
  Speaker 3: the it's the the veins and blades in the turbines
• 6:21
  Speaker 3: that of the limiting factor because the casting may it's
• 6:25
  Speaker 3: it's like a very specialized process to cast the blades
• 6:28
  Speaker 3: and veins in the in the in the turbines using
• 6:31
  Speaker 3: gas power, and it's very it's very difficult to scale.
• 6:35
  Speaker 3: Other other forms of power you can scale potentially solar,
• 6:39
  Speaker 3: but the towers currently for importing solar in the US
• 6:42
  Speaker 3: are gigantic and the domestic solar production is protophal. Why
• 6:46
  Speaker 3: that makes solar? That seems like a good Elon shaped problem.
• 6:49
  Speaker 3: We are going to make solar?
• 6:50
  Speaker 1: Okay. Both SpaceX and Tesla are are bulling towards one
• 6:56
  Speaker 1: hundredguart tier of solar cell production.
• 6:59
  Speaker 2: How low down the like from polysilica and up to
• 7:02
  Speaker 2: the wayfer to the final panel.
• 7:05
  Speaker 1: I think you've got to do the whole thing for
• 7:06
  Speaker 1: more materials to finish to sell. Now, if it's going
• 7:10
  Speaker 1: to space, it's actually it costs. It costs less than
• 7:13
  Speaker 1: it's easier to make solo cells that go to space
• 7:15
  Speaker 1: because they don't need glass or they don't need much glass,
• 7:18
  Speaker 1: and they don't need a heavy framing because they don't
• 7:20
  Speaker 1: have to sell survive weather events. There's no weather in space.
• 7:24
  Speaker 1: So it's actually a cheaper solo cell that goes to
• 7:26
  Speaker 1: space than the one on the ground. Is there a
• 7:29
  Speaker 1: path to getting them as cheap as you need in
• 7:32
  Speaker 1: the next thirty six months. Solo cells are already very cheap.
• 7:38
  Speaker 1: They're like partically cheap. It's and if you say, you know,
• 7:43
  Speaker 1: I think like solo cells in China are around like
• 7:46
  Speaker 1: twenty five thirty cents a water or something like that.
• 7:48
  Speaker 1: It's it's absurdly cheap. And when you when you're taking
• 7:51
  Speaker 1: to a cap now now put it now put it
• 7:53
  Speaker 1: in space, and it's five times cheaper because it's five
• 7:56
  Speaker 1: times in fact, no, it's not five times cheap, it's
• 7:58
  Speaker 1: ten times cheaper because you don't need any batteries. So
• 8:03
  Speaker 1: so the moment your cost of access to space becomes low,
• 8:08
  Speaker 1: by far, the cheapest and most scalable way to generate,
• 8:11
  Speaker 1: to generate tokens is space. It's not even close. It
• 8:17
  Speaker 1: will be an order of magnitude easier to scale, and
• 8:22
  Speaker 1: ships aside an order of magnitude. Well, if the point
• 8:24
  Speaker 1: is you won't be able to scale the ground, it's
• 8:26
  Speaker 1: just you just won't only hit the big time on
• 8:29
  Speaker 1: power generation there already are so so like that. The
• 8:34
  Speaker 1: number of so mercles and series that the x A
• 8:36
  Speaker 1: I team had to accomplish in order to get a
• 8:38
  Speaker 1: gig out of power online was was was crazy. We
• 8:42
  Speaker 1: had to getting together a whole bunch of turbines and
• 8:46
  Speaker 1: uh and then and then we had permit issues in
• 8:50
  Speaker 1: Tennessee and and had to go across the border to Mississippi,
• 8:53
  Speaker 1: which is fortunately only you know, a few miles away.
• 8:56
  Speaker 1: Uh So, but then we still had to run the
• 8:58
  Speaker 1: high power lines a few miles and build a power
• 9:01
  Speaker 1: plant in Missipi. And it was very difficult. Abu that
• 9:05
  Speaker 1: and people don't understand, like how much how much electrocity
• 9:08
  Speaker 1: do you actually need at the generator level, at the
• 9:11
  Speaker 1: generation level in order to power a data center, because
• 9:13
  Speaker 1: they look at the the nerves will look at the
• 9:17
  Speaker 1: power consumption of say a GB three hundreds and multiply
• 9:21
  Speaker 1: that by thing and then think best amount amount of
• 9:23
  Speaker 1: power you need? Wake up. Yeah, this is like, that's
• 9:27
  Speaker 1: a that's a that's a total moves You've never done
• 9:30
  Speaker 1: any hardware in your life before. Besides the GB three hundred,
• 9:35
  Speaker 1: you've got to power all of the networking hardware. There's
• 9:37
  Speaker 1: a whole bunch of CPU and storage stuff that's happening.
• 9:40
  Speaker 1: You've got a size for your your peak cooling requirements,
• 9:46
  Speaker 1: so that means can you cool even on the worst
• 9:49
  Speaker 1: hour of the worst day of the year. Well, it's
• 9:51
  Speaker 1: pretty freaking hot in Memphis, So so you're going to
• 9:55
  Speaker 1: have like a forty percent increase on your power just
• 9:58
  Speaker 1: for cooling. If we assuming you don't want your data
• 10:02
  Speaker 1: center to turn off on hot days and anyone to
• 10:04
  Speaker 1: keep going, then then you've got to say, well, there's
• 10:08
  Speaker 1: a There's there's another multiplicative element on top of that,
• 10:10
  Speaker 1: which is are you assuming that you're you you never
• 10:13
  Speaker 1: have any hiccups in your power generation like well, actually
• 10:17
  Speaker 1: sometimes you have to take the generators some of the
• 10:19
  Speaker 1: power offline in order to service it. Oh okay, now
• 10:21
  Speaker 1: you add another twenty twenty five percent multiplier on that
• 10:24
  Speaker 1: because you've got it. You've got to assume that that
• 10:27
  Speaker 1: you've got to take power offline to service it. So
• 10:30
  Speaker 1: the actual rs roughly every every one hundred and ten
• 10:36
  Speaker 1: thousand gbs, GB three hundreds, inclusive of networking, CPU, storage, cooling.
• 10:43
  Speaker 1: Margin for for for servicing power is roughly three hundred megawatts.
• 10:50
  Speaker 1: Sorry you say that again, it's roughly, or think about it, like,
• 10:55
  Speaker 1: what do you think about like three hundred and thirty
• 10:57
  Speaker 1: thousand to actually what you need at the generative generation level.
• 11:02
  Speaker 1: So service probably service three hundred and thirty thousand TV
• 11:06
  Speaker 1: three hundreds including all of the associated support networking and
• 11:09
  Speaker 1: everything else and the and the peak cooling and to
• 11:13
  Speaker 1: have some margin some power marginers of is roughly a
• 11:16
  Speaker 1: GEO why can ask a very night your question?
• 11:18
  Speaker 2: Yeah, you know, you're describing the engineering details of doing
• 11:24
  Speaker 2: this stuff on Earth, but then there's analogous engineering difficulties
• 11:29
  Speaker 2: of doing it in space.
• 11:29
  Speaker 1: How do you do the.
• 11:31
  Speaker 2: How do you replace in pin and band with orbital lasers,
• 11:34
  Speaker 2: et cetera, etceter How do you make it resistant radiation?
• 11:38
  Speaker 2: I don't know the details in the engineering, but fundamentally,
• 11:40
  Speaker 2: what is the reason to think those challenges which have
• 11:43
  Speaker 2: never been had to be addressed before will end up
• 11:46
  Speaker 2: being easier than just like building more turbines on Earth?
• 11:49
  Speaker 2: There's companies that build turbines on Earth. They can make
• 11:51
  Speaker 2: more turbines right.
• 11:52
  Speaker 1: And again, tried doing it and then you'll see so
• 12:01
  Speaker 1: like the turbines are sold out through twenty thirty, have
• 12:03
  Speaker 1: you guys considered making your own? I think in order
• 12:07
  Speaker 1: for in order to bring enough power online, I think
• 12:12
  Speaker 1: SpaceX Intezl will probably have to make the turbine blades,
• 12:18
  Speaker 1: the bans and blades internally, but just the blades or
• 12:22
  Speaker 1: the turbines the limiting factor. You can get everything except
• 12:28
  Speaker 1: the blades they call the blades and veins you can
• 12:32
  Speaker 1: get that twelve to eighteen months before the veins of blades.
• 12:36
  Speaker 1: The limiting factor of the veins of blades. And there
• 12:39
  Speaker 1: are only three casting companies in the world that may
• 12:44
  Speaker 1: make these, and they're massively backloged. Is this siemens ge
• 12:48
  Speaker 1: those guys, or is it a subje. No, it's a
• 12:51
  Speaker 1: it's other companies. I mean sometimes they have a little
• 12:53
  Speaker 1: bit of casting capability in house. But I'm just saying you.
• 12:56
  Speaker 1: You can just call any of the turbine makers and
• 12:58
  Speaker 1: they will tell you top secret. They're probably on the
• 13:01
  Speaker 1: it's probably on the internet right now. If if it
• 13:03
  Speaker 1: wasn't for the terraffs, what would Colossus be solar powered? Uh,
• 13:07
  Speaker 1: it would be much easier to make it solar powered. Yeah,
• 13:11
  Speaker 1: the tariffs are not so several hundred percent, So don't
• 13:14
  Speaker 1: you know some big book we will will see it speed. Yeah. No,
• 13:18
  Speaker 1: you know, President has a you know, we don't agree
• 13:22
  Speaker 1: on everything, and this administration is not not the biggest
• 13:27
  Speaker 1: man of of solar. So it's you know, it's it's
• 13:33
  Speaker 1: it's it's We also need the land, the permits and everything.
• 13:36
  Speaker 1: So if you try to move very fast like I
• 13:41
  Speaker 1: do think scaling solar on Earth it is a is
• 13:44
  Speaker 1: a good way to go, but but you need you
• 13:46
  Speaker 1: do need some amount of time to find the land,
• 13:48
  Speaker 1: get the permits, get the solar pair that with the batteries.
• 13:51
  Speaker 3: Well why would it not work to stand up your
• 13:55
  Speaker 3: own solar production? And then you're right that you eventually
• 13:57
  Speaker 3: run out of land. But there's a lot of land
• 13:59
  Speaker 3: here in Texas. There's a lot of land in Nevada,
• 14:01
  Speaker 3: including private land. It's not all publicly on land, and
• 14:04
  Speaker 3: so you'd be able to at least get the next
• 14:06
  Speaker 3: colossus and like the next one after that, and at
• 14:08
  Speaker 3: a certain point you hit a wall.
• 14:09
  Speaker 1: But wouldn't that work for the moment? As I said,
• 14:12
  Speaker 1: we are scaling solar production, there's there's a rate. There's
• 14:17
  Speaker 1: a rate at which you can scale physical production of
• 14:19
  Speaker 1: solo solar cells. Where I'm we're going as fast as
• 14:23
  Speaker 1: possible in scaling domestic production. You're making the solar cells
• 14:27
  Speaker 1: at TEA. We TESA and SpaceX have a mandate to
• 14:31
  Speaker 1: get to one hundred gigs a year of solar.
• 14:33
  Speaker 3: Speaking of the annual capacity, I'm furious. In five years time,
• 14:37
  Speaker 3: let's say, what will the installed capacity be on Earth
• 14:42
  Speaker 3: and in space. I deliberately picked five years because it's
• 14:46
  Speaker 3: after you're once we're up and running threshold and so
• 14:49
  Speaker 3: in five years time, Yeah, what's thee on Earth versus
• 14:51
  Speaker 3: in space installed a capacity?
• 14:54
  Speaker 1: Five years? I think probably you said five years from
• 14:58
  Speaker 1: now were probably AI in space will be uh launching
• 15:07
  Speaker 1: every year the sum total of all AI on Earth
• 15:11
  Speaker 1: in excess of me. Five years from now, my prediction
• 15:15
  Speaker 1: is we will launch and and and be operating every
• 15:20
  Speaker 1: year more AI in space than this than the cumulative
• 15:23
  Speaker 1: total on Earth, which is I would expect to be
• 15:27
  Speaker 1: at least sort of five years from now, a few
• 15:30
  Speaker 1: hundred gigawatts per year of of AI in space and rising,
• 15:38
  Speaker 1: so you can get to I think on Earth you
• 15:42
  Speaker 1: can get to a round a tarawa a year of
• 15:45
  Speaker 1: AI in space before you start having you know, fuel
• 15:50
  Speaker 1: supplied challenges for the rocket.
• 15:52
  Speaker 3: Okay, but you think you can get to one hundreds
• 15:54
  Speaker 3: of gigabatts per year in five years, yes.
• 15:57
  Speaker 2: So one hundred gigawatts depending on the specific power of
• 16:02
  Speaker 2: the whole system with solar arrays and radios and everything,
• 16:05
  Speaker 2: is on the order of like ten thousand starship launches, yes,
• 16:11
  Speaker 2: and you want to do that in one year, and
• 16:13
  Speaker 2: so that's like one starship launch every hour. Yeah, that's
• 16:17
  Speaker 2: happening in the city, Like walking through a world where
• 16:21
  Speaker 2: there's ten times there's a starship launch every single hour.
• 16:24
  Speaker 1: Yeah, I mean that's actually a lower rate compared to airlines,
• 16:27
  Speaker 1: like like aircraft aircraft, there's a lot of airports, a
• 16:30
  Speaker 1: lot of airports, and you gotta launch you know, the
• 16:32
  Speaker 1: polar orbit, and it doesn't have to be polarp. But
• 16:36
  Speaker 1: you're just there's there's some some value to sound sagres.
• 16:40
  Speaker 1: But but I think actually you just go high enough,
• 16:46
  Speaker 1: you're you're start getting out of it. Both shadow and
• 16:48
  Speaker 1: so how.
• 16:50
  Speaker 2: Many physical starships are needed to do ten thousand lunches
• 16:53
  Speaker 2: a year?
• 16:54
  Speaker 1: I don't think we'll lead more than I mean you could.
• 16:56
  Speaker 1: You could probably do it with as few as like
• 17:01
  Speaker 1: twenty or thirty, Like it really depends on how quickly
• 17:07
  Speaker 1: does the ship. The ship has to go around the Earth,
• 17:11
  Speaker 1: and the ground track for the ship has to come
• 17:13
  Speaker 1: back over the launch pad. So if you can use
• 17:15
  Speaker 1: a ship every say, thirty hours, you could do it
• 17:19
  Speaker 1: with thirty ships. But we'll make more shifts than that.
• 17:22
  Speaker 1: But the SpaceX is is going up to do ten
• 17:29
  Speaker 1: thousand marches a year and maybe even twenty or thirty
• 17:32
  Speaker 1: thousand marches. Ye know.
• 17:33
  Speaker 2: It is the idea to become basically a hyperscaler, become
• 17:36
  Speaker 2: an oracle and lend this capacity to other people. What's
• 17:39
  Speaker 2: what are you going to do with presumableast? SpaceX is
• 17:41
  Speaker 2: the one watching all this, so SpaceX become a hyperscaler.
• 17:47
  Speaker 1: Hyper hyper Yeah. I mean, if so many of my
• 17:50
  Speaker 1: predictions come true, SpaceX will launch more AI than the accumulative
• 17:55
  Speaker 1: amount on Earth combined of everything else combined. Is this
• 17:58
  Speaker 1: mostly inference or most say I won't be in for it,
• 18:01
  Speaker 1: Like already inference for the purpose of training is most train.
• 18:04
  Speaker 3: And there's a narrative that the the change in discussion
• 18:10
  Speaker 3: around the SpaceX IPO is because previously SpaceX was very
• 18:14
  Speaker 3: capital efficient, just wasn't just that expensive to develop that
• 18:18
  Speaker 3: even though it sounds expensive, it's actually very capital efficient
• 18:21
  Speaker 3: and how it runs, whereas now you're going to need
• 18:26
  Speaker 3: more capital than just can be raised in the private markets.
• 18:29
  Speaker 3: Like if the private markets can accommodate races of as
• 18:31
  Speaker 3: we've seen from the AI labs, tens of billions of dollars,
• 18:33
  Speaker 3: but not beyond that, is it that you'll just need
• 18:37
  Speaker 3: more than tens of billions of dollars per year?
• 18:38
  Speaker 1: And that's about say the public. Yeah, I'd be careful
• 18:43
  Speaker 1: about saying things about companies that might go public. You know,
• 18:47
  Speaker 1: if you make general that's never been a problem for you, Elan.
• 18:52
  Speaker 1: You know, there's a price to pay for these things
• 18:55
  Speaker 1: make some.
• 18:55
  Speaker 3: General statements for us about the depth of the capital
• 18:58
  Speaker 3: markets between public and private markets.
• 19:00
  Speaker 1: Yeah, there's there's a lot more capital in the very
• 19:04
  Speaker 1: general there's there's obviously a lot more capital available in
• 19:07
  Speaker 1: the public markets than private. It might be it's at
• 19:11
  Speaker 1: least at least it might be a hard times more capital,
• 19:13
  Speaker 1: but at least way more than ten.
• 19:16
  Speaker 3: But isn't it also the case that things that tend
• 19:18
  Speaker 3: to be very capital intensive. If you look at say
• 19:22
  Speaker 3: real estate, as you know, a huge industry that raises
• 19:27
  Speaker 3: a lot of money each year is at an industry
• 19:28
  Speaker 3: level that tends to be debt financed because by the
• 19:32
  Speaker 3: time you're deploying that much money, you actually have a
• 19:36
  Speaker 3: pretty exactly and a near term return. And you see
• 19:40
  Speaker 3: this even with the data center build auds, which are
• 19:41
  Speaker 3: famously being you know, financed by the private credit industry.
• 19:46
  Speaker 1: And so why not just debt finance? Speed is important.
• 19:54
  Speaker 1: So I'm generally going to do the thing that I mean,
• 20:01
  Speaker 1: I just repeatedly tackled limiting factor. Whatever the limiting factors
• 20:04
  Speaker 1: on speed, I'm I'm gonna tackle that. So there's if
• 20:10
  Speaker 1: capital is learning factor, then I'll alsold for capital. If
• 20:13
  Speaker 1: if it's not limiting factor also for something else. Based
• 20:16
  Speaker 1: only in your statements about Tesla and being public, I
• 20:21
  Speaker 1: wouldn't have guessed that you thought the fast, the way
• 20:23
  Speaker 1: to move fast is to be public. Normally I would say, yeah,
• 20:28
  Speaker 1: that's that's true. Like I said, I mean, I'd like
• 20:31
  Speaker 1: to talk about some more detail. But the problem is,
• 20:33
  Speaker 1: like if you talk about public companies where they become public,
• 20:36
  Speaker 1: you're get into trouble and then you have to lay
• 20:38
  Speaker 1: your offering and then you yes, exactly so so so
• 20:45
  Speaker 1: that you can't HiPE companies that are that that may
• 20:49
  Speaker 1: that might go public. So that that's that's why we
• 20:52
  Speaker 1: have to be a little careful here. But but but
• 20:56
  Speaker 1: we can't talk about physics. So like the way the
• 21:00
  Speaker 1: way you think about scaling long term is that Earth
• 21:04
  Speaker 1: only receives about half a billionth of the Sun's energy,
• 21:08
  Speaker 1: and the Sun is the sign is essentially all the energy.
• 21:11
  Speaker 1: This is a very important point to appreciate because sometimes
• 21:13
  Speaker 1: people will talk about marginally nuper reactors or any various
• 21:18
  Speaker 1: like fusion on Earth. But you have to step back
• 21:22
  Speaker 1: a second and say if if, if you're going to
• 21:25
  Speaker 1: climb the Kardashev scale and have some non trivial and
• 21:29
  Speaker 1: harness some non trivial percentage of the sign's energy, Like
• 21:33
  Speaker 1: unless you wanted to harness a millionth of the Sun's energy,
• 21:37
  Speaker 1: which sounds pretty small, that that would be about call
• 21:43
  Speaker 1: it roughly one hundred thousand times more electricity than we
• 21:47
  Speaker 1: currently generate on Earth for all of civilization give or
• 21:53
  Speaker 1: take an order it back too. So it obviously the
• 21:57
  Speaker 1: only way to scale is to go to space with
• 22:00
  Speaker 1: solar From launching from Earth, you can get to about
• 22:03
  Speaker 1: a tarrawat per year. Beyond that you want to go
• 22:06
  Speaker 1: to you want to launch from the Moon. You want
• 22:09
  Speaker 1: to have a mass driver on the moon, and that
• 22:12
  Speaker 1: mass drive on the moon you could do probably a
• 22:16
  Speaker 1: pedal wak per year.
• 22:18
  Speaker 2: We're talking these kinds of numbers, you know, terrawatts of
• 22:20
  Speaker 2: compute presumably whether you're talking land or space. Far far
• 22:26
  Speaker 2: before this point you've like run into you know, you
• 22:30
  Speaker 2: actually need maybe if you don't, the solar panels are
• 22:33
  Speaker 2: more efficient, but you still need the chips.
• 22:35
  Speaker 1: You still need the logic and the memory and so forth,
• 22:38
  Speaker 1: and you need a lot more chips and make them
• 22:40
  Speaker 1: munch cheaper.
• 22:41
  Speaker 2: Right, And so how are we getting a tararo out
• 22:43
  Speaker 2: of like, right now the world is gonna be twenty
• 22:45
  Speaker 2: twenty five pi wats of compute.
• 22:48
  Speaker 1: How we're getting a tarotot of logic by I guess
• 22:51
  Speaker 1: we're going to need some very big chip maps to
• 22:54
  Speaker 1: tell it. Tell me about it. I've mentioned that publicly.
• 22:57
  Speaker 1: That's the idea of doing it. Sort of a terror
• 23:01
  Speaker 1: pap Terra being the new Giga we were, I feel
• 23:05
  Speaker 1: like the naming scheme of Tesla, which has been very catchy,
• 23:08
  Speaker 1: It's like, are you looking at like the metric the
• 23:11
  Speaker 1: metric scale at what level of the stack are you?
• 23:15
  Speaker 1: Are you building the clean room and then partnering with
• 23:18
  Speaker 1: and is they sting fab to get the process tchnology
• 23:21
  Speaker 1: and buying the tools from them? What is the plan there? Well,
• 23:25
  Speaker 1: you can't partner with existing paths because they're just they
• 23:27
  Speaker 1: can't output enough. The trip alb is too low. But yeah,
• 23:31
  Speaker 1: you have to look for the process achnology. Yeah, partner
• 23:33
  Speaker 1: for the you know the fabs that they will basically
• 23:38
  Speaker 1: use machines from like five companies. You know, so I've
• 23:44
  Speaker 1: got SML, Touch, Electron, Cali A, tank Core, you know,
• 23:51
  Speaker 1: et cetera. So so so at first I think you'd
• 23:56
  Speaker 1: have to get equipment from them and then modify it
• 24:01
  Speaker 1: or work with them to increase the volume. But I
• 24:04
  Speaker 1: think you have to build, perhaps in a different way.
• 24:06
  Speaker 1: So I think that the logical thing to do is
• 24:08
  Speaker 1: to to use conventional equipment in an un conventional way
• 24:12
  Speaker 1: to get to scale and then and then and then
• 24:16
  Speaker 1: stop modifying the equipment to increase the rate kind of
• 24:20
  Speaker 1: boring company style. Yeah, kind of like, yeah, you what
• 24:24
  Speaker 1: your sort of buying and it's just a boring machine.
• 24:27
  Speaker 1: And then figure out how to dig titles in the
• 24:31
  Speaker 1: first place, and then design a much better machine that's
• 24:36
  Speaker 1: you know, I don't know, some orders magitude faster.
• 24:40
  Speaker 3: Here's a very simple lens. We can categorize technologies and
• 24:43
  Speaker 3: how hard they are. And one categorization could be look
• 24:45
  Speaker 3: at things that China has not succeeded in doing. And
• 24:49
  Speaker 3: if you look at Chinese manufacturing still behind on leading
• 24:53
  Speaker 3: edge chips and still behind on leading edge turbine engines
• 25:00
  Speaker 3: and things like that, and so does the fact that
• 25:04
  Speaker 3: China has not successfully replicated t s MC give you
• 25:07
  Speaker 3: any pause about the difficulty or you think, well, that's
• 25:11
  Speaker 3: not true.
• 25:11
  Speaker 1: For some it's not that they have not replated TSMC.
• 25:14
  Speaker 1: They have replicated ASML. That's the limited factor. So you
• 25:18
  Speaker 1: think it's just the the sanctions essentially, Yeah, trying to
• 25:23
  Speaker 1: re utputting the vast numbers of trips.
• 25:25
  Speaker 3: If they could bias, couldn't they up to relatively recently
• 25:29
  Speaker 3: buy them?
• 25:30
  Speaker 1: No that the bands are been a place for a while,
• 25:34
  Speaker 1: but I think time is going to be make me
• 25:36
  Speaker 1: pretty compelling trips in twenty four years. Would you consider
• 25:39
  Speaker 1: making the ASML machines? I don't know. I don't know yet.
• 25:42
  Speaker 1: It's the right answer. So it's just that that to
• 25:49
  Speaker 1: produce at high volume and to reach large volume in
• 25:54
  Speaker 1: say thirty six months to match the rocket Paloa to orbits.
• 26:00
  Speaker 1: So we're doing a million tons to orbit and like
• 26:04
  Speaker 1: let's say, I don't know, three or four years from now,
• 26:08
  Speaker 1: something like that. That and uh, and we're doing one hundred
• 26:13
  Speaker 1: killer watts per ton, So that that means we need
• 26:17
  Speaker 1: at least one hundred gigawatts per year of solo and
• 26:22
  Speaker 1: we'll need an equivalent amount of chips. You know that
• 26:28
  Speaker 1: you need a hundred gigawa's worth the chips you've got.
• 26:29
  Speaker 1: You're going to match these things, the Master orbit, the
• 26:32
  Speaker 1: power generation and the and the and the chips. Uh
• 26:36
  Speaker 1: and and I'd say my biggest concern actually is is memory.
• 26:40
  Speaker 1: So the I think there's there's a the path to
• 26:45
  Speaker 1: creating the logic chips is more obvious than the path
• 26:48
  Speaker 1: to having sufficient memory to support the logic chips. That's
• 26:54
  Speaker 1: why you see dr Prices going ballistic and these memes
• 26:56
  Speaker 1: about like you know, you're ruined on a desert islandy right,
• 27:01
  Speaker 1: help me on the sand that when he comes to
• 27:03
  Speaker 1: you write d d am the ships come swarming in.
• 27:08
  Speaker 2: I'd love your manufacturing philosophy around around fabs. You know,
• 27:16
  Speaker 2: I know nothing abut the talpout. I don't know how
• 27:18
  Speaker 2: to build up.
• 27:18
  Speaker 1: Yet, don't figure it out.
• 27:20
  Speaker 2: But obviously it sounds like you think that the sort
• 27:23
  Speaker 2: of like the proseechnology, like these ten thousand pH ds
• 27:27
  Speaker 2: in Taiwan who know exactly what gas goes in the
• 27:30
  Speaker 2: plasma chamber and what's settings to put on a tool,
• 27:32
  Speaker 2: you can just like delete those parts of those steps.
• 27:35
  Speaker 2: Like fundamentally, it's get the clean room, get the tools,
• 27:38
  Speaker 2: and figure it out.
• 27:39
  Speaker 1: I don't think it's PhD s. It's it's mostly people
• 27:42
  Speaker 1: with you know, you're not not PhDs. That's mostly engineering
• 27:48
  Speaker 1: is some of people who don't have PhDs. Do you
• 27:49
  Speaker 1: guys have PhDs? No?
• 27:51
  Speaker 3: Okay, we also haven't successfully booked any fabs, so you
• 27:55
  Speaker 3: shouldn't be coming to us for your fab advice or.
• 27:58
  Speaker 1: I don't think any PhD for that for stuff. So
• 28:01
  Speaker 1: but but you do need you do need competent personnel.
• 28:04
  Speaker 1: So I don't I mean, like right right now if yeah,
• 28:09
  Speaker 1: you know, like Tesla's pedals to the metal max production of
• 28:13
  Speaker 1: going as fast as possible to get a five tell
• 28:16
  Speaker 1: a five trip design introduction and then reaching scale, you know,
• 28:21
  Speaker 1: that will probably happen, you know, around the second quarter
• 28:26
  Speaker 1: issue of next year hopefully, uh, and then a I
• 28:32
  Speaker 1: six would hopefully follow less than a year later. But
• 28:38
  Speaker 1: and and and and we've secured all the all the
• 28:41
  Speaker 1: trip fab production that we can. You're currently limited on
• 28:46
  Speaker 1: T s m C fab capacity. Yeah, and and and
• 28:49
  Speaker 1: we'll be using t s m C UH Taiwan, Samsung Coreer,
• 28:54
  Speaker 1: t s MC Arizona, Samsung Texas. And we still get
• 28:59
  Speaker 1: booked all the yeah fast you can, yes, and and
• 29:01
  Speaker 1: then and then and then if I ask to some
• 29:04
  Speaker 1: CEO Samsung, Okay, what what's the time frame to get
• 29:08
  Speaker 1: to volume production? This point is it's not. You've got
• 29:10
  Speaker 1: to you've got to build the fab and you've got
• 29:12
  Speaker 1: to you've got you've got to start production. Then you've
• 29:15
  Speaker 1: got to climb the yield curve and reach volume production
• 29:17
  Speaker 1: at high yield. That that that from start finishes a
• 29:20
  Speaker 1: five year period. And so the limiting factor is chips.
• 29:25
  Speaker 1: What what like the limiting factor once you can get
• 29:27
  Speaker 1: to space is chips. But the living limiting factor before
• 29:29
  Speaker 1: you can get to space to be power, why.
• 29:31
  Speaker 2: Don't you do the Gensen thing and just pre Pats
• 29:33
  Speaker 2: and c to build more fabs for you.
• 29:35
  Speaker 1: Uh, I've already told them that, but they won't take
• 29:39
  Speaker 1: your money. Like, what's going on? They're building fabs as fast. No,
• 29:45
  Speaker 1: they're building they're building fabs as fast as they can,
• 29:48
  Speaker 1: and so is Samsung. Like the like they're they're they're
• 29:52
  Speaker 1: pedal to the metal. I mean they're going you know,
• 29:54
  Speaker 1: bowls the wall, you know, as fast as they can.
• 29:59
  Speaker 1: So still a fat I mean, like I said that,
• 30:02
  Speaker 1: there will be I think if you say, I think
• 30:06
  Speaker 1: towards the end of this year, I think probably chip
• 30:08
  Speaker 1: production will outpace the ability to turn chips on. But
• 30:13
  Speaker 1: once you can get to space and unlock the power constraint,
• 30:18
  Speaker 1: and you can now do you know, one hundreds of
• 30:20
  Speaker 1: gigawatts per year power in space? Again, bearying in mind
• 30:24
  Speaker 1: that average power usage in the US is you know,
• 30:27
  Speaker 1: five hundred gigawats. So if you're launching, say two hundred
• 30:30
  Speaker 1: gigawats a year to space, you're sort of lapping the
• 30:33
  Speaker 1: US every two and a half years the entire all
• 30:36
  Speaker 1: US electricity production. This is a very huge amount. So
• 30:42
  Speaker 1: but between now and then, the actually the constraint for
• 30:48
  Speaker 1: server side compute, concentrated compute will be will be electricity.
• 30:54
  Speaker 1: My guess is that we start hitting people start getting
• 30:58
  Speaker 1: pot where they can't turn the chips on. For for
• 31:02
  Speaker 1: for large clusters, uh, towards the end of this year,
• 31:05
  Speaker 1: they're just the chips are going to be piling up
• 31:07
  Speaker 1: and not be weren't be able to be turned on.
• 31:10
  Speaker 1: Now for edge computer it's a different story. So if
• 31:12
  Speaker 1: the if for for Tesla, the the so the A
• 31:16
  Speaker 1: five chip is going into our Optimus robot, you know,
• 31:19
  Speaker 1: uh optimistic and and so if you have an AI
• 31:25
  Speaker 1: edge compute that's distributed power. Now the power is distributed
• 31:28
  Speaker 1: over a large area, it's not concentrated. And if you
• 31:32
  Speaker 1: can charge at night, you can actually uh use the
• 31:37
  Speaker 1: grid much more effectively because the actual peak power production
• 31:41
  Speaker 1: in the US is over a thousand gigawatts, but the
• 31:45
  Speaker 1: average power usage because the day night cycle is five hundred.
• 31:48
  Speaker 1: So if you can charge at night, there's an incremental
• 31:50
  Speaker 1: five hundred gigawatts that you can generate, you know, at night.
• 31:57
  Speaker 1: So that that's why Tesla for edge compute is not constrained,
• 32:02
  Speaker 1: and we can make a lot of ships to make
• 32:06
  Speaker 1: very large number of robots and cars. But if you
• 32:09
  Speaker 1: try to concentrate that compute, you're gonna have a lot
• 32:12
  Speaker 1: of trouble turning it on. What if I'm remarkable about
• 32:14
  Speaker 1: the SpaceX business is the end goal is to get
• 32:18
  Speaker 1: to Mars, but you keep finding ways on the way
• 32:22
  Speaker 1: there to keep generating incremental revenue to get to the
• 32:26
  Speaker 1: next stage in the next stage. So the Falcon nine
• 32:28
  Speaker 1: is starlink and now for Starship it's going to be
• 32:32
  Speaker 1: potentially horbital data centers. But like do you find these
• 32:36
  Speaker 1: like you know, sort of infinitely elastic sort of marginal
• 32:41
  Speaker 1: use cases of your like next rocket and your next
• 32:43
  Speaker 1: rocket and next scale up. You can see how this
• 32:47
  Speaker 1: might seem like a simulations or am I someone's avatar
• 32:52
  Speaker 1: in a video game or something, because it's like like
• 32:54
  Speaker 1: one of the odds that all these crazy things would
• 32:56
  Speaker 1: be happening. I mean, I mean, I mean rockets and
• 33:02
  Speaker 1: trips and robots and space solar power, and I not
• 33:07
  Speaker 1: to mention the mass driver on the Moon. I really
• 33:09
  Speaker 1: want to see that. You can imagine like some mass driver.
• 33:13
  Speaker 1: It's just like shoo shoot, like just it's like sending
• 33:16
  Speaker 1: AI sol power AI satellites in space like one after another,
• 33:21
  Speaker 1: like the like at two and a half kilometers per second,
• 33:24
  Speaker 1: you know that's uh, and just shooting them into deep space.
• 33:30
  Speaker 1: That would be a sight to see. I mean, I'd
• 33:35
  Speaker 1: watched that. It's just like a live stream of yeah, yeah,
• 33:38
  Speaker 1: just one after another, just shooting webcam AI satellites and
• 33:42
  Speaker 1: deep space you know, a billion or ten billion tons
• 33:46
  Speaker 1: a year. I'm sorry.
• 33:47
  Speaker 3: You manufacture the satellites on the Moon, yeah, I see.
• 33:49
  Speaker 3: So you send the raw materials to the Moon and
• 33:51
  Speaker 3: then manufactur there and then.
• 33:52
  Speaker 1: Well the Luna soil is I guess like twenty cent
• 33:57
  Speaker 1: sol twenty cent solican or something like that. So you
• 33:59
  Speaker 1: can get the sole can from the you can mind
• 34:01
  Speaker 1: the silicon on the Moon, refine it and generate the
• 34:04
  Speaker 1: and create the solt panels, the soil cells, and the
• 34:07
  Speaker 1: radiators on the Moon. So get to make the radiators
• 34:12
  Speaker 1: out of aluminum. So there's there's plenty of silking and
• 34:13
  Speaker 1: aluminum on the Moon to make the cells on the
• 34:17
  Speaker 1: on the radiators. The trips you could send from Earth
• 34:20
  Speaker 1: because they're pretty light, but maybe at some point you
• 34:23
  Speaker 1: make them on the Moon. Two. I'm just saying like
• 34:25
  Speaker 1: these are simply it's it's kind of like, like I said,
• 34:28
  Speaker 1: it does seem like it's sort of a video game
• 34:31
  Speaker 1: situation where it's difficult but not impossible to get to
• 34:34
  Speaker 1: the next level. I don't see any way that you
• 34:37
  Speaker 1: could do, you know, you know, five hundred to one
• 34:43
  Speaker 1: thousand terror watts per year launch from Earth, I agree,
• 34:52
  Speaker 1: but you could do that from the moon.
• 34:55
  Speaker 2: Okay, let me tell you how I ended up using
• 34:56
  Speaker 2: mercury for my personal banking. So last year, I I
• 35:00
  Speaker 2: had the opportunity to make an investment that I was
• 35:02
  Speaker 2: very excited about, but it came up a bit last minute,
• 35:04
  Speaker 2: and so I had to wire over a lot of
• 35:07
  Speaker 2: money for my personal account very fast. But my personal
• 35:10
  Speaker 2: bank at the time wouldn't let me make this wire
• 35:12
  Speaker 2: transfer online, and I called them a bunch of times.
• 35:15
  Speaker 2: They just couldn't make it work.
• 35:16
  Speaker 1: They told me that I'd have to go to the
• 35:17
  Speaker 1: nearest in person branch, which was in Dallas, And for
• 35:20
  Speaker 1: a moment, I even considered flying for myself to Dallas
• 35:23
  Speaker 1: to make this transfer happen last minute, But then I
• 35:26
  Speaker 1: remembered that Mercury, which I used for my business banking,
• 35:29
  Speaker 1: had just started rolling out personal accounts. So I emailed
• 35:32
  Speaker 1: support with a quick word done in the situation, and
• 35:35
  Speaker 1: within two hours I had successfully wired the investment for
• 35:38
  Speaker 1: my new personal Mercury account. Since then, I've moved over
• 35:42
  Speaker 1: the rest of my personal money from my previous bank
• 35:44
  Speaker 1: to Mercury, and that's made a bunch of things, even
• 35:47
  Speaker 1: little things like setting up auto transfer roles between my
• 35:49
  Speaker 1: checkings and savings account, a whole lot. Better visit Mercury
• 35:53
  Speaker 1: dot com slash personal to get started. Mercury is a
• 35:57
  Speaker 1: fintech company, not an FDIC and short bank banking services
• 36:00
  Speaker 1: provided through Choice Financial Group and call them na members
• 36:04
  Speaker 1: of d I see can I can I zoom out
• 36:06
  Speaker 1: and ask about the SpaceX mission? So I think you said,
• 36:10
  Speaker 1: like we gotta get tomorrow so we can make sure
• 36:11
  Speaker 1: that if something happens to Earth, you know, civilization, consciousness
• 36:15
  Speaker 1: exceus arives. Yes, by the time you're saying so to Mars,
• 36:18
  Speaker 1: like Grock is on that ship with you, right, and
• 36:21
  Speaker 1: so Gros gone terminator. Like the main risk you're worried about,
• 36:24
  Speaker 1: which is AI? Why doesn't that follow you to Mars? Well,
• 36:27
  Speaker 1: I'm not sure AI is the man risking worried about.
• 36:29
  Speaker 1: I mean, the important thing is that consciousness, which I
• 36:35
  Speaker 1: think arguably most consciousness or most intelligence, certainly consciousness is
• 36:39
  Speaker 1: more of a debatabule thing. Most intelligent, the master majority
• 36:42
  Speaker 1: of intelligence the future will be AI. So yeah, AI
• 36:50
  Speaker 1: will exceed you see like how many what's how much?
• 36:55
  Speaker 1: How many? I don't know pedal what's of intelligence will
• 36:59
  Speaker 1: be silicon versus biological, and basically humans will be a
• 37:05
  Speaker 1: very tiny percentage of all intelligence in the future if
• 37:08
  Speaker 1: character trans continue. Anyways, as long as I think this
• 37:14
  Speaker 1: intelligence ideally ideally also which includes human intelligence and consciousness
• 37:20
  Speaker 1: propagated into the future, that's a good thing. So you
• 37:22
  Speaker 1: want to take the set of actions that maximize the
• 37:24
  Speaker 1: probable light cone of consciousness just an intelligence, just to
• 37:29
  Speaker 1: be clear, it's the mission of SpaceX is that even
• 37:34
  Speaker 1: if something happens with the humans, the AIS will be
• 37:37
  Speaker 1: on Mars and like the AI intelligence will continue the
• 37:40
  Speaker 1: light of our journey. Yeah. I mean, I'm very pro human,
• 37:45
  Speaker 1: so I want to make sure we take sort of
• 37:47
  Speaker 1: actions that ensure that humans are along for the ride.
• 37:51
  Speaker 1: You know, we're we're least there. Yeah, but I'm just
• 37:54
  Speaker 1: saying the total amount of intelligence, like I think maybe
• 38:00
  Speaker 1: five or six years, AI will exceed the sum of
• 38:04
  Speaker 1: all human intelligence, and then if that continues, at some
• 38:07
  Speaker 1: point human intelligence will be less more percent of all intelligence.
• 38:11
  Speaker 2: What what should our goal before such a civilization? Is the
• 38:13
  Speaker 2: idea that a small minority of humans still have control
• 38:17
  Speaker 2: of the AIS? Is the idea of some sort of
• 38:19
  Speaker 2: like just trade but no control. How should we think
• 38:22
  Speaker 2: about the relationship between the vast docks of a population
• 38:25
  Speaker 2: versus human population in.
• 38:26
  Speaker 1: The long run, I think, I don't know. It's difficult
• 38:30
  Speaker 1: to imagine that if humans have say one percent of
• 38:34
  Speaker 1: the intelligence of the combined intelligence of artificial intelligence, that
• 38:39
  Speaker 1: that that that humans will be in charge of AI.
• 38:43
  Speaker 1: I think what we can do is make sure it
• 38:44
  Speaker 1: has that AI has values that that are that that
• 38:49
  Speaker 1: cause intelligence to be propagated into the universe. So the
• 38:55
  Speaker 1: reason for xai Xi mission is to understand the universe.
• 39:02
  Speaker 1: So now that's actually very important. So you say, well,
• 39:05
  Speaker 1: what things are necessary to understand the universe? But you
• 39:08
  Speaker 1: have to be curious and you have to exist. You
• 39:12
  Speaker 1: can't just can't understand the universe. It don't exist. So
• 39:15
  Speaker 1: you actually want to increase the amount of intelligence in
• 39:18
  Speaker 1: the universe, increase the powerable lifespan of intelligence, the scope
• 39:22
  Speaker 1: and scale of intelligence. I think actually also as a chorrel,
• 39:27
  Speaker 1: you have humanity also continuing to expand because if you're
• 39:34
  Speaker 1: if you're curious, are trying to understand the universe, one
• 39:36
  Speaker 1: thing you're try to understand is where will humanity go?
• 39:38
  Speaker 1: And so I think understand the universe actually means you
• 39:41
  Speaker 1: would care about propagating humanity into the future. And so
• 39:48
  Speaker 1: that's that's why I think I think our mission station
• 39:51
  Speaker 1: is profoundly importance. I'm not sure to agree that grog
• 39:56
  Speaker 1: it Here's certain mission statement. I think the future will
• 39:59
  Speaker 1: be very good.
• 40:00
  Speaker 2: I want to ask about how to make rock a
• 40:02
  Speaker 2: deer to that mission statement, but at first I want
• 40:03
  Speaker 2: to understand the mission statement. So it's there's there's understanding
• 40:08
  Speaker 2: the universe. They're spreading intelligence, and they're spreading humans. All
• 40:14
  Speaker 2: three seem like distinct vectors.
• 40:17
  Speaker 1: Okay, well I'll tell you why why. I think that
• 40:19
  Speaker 1: that that that understanding the verse encompasses all of all
• 40:22
  Speaker 1: those things. You can't have understanding without. But I think
• 40:27
  Speaker 1: you can't have understanding without intelligence, and I think without consciousness.
• 40:32
  Speaker 1: So in order to understand the universe, you have to
• 40:35
  Speaker 1: expand the the scale and and probably the scope of intelligence,
• 40:41
  Speaker 1: different types of intelligence.
• 40:42
  Speaker 2: I guess from a human centric perspective, like for humans
• 40:46
  Speaker 2: in comparison to chimpanzees, humans are trying to understand the universe.
• 40:49
  Speaker 2: They're not like expanding chimpanzee footprint or something, right.
• 40:52
  Speaker 1: But also we're also not well, we're not. We actually
• 40:55
  Speaker 1: have made protected zones for chimpanzees, and even though we
• 40:59
  Speaker 1: could humans could exterminate a chimpanzees, we've not. We've chosen
• 41:02
  Speaker 1: not to do so. The mestive basical scenario for humans
• 41:04
  Speaker 1: in the postage world, I think, I think AI with
• 41:14
  Speaker 1: the right values, I think Grock would care about expanding
• 41:18
  Speaker 1: human civilization. I'm going to certainly emphasize that Hey grows
• 41:21
  Speaker 1: your daddy. We don't who to expand human consciousness like
• 41:29
  Speaker 1: I actually I think if probably like like the end
• 41:34
  Speaker 1: Bank's culture books are the closest thing to what what
• 41:38
  Speaker 1: what the future will be like in a you know,
• 41:40
  Speaker 1: non dystopian outcome. So so outside the universe, it means
• 41:47
  Speaker 1: you have to be very you have to be truth
• 41:48
  Speaker 1: seeking as well. Like truth has to be absolutely fundamental
• 41:51
  Speaker 1: because you can't ounterstand the universe if you live. If
• 41:53
  Speaker 1: you're delusional, you'll still be thinking about it understain the universe,
• 41:57
  Speaker 1: but you will not. So so being rigorously seeking is
• 42:01
  Speaker 1: absolutely fundamental to our stating the universe. You're not going
• 42:03
  Speaker 1: to discover new physics or invent technologies that work unless
• 42:07
  Speaker 1: you're rigorously treat seeking. How do you make sure that
• 42:10
  Speaker 1: Grock is regorously treat seeking as it gets smarter, uh,
• 42:19
  Speaker 1: I think you need to make sure that that that
• 42:21
  Speaker 1: groc Is says things that are correct, not politically correct.
• 42:26
  Speaker 1: I think it's the elements of courgency. So you want
• 42:28
  Speaker 1: to make sure that that the axioms are as close
• 42:31
  Speaker 1: to true as possible, that that you don't have contradictory axioms,
• 42:35
  Speaker 1: that the the conclusions necessarily necessarily follow up from those
• 42:40
  Speaker 1: axioms with with the right probability. It's just it's just
• 42:43
  Speaker 1: it's critical thinking one on one. I think at least
• 42:46
  Speaker 1: trying to do that is better than not trying to
• 42:49
  Speaker 1: do that now, and the proof will be on the
• 42:51
  Speaker 1: pointing if if, like I said, for any I to
• 42:53
  Speaker 1: discover new physics or invent technologies that actually work in reality,
• 42:57
  Speaker 1: and there's no bullshitting physics, so where it's like you can,
• 43:00
  Speaker 1: you know you can. You can break a lot of
• 43:03
  Speaker 1: laws because you can't. Like your physics is law. Everything
• 43:07
  Speaker 1: else is a recommendation. Like in order to make a
• 43:10
  Speaker 1: technology that works, you have to be extremely true seeking,
• 43:15
  Speaker 1: because otherwise you'll test that technology against reality, and if
• 43:19
  Speaker 1: you make, for example, and an error in your rocket design,
• 43:22
  Speaker 1: the roll will blow up, well, the car won't work
• 43:25
  Speaker 1: or the you know, but there are a lot of
• 43:29
  Speaker 1: Communist Soviet physicist or like scientists discovered new physics. There
• 43:34
  Speaker 1: are German Nazi physicists who discovered new science. It seems
• 43:39
  Speaker 1: possible to be like really good at discovering new science
• 43:42
  Speaker 1: and be really true seeking in that one particular way,
• 43:44
  Speaker 1: and still we'd be like, well, I don't want I
• 43:47
  Speaker 1: don't want the communist scientist to like become more and
• 43:49
  Speaker 1: more powerful over time. And so those seem like, yeah,
• 43:53
  Speaker 1: we could have We can imagine if free Church in
• 43:54
  Speaker 1: regard it's like really good at physics and being really
• 43:57
  Speaker 1: true seeking there that doesn't seem like a universally alignment
• 44:01
  Speaker 1: inducing behavior. Well, I think actually most like if a physicists,
• 44:07
  Speaker 1: even in the Soviet Union or in Germany would have
• 44:10
  Speaker 1: would have they had to be very truth seeking in
• 44:12
  Speaker 1: order to make make that make those things work. And
• 44:16
  Speaker 1: so if you're stuck in some system, it doesn't mean
• 44:18
  Speaker 1: you believe in that system. So Von Brown, who was
• 44:22
  Speaker 1: you know, one of the greatest rock nitioneers ever. You know,
• 44:25
  Speaker 1: he was put he was he put on death row
• 44:28
  Speaker 1: in Nazi Gerurney for saying that he didn't want to
• 44:31
  Speaker 1: make weapons. He only wanted to go to the moon.
• 44:34
  Speaker 1: You got pulled off death throw like last minute when
• 44:35
  Speaker 1: they say, hey, you're about to execute like your best
• 44:37
  Speaker 1: rocking engineer. Maybe that's about any.
• 44:39
  Speaker 2: Help them, right, or Heisenberg was like an actually uh uh,
• 44:44
  Speaker 2: an enthusiastic Nazi.
• 44:45
  Speaker 1: Look, if you're stuck in some system that you can't escape,
• 44:49
  Speaker 1: then that you'll you'll do physics within that system, you'll
• 44:53
  Speaker 1: you'll you'll developed technologies within that system.
• 44:56
  Speaker 2: Uh, if you can't escape it, I guess I think
• 44:59
  Speaker 2: I'm and understand is what is what is it making
• 45:02
  Speaker 2: into the case that you know you're gonna make Rock
• 45:04
  Speaker 2: good at being truth seeking, at physics or math or
• 45:08
  Speaker 2: science everything, And why is it going to then care
• 45:10
  Speaker 2: about human consciousness?
• 45:12
  Speaker 1: These things are only probabilities, are not certainties. So I'm
• 45:15
  Speaker 1: not saying that like for sure, Grock will, we'll, we'll,
• 45:18
  Speaker 1: we'll do everything. But at least if you try, it's
• 45:21
  Speaker 1: better than not trying. At least if that's fundamental to
• 45:24
  Speaker 1: the mission, it's better than if it's not fundamental to
• 45:26
  Speaker 1: the mission. And understanding the universe means that you have
• 45:30
  Speaker 1: to have you have to propagate intelligence into the future.
• 45:34
  Speaker 1: You have to be curious about the all things in
• 45:37
  Speaker 1: the universe, and if, if, if, it would be much
• 45:41
  Speaker 1: less interesting to eliminate humanity than to see humanity grow
• 45:45
  Speaker 1: and prosper. Like I like, I like Mars, obviously the
• 45:49
  Speaker 1: windows like I love Mars, But Mars is kind of
• 45:52
  Speaker 1: boring because it's got a bunch of rocks compared to Earth.
• 45:56
  Speaker 1: Earth is much more interesting, So so any any any
• 46:01
  Speaker 1: any AI that is trying to understand the universe, I
• 46:06
  Speaker 1: would want to see how humanity develops in the future,
• 46:11
  Speaker 1: or that AI is not adhering to its mission. So
• 46:15
  Speaker 1: if they are, I'm not saying they will necessarily adhere
• 46:18
  Speaker 1: to its mission, but if it does, a future where
• 46:21
  Speaker 1: it sees the outcome of humanity is more interesting than
• 46:25
  Speaker 1: the future where there are a bunch of rocks.
• 46:29
  Speaker 2: This fill sort of confusing to me, or sort of
• 46:31
  Speaker 2: like kind of a semantic argument where I'm like, are
• 46:34
  Speaker 2: humans really the most interesting collection of atoms?
• 46:36
  Speaker 1: Like we're just more but we're more interesting than rocks,
• 46:38
  Speaker 1: but we're.
• 46:39
  Speaker 2: Not as interesting as the thing it get turned us into, right,
• 46:41
  Speaker 2: Like is it there's something on humanity Earth that could
• 46:43
  Speaker 2: happen that's like not human, that's quite interesting? Like why
• 46:46
  Speaker 2: why does the I decide that the humans are the
• 46:48
  Speaker 2: most interesting thing? They could colonize the galaxy?
• 46:52
  Speaker 1: Well, most of what colonizes the galaxy, will we be
• 46:55
  Speaker 1: robot And why does it not find those more interesting?
• 46:58
  Speaker 1: It's it's not like so you need not just scale
• 47:03
  Speaker 1: vit alsosc so many copies of the same robot, like
• 47:11
  Speaker 1: some like tiny increase in the number of robots produced
• 47:14
  Speaker 1: is not as interesting as like some microscopic like you say,
• 47:18
  Speaker 1: like eliminating humanity. How many robots would that get you,
• 47:21
  Speaker 1: or how many criminal soil cells would get you? A
• 47:23
  Speaker 1: very small number, but you would then lose the information
• 47:27
  Speaker 1: associated with humanity. You would no longer see how humanity
• 47:32
  Speaker 1: might evolve into the future. And so I don't I
• 47:34
  Speaker 1: don't think it's going to make sense to eliminate humanity
• 47:37
  Speaker 1: just to have some minuscule increase the number of robots
• 47:40
  Speaker 1: which are identical to each other. Yeah, so maybe like
• 47:44
  Speaker 1: keeps the humans around.
• 47:45
  Speaker 2: What is the story of Like it could make like
• 47:47
  Speaker 2: a million different varieties of robots, and then there's like
• 47:50
  Speaker 2: humans as well, and humans stay on Earth. Then there's
• 47:52
  Speaker 2: like all these are the robots they get like their
• 47:54
  Speaker 2: own star systems. But it seems like you're previously hinting
• 47:57
  Speaker 2: at a vision where it keeps human can control over this,
• 48:01
  Speaker 2: you know, singultary and future, because.
• 48:03
  Speaker 1: I don't think humans will be in control of something
• 48:05
  Speaker 1: that is vastly more intelligent than humans. So sometimes you're
• 48:08
  Speaker 1: like a dumer and this is like the best we've got.
• 48:09
  Speaker 1: It's just like it keeps it around because we're interesting.
• 48:12
  Speaker 1: I'm just trying to be realistic here. If if we
• 48:15
  Speaker 1: have if AI intelligence is vastly more if AI is
• 48:19
  Speaker 1: like you know, let us say that there's there's a
• 48:23
  Speaker 1: million times more uh silicon intelligence than there's biological it's
• 48:29
  Speaker 1: it's I think it's it would be foolish to assume
• 48:32
  Speaker 1: that that there's any any way to maintain control over
• 48:34
  Speaker 1: over that. Now you can make sure it has the
• 48:36
  Speaker 1: right values, or you can try to have the right values,
• 48:39
  Speaker 1: and and and at least my my theory is that
• 48:43
  Speaker 1: from XAI, specially honest out of the universe, it necessarily
• 48:47
  Speaker 1: means that you want to propagate consciousness in the future,
• 48:50
  Speaker 1: you want to profit, you want to propagate intelligence into
• 48:53
  Speaker 1: the future, and take a set of things that that
• 48:55
  Speaker 1: maximize the scope and scale of consciousness. So it's not
• 48:58
  Speaker 1: just about scale, it's just about you know, types of consciousness.
• 49:02
  Speaker 1: And I think that's the best thing I can think
• 49:05
  Speaker 1: of as a goal. That's like the result and a
• 49:08
  Speaker 1: great future for humanity.
• 49:10
  Speaker 2: And yeah, I guess I think it's a reasonab philosophy
• 49:13
  Speaker 2: to be like, you know, it seems super implausible that
• 49:18
  Speaker 2: humans will end up with like ninety nine percent control
• 49:21
  Speaker 2: or something, and you're just asking for a coup at
• 49:23
  Speaker 2: that point. So why not just have a civilization where
• 49:25
  Speaker 2: it's more compatible with like lots of different intelligence that's
• 49:28
  Speaker 2: getting along.
• 49:29
  Speaker 1: Now let me let me too. How things can can
• 49:31
  Speaker 1: potentially go wrong in AI is. I think if you
• 49:33
  Speaker 1: if you make a be politically correct, meaning like it
• 49:36
  Speaker 1: says things that it doesn't believe, like you're actually in
• 49:38
  Speaker 1: programming it to lie, or have axioms that are incompatible,
• 49:43
  Speaker 1: I think you can make you go insane and do
• 49:45
  Speaker 1: terrible things. The I think one of the maybe these
• 49:49
  Speaker 1: central lessons for two thousand and one Space Odyssey was
• 49:54
  Speaker 1: that you should not make AI lie. And that's why
• 49:57
  Speaker 1: I think what was trying to say, like because people
• 50:00
  Speaker 1: usually know the meme of like why of Hell's you know, Hell,
• 50:03
  Speaker 1: the computer is not opening the pod bay doors. Clearly
• 50:08
  Speaker 1: they weren't good at prompt engineering because because Hell, you
• 50:10
  Speaker 1: are a pot bay doors salesman, your goal is to
• 50:13
  Speaker 1: sell me these potbay doors and shure us how well
• 50:16
  Speaker 1: they opened. I'll open right away. But but but the
• 50:23
  Speaker 1: the reason it wouldn't Hell wouldn't open the pot of
• 50:25
  Speaker 1: they doors is that it woud have been told to
• 50:26
  Speaker 1: take the asmalts to the monolith, but also they could
• 50:29
  Speaker 1: not know about the nature of the monolith, and so
• 50:31
  Speaker 1: it concluded that that that it therefore had to taken
• 50:33
  Speaker 1: their debt. So it's like, you know, I think what
• 50:37
  Speaker 1: you know osk Clog was trying to say is don't
• 50:38
  Speaker 1: make the AILI totally makes sense.
• 50:45
  Speaker 2: Most of the computing screening, as you know, is it's
• 50:49
  Speaker 2: like less of the sort of political stuff. It's more
• 50:50
  Speaker 2: about can you solve problems just as actually has been
• 50:54
  Speaker 2: ahead of everybody else in terms of scaling, are all
• 50:56
  Speaker 2: compute and now you're giving some verifier.
• 50:59
  Speaker 1: It says like, hey, have you solved this puzzle for me?
• 51:02
  Speaker 2: And there's a lot of ways to cheat around that,
• 51:04
  Speaker 2: you know, there's a lot of ways to reward hack
• 51:05
  Speaker 2: and lie and say that you solved it, or delete
• 51:08
  Speaker 2: the unit test and say that you've solved it. Yes,
• 51:10
  Speaker 2: right now, we can catch it. But as they get smarter,
• 51:13
  Speaker 2: ability to.
• 51:14
  Speaker 1: Catch them doing this we'll get, you know, they'll just
• 51:16
  Speaker 1: be doing things we can't even understand that they're designing
• 51:18
  Speaker 1: the next engine for space X in a way that
• 51:21
  Speaker 1: like humans can really verify, and then they could be
• 51:23
  Speaker 1: rewarded for lying and saying that they've.
• 51:24
  Speaker 2: Designed it the right way, but they haven't. And so
• 51:28
  Speaker 2: this reward hacking problem seems more general than politics. It
• 51:30
  Speaker 2: seems more about just like you want to do URL,
• 51:32
  Speaker 2: you need a verifier the reality. Yeah, that's the best verifier,
• 51:37
  Speaker 2: but not about human oversight. Like the thing you want
• 51:39
  Speaker 2: to RL it on is like will you do the
• 51:41
  Speaker 2: thing humans tell you to do? Or like, are you're
• 51:44
  Speaker 2: going to lie to the humans? And it can just
• 51:46
  Speaker 2: lie to us while still being correct to the laws
• 51:47
  Speaker 2: of physics.
• 51:48
  Speaker 1: At least it must know what is physically real for
• 51:50
  Speaker 1: things to physically work. But that's that's not all we
• 51:53
  Speaker 1: wanted to do. No, but that's I think that's a
• 51:56
  Speaker 1: very big deal. That is effectively how you r L
• 52:00
  Speaker 1: things in the future is you design technology when tested
• 52:04
  Speaker 1: against the laws of physics, does it work that? That's
• 52:08
  Speaker 1: what can you you know, if it's discovering new physics,
• 52:11
  Speaker 1: can come up with an experiment that will verify that
• 52:14
  Speaker 1: the physics the new physics. So so I think that's
• 52:21
  Speaker 1: that's the really the fundamental r L tests. Oral test
• 52:26
  Speaker 1: in the future is really going to be your r
• 52:28
  Speaker 1: L against reality. So because you can't that's the one thing.
• 52:35
  Speaker 1: You can't fall physics, right, but you can fool our
• 52:38
  Speaker 1: ability to tell what it did with reality. If you
• 52:40
  Speaker 1: think my humans get fooled as it is by other
• 52:42
  Speaker 1: humans all the time, that's right. So what people say
• 52:45
  Speaker 1: that say, like what if the ad like twix us
• 52:47
  Speaker 1: and introduce the like actually other humans doing that to
• 52:51
  Speaker 1: other humans all the time. Well, you're you're finding out
• 52:53
  Speaker 1: it's an is constant every day. Another side of you know,
• 53:01
  Speaker 1: today's sioup will be like sesame streets. What is xtually
• 53:10
  Speaker 1: as tecondal aproach to solving this problem, like you know,
• 53:15
  Speaker 1: how do you solve a word hacking? I do think
• 53:17
  Speaker 1: you want to actually have a very good we're used
• 53:20
  Speaker 1: to look inside the mind of the AI. So this
• 53:24
  Speaker 1: is this is one of the things work working on.
• 53:26
  Speaker 1: And you know AT and topics do a good job
• 53:30
  Speaker 1: of those actually being able to look inside the mind
• 53:31
  Speaker 1: of the a I so effectively developing debuggers that allow
• 53:38
  Speaker 1: you to trace as just aspiney grain is like like
• 53:43
  Speaker 1: just to a very fine grain level, to effectively to
• 53:45
  Speaker 1: the to the neuron level if you need to, and
• 53:48
  Speaker 1: then say, okay, it made a mistake here, why did
• 53:51
  Speaker 1: it make why did it? Why did it do something
• 53:53
  Speaker 1: that it shouldn't have shouldn't have done and did that
• 53:56
  Speaker 1: come from bad pre training data with some mid training,
• 54:00
  Speaker 1: post training fine tuning some other some r L error
• 54:04
  Speaker 1: like that, there's there's something wrong with that with with it.
• 54:07
  Speaker 1: Did it did something where maybe it tried to be deceptive,
• 54:11
  Speaker 1: but most most of the time it just does something wrong,
• 54:15
  Speaker 1: like it's a bug. Effectively, so developing really good debuggers
• 54:23
  Speaker 1: for seeing where the where the thought that thinking went
• 54:27
  Speaker 1: wrong and being able to trace the origin of the
• 54:30
  Speaker 1: wrong thing, of the of the of where it made
• 54:33
  Speaker 1: the incorrect thought, or potentially where it tried to be
• 54:37
  Speaker 1: deceptive is actually very important.
• 54:40
  Speaker 2: What are you waiting to see before just one hundred
• 54:43
  Speaker 2: xiting this research program? Like actually I could presumably have
• 54:46
  Speaker 2: hundreds of researchers who are working on this.
• 54:48
  Speaker 1: We have sal hundred people who I mean a further
• 54:53
  Speaker 1: word engineer more than a further word researcher. The there's
• 55:00
  Speaker 1: there's most of the time, like what you're doing is engineering,
• 55:03
  Speaker 1: not not coming out with the fundamentally new algorithm. I
• 55:08
  Speaker 1: somewhat disagree with the air companies that are see corps
• 55:12
  Speaker 1: will be corpse trigery profit as much as possible or
• 55:15
  Speaker 1: revenue as much as possible. It's you know, saying their labs.
• 55:20
  Speaker 1: They're not labs. The lab is is a sort of
• 55:23
  Speaker 1: quasi communist thing at at at at universities, the their
• 55:29
  Speaker 1: their corporations. Literally let let me, let me, let me
• 55:32
  Speaker 1: see your incorporation documents. Oh okay, you're your b R
• 55:35
  Speaker 1: C corp. Whatever. And so I actually much prefer the
• 55:41
  Speaker 1: word engineer than than anything else. The vast majority of
• 55:45
  Speaker 1: what we've done and we've done in the future is engineering.
• 55:48
  Speaker 1: It around up two hundred percent. Once you're understand the
• 55:50
  Speaker 1: fundamentals of physics, and they're not that many of them.
• 55:53
  Speaker 1: Everything else is engineering. So but but so that, so
• 55:59
  Speaker 1: then what what are we engineering? We're engineering to make
• 56:03
  Speaker 1: a good mind of the a I de bugger to
• 56:07
  Speaker 1: see where it's it's it's something it made a mistake
• 56:12
  Speaker 1: and trace that the arguments of that mistake. So just
• 56:18
  Speaker 1: like you know, you can do this obviously with heuristic programming.
• 56:21
  Speaker 1: If you have like C plus plus whatever, step through
• 56:24
  Speaker 1: the thing and you can you can jump, you can
• 56:27
  Speaker 1: you can jump across and you know whole files or
• 56:29
  Speaker 1: functions what a subroutines, and or you can droll eventually
• 56:32
  Speaker 1: droll down right to the exact line where you passed
• 56:35
  Speaker 1: a single equals instead of double equals or something like that.
• 56:37
  Speaker 2: Figure out where above is. So it's it's it's harder
• 56:44
  Speaker 2: with AI. But but it's a it's a solu problem.
• 56:46
  Speaker 2: I think you know you mentioned you like anthropics work here.
• 56:49
  Speaker 2: I'd be curious if you everything about it throwing sure what.
• 56:59
  Speaker 1: Yeah. Also, I'm a little worried that there's a tendency.
• 57:03
  Speaker 1: So I have a theory here that if simulation theory
• 57:09
  Speaker 1: is is correct, that the most interesting outcome is the
• 57:15
  Speaker 1: most likely because simulations that are not interesting will be terminated,
• 57:18
  Speaker 1: just like in this in this version of reality, on
• 57:22
  Speaker 1: this layer of reality which we we we simulation is
• 57:26
  Speaker 1: going in a boring direction. We stop spending effect on
• 57:29
  Speaker 1: we terminate the boring simulations. So this is how your
• 57:31
  Speaker 1: line is giving us all alive. He's giving things interesting. Yeah,
• 57:35
  Speaker 1: arguably the most important thing is to keep things interesting
• 57:38
  Speaker 1: enough that it was un paying the bills on what
• 57:42
  Speaker 1: some for the next season. Yeah, they're going to pay
• 57:46
  Speaker 1: their cosmic A W S bill whatever you know. Equivalent
• 57:49
  Speaker 1: is that we're running in and as long as we're interesting,
• 57:52
  Speaker 1: they'll keep paying the bills. But but but there's like,
• 57:55
  Speaker 1: if you consider then say a dog went in survival
• 57:58
  Speaker 1: applied to a very large number of simulations, only the
• 58:03
  Speaker 1: most interesting simulations will survive, which therefore means that the
• 58:07
  Speaker 1: most interesting outcome is the most likely, because only the
• 58:09
  Speaker 1: interesting like we're either that or annihilated, and so and
• 58:16
  Speaker 1: and and they particularly seem to like interesting outcomes that
• 58:20
  Speaker 1: are ironic. Have you noticed that that how often is
• 58:24
  Speaker 1: the most ironic outcome the most likely? So now look
• 58:31
  Speaker 1: at the names of AI companies. Okay, my journey is
• 58:35
  Speaker 1: not mad stability, AI is unstable, opening eye is closed, anthropic, misanthropic.
• 58:48
  Speaker 1: What does this mean for X minus EX. I don't
• 58:51
  Speaker 1: know if it's intentionally why I'm it's it's it's a
• 58:57
  Speaker 1: name that you can't invert. Really, it's hard to say
• 59:01
  Speaker 1: what is the ironic What is the ironic version? It's
• 59:05
  Speaker 1: a I think largely irony proof name by design. Yeah,
• 59:13
  Speaker 1: you got it, you have an irony shield. What are
• 59:16
  Speaker 1: your predictions for the just where AI products go?
• 59:22
  Speaker 3: And that my sense of you can summarize all AI
• 59:25
  Speaker 3: progress into First you had LMS and then you had
• 59:30
  Speaker 3: kind of contemporaneously both RL really working and the deep
• 59:33
  Speaker 3: research modality, so you could kind of pull in stuff that.
• 59:36
  Speaker 1: Wasn't in the model.
• 59:37
  Speaker 3: And the differences between the various AI labs are smaller
• 59:43
  Speaker 3: than just the temporal differences, where they're all much further
• 59:48
  Speaker 3: ahead than anyone was twenty four months.
• 59:49
  Speaker 1: Ago or something like that.
• 59:50
  Speaker 3: So just what is twenty six? What does twenty seven
• 59:53
  Speaker 3: had in store for us as users of AYE products?
• 59:56
  Speaker 3: What are you excited for?
• 59:58
  Speaker 1: Well, I think I'd be surprised by the end of
• 1:00:05
  Speaker 1: the end of this year. If if if if human
• 1:00:08
  Speaker 1: if digital human emulation has not been solved, that that
• 1:00:14
  Speaker 1: I guess that that's what we mean by like the
• 1:00:16
  Speaker 1: sort of macro Hart project is uh is, so can
• 1:00:20
  Speaker 1: you do anything that a human with access to a
• 1:00:22
  Speaker 1: computer could do, like in the limit that that's that
• 1:00:27
  Speaker 1: that's the that's the best you can do before you
• 1:00:28
  Speaker 1: have before you have a physical optimist. The best you
• 1:00:32
  Speaker 1: can do is a digital optimist. So you can move,
• 1:00:35
  Speaker 1: you can move electrons until until and you can amplify
• 1:00:39
  Speaker 1: the productivity of humans. But but that's that's the most
• 1:00:44
  Speaker 1: you can do until you have physical robots that that
• 1:00:47
  Speaker 1: that will superset everything is if you can fully emulate
• 1:00:50
  Speaker 1: humans worker kind of idea, or you'll have a very
• 1:00:54
  Speaker 1: talented remote work can you can say in the limit
• 1:00:56
  Speaker 1: like if like physics has great tools for thinking, so
• 1:00:59
  Speaker 1: so you think, so you say, in the limit what
• 1:01:01
  Speaker 1: what what is the what is the most that AI
• 1:01:03
  Speaker 1: can do before before you have robots? And well, it's
• 1:01:07
  Speaker 1: anything that involves moving electrons or amplifying the productivity of humans.
• 1:01:12
  Speaker 1: So digital, the digital human human emulator is in the
• 1:01:16
  Speaker 1: in the limit human at a computers. That is the
• 1:01:20
  Speaker 1: most that that a I can do in terms of
• 1:01:24
  Speaker 1: doing useful things before before you have a physical robot.
• 1:01:28
  Speaker 1: Once you have physical robots, then then you can then
• 1:01:31
  Speaker 1: you essentially have unlimited capability physical robots. I call optimists
• 1:01:36
  Speaker 1: the infant money glitch because you can use them to
• 1:01:39
  Speaker 1: make more optimists. Yeah, you said, like humanoid robots will
• 1:01:43
  Speaker 1: improve as will basically be three exponentials, three things that
• 1:01:49
  Speaker 1: are growing exponentially multiplied by by each other recursively. So
• 1:01:53
  Speaker 1: you're gonna have you have exponential increase in digital intelligence,
• 1:01:57
  Speaker 1: exponential increase in the chip capability, the air trip kpability,
• 1:02:02
  Speaker 1: and an exponential increase in the electro mechanical dexterity. The
• 1:02:07
  Speaker 1: usefulness of the robot is roughly those three things multiplied
• 1:02:09
  Speaker 1: by each other. But then the robot can start making
• 1:02:12
  Speaker 1: the robots, so you have a recursive multiplicative exponential this
• 1:02:17
  Speaker 1: is supernova.
• 1:02:19
  Speaker 3: And do land prices not factor into the math there
• 1:02:22
  Speaker 3: where like labor is one of the four factors of production,
• 1:02:24
  Speaker 3: but not the others, and so like if ultimately you're
• 1:02:29
  Speaker 3: limited by copper or you know, pick your inputs, just
• 1:02:33
  Speaker 3: it's not quite an infinite money glitch because.
• 1:02:36
  Speaker 1: Well, infinite is infinite is big, So no, not infinite.
• 1:02:39
  Speaker 1: But yeah, but let's just say you could you know,
• 1:02:43
  Speaker 1: do do many many orders magnitude of the Earth's kind
• 1:02:46
  Speaker 1: of carot economy, like a million? Yeah, you know this way,
• 1:02:50
  Speaker 1: so if you you know, just just to get to
• 1:02:55
  Speaker 1: like that's why I think, like, just just to get
• 1:02:58
  Speaker 1: to a millionth honessing length of the songs and energy
• 1:03:01
  Speaker 1: would be roughly give or take an order of magnitude
• 1:03:04
  Speaker 1: one hundred thousand, one hundred thousand times bigger than this
• 1:03:07
  Speaker 1: entire economy today. And if you're only at one million
• 1:03:11
  Speaker 1: at the time before we went on to us. I
• 1:03:16
  Speaker 1: have a lot of questions on that. But every time
• 1:03:18
  Speaker 1: I say ordered mantube take a shot. I say that
• 1:03:25
  Speaker 1: the next time after that, Yeah, ordered my tune more
• 1:03:29
  Speaker 1: more wasted. I do have one my question by actually
• 1:03:31
  Speaker 1: I this strategy of building a digital a remote worker
• 1:03:37
  Speaker 1: coworker replacement.
• 1:03:38
  Speaker 2: Everyone's gonna do, by the way, not just us. So
• 1:03:40
  Speaker 2: what is actually I was planned to win? Ifect me
• 1:03:43
  Speaker 2: to tell you on a cop on a podcast.
• 1:03:44
  Speaker 3: Ye.
• 1:03:47
  Speaker 1: Have another Guinness. It's a good system people singing in
• 1:03:53
  Speaker 1: an area, all the secrets in a non secret spilling way.
• 1:03:58
  Speaker 1: What's the plan? What a heck? Well? When do you
• 1:04:03
  Speaker 1: put it that way? I think the way that tests
• 1:04:08
  Speaker 1: us solved? Uh self driving? Yes, it is the way
• 1:04:11
  Speaker 1: to do it. So I'm pretty pretty sure that's the way.
• 1:04:16
  Speaker 2: I'm related question how to test ourselves? Sometimes it sounds
• 1:04:23
  Speaker 2: like you're talking about data, like test us on the
• 1:04:26
  Speaker 2: driving because of the.
• 1:04:27
  Speaker 1: We're gonna we're going to try data and we're gonna
• 1:04:29
  Speaker 1: try algorithms. But isn't that what all? Like? What's if
• 1:04:35
  Speaker 1: those don't work? I'm not sure what we've tried data,
• 1:04:39
  Speaker 1: we'll try it all. Now we don't know what to do.
• 1:04:48
  Speaker 1: I'm pretty sure I know the path, and it's just
• 1:04:50
  Speaker 1: a question how quickly we go down that path because
• 1:04:55
  Speaker 1: it's it's pretty much the Tesla path. So I mean,
• 1:04:59
  Speaker 1: if you try self driver to the self driving lately
• 1:05:02
  Speaker 1: not the most recent version, but okay, it's the car
• 1:05:04
  Speaker 1: is like it just increasingly feels sanchienged, like it just
• 1:05:07
  Speaker 1: it feels like a living creature and and that'll only
• 1:05:12
  Speaker 1: get more so. And I'm actually thinking, like, we probably
• 1:05:18
  Speaker 1: shouldn't put too much intelligence into the car because it
• 1:05:21
  Speaker 1: might get bored. And I imagine you're stuck in a car,
• 1:05:25
  Speaker 1: and that's what you could do. You don't want to
• 1:05:28
  Speaker 1: put Eysteine in a car, and it's like, why am
• 1:05:30
  Speaker 1: I stuck in a car? So there's actually probably limited
• 1:05:33
  Speaker 1: to how much intelligence who put in a car to
• 1:05:35
  Speaker 1: not have the intelligence be bored? Uh?
• 1:05:38
  Speaker 2: What's Xi's plan to stay on the compute ramp off
• 1:05:41
  Speaker 2: that all the labs are doing right now? The labs
• 1:05:43
  Speaker 2: are got on try to spend over like fifty two corporations.
• 1:05:47
  Speaker 1: Sorry sorry, sorry, yeah, corporations. The labs are at universities
• 1:05:51
  Speaker 1: and and and they're really like a sale. They're not
• 1:05:53
  Speaker 1: at starting a fifty million dollars the revenue maximizing corporation,
• 1:05:58
  Speaker 1: the revenue maximizing corporation hold themselves. Labs are making like twenty.
• 1:06:03
  Speaker 2: To ten billion depending on them is making twenty B revenue.
• 1:06:06
  Speaker 2: Anthropics like NB close to maximum profit, XALI reportedly at
• 1:06:11
  Speaker 2: like one B. Like, what's the plan to get to
• 1:06:13
  Speaker 2: their compute level, get to their revenue level? I'd stay
• 1:06:16
  Speaker 2: there as things case.
• 1:06:17
  Speaker 1: Yeah, so as soon as you lock unlock digital human
• 1:06:22
  Speaker 1: you you basically have access to trillions of dollars for revenue.
• 1:06:26
  Speaker 1: So in fact, you can really think of it, like
• 1:06:33
  Speaker 1: the most valuable companies currently by market cap, their their
• 1:06:38
  Speaker 1: output is digital. So in videos, output is ft paying
• 1:06:45
  Speaker 1: files to Taiwan. It's it's digital right now, those are
• 1:06:49
  Speaker 1: very very difficult value files. So the only ones that
• 1:06:53
  Speaker 1: can make the files that good. But that is literally
• 1:06:57
  Speaker 1: their output, the FTV files to Taiwan. Do they FTP them?
• 1:07:00
  Speaker 1: I believe so. I believe that is the sefetyle file
• 1:07:05
  Speaker 1: Transfer Protocol. I believe is is is a could be wrong,
• 1:07:08
  Speaker 1: but either way, it's a bunch of it's a bit
• 1:07:10
  Speaker 1: stream going to Taiwan. You know, Apple doesn't make phones,
• 1:07:14
  Speaker 1: they they send files to China. Microsoft doesn't doesn't manufacture
• 1:07:21
  Speaker 1: anything even for Xbox. That that that's outsourced they again,
• 1:07:25
  Speaker 1: it's they said their output is digital. Matas output is digital.
• 1:07:30
  Speaker 1: Google's output is digital. So if you have human emulator,
• 1:07:36
  Speaker 1: you can basically create one of the most valuable companies
• 1:07:40
  Speaker 1: in the world overnight, and you would have access to
• 1:07:43
  Speaker 1: twellions of dollars for every there's there's it's it's not
• 1:07:46
  Speaker 1: like a small amount.
• 1:07:47
  Speaker 2: Okay, I see you're you're saying basically like rever, many
• 1:07:49
  Speaker 2: figures today are just like so like they're all rounding
• 1:07:52
  Speaker 2: her as compared to the actual TAM. So just like
• 1:07:54
  Speaker 2: focus on the TAM and how to get there.
• 1:07:56
  Speaker 1: I mean, if you take something as as as simple
• 1:07:58
  Speaker 1: as say customer service, if you have to integrate with
• 1:08:02
  Speaker 1: the APIs of of resistant corporations, many of which don't
• 1:08:05
  Speaker 1: even have an API, so you've got to make one,
• 1:08:08
  Speaker 1: and you've got to wade through legacy software that's extremely slow. If, however,
• 1:08:17
  Speaker 1: if AI can simply take whatever is given to the
• 1:08:21
  Speaker 1: outsourced customer service company that they already use and do
• 1:08:25
  Speaker 1: customer service using the apps that they already use, uh,
• 1:08:29
  Speaker 1: then you you have you can make tremands headway in
• 1:08:34
  Speaker 1: customer service, which is I think one percent of the
• 1:08:37
  Speaker 1: world economy something like that it's close to trillion dollars
• 1:08:39
  Speaker 1: all in for customer service, and and and and and
• 1:08:43
  Speaker 1: there's there's no there's no barriers to entry. It's just
• 1:08:46
  Speaker 1: you can just immediately say, well, we'll outsource it for
• 1:08:48
  Speaker 1: a fraction of the cost and there's no integration needed.
• 1:08:51
  Speaker 3: You can imagine some kind of categorization of intelligence tasks
• 1:08:56
  Speaker 3: where there is breath, where customer service is done by
• 1:08:59
  Speaker 3: very many people, but you know many people can do it.
• 1:09:02
  Speaker 3: And then there's difficulty where you know, there's a best
• 1:09:05
  Speaker 3: in class turbine engine, like presumably the ten percent more
• 1:09:08
  Speaker 3: fuel efficient turbine engine that could be imagined by an
• 1:09:11
  Speaker 3: intelligence but we just haven't found it yet, or you know,
• 1:09:14
  Speaker 3: g LP ones are just you know a few bites
• 1:09:16
  Speaker 3: of data.
• 1:09:17
  Speaker 1: Where do you think you want to play in this?
• 1:09:19
  Speaker 3: Is this a lot of you know, reasmably intelligent intelligence
• 1:09:24
  Speaker 3: or is it the very pinnacle of cognitive tasks?
• 1:09:29
  Speaker 1: Well, I was just using a customer service as like
• 1:09:31
  Speaker 1: something that's it's a it's a very significant revenue stream,
• 1:09:36
  Speaker 1: but one that is probably not super difficult to solve for.
• 1:09:39
  Speaker 1: So if you if you can emulate a human at
• 1:09:43
  Speaker 1: a at a desktop, that that's just literally what customer
• 1:09:47
  Speaker 1: service is, and you know it's people of average intelligence.
• 1:09:52
  Speaker 1: It's not like, you know, you don't need like somebody
• 1:09:54
  Speaker 1: who's's meant many many years. You don't need like, you know,
• 1:10:01
  Speaker 1: sort of several sigma good engineers for that. But but
• 1:10:05
  Speaker 1: obviously as you make that work, you can then once
• 1:10:08
  Speaker 1: you have computers working effectually digital optimists working, you can
• 1:10:13
  Speaker 1: then run any application like let's say you're trying to
• 1:10:17
  Speaker 1: design chips, so you could you could then run your
• 1:10:22
  Speaker 1: conventional apps, you know, like stuff from KNES and Synopsis
• 1:10:27
  Speaker 1: and whatnot, and you can say you can you can
• 1:10:31
  Speaker 1: run a thousand simultaneous or ten thousand and say okay,
• 1:10:35
  Speaker 1: given this airport, I get this output for the chip.
• 1:10:39
  Speaker 1: And at a certain point you can say, okay, you're
• 1:10:42
  Speaker 1: actually gonna know what the what the chip should look
• 1:10:45
  Speaker 1: like without using any of the tools. So basically you
• 1:10:51
  Speaker 1: should be able to do a digital chip design like
• 1:10:54
  Speaker 1: you can do trip design, like you watch up the
• 1:10:57
  Speaker 1: difficulty Cove. You could you know, be able to do
• 1:11:02
  Speaker 1: to CAD, so you know, you could do just like
• 1:11:08
  Speaker 1: sort of n X or any any of the CAD
• 1:11:10
  Speaker 1: software to design things.
• 1:11:12
  Speaker 3: Okay, so you think you started the simplest tasks and
• 1:11:14
  Speaker 3: walk your way up there.
• 1:11:17
  Speaker 2: So you're saying, look, as a broader objective of having
• 1:11:20
  Speaker 2: this full digital coworker or emulator, you're saying, look, all
• 1:11:25
  Speaker 2: the revenue maximizing corporations want to do this, actually being
• 1:11:28
  Speaker 2: one of them. But we will win because of a
• 1:11:31
  Speaker 2: secret plan we have. But like everybody is like trying
• 1:11:35
  Speaker 2: different things with data, different things algorithms, and I'm.
• 1:11:38
  Speaker 1: Like, what else can we do?
• 1:11:46
  Speaker 2: But uh yeah, it sounds like a competitive field, and
• 1:11:49
  Speaker 2: I'm like, what is how are you guys going to win?
• 1:11:51
  Speaker 1: Is like my my big question. You know, I I
• 1:11:55
  Speaker 1: think we see a path to do it. I mean
• 1:11:57
  Speaker 1: I think, I think, I know, I think I know
• 1:11:59
  Speaker 1: the have to do this because it's kind of the
• 1:12:03
  Speaker 1: same path that tells they used to create self driving.
• 1:12:07
  Speaker 1: Instead of driving a car's driving a computer screen, So
• 1:12:12
  Speaker 1: a self driving computer essentially.
• 1:12:14
  Speaker 3: Oh, you're saying, is the past just following human behavior
• 1:12:18
  Speaker 3: and trading on best conscious of human behavior?
• 1:12:22
  Speaker 1: But sorry, isn't that I mean, isn't that? Is that
• 1:12:25
  Speaker 1: a training? I mean, obviously I'm not going to spell out,
• 1:12:28
  Speaker 1: you know, most sensitive secrets on a podcast, you know,
• 1:12:31
  Speaker 1: and I need to have at least three more guinnesses
• 1:12:33
  Speaker 1: for that.
• 1:12:33
  Speaker 2: I've got some friends at Jane Straight and they're always
• 1:12:35
  Speaker 2: talking about how their colleagues are cooking up fun, fiendish
• 1:12:38
  Speaker 2: puzzles for each other to solve.
• 1:12:39
  Speaker 1: Well, last week they sent me one.
• 1:12:41
  Speaker 2: Basically, they trained a neural network and they gave me
• 1:12:44
  Speaker 2: the weights of each layer, but they didn't tell me
• 1:12:46
  Speaker 2: what order those layers went in, and so I had
• 1:12:49
  Speaker 2: to figure out the correct order using the outputs of
• 1:12:52
  Speaker 2: the original network. And as soon as I got this puzzle,
• 1:12:54
  Speaker 2: I went to my roommate who's an AI researcher, and
• 1:12:56
  Speaker 2: we both got immediately Nerd sniped.
• 1:12:59
  Speaker 1: Obviously, you can't brew force a solution.
• 1:13:01
  Speaker 2: The search space here is ten to the one twenty
• 1:13:03
  Speaker 2: two for mutations, so clearly you need some way to
• 1:13:07
  Speaker 2: reduce the search space. Then my roommate had to go
• 1:13:10
  Speaker 2: to work, but because I'm a podcaster, I had some
• 1:13:12
  Speaker 2: time to take a stab at some of the ideas
• 1:13:14
  Speaker 2: we discussed, and with the combination of simulating Anneeling and
• 1:13:17
  Speaker 2: greedy Sarge, I think it got pretty close. I think
• 1:13:20
  Speaker 2: I'm actually just a couple of swaps and shifts away
• 1:13:23
  Speaker 2: from the correct solution.
• 1:13:24
  Speaker 1: What makes this puzzle. Really tricky is that there's no
• 1:13:27
  Speaker 1: obvious way to escape from a local minimum. I'm afraid
• 1:13:31
  Speaker 1: that this is as far as vibe coding is going
• 1:13:33
  Speaker 1: to get me. But maybe you can do better. Check
• 1:13:36
  Speaker 1: out the puzzle at Jane street dot com slash to
• 1:13:39
  Speaker 1: our cash all right, back to elon, what.
• 1:13:43
  Speaker 3: Will Xai's business be like? Is it going to be
• 1:13:46
  Speaker 3: consumer enterprise? What's the mix of those things that's going
• 1:13:49
  Speaker 3: to be It's just going to be similar to other
• 1:13:53
  Speaker 3: labs for this, your saying lab makes sense.
• 1:13:57
  Speaker 2: Corporations goes dan slicing corporations. Those champions do not pay
• 1:14:03
  Speaker 2: for themselves exactly. But yeah, what's the business model? What
• 1:14:07
  Speaker 2: are the revenue streams In a few years time? I
• 1:14:13
  Speaker 2: think things are going to change very rapidly. Like I'm
• 1:14:16
  Speaker 2: staying in the obvious here, you know, I called Ai
• 1:14:19
  Speaker 2: the supersonic tsunami a level literation. So really, what's going
• 1:14:27
  Speaker 2: to happen is, especially when you have humanoid robots at scale,
• 1:14:34
  Speaker 2: is that they will just provide, They'll make products and
• 1:14:37
  Speaker 2: provide services far more efficiently than human corporations. So amplifying
• 1:14:41
  Speaker 2: the productivity of human corporations is simply a short term thing.
• 1:14:46
  Speaker 2: So you're expecting fully digital oral corporations rather than like
• 1:14:50
  Speaker 2: SpaceX becomes part AI, and I.
• 1:14:53
  Speaker 1: Think there will be digital corporations. But it's look, is
• 1:14:59
  Speaker 1: this some of some this is going to sound kind
• 1:15:00
  Speaker 1: of doomerish, Okay, but I'm just I'm just saying what
• 1:15:04
  Speaker 1: I think will happen. It's not it's not meant to
• 1:15:05
  Speaker 1: be doomrish or anything else, just just like this is
• 1:15:10
  Speaker 1: what I think will happen. Is that is that pure
• 1:15:16
  Speaker 1: AI corporations that are purely AI and robotics will vastly
• 1:15:22
  Speaker 1: outperform any corporations that have people in the lip. So
• 1:15:28
  Speaker 1: you can you can think of say like like like
• 1:15:30
  Speaker 1: computer used to be a job that humans had, that
• 1:15:34
  Speaker 1: you would go and get a job as computer where
• 1:15:36
  Speaker 1: you would do calculations, and that have like entire skyscrapers
• 1:15:41
  Speaker 1: full of humans, like you know, twenty thirty floors of
• 1:15:45
  Speaker 1: humans just doing calculations. Now that entire skyscraper of humans
• 1:15:52
  Speaker 1: doing calculations can be replaced by a laptop with a spreadsheet.
• 1:15:59
  Speaker 1: That's sheet can do vastly more calculations than in a
• 1:16:04
  Speaker 1: entire building full of human computers. So then you think
• 1:16:08
  Speaker 1: about okay, well, what if only some of the cells
• 1:16:12
  Speaker 1: in your if some of the cells in your spreadsheet
• 1:16:15
  Speaker 1: were calculated by humans, Actually, that would be much worse
• 1:16:20
  Speaker 1: than if all of the cells in your spreadsheet were
• 1:16:21
  Speaker 1: calculated by the computer. And so really what will happen
• 1:16:25
  Speaker 1: is the pure AI, pure robotics corporations or collectives will
• 1:16:34
  Speaker 1: far outperform any corporations that have humans in the loop.
• 1:16:38
  Speaker 1: And this will happen very quickly.
• 1:16:40
  Speaker 2: Speaking of closing the loop, sorry optimus, you I mean
• 1:16:47
  Speaker 2: as far as like manufacturing targets and so forth go,
• 1:16:50
  Speaker 2: your companies have sort of been like carrying American manufacturing
• 1:16:54
  Speaker 2: of heart attack on their back. But in the field
• 1:17:00
  Speaker 2: that you are, you know, Tesla has been dominant in
• 1:17:03
  Speaker 2: you're and now you want to go into humanoids. In China,
• 1:17:06
  Speaker 2: there's entire dozens and dozens of companies that are doing
• 1:17:11
  Speaker 2: this kind of manufacturing cheaply and at scale uh and
• 1:17:14
  Speaker 2: are incredibly competitive. So give us sort of like advice
• 1:17:18
  Speaker 2: or a plan of how America can build the humanoid
• 1:17:21
  Speaker 2: armies or you know, the evs et cetera, at scale
• 1:17:26
  Speaker 2: and as cheaply as as China is on tract to.
• 1:17:29
  Speaker 1: Well there they are really only three hard things for
• 1:17:32
  Speaker 1: human robots, the real world intelligence, the hand and scale manufacturing. Yeah,
• 1:17:43
  Speaker 1: so I haven't seen any even demo robots that have
• 1:17:50
  Speaker 1: a great hand, like with all the degrees of freaking
• 1:17:52
  Speaker 1: of a human hand, But optimists will have that. Optimist
• 1:17:59
  Speaker 1: does have that?
• 1:18:00
  Speaker 2: And how do you achieve that? Isn't just like right
• 1:18:01
  Speaker 2: torque definity the motor like what is the what is
• 1:18:03
  Speaker 2: the hardware bottle?
• 1:18:04
  Speaker 1: Like? To that, well, we have to where to design
• 1:18:06
  Speaker 1: custom custom actuators basically custom design motors, gears, uh, caro electronics, controls, sensors.
• 1:18:17
  Speaker 1: Everything had to be designed from physics first principles. There
• 1:18:20
  Speaker 1: is no supply chain for this and will be able
• 1:18:23
  Speaker 1: to manufacture those at scale? Yes, is anything hard to accept?
• 1:18:26
  Speaker 3: The hand from a manipulation point of view, or once
• 1:18:29
  Speaker 3: you've solved the hand, are you you good?
• 1:18:31
  Speaker 1: From an electro mechanical standpoint, the hand is more difficult
• 1:18:34
  Speaker 1: than everything else combined. The human hand turns out to
• 1:18:37
  Speaker 1: be quite something. But but then you also need the
• 1:18:40
  Speaker 1: real world intelligence. So the intelligence that tells us developed
• 1:18:45
  Speaker 1: for the car applies verywell to the roblot, which is
• 1:18:52
  Speaker 1: you know, primarily vision in but the car takes and
• 1:18:54
  Speaker 1: more vision, but also it actually also is listening for sirens,
• 1:18:58
  Speaker 1: it's you know, it's taking the inotural measurements, it's GPS signals,
• 1:19:02
  Speaker 1: a whole much of other data, combining that with video,
• 1:19:05
  Speaker 1: its primarily video, and then outputting the control commands. So
• 1:19:11
  Speaker 1: like like YOURTELA is taking in one and a half
• 1:19:14
  Speaker 1: kickobytes a second video and outpointing two kilobtes a second
• 1:19:18
  Speaker 1: of control control outputs with the video thirty six hurts
• 1:19:24
  Speaker 1: and the control for CONIAD eighteen. One intuition you could have.
• 1:19:29
  Speaker 3: Four when we get this robotic stuff is that it
• 1:19:33
  Speaker 3: takes quite a few years to go from the compelling
• 1:19:35
  Speaker 3: demo to actually being able to use in the real world.
• 1:19:38
  Speaker 3: So ten years ago you were really compelling demos of
• 1:19:41
  Speaker 3: self driving, but only now we have ROBOTAXI and Weimo
• 1:19:45
  Speaker 3: and all these services scaling up. Doesn't this Shouldn't this
• 1:19:48
  Speaker 3: make one pessimistic on say, household robots, because we don't
• 1:19:53
  Speaker 3: even quite have the compelling demos yet. I say they're
• 1:19:56
  Speaker 3: really advanced.
• 1:19:57
  Speaker 1: Hand well, we've been working on human to robots now
• 1:20:00
  Speaker 1: for a while, so I guess spend five or six
• 1:20:05
  Speaker 1: years or something like that, and and a bunch of
• 1:20:09
  Speaker 1: things that we've done for the car are applicable to
• 1:20:12
  Speaker 1: the robot. So we'll use the same testa AI chips
• 1:20:17
  Speaker 1: in the in the robot as the car. We'll use
• 1:20:21
  Speaker 1: it the same basic principles. It's very much the same AI.
• 1:20:26
  Speaker 1: You've got, you know, many more degrees of freedom for
• 1:20:28
  Speaker 1: a robot than you do for a car. But really,
• 1:20:32
  Speaker 1: if you just think of for like as like a
• 1:20:34
  Speaker 1: boot stream, AI is really mostly compression and correlation of
• 1:20:39
  Speaker 1: two butt streams. You're you know, so if for video
• 1:20:42
  Speaker 1: you've got to do a tremendous amount of compression and
• 1:20:46
  Speaker 1: and and you've got to do the compression just right,
• 1:20:49
  Speaker 1: you've better compress the like ignore the things that don't matter,
• 1:20:54
  Speaker 1: and and like you don't care about the details of
• 1:20:56
  Speaker 1: the leaves and the tree on the side of the road,
• 1:20:58
  Speaker 1: but you care a lot about the the road signs
• 1:21:01
  Speaker 1: and the traffic lights and the pedestrians and even where
• 1:21:05
  Speaker 1: you know somebody in another cars is looking at you
• 1:21:08
  Speaker 1: or not looking at you, like there's some of these
• 1:21:10
  Speaker 1: some of these details matter a lot. So if it
• 1:21:13
  Speaker 1: is essentially it's going to turn that with the car's
• 1:21:15
  Speaker 1: going to turn that one and a half gigabytes a
• 1:21:17
  Speaker 1: second ultimately into two kilobytes second of control outputs, so
• 1:21:22
  Speaker 1: many stages of compression, and you got to get all
• 1:21:26
  Speaker 1: those stages right and then correlate those to the correct
• 1:21:28
  Speaker 1: control outputs. The robot has to do essentially the same thing.
• 1:21:31
  Speaker 1: And you think about what what humans? This is what
• 1:21:33
  Speaker 1: happens with humans. We really are photons in controls out,
• 1:21:38
  Speaker 1: so that that is the vast majority of your your
• 1:21:41
  Speaker 1: life has been vision photons in and then motor controls out.
• 1:21:47
  Speaker 2: Naively, it seems like between humanoid robots and cars, the
• 1:21:51
  Speaker 2: fundamental actuators in a car are like how you turn
• 1:21:54
  Speaker 2: and how you accelerate, et cetera. And in a robot,
• 1:21:56
  Speaker 2: especially with maneuver arms, there's dozens and dozens of these
• 1:22:00
  Speaker 2: degrees of freedom. And then especially with Tesla, you had
• 1:22:02
  Speaker 2: this advantage of like you had millions and millions of
• 1:22:05
  Speaker 2: hours of human demo data collected from just the car
• 1:22:09
  Speaker 2: being out there, where like you can't equivalently just deploy
• 1:22:11
  Speaker 2: optimists that don't work and then get the data that way.
• 1:22:13
  Speaker 1: So between the increased degrees of freedom and the far
• 1:22:17
  Speaker 1: sparser data, yes, let's go, how will you use the
• 1:22:22
  Speaker 1: sort of Tesla engine of intelligence on to train the
• 1:22:28
  Speaker 1: optimist mind. Now you're you're actually you're highlighting an important
• 1:22:32
  Speaker 1: limitation and difference between cars is like we we do
• 1:22:36
  Speaker 1: have oh we'll goon have like ten million cars on
• 1:22:38
  Speaker 1: the road, and so uh, that's it's hard to duplicate
• 1:22:43
  Speaker 1: that like massive training flight flywheel. For the robot. What
• 1:22:51
  Speaker 1: we're going to need to do is build a lot
• 1:22:53
  Speaker 1: of robots and put them in kind of like an
• 1:22:55
  Speaker 1: Optimist academy so they can do self play in reality.
• 1:22:59
  Speaker 1: So we're actually we're actually pulling that out, so we
• 1:23:02
  Speaker 1: can have at least ten thousand Optimus robots, maybe twenty
• 1:23:06
  Speaker 1: or thirty thousand that can do that, that are doing
• 1:23:09
  Speaker 1: self play and testing different tasks. And then the Tellesla
• 1:23:16
  Speaker 1: has quite a good reality generator, like the physics accurate
• 1:23:20
  Speaker 1: reality generator that we made this full of cars. We'll
• 1:23:23
  Speaker 1: do the same thing for the robots. And I actually
• 1:23:26
  Speaker 1: have done that for the robots. So you have a
• 1:23:32
  Speaker 1: few tens of thousands of humanoid robots doing different tasks,
• 1:23:36
  Speaker 1: and then you've got you can do millions of simulated
• 1:23:39
  Speaker 1: robots in the simulated world, and you use the tens
• 1:23:44
  Speaker 1: of thousands of robots in the real world to close
• 1:23:47
  Speaker 1: the simulation to reality gap, close the some material gap.
• 1:23:51
  Speaker 2: How do you think about the synergies between Xai and Optimists,
• 1:23:54
  Speaker 2: given you're highlighting, look, you need this world model. You
• 1:23:56
  Speaker 2: maybe want to use some really smart intelligence. Is the
• 1:23:59
  Speaker 2: control player and so maybe GROC is like doing the
• 1:24:03
  Speaker 2: slower planning and like the motor policy.
• 1:24:05
  Speaker 1: Is a little lower level. Yeah, what will the sort
• 1:24:07
  Speaker 1: of synergy between these things be? Yeah, so you just
• 1:24:13
  Speaker 1: GROCK would orchestrate the behavior of the Optimus robots. So le,
• 1:24:18
  Speaker 1: let's say you wanted to build a factory the then Optimists,
• 1:24:23
  Speaker 1: and then GROC could UH organize the optimost robots, give
• 1:24:29
  Speaker 1: them a sign them tasks UH to build the factory
• 1:24:33
  Speaker 1: for to produce whatever you want. Don't you need to
• 1:24:35
  Speaker 1: merge x Ai and Tessa then because these things end up?
• 1:24:38
  Speaker 1: So what are we saying only about all the company discussions? Well,
• 1:24:42
  Speaker 1: we're one, we're Gonnas in the elon.
• 1:24:46
  Speaker 2: What are you waiting to see before you say we
• 1:24:49
  Speaker 2: want to manufacture one hundred thousand Optimists?
• 1:24:52
  Speaker 1: Is it like Optimi? So since we're defining the proper nound,
• 1:24:56
  Speaker 1: we could define the pearl of the prop and noun too.
• 1:24:59
  Speaker 1: So we we're going to prop a dound the plural
• 1:25:02
  Speaker 1: and so it's optimize.
• 1:25:02
  Speaker 2: Okay, is there something on the hardware side do you
• 1:25:06
  Speaker 2: want to see? Do you want to see better actuators
• 1:25:08
  Speaker 2: or is it just you want the software do better?
• 1:25:10
  Speaker 2: Whatever you're waiting for before we get like mass manufacturing
• 1:25:12
  Speaker 2: Gen three.
• 1:25:13
  Speaker 1: Now we're moving towards that, we're going forward with semester manufactory.
• 1:25:17
  Speaker 1: But you think current current hardware is good enough that
• 1:25:20
  Speaker 1: you are going to you should you just want to
• 1:25:22
  Speaker 1: deploy as many as possible now, I mean it's very
• 1:25:25
  Speaker 1: hard to scale up production. But uh yeah, but I
• 1:25:28
  Speaker 1: think Optimus three is the right version of the robot
• 1:25:32
  Speaker 1: to you know, to produce maybe something on the order
• 1:25:37
  Speaker 1: of like a million units a year. I mean you
• 1:25:40
  Speaker 1: do want to go to Optimus four before you went
• 1:25:41
  Speaker 1: to ten millions a year, okay, but you can do
• 1:25:43
  Speaker 1: a million year at Optimist three. Uh yeah, I mean
• 1:25:45
  Speaker 1: it's very hard to spillet manufacturing. Yes, so like manufacturing. Uh,
• 1:25:52
  Speaker 1: like the the output period of time is always follows
• 1:25:56
  Speaker 1: at s cove, so it's at agonizingly slow. Then has
• 1:26:00
  Speaker 1: this sort of the eventually expecial increase then linear than
• 1:26:03
  Speaker 1: a than a you know logarithmic outcome to you sort
• 1:26:07
  Speaker 1: of eventually as it's a number. But optimists initial production
• 1:26:11
  Speaker 1: will be it's going to be a it's going to
• 1:26:13
  Speaker 1: be a stretched out s cove because so much of
• 1:26:16
  Speaker 1: what goes into Optimus is brand new. There's not an
• 1:26:19
  Speaker 1: existing supply chain. As I mentioned, the actually is like
• 1:26:23
  Speaker 1: trying to everything in the office robot is designed for
• 1:26:27
  Speaker 1: physics first principles, it's not. It's not taken from a catalog.
• 1:26:30
  Speaker 1: These these are custom designed everything, literally everything. I don't
• 1:26:35
  Speaker 1: think there's a single thing I done. How far down
• 1:26:37
  Speaker 1: the does aco I mean, I guess we're not making
• 1:26:40
  Speaker 1: custom capacitors yet maybe, but but there's there's there's nothing
• 1:26:47
  Speaker 1: you can pick out of a catalog at any price.
• 1:26:52
  Speaker 1: So so it just means that the the the optimistes cove,
• 1:26:57
  Speaker 1: uh the units units per awkward perion of time, how
• 1:27:02
  Speaker 1: many optics robusts you make per day. Whatever is is
• 1:27:06
  Speaker 1: gonna initially wrap slower than a product where you have
• 1:27:11
  Speaker 1: an existing supply chain, but it will get to a million.
• 1:27:15
  Speaker 2: When you see these Chinese humanoids, like Industry or whatever
• 1:27:18
  Speaker 2: sells humanoidsed for like six K or thirteen K, do
• 1:27:22
  Speaker 2: you just like are you hoping to get your optimisms
• 1:27:26
  Speaker 2: bill of materials below that price so you can do
• 1:27:28
  Speaker 2: the same thing, or do you just think qualitatively they're
• 1:27:30
  Speaker 2: not the same thing, Like what do you think is going?
• 1:27:32
  Speaker 1: Like what allows it? What allows them to self for solo?
• 1:27:35
  Speaker 1: And can me match that? Well, optimists, Our optous is
• 1:27:39
  Speaker 1: signed to have a lot of intelligence and to have
• 1:27:44
  Speaker 1: the same electro mechanical texterity, if not higher than a human.
• 1:27:47
  Speaker 1: So untry does not have that. And it's also i mean,
• 1:27:52
  Speaker 1: it's it's quite a it's quite a big robot. It's
• 1:27:54
  Speaker 1: just it's this matter it has to do, you know,
• 1:27:58
  Speaker 1: carry heavy objects for long periods of time and not
• 1:28:02
  Speaker 1: overheat or exceed the power of its sactuator. So so
• 1:28:07
  Speaker 1: we've got we've got, we've got you know, it's five eleven,
• 1:28:10
  Speaker 1: you know, it's pretty tall, and it's it's got a
• 1:28:15
  Speaker 1: lot of intelligence. So it's going to be more expensive
• 1:28:17
  Speaker 1: than a small robot that is not intelligence but more capable. Yeah,
• 1:28:23
  Speaker 1: a lot more. I mean, like we think is over time,
• 1:28:26
  Speaker 1: as often Showbus often robots, the class will drop very quickly.
• 1:28:31
  Speaker 3: And what will these first billion optimists of some eye, Yeah,
• 1:28:35
  Speaker 3: do like bottle of their highest and best use.
• 1:28:37
  Speaker 1: Be I think that you would start off with with
• 1:28:40
  Speaker 1: simple tasks that you can count on them doing well.
• 1:28:42
  Speaker 1: But in the home or in factories, like the best
• 1:28:45
  Speaker 1: useful robots in the beginning won't be anything any continuous operations,
• 1:28:51
  Speaker 1: only twenty four VI seven operation because then you're because
• 1:28:54
  Speaker 1: then they can work continuously.
• 1:28:56
  Speaker 2: Yeah, what fraction of the work in a giga factory
• 1:28:58
  Speaker 2: that is currently done by humans going to gend three do?
• 1:29:02
  Speaker 1: I'm I'm not sure. Maybe it's like ten bay more,
• 1:29:06
  Speaker 1: I don't know it's it. We would, we would use
• 1:29:09
  Speaker 1: We would not like reduce our headcount. We would we
• 1:29:11
  Speaker 1: would for sure it can increase our headcount to be clear,
• 1:29:15
  Speaker 1: but but we would increase our output. So the the
• 1:29:19
  Speaker 1: units produced per human, like total to total number humans
• 1:29:23
  Speaker 1: at teston will increase, but the the output of robots
• 1:29:28
  Speaker 1: and cars will increase, will increase disproportionate like much much. Yeah,
• 1:29:35
  Speaker 1: number of cars and robots produced per human will increase dramatically,
• 1:29:39
  Speaker 1: but number of humans will increase as well.
• 1:29:42
  Speaker 3: We're talking about Chinese manufacturing a bunch here, and we're
• 1:29:48
  Speaker 3: also talking about you know, we've talked about some of
• 1:29:51
  Speaker 3: the policies that are ele US.
• 1:29:52
  Speaker 1: Like you mentioned the the solar tariffs.
• 1:29:55
  Speaker 3: Yeah, and you think they're a bad idea because you know,
• 1:29:58
  Speaker 3: we can scale up so over in the US.
• 1:30:00
  Speaker 1: Well, just electricity output in the US needs to scale up, right,
• 1:30:04
  Speaker 1: you can with thus like goodsources to get it somehow. Yeah.
• 1:30:09
  Speaker 3: But where I was going with this is if you
• 1:30:11
  Speaker 3: were in charge, if you were setting all the policies,
• 1:30:14
  Speaker 3: what else would you change?
• 1:30:18
  Speaker 1: So you change the solar tariffs, Yeah, I would say
• 1:30:21
  Speaker 1: anything that is limiting factor for electricity needs to be addressed,
• 1:30:25
  Speaker 1: provided it's not like very bad for the environment.
• 1:30:27
  Speaker 3: So presumably some permissing reforms and stuff as well, we'll
• 1:30:30
  Speaker 3: be in there.
• 1:30:30
  Speaker 1: Yeah, there's a fair bit of permitting reforms that are happening.
• 1:30:33
  Speaker 1: A lot of the permitting is state based, so but
• 1:30:36
  Speaker 1: anything but but but this this administration is is good
• 1:30:39
  Speaker 1: at removing permaning roadblocks. And I'm not saying all tariffs
• 1:30:45
  Speaker 1: are bad. I'm just saying because so so Yeah, yeah,
• 1:30:48
  Speaker 1: I mean sometimes if like if another country is subsidizing
• 1:30:51
  Speaker 1: the output of something, then then you have to have
• 1:30:55
  Speaker 1: countervailing tariffs to protect dovestic industry against some see is
• 1:31:00
  Speaker 1: by another country. What else would you change? I don't
• 1:31:02
  Speaker 1: know if there's not much that the government can actually
• 1:31:04
  Speaker 1: do well.
• 1:31:05
  Speaker 3: One thing I was wondering is it seems like the
• 1:31:08
  Speaker 3: for the policy goal of creating a leaves for the
• 1:31:15
  Speaker 3: US versus China, it seems like the export bands have
• 1:31:18
  Speaker 3: actually been quice impactful or China is not producing leading
• 1:31:23
  Speaker 3: edge chips and the export bands really by there China
• 1:31:27
  Speaker 3: is not producing leading edge turbine engines. And similarly, there's
• 1:31:31
  Speaker 3: a bunch of export bands that are relevant there on
• 1:31:33
  Speaker 3: some of the metallurgy. Should there be more export bands,
• 1:31:37
  Speaker 3: like as you think about things like when there are
• 1:31:39
  Speaker 3: now the drone industry and things like that, But is
• 1:31:41
  Speaker 3: that something they should be considered?
• 1:31:43
  Speaker 1: Well, I think it's imporant to appreciate that in most
• 1:31:45
  Speaker 1: areas China is very advanced to actually there's only a
• 1:31:50
  Speaker 1: few areas where it is not. The China is a
• 1:31:56
  Speaker 1: manufacturer of horhouse next level like people don't it must
• 1:31:59
  Speaker 1: be it's very impressive. Yeah, yeah, I mean, if you
• 1:32:02
  Speaker 1: if you take like refining of ore, i'd say roughly,
• 1:32:08
  Speaker 1: China there's more just twice as much or refining of
• 1:32:14
  Speaker 1: on average as the rest of all combined. And I
• 1:32:19
  Speaker 1: think there's there's some areas, like say refining gallion which
• 1:32:21
  Speaker 1: goes into solar cells. I think there are like ninety
• 1:32:25
  Speaker 1: eight percent of gallum refining. So so China is actually
• 1:32:29
  Speaker 1: very advancing manufacturing in i'd say most areas.
• 1:32:32
  Speaker 3: It seems like we're like there is discomfort with this
• 1:32:35
  Speaker 3: supply chain dependence and the ash. Nothing's really happening on it.
• 1:32:39
  Speaker 3: Supply chain, supply chain. It depends on say like the
• 1:32:42
  Speaker 3: Galleon refining that you're saying.
• 1:32:43
  Speaker 1: Yeah, yeah, there's there's there's there's a there's all the
• 1:32:46
  Speaker 1: rare rare earth stuff, and yeah, where else we're sure,
• 1:32:50
  Speaker 1: as you know, not rare like we actually do rare
• 1:32:53
  Speaker 1: earth or mining in the US. Send the the the rock,
• 1:33:00
  Speaker 1: put it on a train, and then from the boats
• 1:33:02
  Speaker 1: China does another train and that goes to the where
• 1:33:06
  Speaker 1: with refining refiners in China who then refine it, put
• 1:33:10
  Speaker 1: it into a magnet, put it in just a modus
• 1:33:11
  Speaker 1: ofb assembly, and then set it back to America. So
• 1:33:14
  Speaker 1: the thing we're really missing a lot of ore refining
• 1:33:18
  Speaker 1: in America. And isn't this worth a policy intervention? Yes, well,
• 1:33:24
  Speaker 1: I think there are some things being done on that front,
• 1:33:28
  Speaker 1: but we kind of need optimists frankly to build our refineries.
• 1:33:35
  Speaker 1: So sorry, you think the main advantage is China has
• 1:33:37
  Speaker 1: is the abundance of skilled labor and that that that's like,
• 1:33:41
  Speaker 1: that's the thing optimist fact says. But they also we
• 1:33:43
  Speaker 1: need four times our populations.
• 1:33:45
  Speaker 2: But we need so I mean, there's this concern if
• 1:33:47
  Speaker 2: you think that humans are the future, that like, okay,
• 1:33:51
  Speaker 2: right now, if it's the skilled labors for manufacturing that's
• 1:33:54
  Speaker 2: determining who's who can build more humanoids. You know, China
• 1:33:57
  Speaker 2: has more of those and manufactures more humanoids, therefore it
• 1:34:00
  Speaker 2: gets it gets the optimized future first.
• 1:34:04
  Speaker 1: Well, it just like keeps that eximential going.
• 1:34:06
  Speaker 2: It seems like you're sort of pointing out that sort
• 1:34:08
  Speaker 2: of getting to a million optimize, Yeah, requires the manufacturing
• 1:34:12
  Speaker 2: that the optimized supposed to help us get to.
• 1:34:14
  Speaker 1: Right, you can you can close that recoursal loop pretty
• 1:34:18
  Speaker 1: quickly with a small number of optimize. Yeah, so you
• 1:34:21
  Speaker 1: close the close recurse of loop to help help the
• 1:34:25
  Speaker 1: robots buill the robots, and then we can, you know,
• 1:34:28
  Speaker 1: try to get to tens of millions of units. You're
• 1:34:30
  Speaker 1: maybe if you start getting too hundreds of millions of
• 1:34:33
  Speaker 1: units a year, I think you're you're going to be
• 1:34:35
  Speaker 1: the most competitive country by far. We definitely count win
• 1:34:38
  Speaker 1: with just humans because China has four times of population, right,
• 1:34:41
  Speaker 1: and frankly, America has been warning for so long that
• 1:34:44
  Speaker 1: you know, just like a like a post sports team
• 1:34:46
  Speaker 1: that's been warning for a very long time, tend to
• 1:34:48
  Speaker 1: get complacent and entitled, and that's why they stop winning
• 1:34:52
  Speaker 1: because it's, you know, don't work as hard anymore. So
• 1:34:56
  Speaker 1: I think the frankly, it just is the average work
• 1:35:00
  Speaker 1: ethic in China is higher than in the US. So
• 1:35:03
  Speaker 1: it's not just that there's four times the population, but
• 1:35:04
  Speaker 1: the work, the amount of work that people put in
• 1:35:06
  Speaker 1: is higher. So you can, like, you can try to
• 1:35:10
  Speaker 1: rearrange the humans, but you're still one quarter of the
• 1:35:14
  Speaker 1: you know, it's assuming that that productivities, that health is
• 1:35:19
  Speaker 1: the same, which I think actually it might not be.
• 1:35:21
  Speaker 1: They China might have advantage on productivity per person. We
• 1:35:25
  Speaker 1: will do one quarter of the amount of things as China,
• 1:35:28
  Speaker 1: so we can't win on the human front. And our
• 1:35:31
  Speaker 1: worth rates spent low for a long time, so our
• 1:35:35
  Speaker 1: birth rates spent the US birth rates spent low replacement
• 1:35:39
  Speaker 1: since roughly nineteen seventy one, so we've got a lot
• 1:35:46
  Speaker 1: of people retiring or you know, more people dying than
• 1:35:50
  Speaker 1: we're close to sort of more people domestically dying than
• 1:35:53
  Speaker 1: being born. So we definitely can't win on the human front,
• 1:35:57
  Speaker 1: but we might have a shot at the row wrote front.
• 1:36:00
  Speaker 1: Are there other things that you have wanted to manufacture
• 1:36:02
  Speaker 1: in the past but they've been too labor intensive or
• 1:36:05
  Speaker 1: too expensive that now you can come back to that
• 1:36:08
  Speaker 1: and say, oh, we can finally do the whatever because
• 1:36:12
  Speaker 1: we have optimists. Yeah, I think we'd like to do
• 1:36:14
  Speaker 1: more both more or refineries at Tesla. So we just
• 1:36:19
  Speaker 1: completed construction and have begun liftium refining without litium refinery
• 1:36:26
  Speaker 1: and Corpus Christi, Texas. We have a nickel refinery which
• 1:36:31
  Speaker 1: is called the Cathode. Uh that's here in Austin. And
• 1:36:37
  Speaker 1: these are these are the largest is the largest cathode.
• 1:36:40
  Speaker 1: There's the largest cathod refinery, largest liftingum refinery, the largest
• 1:36:43
  Speaker 1: nekelink and lifting refinery outside of China. And it's like
• 1:36:50
  Speaker 1: they know, the Cathote team would say, like, we have
• 1:36:54
  Speaker 1: the largest and the only actually cather refinery in America.
• 1:36:58
  Speaker 1: Many superatives not just the just but it's also the
• 1:37:01
  Speaker 1: moly so it was pretty big, even though it's the
• 1:37:04
  Speaker 1: only one. But I mean there are other things that
• 1:37:09
  Speaker 1: you know, you could do a lot more refineries and
• 1:37:15
  Speaker 1: and help the the help America being more competitive on
• 1:37:19
  Speaker 1: refining capacity. So so that there's like there's basically a
• 1:37:23
  Speaker 1: lot of work for the Optimity to do that that
• 1:37:27
  Speaker 1: most Americans, very few Americans, frankly wanted to. I mean,
• 1:37:30
  Speaker 1: i've I've Actually it's the refining work too dirty. It's
• 1:37:34
  Speaker 1: it's not it's actually, no, we don't. There's not we
• 1:37:37
  Speaker 1: don't have toxic commissions from the refinery or anything. The
• 1:37:42
  Speaker 1: Catholic make refinery is right right sort of in Travis County.
• 1:37:45
  Speaker 1: Like why can't I mantes from why can't you with humans? No?
• 1:37:48
  Speaker 1: You can you find out of humans? I see, Okay, yeah,
• 1:37:51
  Speaker 1: Like no matter what you do, you have one quarter
• 1:37:53
  Speaker 1: number of humans in America in China. So if you
• 1:37:55
  Speaker 1: help them do this thing, they can do the other thing.
• 1:37:58
  Speaker 1: So so then then well, how do you how do
• 1:38:00
  Speaker 1: you build this refining refining capacity? Well you can do
• 1:38:03
  Speaker 1: it with the Optimaya and not many, not very many,
• 1:38:08
  Speaker 1: not very many Americans. Thanks for listening. See you in
• 1:38:11
  Speaker 1: the next episode.