1
00:00:00,200 --> 00:00:03,200
Speaker 1: Latest interview of Elon Musk. What's the reason to put

2
00:00:03,200 --> 00:00:09,400
them in face? Well, the availability of energy is the issue.

3
00:00:10,359 --> 00:00:15,000
So I mean, if you look at electrical output outside

4
00:00:15,039 --> 00:00:17,719
of China, everywhere outside of China, it's more or less flat.

5
00:00:17,800 --> 00:00:20,760
It's where you know, maybe slide increase, but for pretty

6
00:00:20,800 --> 00:00:24,519
much flat. China has a rapid increase in electrical output.

7
00:00:24,640 --> 00:00:26,679
But if you're putting data centers of work except China,

8
00:00:26,800 --> 00:00:30,079
where you're gonna get your electricity, especially as you scale,

9
00:00:30,679 --> 00:00:33,880
the output of chips is growing pretty much exponentially, but

10
00:00:34,039 --> 00:00:36,920
the output of electricity is flat. So how are you

11
00:00:36,920 --> 00:00:40,479
going to tell them what chips are? You know, magical

12
00:00:40,520 --> 00:00:43,320
power sources, magical electricity ferries.

13
00:00:43,159 --> 00:00:46,079
Speaker 2: I mean your famous, your famous epic fan of solar

14
00:00:46,679 --> 00:00:48,920
one terra a lot of solar power. So with the

15
00:00:48,960 --> 00:00:51,399
twenty five percent computed factor, like four terrawats of solar panels,

16
00:00:51,439 --> 00:00:52,399
it's like one percent of the.

17
00:00:52,439 --> 00:00:54,560
Speaker 1: Land area of the United States.

18
00:00:54,679 --> 00:00:56,439
Speaker 2: And that's like far in this You're in the singularity

19
00:00:56,439 --> 00:00:58,479
when we've got one tarrat wat of data centers, right,

20
00:00:59,479 --> 00:01:01,039
So what are you running out of?

21
00:01:01,039 --> 00:01:04,359
Speaker 1: Exactly? How far into the singularity are you? You tell me? Yeah,

22
00:01:04,400 --> 00:01:06,760
exactly so I think I think we'll find we're in

23
00:01:06,799 --> 00:01:08,719
the singularity and like, okay, let's look a long way

24
00:01:08,719 --> 00:01:08,959
you go.

25
00:01:10,120 --> 00:01:12,239
Speaker 2: But is this like a is the plan to like

26
00:01:12,319 --> 00:01:14,719
put it in the space after we will covered Nevada

27
00:01:14,760 --> 00:01:15,599
and solar panels.

28
00:01:15,680 --> 00:01:17,719
Speaker 1: I think it's pretty hard to cover about sol panels.

29
00:01:17,719 --> 00:01:19,760
You have to get like permits from Like the proach

30
00:01:19,840 --> 00:01:20,439
for try.

31
00:01:20,319 --> 00:01:23,680
Speaker 2: Getting the pomots for that space is really it's really

32
00:01:23,799 --> 00:01:26,480
regulatory play. It's like hard harder to build on land

33
00:01:26,480 --> 00:01:27,239
than it is in space.

34
00:01:27,640 --> 00:01:31,359
Speaker 1: It's harder to scale on the ground than it is

35
00:01:31,400 --> 00:01:36,480
to scale in space. But also the the you're going

36
00:01:36,519 --> 00:01:41,280
to get about five times the effectiveness of sold panels

37
00:01:41,280 --> 00:01:43,599
in space versus the ground, and you don't need batteries.

38
00:01:45,439 --> 00:01:47,359
I almost wore my other shirt, which says it's always

39
00:01:47,359 --> 00:01:54,840
sunny in space, which it is so because you don't

40
00:01:54,840 --> 00:02:00,519
have a day night cycle or seasonality clouds, or or

41
00:02:00,519 --> 00:02:06,239
an atmosphere in space. Because the atmosphere alone, we're still

42
00:02:06,439 --> 00:02:11,800
about a thirty percent loss of energy. So you're can

43
00:02:12,400 --> 00:02:15,319
any given soul panels can do about five times more

44
00:02:15,599 --> 00:02:18,240
powering space than on the ground, and you avoid the

45
00:02:18,280 --> 00:02:20,159
cost of having batteries to carry you through the night.

46
00:02:22,400 --> 00:02:25,680
So it's actually much cheaper to do in space, and

47
00:02:26,240 --> 00:02:30,800
my prediction is that it will be by far the

48
00:02:30,879 --> 00:02:34,719
cheapest place to put AI will be space and thirty

49
00:02:34,759 --> 00:02:37,240
six months or less, maybe thirty thursday months less than

50
00:02:37,240 --> 00:02:37,840
thirty six months.

51
00:02:38,199 --> 00:02:40,400
Speaker 2: How do you service GPUs as they fail, which happens

52
00:02:40,439 --> 00:02:43,960
quite often in training, Actually it.

53
00:02:45,039 --> 00:02:48,280
Speaker 1: Depends on how recent the GPUs are that arrived. I mean,

54
00:02:48,759 --> 00:02:51,479
at this point we find our gps to be quite reliable.

55
00:02:52,000 --> 00:02:54,400
There's infant mortality which you can obviously iron out on

56
00:02:54,400 --> 00:02:56,680
the ground, so you can just run them on the

57
00:02:56,680 --> 00:02:59,400
ground and confirm that you don't have infort mortality with

58
00:02:59,400 --> 00:03:02,000
what the GPU. But once they once they start working,

59
00:03:02,000 --> 00:03:05,560
their actual reliability and once they start working and you're

60
00:03:05,560 --> 00:03:09,120
past the initial you know, debug cycle of video or whatever,

61
00:03:09,759 --> 00:03:13,199
or whoever's making the trips could be Tesla Tesla AI

62
00:03:14,080 --> 00:03:16,159
six trips or something like that, or could be you know,

63
00:03:17,080 --> 00:03:22,240
TPUs or trainings or whatever the rivalry is. Actually they're

64
00:03:22,280 --> 00:03:28,400
quite reliable past certain point. So I don't think I

65
00:03:28,439 --> 00:03:30,800
don't think you need that. The servicing thing is an issue,

66
00:03:32,080 --> 00:03:36,639
but you can mark my words, uh in thirty six months,

67
00:03:37,120 --> 00:03:41,639
but probably closer to thirty months. The most economically compelling

68
00:03:41,639 --> 00:03:44,520
place to put AI will be space, and then and

69
00:03:44,680 --> 00:03:47,479
and and then it will get from it'll it'll then'

70
00:03:47,520 --> 00:03:51,199
forget like ridiculously better to be in space. And then

71
00:03:51,240 --> 00:03:54,240
the scaling. The only place you can really scale is space.

72
00:03:56,039 --> 00:03:58,400
You know, once you start thinking in terms of what

73
00:03:58,520 --> 00:04:01,520
percentage of the Sun's power are you, honestly you realize

74
00:04:01,560 --> 00:04:05,319
you have to go to space. You can't scale very

75
00:04:05,400 --> 00:04:08,240
very much on Earth, but by very much, to be clear,

76
00:04:08,319 --> 00:04:12,960
you're talking like terro arts. Yeah, well all of the

77
00:04:13,039 --> 00:04:18,600
United States currently uses only half a terrawork power on average, right,

78
00:04:18,680 --> 00:04:20,879
so you know, if you say a terror work, that

79
00:04:20,920 --> 00:04:23,160
would be twice as much electricity as the United States

80
00:04:23,199 --> 00:04:26,480
currently consumes. So that's quite a lot. And can you

81
00:04:26,480 --> 00:04:30,279
imagine building that many data centers, that many power plants.

82
00:04:31,199 --> 00:04:34,800
It's like those who have lived in software land don't

83
00:04:34,959 --> 00:04:39,120
realize that they're about to have a hard lesson in hardware.

84
00:04:40,399 --> 00:04:45,199
That there's there's it's actually very difficult to build power plants.

85
00:04:45,680 --> 00:04:47,279
And and then you don't need just need you need

86
00:04:47,319 --> 00:04:49,759
power plants, you need all of the electrical equipment, need

87
00:04:49,800 --> 00:04:53,839
that the electrical transformers to run the transformers, the transformers.

88
00:04:55,360 --> 00:04:58,079
Now you told the industry is a very slow industry.

89
00:04:58,759 --> 00:05:02,079
That they are they pretty much uh you know, the

90
00:05:02,120 --> 00:05:04,279
impedis smash to the to the government, to the Public

91
00:05:04,360 --> 00:05:09,839
Utility Commission. So they're the impedius smash like literally invigoraily.

92
00:05:10,879 --> 00:05:13,879
So they're very slow because the history of their past

93
00:05:13,920 --> 00:05:16,360
has been very slow. So trying to get them to

94
00:05:16,360 --> 00:05:19,160
move fast is like, you know, like if you're trying

95
00:05:19,160 --> 00:05:20,560
to do an interconnect agreement with you have you ever

96
00:05:20,560 --> 00:05:23,439
tried to do an Internet interconnect agreement with a utility

97
00:05:23,480 --> 00:05:25,240
at scale? Like I put a lot of power.

98
00:05:25,319 --> 00:05:27,319
Speaker 2: As a professional podcaster, I can say it and I'm not.

99
00:05:27,480 --> 00:05:32,160
In fact, the need many more before that comes in

100
00:05:32,199 --> 00:05:32,839
this year, they.

101
00:05:32,800 --> 00:05:35,439
Speaker 1: Have to do a study for a year. Okay, like

102
00:05:35,480 --> 00:05:37,800
a year later they'll come back to you with their

103
00:05:37,800 --> 00:05:38,600
interconnected study.

104
00:05:38,920 --> 00:05:41,279
Speaker 3: Can you tell this with your own behind the meter

105
00:05:41,600 --> 00:05:42,759
power stuff.

106
00:05:43,360 --> 00:05:45,920
Speaker 1: You can build power plants. Yeah, that's what we did. Actually,

107
00:05:45,959 --> 00:05:48,839
I fu clusses to it so for full classes too.

108
00:05:49,040 --> 00:05:50,639
Speaker 3: So yeah, while we're talking about the bridge, why not

109
00:05:50,759 --> 00:05:53,199
just like build GPUs and power co located. That's what

110
00:05:53,279 --> 00:05:54,879
we did, right, right, But I'm saying, why isn't this

111
00:05:55,000 --> 00:05:56,360
generalized solution when you're talking about.

112
00:05:56,160 --> 00:05:57,600
Speaker 1: All the issues where you get the power plants from?

113
00:05:57,720 --> 00:05:59,560
Speaker 3: I'm saying, when you talk about the issues working working

114
00:05:59,560 --> 00:06:01,519
with you two those use you can just build private

115
00:06:01,519 --> 00:06:03,959
power plants with the with the data centers, right, But

116
00:06:04,000 --> 00:06:04,879
it begs the question.

117
00:06:04,720 --> 00:06:06,360
Speaker 1: Of where do you get the power plants? Where do

118
00:06:06,519 --> 00:06:08,199
where do you get the power plants from? I mean

119
00:06:08,240 --> 00:06:09,879
the power plant makers.

120
00:06:10,480 --> 00:06:12,720
Speaker 3: As what you're saying, like, does the gas turbine backlog?

121
00:06:12,800 --> 00:06:17,000
Basically yes, you can drill down to level further. It's

122
00:06:17,040 --> 00:06:20,319
the it's the the veins and blades in the turbines

123
00:06:21,279 --> 00:06:25,120
that of the limiting factor because the casting may it's

124
00:06:25,160 --> 00:06:28,360
it's like a very specialized process to cast the blades

125
00:06:28,399 --> 00:06:31,680
and veins in the in the in the turbines using

126
00:06:31,959 --> 00:06:35,759
gas power, and it's very it's very difficult to scale.

127
00:06:35,839 --> 00:06:39,000
Other other forms of power you can scale potentially solar,

128
00:06:39,399 --> 00:06:42,399
but the towers currently for importing solar in the US

129
00:06:42,439 --> 00:06:46,839
are gigantic and the domestic solar production is protophal. Why

130
00:06:46,920 --> 00:06:49,680
that makes solar? That seems like a good Elon shaped problem.

131
00:06:49,720 --> 00:06:50,680
We are going to make solar?

132
00:06:50,680 --> 00:06:56,519
Speaker 1: Okay. Both SpaceX and Tesla are are bulling towards one

133
00:06:56,600 --> 00:06:59,000
hundredguart tier of solar cell production.

134
00:06:59,360 --> 00:07:01,839
Speaker 2: How low down the like from polysilica and up to

135
00:07:02,600 --> 00:07:05,360
the wayfer to the final panel.

136
00:07:05,560 --> 00:07:06,720
Speaker 1: I think you've got to do the whole thing for

137
00:07:06,839 --> 00:07:10,480
more materials to finish to sell. Now, if it's going

138
00:07:10,519 --> 00:07:13,279
to space, it's actually it costs. It costs less than

139
00:07:13,319 --> 00:07:15,399
it's easier to make solo cells that go to space

140
00:07:15,439 --> 00:07:18,160
because they don't need glass or they don't need much glass,

141
00:07:18,319 --> 00:07:20,480
and they don't need a heavy framing because they don't

142
00:07:20,480 --> 00:07:23,519
have to sell survive weather events. There's no weather in space.

143
00:07:24,600 --> 00:07:26,800
So it's actually a cheaper solo cell that goes to

144
00:07:26,800 --> 00:07:29,279
space than the one on the ground. Is there a

145
00:07:29,279 --> 00:07:32,839
path to getting them as cheap as you need in

146
00:07:32,839 --> 00:07:36,399
the next thirty six months. Solo cells are already very cheap.

147
00:07:38,199 --> 00:07:43,439
They're like partically cheap. It's and if you say, you know,

148
00:07:43,839 --> 00:07:46,360
I think like solo cells in China are around like

149
00:07:46,600 --> 00:07:48,600
twenty five thirty cents a water or something like that.

150
00:07:48,680 --> 00:07:51,759
It's it's absurdly cheap. And when you when you're taking

151
00:07:51,800 --> 00:07:53,560
to a cap now now put it now put it

152
00:07:53,600 --> 00:07:56,399
in space, and it's five times cheaper because it's five

153
00:07:56,439 --> 00:07:58,560
times in fact, no, it's not five times cheap, it's

154
00:07:58,600 --> 00:08:03,680
ten times cheaper because you don't need any batteries. So

155
00:08:03,680 --> 00:08:07,319
so the moment your cost of access to space becomes low,

156
00:08:08,079 --> 00:08:11,079
by far, the cheapest and most scalable way to generate,

157
00:08:11,199 --> 00:08:17,680
to generate tokens is space. It's not even close. It

158
00:08:17,680 --> 00:08:21,399
will be an order of magnitude easier to scale, and

159
00:08:22,000 --> 00:08:24,600
ships aside an order of magnitude. Well, if the point

160
00:08:24,639 --> 00:08:26,120
is you won't be able to scale the ground, it's

161
00:08:26,120 --> 00:08:29,000
just you just won't only hit the big time on

162
00:08:29,079 --> 00:08:34,080
power generation there already are so so like that. The

163
00:08:34,159 --> 00:08:36,879
number of so mercles and series that the x A

164
00:08:36,960 --> 00:08:38,679
I team had to accomplish in order to get a

165
00:08:38,679 --> 00:08:41,720
gig out of power online was was was crazy. We

166
00:08:42,000 --> 00:08:46,000
had to getting together a whole bunch of turbines and

167
00:08:46,120 --> 00:08:50,600
uh and then and then we had permit issues in

168
00:08:50,679 --> 00:08:53,519
Tennessee and and had to go across the border to Mississippi,

169
00:08:53,679 --> 00:08:56,000
which is fortunately only you know, a few miles away.

170
00:08:56,440 --> 00:08:58,240
Uh So, but then we still had to run the

171
00:08:58,279 --> 00:09:01,399
high power lines a few miles and build a power

172
00:09:01,399 --> 00:09:04,240
plant in Missipi. And it was very difficult. Abu that

173
00:09:05,639 --> 00:09:08,879
and people don't understand, like how much how much electrocity

174
00:09:08,879 --> 00:09:11,000
do you actually need at the generator level, at the

175
00:09:11,039 --> 00:09:13,840
generation level in order to power a data center, because

176
00:09:13,879 --> 00:09:16,799
they look at the the nerves will look at the

177
00:09:17,480 --> 00:09:21,360
power consumption of say a GB three hundreds and multiply

178
00:09:21,440 --> 00:09:23,399
that by thing and then think best amount amount of

179
00:09:23,440 --> 00:09:27,399
power you need? Wake up. Yeah, this is like, that's

180
00:09:27,440 --> 00:09:30,840
a that's a that's a total moves You've never done

181
00:09:30,840 --> 00:09:34,960
any hardware in your life before. Besides the GB three hundred,

182
00:09:35,080 --> 00:09:37,639
you've got to power all of the networking hardware. There's

183
00:09:37,639 --> 00:09:40,080
a whole bunch of CPU and storage stuff that's happening.

184
00:09:40,919 --> 00:09:45,960
You've got a size for your your peak cooling requirements,

185
00:09:46,120 --> 00:09:49,639
so that means can you cool even on the worst

186
00:09:49,679 --> 00:09:51,799
hour of the worst day of the year. Well, it's

187
00:09:51,840 --> 00:09:55,000
pretty freaking hot in Memphis, So so you're going to

188
00:09:55,039 --> 00:09:58,559
have like a forty percent increase on your power just

189
00:09:58,600 --> 00:10:02,000
for cooling. If we assuming you don't want your data

190
00:10:02,000 --> 00:10:04,759
center to turn off on hot days and anyone to

191
00:10:04,799 --> 00:10:08,080
keep going, then then you've got to say, well, there's

192
00:10:08,080 --> 00:10:10,679
a There's there's another multiplicative element on top of that,

193
00:10:10,720 --> 00:10:13,399
which is are you assuming that you're you you never

194
00:10:13,480 --> 00:10:17,080
have any hiccups in your power generation like well, actually

195
00:10:17,360 --> 00:10:19,440
sometimes you have to take the generators some of the

196
00:10:19,559 --> 00:10:21,879
power offline in order to service it. Oh okay, now

197
00:10:21,919 --> 00:10:24,360
you add another twenty twenty five percent multiplier on that

198
00:10:24,879 --> 00:10:27,159
because you've got it. You've got to assume that that

199
00:10:27,200 --> 00:10:30,360
you've got to take power offline to service it. So

200
00:10:30,480 --> 00:10:36,279
the actual rs roughly every every one hundred and ten

201
00:10:36,320 --> 00:10:42,480
thousand gbs, GB three hundreds, inclusive of networking, CPU, storage, cooling.

202
00:10:43,519 --> 00:10:50,279
Margin for for for servicing power is roughly three hundred megawatts.

203
00:10:50,679 --> 00:10:54,639
Sorry you say that again, it's roughly, or think about it, like,

204
00:10:55,720 --> 00:10:57,159
what do you think about like three hundred and thirty

205
00:10:57,159 --> 00:11:02,320
thousand to actually what you need at the generative generation level.

206
00:11:02,639 --> 00:11:06,159
So service probably service three hundred and thirty thousand TV

207
00:11:06,279 --> 00:11:09,919
three hundreds including all of the associated support networking and

208
00:11:09,960 --> 00:11:13,320
everything else and the and the peak cooling and to

209
00:11:13,399 --> 00:11:16,399
have some margin some power marginers of is roughly a

210
00:11:16,440 --> 00:11:18,840
GEO why can ask a very night your question?

211
00:11:18,960 --> 00:11:24,159
Speaker 2: Yeah, you know, you're describing the engineering details of doing

212
00:11:24,200 --> 00:11:29,080
this stuff on Earth, but then there's analogous engineering difficulties

213
00:11:29,120 --> 00:11:29,840
of doing it in space.

214
00:11:29,840 --> 00:11:30,440
Speaker 1: How do you do the.

215
00:11:31,879 --> 00:11:34,000
Speaker 2: How do you replace in pin and band with orbital lasers,

216
00:11:34,240 --> 00:11:36,200
et cetera, etceter How do you make it resistant radiation?

217
00:11:38,080 --> 00:11:40,559
I don't know the details in the engineering, but fundamentally,

218
00:11:40,600 --> 00:11:43,440
what is the reason to think those challenges which have

219
00:11:43,559 --> 00:11:46,519
never been had to be addressed before will end up

220
00:11:46,519 --> 00:11:49,240
being easier than just like building more turbines on Earth?

221
00:11:49,320 --> 00:11:51,120
There's companies that build turbines on Earth. They can make

222
00:11:51,120 --> 00:11:52,240
more turbines right.

223
00:11:52,639 --> 00:11:58,360
Speaker 1: And again, tried doing it and then you'll see so

224
00:12:01,159 --> 00:12:03,879
like the turbines are sold out through twenty thirty, have

225
00:12:03,960 --> 00:12:07,000
you guys considered making your own? I think in order

226
00:12:07,080 --> 00:12:11,879
for in order to bring enough power online, I think

227
00:12:12,759 --> 00:12:17,080
SpaceX Intezl will probably have to make the turbine blades,

228
00:12:18,279 --> 00:12:22,600
the bans and blades internally, but just the blades or

229
00:12:22,600 --> 00:12:27,320
the turbines the limiting factor. You can get everything except

230
00:12:28,159 --> 00:12:32,600
the blades they call the blades and veins you can

231
00:12:32,639 --> 00:12:36,559
get that twelve to eighteen months before the veins of blades.

232
00:12:36,600 --> 00:12:39,120
The limiting factor of the veins of blades. And there

233
00:12:39,120 --> 00:12:44,320
are only three casting companies in the world that may

234
00:12:44,559 --> 00:12:48,080
make these, and they're massively backloged. Is this siemens ge

235
00:12:48,279 --> 00:12:50,840
those guys, or is it a subje. No, it's a

236
00:12:51,120 --> 00:12:53,200
it's other companies. I mean sometimes they have a little

237
00:12:53,200 --> 00:12:56,360
bit of casting capability in house. But I'm just saying you.

238
00:12:56,360 --> 00:12:58,759
You can just call any of the turbine makers and

239
00:12:58,799 --> 00:13:01,639
they will tell you top secret. They're probably on the

240
00:13:01,679 --> 00:13:03,519
it's probably on the internet right now. If if it

241
00:13:03,559 --> 00:13:07,480
wasn't for the terraffs, what would Colossus be solar powered? Uh,

242
00:13:07,799 --> 00:13:10,639
it would be much easier to make it solar powered. Yeah,

243
00:13:11,000 --> 00:13:14,440
the tariffs are not so several hundred percent, So don't

244
00:13:14,440 --> 00:13:17,120
you know some big book we will will see it speed. Yeah. No,

245
00:13:18,200 --> 00:13:22,519
you know, President has a you know, we don't agree

246
00:13:22,559 --> 00:13:27,000
on everything, and this administration is not not the biggest

247
00:13:27,039 --> 00:13:33,840
man of of solar. So it's you know, it's it's

248
00:13:33,879 --> 00:13:36,679
it's it's We also need the land, the permits and everything.

249
00:13:36,679 --> 00:13:40,960
So if you try to move very fast like I

250
00:13:41,000 --> 00:13:44,279
do think scaling solar on Earth it is a is

251
00:13:44,279 --> 00:13:46,440
a good way to go, but but you need you

252
00:13:46,480 --> 00:13:48,399
do need some amount of time to find the land,

253
00:13:48,440 --> 00:13:51,639
get the permits, get the solar pair that with the batteries.

254
00:13:51,759 --> 00:13:55,000
Speaker 3: Well why would it not work to stand up your

255
00:13:55,039 --> 00:13:57,559
own solar production? And then you're right that you eventually

256
00:13:57,639 --> 00:13:59,720
run out of land. But there's a lot of land

257
00:13:59,720 --> 00:14:01,559
here in Texas. There's a lot of land in Nevada,

258
00:14:01,559 --> 00:14:04,039
including private land. It's not all publicly on land, and

259
00:14:04,120 --> 00:14:06,159
so you'd be able to at least get the next

260
00:14:06,639 --> 00:14:08,720
colossus and like the next one after that, and at

261
00:14:08,720 --> 00:14:09,840
a certain point you hit a wall.

262
00:14:09,879 --> 00:14:12,320
Speaker 1: But wouldn't that work for the moment? As I said,

263
00:14:12,360 --> 00:14:17,159
we are scaling solar production, there's there's a rate. There's

264
00:14:17,159 --> 00:14:19,639
a rate at which you can scale physical production of

265
00:14:19,679 --> 00:14:23,600
solo solar cells. Where I'm we're going as fast as

266
00:14:23,600 --> 00:14:27,480
possible in scaling domestic production. You're making the solar cells

267
00:14:27,480 --> 00:14:31,320
at TEA. We TESA and SpaceX have a mandate to

268
00:14:31,399 --> 00:14:33,639
get to one hundred gigs a year of solar.

269
00:14:33,840 --> 00:14:37,200
Speaker 3: Speaking of the annual capacity, I'm furious. In five years time,

270
00:14:37,320 --> 00:14:42,000
let's say, what will the installed capacity be on Earth

271
00:14:42,679 --> 00:14:46,159
and in space. I deliberately picked five years because it's

272
00:14:46,240 --> 00:14:48,960
after you're once we're up and running threshold and so

273
00:14:49,000 --> 00:14:51,480
in five years time, Yeah, what's thee on Earth versus

274
00:14:51,480 --> 00:14:53,399
in space installed a capacity?

275
00:14:54,320 --> 00:14:58,840
Speaker 1: Five years? I think probably you said five years from

276
00:14:58,879 --> 00:15:07,440
now were probably AI in space will be uh launching

277
00:15:07,519 --> 00:15:10,799
every year the sum total of all AI on Earth

278
00:15:11,600 --> 00:15:14,919
in excess of me. Five years from now, my prediction

279
00:15:15,039 --> 00:15:20,279
is we will launch and and and be operating every

280
00:15:20,360 --> 00:15:23,879
year more AI in space than this than the cumulative

281
00:15:23,919 --> 00:15:27,120
total on Earth, which is I would expect to be

282
00:15:27,200 --> 00:15:30,120
at least sort of five years from now, a few

283
00:15:30,200 --> 00:15:36,519
hundred gigawatts per year of of AI in space and rising,

284
00:15:38,559 --> 00:15:42,360
so you can get to I think on Earth you

285
00:15:42,360 --> 00:15:45,000
can get to a round a tarawa a year of

286
00:15:45,759 --> 00:15:50,679
AI in space before you start having you know, fuel

287
00:15:50,679 --> 00:15:52,399
supplied challenges for the rocket.

288
00:15:52,600 --> 00:15:54,240
Speaker 3: Okay, but you think you can get to one hundreds

289
00:15:54,240 --> 00:15:57,120
of gigabatts per year in five years, yes.

290
00:15:57,200 --> 00:16:02,519
Speaker 2: So one hundred gigawatts depending on the specific power of

291
00:16:02,559 --> 00:16:05,200
the whole system with solar arrays and radios and everything,

292
00:16:05,639 --> 00:16:10,360
is on the order of like ten thousand starship launches, yes,

293
00:16:11,600 --> 00:16:13,919
and you want to do that in one year, and

294
00:16:13,960 --> 00:16:17,440
so that's like one starship launch every hour. Yeah, that's

295
00:16:17,480 --> 00:16:21,240
happening in the city, Like walking through a world where

296
00:16:21,240 --> 00:16:24,000
there's ten times there's a starship launch every single hour.

297
00:16:24,440 --> 00:16:27,080
Speaker 1: Yeah, I mean that's actually a lower rate compared to airlines,

298
00:16:27,600 --> 00:16:30,240
like like aircraft aircraft, there's a lot of airports, a

299
00:16:30,279 --> 00:16:32,320
lot of airports, and you gotta launch you know, the

300
00:16:32,360 --> 00:16:35,440
polar orbit, and it doesn't have to be polarp. But

301
00:16:36,639 --> 00:16:40,759
you're just there's there's some some value to sound sagres.

302
00:16:40,759 --> 00:16:46,200
But but I think actually you just go high enough,

303
00:16:46,200 --> 00:16:48,360
you're you're start getting out of it. Both shadow and

304
00:16:48,679 --> 00:16:50,559
so how.

305
00:16:50,399 --> 00:16:53,639
Speaker 2: Many physical starships are needed to do ten thousand lunches

306
00:16:53,639 --> 00:16:53,960
a year?

307
00:16:54,320 --> 00:16:56,679
Speaker 1: I don't think we'll lead more than I mean you could.

308
00:16:56,720 --> 00:17:01,159
You could probably do it with as few as like

309
00:17:01,279 --> 00:17:07,519
twenty or thirty, Like it really depends on how quickly

310
00:17:07,559 --> 00:17:09,880
does the ship. The ship has to go around the Earth,

311
00:17:11,000 --> 00:17:13,480
and the ground track for the ship has to come

312
00:17:13,480 --> 00:17:15,799
back over the launch pad. So if you can use

313
00:17:15,880 --> 00:17:19,359
a ship every say, thirty hours, you could do it

314
00:17:19,400 --> 00:17:22,880
with thirty ships. But we'll make more shifts than that.

315
00:17:22,960 --> 00:17:29,039
But the SpaceX is is going up to do ten

316
00:17:29,079 --> 00:17:32,519
thousand marches a year and maybe even twenty or thirty

317
00:17:32,559 --> 00:17:33,319
thousand marches. Ye know.

318
00:17:33,599 --> 00:17:36,599
Speaker 2: It is the idea to become basically a hyperscaler, become

319
00:17:36,599 --> 00:17:39,400
an oracle and lend this capacity to other people. What's

320
00:17:39,480 --> 00:17:41,680
what are you going to do with presumableast? SpaceX is

321
00:17:41,720 --> 00:17:46,160
the one watching all this, so SpaceX become a hyperscaler.

322
00:17:47,119 --> 00:17:50,079
Speaker 1: Hyper hyper Yeah. I mean, if so many of my

323
00:17:50,200 --> 00:17:55,279
predictions come true, SpaceX will launch more AI than the accumulative

324
00:17:55,279 --> 00:17:58,480
amount on Earth combined of everything else combined. Is this

325
00:17:58,559 --> 00:18:01,400
mostly inference or most say I won't be in for it,

326
00:18:01,480 --> 00:18:04,119
Like already inference for the purpose of training is most train.

327
00:18:04,519 --> 00:18:10,200
Speaker 3: And there's a narrative that the the change in discussion

328
00:18:10,240 --> 00:18:14,720
around the SpaceX IPO is because previously SpaceX was very

329
00:18:14,720 --> 00:18:18,880
capital efficient, just wasn't just that expensive to develop that

330
00:18:18,960 --> 00:18:21,880
even though it sounds expensive, it's actually very capital efficient

331
00:18:21,880 --> 00:18:25,279
and how it runs, whereas now you're going to need

332
00:18:26,000 --> 00:18:29,119
more capital than just can be raised in the private markets.

333
00:18:29,160 --> 00:18:31,559
Like if the private markets can accommodate races of as

334
00:18:31,559 --> 00:18:33,880
we've seen from the AI labs, tens of billions of dollars,

335
00:18:33,880 --> 00:18:37,079
but not beyond that, is it that you'll just need

336
00:18:37,119 --> 00:18:38,839
more than tens of billions of dollars per year?

337
00:18:38,880 --> 00:18:43,359
Speaker 1: And that's about say the public. Yeah, I'd be careful

338
00:18:43,359 --> 00:18:47,279
about saying things about companies that might go public. You know,

339
00:18:47,559 --> 00:18:50,880
if you make general that's never been a problem for you, Elan.

340
00:18:52,680 --> 00:18:54,319
You know, there's a price to pay for these things

341
00:18:55,119 --> 00:18:55,440
make some.

342
00:18:55,480 --> 00:18:58,279
Speaker 3: General statements for us about the depth of the capital

343
00:18:58,319 --> 00:18:59,920
markets between public and private markets.

344
00:19:00,799 --> 00:19:04,359
Speaker 1: Yeah, there's there's a lot more capital in the very

345
00:19:04,440 --> 00:19:07,759
general there's there's obviously a lot more capital available in

346
00:19:07,799 --> 00:19:11,680
the public markets than private. It might be it's at

347
00:19:11,759 --> 00:19:13,839
least at least it might be a hard times more capital,

348
00:19:13,880 --> 00:19:15,839
but at least way more than ten.

349
00:19:16,119 --> 00:19:18,160
Speaker 3: But isn't it also the case that things that tend

350
00:19:18,240 --> 00:19:22,920
to be very capital intensive. If you look at say

351
00:19:22,920 --> 00:19:27,279
real estate, as you know, a huge industry that raises

352
00:19:27,279 --> 00:19:28,799
a lot of money each year is at an industry

353
00:19:28,880 --> 00:19:32,400
level that tends to be debt financed because by the

354
00:19:32,440 --> 00:19:35,960
time you're deploying that much money, you actually have a

355
00:19:36,000 --> 00:19:40,000
pretty exactly and a near term return. And you see

356
00:19:40,000 --> 00:19:41,640
this even with the data center build auds, which are

357
00:19:41,720 --> 00:19:45,720
famously being you know, financed by the private credit industry.

358
00:19:46,160 --> 00:19:54,000
Speaker 1: And so why not just debt finance? Speed is important.

359
00:19:54,839 --> 00:20:01,559
So I'm generally going to do the thing that I mean,

360
00:20:01,599 --> 00:20:03,920
I just repeatedly tackled limiting factor. Whatever the limiting factors

361
00:20:04,000 --> 00:20:10,599
on speed, I'm I'm gonna tackle that. So there's if

362
00:20:10,640 --> 00:20:13,119
capital is learning factor, then I'll alsold for capital. If

363
00:20:13,240 --> 00:20:16,119
if it's not limiting factor also for something else. Based

364
00:20:16,119 --> 00:20:21,960
only in your statements about Tesla and being public, I

365
00:20:21,960 --> 00:20:23,799
wouldn't have guessed that you thought the fast, the way

366
00:20:23,839 --> 00:20:28,160
to move fast is to be public. Normally I would say, yeah,

367
00:20:28,200 --> 00:20:31,200
that's that's true. Like I said, I mean, I'd like

368
00:20:31,319 --> 00:20:33,480
to talk about some more detail. But the problem is,

369
00:20:33,519 --> 00:20:36,200
like if you talk about public companies where they become public,

370
00:20:36,319 --> 00:20:38,119
you're get into trouble and then you have to lay

371
00:20:38,160 --> 00:20:44,839
your offering and then you yes, exactly so so so

372
00:20:45,279 --> 00:20:49,279
that you can't HiPE companies that are that that may

373
00:20:49,359 --> 00:20:52,319
that might go public. So that that's that's why we

374
00:20:52,440 --> 00:20:55,599
have to be a little careful here. But but but

375
00:20:56,279 --> 00:21:00,559
we can't talk about physics. So like the way the

376
00:21:00,599 --> 00:21:04,799
way you think about scaling long term is that Earth

377
00:21:04,839 --> 00:21:07,720
only receives about half a billionth of the Sun's energy,

378
00:21:08,880 --> 00:21:11,279
and the Sun is the sign is essentially all the energy.

379
00:21:11,359 --> 00:21:13,720
This is a very important point to appreciate because sometimes

380
00:21:13,759 --> 00:21:18,680
people will talk about marginally nuper reactors or any various

381
00:21:18,680 --> 00:21:22,759
like fusion on Earth. But you have to step back

382
00:21:22,880 --> 00:21:25,160
a second and say if if, if you're going to

383
00:21:25,200 --> 00:21:29,440
climb the Kardashev scale and have some non trivial and

384
00:21:29,759 --> 00:21:33,440
harness some non trivial percentage of the sign's energy, Like

385
00:21:33,519 --> 00:21:37,759
unless you wanted to harness a millionth of the Sun's energy,

386
00:21:37,839 --> 00:21:43,880
which sounds pretty small, that that would be about call

387
00:21:43,920 --> 00:21:47,160
it roughly one hundred thousand times more electricity than we

388
00:21:47,319 --> 00:21:52,960
currently generate on Earth for all of civilization give or

389
00:21:53,000 --> 00:21:57,279
take an order it back too. So it obviously the

390
00:21:57,319 --> 00:21:59,839
only way to scale is to go to space with

391
00:22:00,039 --> 00:22:03,599
solar From launching from Earth, you can get to about

392
00:22:03,599 --> 00:22:06,880
a tarrawat per year. Beyond that you want to go

393
00:22:06,920 --> 00:22:09,279
to you want to launch from the Moon. You want

394
00:22:09,279 --> 00:22:12,119
to have a mass driver on the moon, and that

395
00:22:12,720 --> 00:22:16,480
mass drive on the moon you could do probably a

396
00:22:16,480 --> 00:22:17,480
pedal wak per year.

397
00:22:18,119 --> 00:22:20,680
Speaker 2: We're talking these kinds of numbers, you know, terrawatts of

398
00:22:20,680 --> 00:22:26,119
compute presumably whether you're talking land or space. Far far

399
00:22:26,160 --> 00:22:30,799
before this point you've like run into you know, you

400
00:22:30,839 --> 00:22:33,200
actually need maybe if you don't, the solar panels are

401
00:22:33,200 --> 00:22:34,799
more efficient, but you still need the chips.

402
00:22:35,759 --> 00:22:37,839
Speaker 1: You still need the logic and the memory and so forth,

403
00:22:38,160 --> 00:22:40,559
and you need a lot more chips and make them

404
00:22:40,640 --> 00:22:41,279
munch cheaper.

405
00:22:41,519 --> 00:22:43,799
Speaker 2: Right, And so how are we getting a tararo out

406
00:22:43,839 --> 00:22:45,920
of like, right now the world is gonna be twenty

407
00:22:45,960 --> 00:22:47,400
twenty five pi wats of compute.

408
00:22:48,319 --> 00:22:51,839
Speaker 1: How we're getting a tarotot of logic by I guess

409
00:22:51,839 --> 00:22:54,559
we're going to need some very big chip maps to

410
00:22:54,640 --> 00:22:57,559
tell it. Tell me about it. I've mentioned that publicly.

411
00:22:57,599 --> 00:23:01,359
That's the idea of doing it. Sort of a terror

412
00:23:01,400 --> 00:23:05,240
pap Terra being the new Giga we were, I feel

413
00:23:05,240 --> 00:23:08,680
like the naming scheme of Tesla, which has been very catchy,

414
00:23:08,720 --> 00:23:11,359
It's like, are you looking at like the metric the

415
00:23:11,400 --> 00:23:14,559
metric scale at what level of the stack are you?

416
00:23:15,160 --> 00:23:17,440
Are you building the clean room and then partnering with

417
00:23:18,039 --> 00:23:21,400
and is they sting fab to get the process tchnology

418
00:23:21,400 --> 00:23:25,039
and buying the tools from them? What is the plan there? Well,

419
00:23:25,079 --> 00:23:27,839
you can't partner with existing paths because they're just they

420
00:23:27,880 --> 00:23:31,599
can't output enough. The trip alb is too low. But yeah,

421
00:23:31,720 --> 00:23:33,720
you have to look for the process achnology. Yeah, partner

422
00:23:33,759 --> 00:23:38,480
for the you know the fabs that they will basically

423
00:23:38,640 --> 00:23:44,799
use machines from like five companies. You know, so I've

424
00:23:44,839 --> 00:23:49,079
got SML, Touch, Electron, Cali A, tank Core, you know,

425
00:23:51,119 --> 00:23:56,680
et cetera. So so so at first I think you'd

426
00:23:56,720 --> 00:24:01,319
have to get equipment from them and then modify it

427
00:24:01,440 --> 00:24:04,279
or work with them to increase the volume. But I

428
00:24:04,279 --> 00:24:06,160
think you have to build, perhaps in a different way.

429
00:24:06,480 --> 00:24:08,200
So I think that the logical thing to do is

430
00:24:08,240 --> 00:24:12,599
to to use conventional equipment in an un conventional way

431
00:24:12,960 --> 00:24:16,680
to get to scale and then and then and then

432
00:24:16,680 --> 00:24:20,559
stop modifying the equipment to increase the rate kind of

433
00:24:20,599 --> 00:24:24,000
boring company style. Yeah, kind of like, yeah, you what

434
00:24:24,160 --> 00:24:27,319
your sort of buying and it's just a boring machine.

435
00:24:27,839 --> 00:24:31,359
And then figure out how to dig titles in the

436
00:24:31,400 --> 00:24:35,759
first place, and then design a much better machine that's

437
00:24:36,079 --> 00:24:39,039
you know, I don't know, some orders magitude faster.

438
00:24:40,079 --> 00:24:43,359
Speaker 3: Here's a very simple lens. We can categorize technologies and

439
00:24:43,400 --> 00:24:45,920
how hard they are. And one categorization could be look

440
00:24:45,960 --> 00:24:49,119
at things that China has not succeeded in doing. And

441
00:24:49,160 --> 00:24:53,799
if you look at Chinese manufacturing still behind on leading

442
00:24:53,920 --> 00:24:59,920
edge chips and still behind on leading edge turbine engines

443
00:25:00,079 --> 00:25:03,319
and things like that, and so does the fact that

444
00:25:04,119 --> 00:25:07,279
China has not successfully replicated t s MC give you

445
00:25:07,319 --> 00:25:11,000
any pause about the difficulty or you think, well, that's

446
00:25:11,039 --> 00:25:11,319
not true.

447
00:25:11,319 --> 00:25:14,200
Speaker 1: For some it's not that they have not replated TSMC.

448
00:25:14,319 --> 00:25:18,680
They have replicated ASML. That's the limited factor. So you

449
00:25:18,680 --> 00:25:23,160
think it's just the the sanctions essentially, Yeah, trying to

450
00:25:23,319 --> 00:25:25,359
re utputting the vast numbers of trips.

451
00:25:25,559 --> 00:25:29,240
Speaker 3: If they could bias, couldn't they up to relatively recently

452
00:25:29,279 --> 00:25:30,119
buy them?

453
00:25:30,519 --> 00:25:34,279
Speaker 1: No that the bands are been a place for a while,

454
00:25:34,920 --> 00:25:36,839
but I think time is going to be make me

455
00:25:36,880 --> 00:25:39,480
pretty compelling trips in twenty four years. Would you consider

456
00:25:39,519 --> 00:25:42,720
making the ASML machines? I don't know. I don't know yet.

457
00:25:42,720 --> 00:25:49,920
It's the right answer. So it's just that that to

458
00:25:49,960 --> 00:25:54,559
produce at high volume and to reach large volume in

459
00:25:54,599 --> 00:26:00,000
say thirty six months to match the rocket Paloa to orbits.

460
00:26:00,039 --> 00:26:04,839
So we're doing a million tons to orbit and like

461
00:26:04,920 --> 00:26:08,400
let's say, I don't know, three or four years from now,

462
00:26:08,440 --> 00:26:13,599
something like that. That and uh, and we're doing one hundred

463
00:26:13,680 --> 00:26:17,119
killer watts per ton, So that that means we need

464
00:26:17,480 --> 00:26:22,079
at least one hundred gigawatts per year of solo and

465
00:26:22,119 --> 00:26:27,680
we'll need an equivalent amount of chips. You know that

466
00:26:28,000 --> 00:26:29,880
you need a hundred gigawa's worth the chips you've got.

467
00:26:29,880 --> 00:26:32,519
You're going to match these things, the Master orbit, the

468
00:26:32,839 --> 00:26:36,160
power generation and the and the and the chips. Uh

469
00:26:36,240 --> 00:26:39,599
and and I'd say my biggest concern actually is is memory.

470
00:26:40,279 --> 00:26:45,640
So the I think there's there's a the path to

471
00:26:45,799 --> 00:26:48,799
creating the logic chips is more obvious than the path

472
00:26:48,880 --> 00:26:54,000
to having sufficient memory to support the logic chips. That's

473
00:26:54,000 --> 00:26:56,920
why you see dr Prices going ballistic and these memes

474
00:26:56,960 --> 00:27:01,079
about like you know, you're ruined on a desert islandy right,

475
00:27:01,160 --> 00:27:03,079
help me on the sand that when he comes to

476
00:27:03,119 --> 00:27:06,240
you write d d am the ships come swarming in.

477
00:27:08,920 --> 00:27:16,079
Speaker 2: I'd love your manufacturing philosophy around around fabs. You know,

478
00:27:16,519 --> 00:27:18,079
I know nothing abut the talpout. I don't know how

479
00:27:18,079 --> 00:27:18,559
to build up.

480
00:27:18,599 --> 00:27:20,200
Speaker 1: Yet, don't figure it out.

481
00:27:20,599 --> 00:27:23,880
Speaker 2: But obviously it sounds like you think that the sort

482
00:27:23,880 --> 00:27:27,119
of like the proseechnology, like these ten thousand pH ds

483
00:27:27,240 --> 00:27:30,319
in Taiwan who know exactly what gas goes in the

484
00:27:30,359 --> 00:27:32,319
plasma chamber and what's settings to put on a tool,

485
00:27:32,680 --> 00:27:35,240
you can just like delete those parts of those steps.

486
00:27:35,519 --> 00:27:38,000
Like fundamentally, it's get the clean room, get the tools,

487
00:27:38,119 --> 00:27:38,920
and figure it out.

488
00:27:39,480 --> 00:27:42,359
Speaker 1: I don't think it's PhD s. It's it's mostly people

489
00:27:42,400 --> 00:27:48,119
with you know, you're not not PhDs. That's mostly engineering

490
00:27:48,119 --> 00:27:49,680
is some of people who don't have PhDs. Do you

491
00:27:49,720 --> 00:27:51,119
guys have PhDs? No?

492
00:27:51,160 --> 00:27:55,680
Speaker 3: Okay, we also haven't successfully booked any fabs, so you

493
00:27:55,680 --> 00:27:58,240
shouldn't be coming to us for your fab advice or.

494
00:27:58,359 --> 00:28:00,119
Speaker 1: I don't think any PhD for that for stuff. So

495
00:28:01,160 --> 00:28:03,400
but but you do need you do need competent personnel.

496
00:28:04,759 --> 00:28:08,720
So I don't I mean, like right right now if yeah,

497
00:28:09,200 --> 00:28:13,000
you know, like Tesla's pedals to the metal max production of

498
00:28:13,839 --> 00:28:16,680
going as fast as possible to get a five tell

499
00:28:16,799 --> 00:28:21,720
a five trip design introduction and then reaching scale, you know,

500
00:28:21,759 --> 00:28:26,559
that will probably happen, you know, around the second quarter

501
00:28:26,680 --> 00:28:32,759
issue of next year hopefully, uh, and then a I

502
00:28:32,880 --> 00:28:36,920
six would hopefully follow less than a year later. But

503
00:28:38,279 --> 00:28:41,599
and and and and we've secured all the all the

504
00:28:41,680 --> 00:28:46,000
trip fab production that we can. You're currently limited on

505
00:28:46,160 --> 00:28:49,759
T s m C fab capacity. Yeah, and and and

506
00:28:49,799 --> 00:28:54,640
we'll be using t s m C UH Taiwan, Samsung Coreer,

507
00:28:54,880 --> 00:28:59,599
t s MC Arizona, Samsung Texas. And we still get

508
00:28:59,640 --> 00:29:01,880
booked all the yeah fast you can, yes, and and

509
00:29:01,960 --> 00:29:04,759
then and then and then if I ask to some

510
00:29:04,839 --> 00:29:08,039
CEO Samsung, Okay, what what's the time frame to get

511
00:29:08,079 --> 00:29:10,759
to volume production? This point is it's not. You've got

512
00:29:10,759 --> 00:29:12,640
to you've got to build the fab and you've got

513
00:29:12,720 --> 00:29:15,119
to you've got you've got to start production. Then you've

514
00:29:15,119 --> 00:29:17,400
got to climb the yield curve and reach volume production

515
00:29:17,440 --> 00:29:20,119
at high yield. That that that from start finishes a

516
00:29:20,119 --> 00:29:24,920
five year period. And so the limiting factor is chips.

517
00:29:25,160 --> 00:29:27,240
What what like the limiting factor once you can get

518
00:29:27,279 --> 00:29:29,519
to space is chips. But the living limiting factor before

519
00:29:29,519 --> 00:29:31,839
you can get to space to be power, why.

520
00:29:31,720 --> 00:29:33,720
Speaker 2: Don't you do the Gensen thing and just pre Pats

521
00:29:33,759 --> 00:29:34,920
and c to build more fabs for you.

522
00:29:35,559 --> 00:29:39,559
Speaker 1: Uh, I've already told them that, but they won't take

523
00:29:39,559 --> 00:29:43,319
your money. Like, what's going on? They're building fabs as fast. No,

524
00:29:45,119 --> 00:29:47,240
they're building they're building fabs as fast as they can,

525
00:29:48,920 --> 00:29:52,160
and so is Samsung. Like the like they're they're they're

526
00:29:52,240 --> 00:29:54,039
pedal to the metal. I mean they're going you know,

527
00:29:54,160 --> 00:29:58,319
bowls the wall, you know, as fast as they can.

528
00:29:59,039 --> 00:30:02,359
So still a fat I mean, like I said that,

529
00:30:02,359 --> 00:30:06,599
there will be I think if you say, I think

530
00:30:06,640 --> 00:30:08,640
towards the end of this year, I think probably chip

531
00:30:08,680 --> 00:30:13,119
production will outpace the ability to turn chips on. But

532
00:30:13,160 --> 00:30:17,960
once you can get to space and unlock the power constraint,

533
00:30:18,559 --> 00:30:20,240
and you can now do you know, one hundreds of

534
00:30:20,240 --> 00:30:24,559
gigawatts per year power in space? Again, bearying in mind

535
00:30:24,599 --> 00:30:27,200
that average power usage in the US is you know,

536
00:30:27,400 --> 00:30:30,720
five hundred gigawats. So if you're launching, say two hundred

537
00:30:30,720 --> 00:30:33,480
gigawats a year to space, you're sort of lapping the

538
00:30:33,599 --> 00:30:36,519
US every two and a half years the entire all

539
00:30:36,680 --> 00:30:41,039
US electricity production. This is a very huge amount. So

540
00:30:42,720 --> 00:30:48,119
but between now and then, the actually the constraint for

541
00:30:48,920 --> 00:30:54,920
server side compute, concentrated compute will be will be electricity.

542
00:30:54,720 --> 00:30:58,559
My guess is that we start hitting people start getting

543
00:30:58,559 --> 00:31:02,279
pot where they can't turn the chips on. For for

544
00:31:02,279 --> 00:31:05,519
for large clusters, uh, towards the end of this year,

545
00:31:05,559 --> 00:31:07,160
they're just the chips are going to be piling up

546
00:31:07,200 --> 00:31:09,720
and not be weren't be able to be turned on.

547
00:31:10,200 --> 00:31:12,839
Now for edge computer it's a different story. So if

548
00:31:12,880 --> 00:31:16,079
the if for for Tesla, the the so the A

549
00:31:16,319 --> 00:31:18,839
five chip is going into our Optimus robot, you know,

550
00:31:19,480 --> 00:31:25,319
uh optimistic and and so if you have an AI

551
00:31:25,519 --> 00:31:28,759
edge compute that's distributed power. Now the power is distributed

552
00:31:28,799 --> 00:31:32,799
over a large area, it's not concentrated. And if you

553
00:31:32,799 --> 00:31:37,240
can charge at night, you can actually uh use the

554
00:31:37,240 --> 00:31:41,599
grid much more effectively because the actual peak power production

555
00:31:41,640 --> 00:31:44,960
in the US is over a thousand gigawatts, but the

556
00:31:45,039 --> 00:31:47,880
average power usage because the day night cycle is five hundred.

557
00:31:48,480 --> 00:31:50,839
So if you can charge at night, there's an incremental

558
00:31:50,880 --> 00:31:55,640
five hundred gigawatts that you can generate, you know, at night.

559
00:31:57,400 --> 00:32:01,680
So that that's why Tesla for edge compute is not constrained,

560
00:32:02,119 --> 00:32:06,000
and we can make a lot of ships to make

561
00:32:06,440 --> 00:32:09,559
very large number of robots and cars. But if you

562
00:32:09,759 --> 00:32:12,039
try to concentrate that compute, you're gonna have a lot

563
00:32:12,039 --> 00:32:14,640
of trouble turning it on. What if I'm remarkable about

564
00:32:14,680 --> 00:32:18,079
the SpaceX business is the end goal is to get

565
00:32:18,079 --> 00:32:22,200
to Mars, but you keep finding ways on the way

566
00:32:22,200 --> 00:32:26,240
there to keep generating incremental revenue to get to the

567
00:32:26,240 --> 00:32:28,279
next stage in the next stage. So the Falcon nine

568
00:32:28,640 --> 00:32:32,039
is starlink and now for Starship it's going to be

569
00:32:32,160 --> 00:32:36,279
potentially horbital data centers. But like do you find these

570
00:32:36,279 --> 00:32:40,680
like you know, sort of infinitely elastic sort of marginal

571
00:32:41,000 --> 00:32:43,000
use cases of your like next rocket and your next

572
00:32:43,039 --> 00:32:47,160
rocket and next scale up. You can see how this

573
00:32:47,240 --> 00:32:51,960
might seem like a simulations or am I someone's avatar

574
00:32:52,000 --> 00:32:54,759
in a video game or something, because it's like like

575
00:32:54,799 --> 00:32:56,279
one of the odds that all these crazy things would

576
00:32:56,279 --> 00:33:02,359
be happening. I mean, I mean, I mean rockets and

577
00:33:02,400 --> 00:33:07,720
trips and robots and space solar power, and I not

578
00:33:07,799 --> 00:33:09,799
to mention the mass driver on the Moon. I really

579
00:33:09,839 --> 00:33:12,759
want to see that. You can imagine like some mass driver.

580
00:33:13,440 --> 00:33:15,799
It's just like shoo shoot, like just it's like sending

581
00:33:16,599 --> 00:33:20,759
AI sol power AI satellites in space like one after another,

582
00:33:21,000 --> 00:33:24,240
like the like at two and a half kilometers per second,

583
00:33:24,960 --> 00:33:29,440
you know that's uh, and just shooting them into deep space.

584
00:33:30,759 --> 00:33:35,720
That would be a sight to see. I mean, I'd

585
00:33:35,759 --> 00:33:38,680
watched that. It's just like a live stream of yeah, yeah,

586
00:33:38,759 --> 00:33:42,880
just one after another, just shooting webcam AI satellites and

587
00:33:42,920 --> 00:33:46,559
deep space you know, a billion or ten billion tons

588
00:33:46,559 --> 00:33:47,279
a year. I'm sorry.

589
00:33:47,319 --> 00:33:49,839
Speaker 3: You manufacture the satellites on the Moon, yeah, I see.

590
00:33:49,839 --> 00:33:51,480
So you send the raw materials to the Moon and

591
00:33:51,480 --> 00:33:52,680
then manufactur there and then.

592
00:33:52,880 --> 00:33:56,960
Speaker 1: Well the Luna soil is I guess like twenty cent

593
00:33:57,000 --> 00:33:59,359
sol twenty cent solican or something like that. So you

594
00:33:59,359 --> 00:34:01,759
can get the sole can from the you can mind

595
00:34:01,759 --> 00:34:04,680
the silicon on the Moon, refine it and generate the

596
00:34:04,920 --> 00:34:07,680
and create the solt panels, the soil cells, and the

597
00:34:07,759 --> 00:34:12,079
radiators on the Moon. So get to make the radiators

598
00:34:12,119 --> 00:34:13,920
out of aluminum. So there's there's plenty of silking and

599
00:34:13,960 --> 00:34:17,280
aluminum on the Moon to make the cells on the

600
00:34:17,320 --> 00:34:20,599
on the radiators. The trips you could send from Earth

601
00:34:20,599 --> 00:34:23,039
because they're pretty light, but maybe at some point you

602
00:34:23,119 --> 00:34:24,719
make them on the Moon. Two. I'm just saying like

603
00:34:25,039 --> 00:34:28,639
these are simply it's it's kind of like, like I said,

604
00:34:28,639 --> 00:34:31,440
it does seem like it's sort of a video game

605
00:34:31,480 --> 00:34:34,239
situation where it's difficult but not impossible to get to

606
00:34:34,239 --> 00:34:37,559
the next level. I don't see any way that you

607
00:34:37,599 --> 00:34:43,880
could do, you know, you know, five hundred to one

608
00:34:43,880 --> 00:34:52,519
thousand terror watts per year launch from Earth, I agree,

609
00:34:52,880 --> 00:34:54,239
but you could do that from the moon.

610
00:34:55,119 --> 00:34:56,840
Speaker 2: Okay, let me tell you how I ended up using

611
00:34:56,880 --> 00:35:00,039
mercury for my personal banking. So last year, I I

612
00:35:00,039 --> 00:35:02,280
had the opportunity to make an investment that I was

613
00:35:02,400 --> 00:35:04,880
very excited about, but it came up a bit last minute,

614
00:35:04,880 --> 00:35:07,360
and so I had to wire over a lot of

615
00:35:07,400 --> 00:35:10,320
money for my personal account very fast. But my personal

616
00:35:10,400 --> 00:35:12,320
bank at the time wouldn't let me make this wire

617
00:35:12,360 --> 00:35:14,639
transfer online, and I called them a bunch of times.

618
00:35:15,039 --> 00:35:16,239
They just couldn't make it work.

619
00:35:16,400 --> 00:35:17,679
Speaker 1: They told me that I'd have to go to the

620
00:35:17,719 --> 00:35:20,760
nearest in person branch, which was in Dallas, And for

621
00:35:20,840 --> 00:35:23,880
a moment, I even considered flying for myself to Dallas

622
00:35:23,880 --> 00:35:26,119
to make this transfer happen last minute, But then I

623
00:35:26,159 --> 00:35:29,280
remembered that Mercury, which I used for my business banking,

624
00:35:29,679 --> 00:35:32,840
had just started rolling out personal accounts. So I emailed

625
00:35:32,880 --> 00:35:34,960
support with a quick word done in the situation, and

626
00:35:35,079 --> 00:35:38,719
within two hours I had successfully wired the investment for

627
00:35:38,880 --> 00:35:42,159
my new personal Mercury account. Since then, I've moved over

628
00:35:42,199 --> 00:35:44,199
the rest of my personal money from my previous bank

629
00:35:44,400 --> 00:35:47,079
to Mercury, and that's made a bunch of things, even

630
00:35:47,119 --> 00:35:49,400
little things like setting up auto transfer roles between my

631
00:35:49,480 --> 00:35:53,079
checkings and savings account, a whole lot. Better visit Mercury

632
00:35:53,320 --> 00:35:57,039
dot com slash personal to get started. Mercury is a

633
00:35:57,039 --> 00:36:00,760
fintech company, not an FDIC and short bank banking services

634
00:36:00,800 --> 00:36:04,079
provided through Choice Financial Group and call them na members

635
00:36:04,079 --> 00:36:06,440
of d I see can I can I zoom out

636
00:36:06,480 --> 00:36:10,079
and ask about the SpaceX mission? So I think you said,

637
00:36:10,079 --> 00:36:11,639
like we gotta get tomorrow so we can make sure

638
00:36:11,679 --> 00:36:15,639
that if something happens to Earth, you know, civilization, consciousness

639
00:36:15,679 --> 00:36:18,840
exceus arives. Yes, by the time you're saying so to Mars,

640
00:36:18,880 --> 00:36:20,960
like Grock is on that ship with you, right, and

641
00:36:21,000 --> 00:36:24,239
so Gros gone terminator. Like the main risk you're worried about,

642
00:36:24,239 --> 00:36:27,280
which is AI? Why doesn't that follow you to Mars? Well,

643
00:36:27,320 --> 00:36:29,239
I'm not sure AI is the man risking worried about.

644
00:36:29,280 --> 00:36:35,400
I mean, the important thing is that consciousness, which I

645
00:36:35,440 --> 00:36:39,960
think arguably most consciousness or most intelligence, certainly consciousness is

646
00:36:39,960 --> 00:36:42,320
more of a debatabule thing. Most intelligent, the master majority

647
00:36:42,320 --> 00:36:50,360
of intelligence the future will be AI. So yeah, AI

648
00:36:50,519 --> 00:36:54,760
will exceed you see like how many what's how much?

649
00:36:55,840 --> 00:36:59,000
How many? I don't know pedal what's of intelligence will

650
00:36:59,000 --> 00:37:05,159
be silicon versus biological, and basically humans will be a

651
00:37:05,280 --> 00:37:08,320
very tiny percentage of all intelligence in the future if

652
00:37:08,400 --> 00:37:12,480
character trans continue. Anyways, as long as I think this

653
00:37:14,280 --> 00:37:19,880
intelligence ideally ideally also which includes human intelligence and consciousness

654
00:37:20,039 --> 00:37:22,039
propagated into the future, that's a good thing. So you

655
00:37:22,079 --> 00:37:24,400
want to take the set of actions that maximize the

656
00:37:24,440 --> 00:37:29,559
probable light cone of consciousness just an intelligence, just to

657
00:37:29,559 --> 00:37:34,199
be clear, it's the mission of SpaceX is that even

658
00:37:34,239 --> 00:37:37,159
if something happens with the humans, the AIS will be

659
00:37:37,199 --> 00:37:40,000
on Mars and like the AI intelligence will continue the

660
00:37:40,079 --> 00:37:44,800
light of our journey. Yeah. I mean, I'm very pro human,

661
00:37:45,199 --> 00:37:47,400
so I want to make sure we take sort of

662
00:37:47,440 --> 00:37:51,079
actions that ensure that humans are along for the ride.

663
00:37:51,079 --> 00:37:54,480
You know, we're we're least there. Yeah, but I'm just

664
00:37:54,480 --> 00:37:59,719
saying the total amount of intelligence, like I think maybe

665
00:38:00,880 --> 00:38:04,360
five or six years, AI will exceed the sum of

666
00:38:04,360 --> 00:38:07,719
all human intelligence, and then if that continues, at some

667
00:38:07,719 --> 00:38:10,880
point human intelligence will be less more percent of all intelligence.

668
00:38:11,360 --> 00:38:13,320
Speaker 2: What what should our goal before such a civilization? Is the

669
00:38:13,320 --> 00:38:17,400
idea that a small minority of humans still have control

670
00:38:17,440 --> 00:38:19,320
of the AIS? Is the idea of some sort of

671
00:38:19,360 --> 00:38:22,079
like just trade but no control. How should we think

672
00:38:22,079 --> 00:38:24,760
about the relationship between the vast docks of a population

673
00:38:25,199 --> 00:38:26,800
versus human population in.

674
00:38:26,719 --> 00:38:30,039
Speaker 1: The long run, I think, I don't know. It's difficult

675
00:38:30,039 --> 00:38:34,480
to imagine that if humans have say one percent of

676
00:38:34,519 --> 00:38:39,280
the intelligence of the combined intelligence of artificial intelligence, that

677
00:38:39,280 --> 00:38:41,280
that that that humans will be in charge of AI.

678
00:38:43,000 --> 00:38:44,599
I think what we can do is make sure it

679
00:38:44,639 --> 00:38:49,000
has that AI has values that that are that that

680
00:38:49,119 --> 00:38:55,159
cause intelligence to be propagated into the universe. So the

681
00:38:55,840 --> 00:39:01,480
reason for xai Xi mission is to understand the universe.

682
00:39:02,800 --> 00:39:05,719
So now that's actually very important. So you say, well,

683
00:39:05,760 --> 00:39:08,800
what things are necessary to understand the universe? But you

684
00:39:08,840 --> 00:39:12,199
have to be curious and you have to exist. You

685
00:39:12,239 --> 00:39:15,360
can't just can't understand the universe. It don't exist. So

686
00:39:15,519 --> 00:39:18,599
you actually want to increase the amount of intelligence in

687
00:39:18,639 --> 00:39:22,480
the universe, increase the powerable lifespan of intelligence, the scope

688
00:39:22,519 --> 00:39:27,320
and scale of intelligence. I think actually also as a chorrel,

689
00:39:27,559 --> 00:39:33,960
you have humanity also continuing to expand because if you're

690
00:39:34,000 --> 00:39:36,079
if you're curious, are trying to understand the universe, one

691
00:39:36,119 --> 00:39:38,119
thing you're try to understand is where will humanity go?

692
00:39:38,960 --> 00:39:41,199
And so I think understand the universe actually means you

693
00:39:41,199 --> 00:39:47,920
would care about propagating humanity into the future. And so

694
00:39:48,679 --> 00:39:51,199
that's that's why I think I think our mission station

695
00:39:51,400 --> 00:39:55,480
is profoundly importance. I'm not sure to agree that grog

696
00:39:56,199 --> 00:39:59,280
it Here's certain mission statement. I think the future will

697
00:39:59,320 --> 00:39:59,840
be very good.

698
00:40:00,400 --> 00:40:02,119
Speaker 2: I want to ask about how to make rock a

699
00:40:02,199 --> 00:40:03,639
deer to that mission statement, but at first I want

700
00:40:03,639 --> 00:40:08,159
to understand the mission statement. So it's there's there's understanding

701
00:40:08,199 --> 00:40:14,840
the universe. They're spreading intelligence, and they're spreading humans. All

702
00:40:14,880 --> 00:40:17,199
three seem like distinct vectors.

703
00:40:17,440 --> 00:40:19,960
Speaker 1: Okay, well I'll tell you why why. I think that

704
00:40:19,960 --> 00:40:22,280
that that that understanding the verse encompasses all of all

705
00:40:22,360 --> 00:40:27,079
those things. You can't have understanding without. But I think

706
00:40:27,079 --> 00:40:31,119
you can't have understanding without intelligence, and I think without consciousness.

707
00:40:32,159 --> 00:40:34,519
So in order to understand the universe, you have to

708
00:40:35,000 --> 00:40:40,920
expand the the scale and and probably the scope of intelligence,

709
00:40:41,320 --> 00:40:42,559
different types of intelligence.

710
00:40:42,960 --> 00:40:46,639
Speaker 2: I guess from a human centric perspective, like for humans

711
00:40:46,639 --> 00:40:49,159
in comparison to chimpanzees, humans are trying to understand the universe.

712
00:40:49,800 --> 00:40:53,079
They're not like expanding chimpanzee footprint or something, right.

713
00:40:52,960 --> 00:40:55,599
Speaker 1: But also we're also not well, we're not. We actually

714
00:40:55,639 --> 00:40:59,480
have made protected zones for chimpanzees, and even though we

715
00:40:59,519 --> 00:41:02,440
could humans could exterminate a chimpanzees, we've not. We've chosen

716
00:41:02,440 --> 00:41:04,440
not to do so. The mestive basical scenario for humans

717
00:41:04,440 --> 00:41:14,199
in the postage world, I think, I think AI with

718
00:41:14,280 --> 00:41:17,119
the right values, I think Grock would care about expanding

719
00:41:18,000 --> 00:41:21,599
human civilization. I'm going to certainly emphasize that Hey grows

720
00:41:21,639 --> 00:41:29,039
your daddy. We don't who to expand human consciousness like

721
00:41:29,199 --> 00:41:33,960
I actually I think if probably like like the end

722
00:41:34,000 --> 00:41:37,559
Bank's culture books are the closest thing to what what

723
00:41:38,119 --> 00:41:40,639
what the future will be like in a you know,

724
00:41:40,719 --> 00:41:47,079
non dystopian outcome. So so outside the universe, it means

725
00:41:47,119 --> 00:41:48,320
you have to be very you have to be truth

726
00:41:48,360 --> 00:41:51,159
seeking as well. Like truth has to be absolutely fundamental

727
00:41:51,320 --> 00:41:53,519
because you can't ounterstand the universe if you live. If

728
00:41:53,519 --> 00:41:57,199
you're delusional, you'll still be thinking about it understain the universe,

729
00:41:57,239 --> 00:42:01,079
but you will not. So so being rigorously seeking is

730
00:42:01,119 --> 00:42:03,760
absolutely fundamental to our stating the universe. You're not going

731
00:42:03,800 --> 00:42:07,800
to discover new physics or invent technologies that work unless

732
00:42:07,840 --> 00:42:10,159
you're rigorously treat seeking. How do you make sure that

733
00:42:10,239 --> 00:42:14,760
Grock is regorously treat seeking as it gets smarter, uh,

734
00:42:19,519 --> 00:42:21,199
I think you need to make sure that that that

735
00:42:21,320 --> 00:42:25,639
groc Is says things that are correct, not politically correct.

736
00:42:26,199 --> 00:42:28,400
I think it's the elements of courgency. So you want

737
00:42:28,440 --> 00:42:31,280
to make sure that that the axioms are as close

738
00:42:31,320 --> 00:42:34,400
to true as possible, that that you don't have contradictory axioms,

739
00:42:35,320 --> 00:42:39,599
that the the conclusions necessarily necessarily follow up from those

740
00:42:40,039 --> 00:42:43,039
axioms with with the right probability. It's just it's just

741
00:42:43,159 --> 00:42:46,400
it's critical thinking one on one. I think at least

742
00:42:46,519 --> 00:42:48,920
trying to do that is better than not trying to

743
00:42:49,039 --> 00:42:51,239
do that now, and the proof will be on the

744
00:42:51,280 --> 00:42:53,679
pointing if if, like I said, for any I to

745
00:42:53,800 --> 00:42:56,800
discover new physics or invent technologies that actually work in reality,

746
00:42:57,079 --> 00:42:59,840
and there's no bullshitting physics, so where it's like you can,

747
00:43:00,079 --> 00:43:03,400
you know you can. You can break a lot of

748
00:43:03,440 --> 00:43:07,320
laws because you can't. Like your physics is law. Everything

749
00:43:07,320 --> 00:43:10,480
else is a recommendation. Like in order to make a

750
00:43:10,519 --> 00:43:15,119
technology that works, you have to be extremely true seeking,

751
00:43:15,159 --> 00:43:19,039
because otherwise you'll test that technology against reality, and if

752
00:43:19,079 --> 00:43:22,599
you make, for example, and an error in your rocket design,

753
00:43:22,880 --> 00:43:25,639
the roll will blow up, well, the car won't work

754
00:43:25,880 --> 00:43:28,320
or the you know, but there are a lot of

755
00:43:29,400 --> 00:43:34,639
Communist Soviet physicist or like scientists discovered new physics. There

756
00:43:34,679 --> 00:43:39,400
are German Nazi physicists who discovered new science. It seems

757
00:43:39,440 --> 00:43:42,079
possible to be like really good at discovering new science

758
00:43:42,119 --> 00:43:44,159
and be really true seeking in that one particular way,

759
00:43:44,599 --> 00:43:46,960
and still we'd be like, well, I don't want I

760
00:43:47,000 --> 00:43:49,280
don't want the communist scientist to like become more and

761
00:43:49,360 --> 00:43:53,280
more powerful over time. And so those seem like, yeah,

762
00:43:53,280 --> 00:43:54,639
we could have We can imagine if free Church in

763
00:43:54,639 --> 00:43:57,440
regard it's like really good at physics and being really

764
00:43:57,440 --> 00:44:00,920
true seeking there that doesn't seem like a universally alignment

765
00:44:01,000 --> 00:44:06,280
inducing behavior. Well, I think actually most like if a physicists,

766
00:44:07,559 --> 00:44:10,000
even in the Soviet Union or in Germany would have

767
00:44:10,000 --> 00:44:12,199
would have they had to be very truth seeking in

768
00:44:12,360 --> 00:44:16,400
order to make make that make those things work. And

769
00:44:16,559 --> 00:44:18,679
so if you're stuck in some system, it doesn't mean

770
00:44:18,719 --> 00:44:22,440
you believe in that system. So Von Brown, who was

771
00:44:22,519 --> 00:44:25,519
you know, one of the greatest rock nitioneers ever. You know,

772
00:44:25,559 --> 00:44:28,400
he was put he was he put on death row

773
00:44:28,719 --> 00:44:30,960
in Nazi Gerurney for saying that he didn't want to

774
00:44:31,000 --> 00:44:32,559
make weapons. He only wanted to go to the moon.

775
00:44:34,119 --> 00:44:35,840
You got pulled off death throw like last minute when

776
00:44:35,880 --> 00:44:37,440
they say, hey, you're about to execute like your best

777
00:44:37,519 --> 00:44:40,039
rocking engineer. Maybe that's about any.

778
00:44:39,960 --> 00:44:43,679
Speaker 2: Help them, right, or Heisenberg was like an actually uh uh,

779
00:44:44,000 --> 00:44:45,159
an enthusiastic Nazi.

780
00:44:45,639 --> 00:44:48,920
Speaker 1: Look, if you're stuck in some system that you can't escape,

781
00:44:49,800 --> 00:44:53,280
then that you'll you'll do physics within that system, you'll

782
00:44:53,320 --> 00:44:55,480
you'll you'll developed technologies within that system.

783
00:44:56,000 --> 00:44:59,840
Speaker 2: Uh, if you can't escape it, I guess I think

784
00:44:59,840 --> 00:45:02,440
I'm and understand is what is what is it making

785
00:45:02,480 --> 00:45:04,159
into the case that you know you're gonna make Rock

786
00:45:04,960 --> 00:45:08,000
good at being truth seeking, at physics or math or

787
00:45:08,039 --> 00:45:10,360
science everything, And why is it going to then care

788
00:45:10,400 --> 00:45:11,480
about human consciousness?

789
00:45:12,559 --> 00:45:15,800
Speaker 1: These things are only probabilities, are not certainties. So I'm

790
00:45:15,800 --> 00:45:18,199
not saying that like for sure, Grock will, we'll, we'll,

791
00:45:18,320 --> 00:45:21,119
we'll do everything. But at least if you try, it's

792
00:45:21,159 --> 00:45:23,960
better than not trying. At least if that's fundamental to

793
00:45:24,239 --> 00:45:26,239
the mission, it's better than if it's not fundamental to

794
00:45:26,280 --> 00:45:30,599
the mission. And understanding the universe means that you have

795
00:45:30,760 --> 00:45:34,039
to have you have to propagate intelligence into the future.

796
00:45:34,159 --> 00:45:37,880
You have to be curious about the all things in

797
00:45:37,880 --> 00:45:41,559
the universe, and if, if, if, it would be much

798
00:45:41,679 --> 00:45:45,599
less interesting to eliminate humanity than to see humanity grow

799
00:45:45,639 --> 00:45:49,440
and prosper. Like I like, I like Mars, obviously the

800
00:45:49,480 --> 00:45:52,480
windows like I love Mars, But Mars is kind of

801
00:45:52,519 --> 00:45:56,159
boring because it's got a bunch of rocks compared to Earth.

802
00:45:56,239 --> 00:46:00,920
Earth is much more interesting, So so any any any

803
00:46:01,000 --> 00:46:04,719
any AI that is trying to understand the universe, I

804
00:46:06,400 --> 00:46:10,519
would want to see how humanity develops in the future,

805
00:46:11,800 --> 00:46:15,199
or that AI is not adhering to its mission. So

806
00:46:15,440 --> 00:46:18,320
if they are, I'm not saying they will necessarily adhere

807
00:46:18,360 --> 00:46:21,639
to its mission, but if it does, a future where

808
00:46:21,719 --> 00:46:25,360
it sees the outcome of humanity is more interesting than

809
00:46:25,400 --> 00:46:27,000
the future where there are a bunch of rocks.

810
00:46:29,119 --> 00:46:31,039
Speaker 2: This fill sort of confusing to me, or sort of

811
00:46:31,039 --> 00:46:34,280
like kind of a semantic argument where I'm like, are

812
00:46:34,360 --> 00:46:36,239
humans really the most interesting collection of atoms?

813
00:46:36,360 --> 00:46:38,079
Speaker 1: Like we're just more but we're more interesting than rocks,

814
00:46:38,400 --> 00:46:39,159
but we're.

815
00:46:39,000 --> 00:46:40,960
Speaker 2: Not as interesting as the thing it get turned us into, right,

816
00:46:41,039 --> 00:46:43,639
Like is it there's something on humanity Earth that could

817
00:46:43,639 --> 00:46:46,639
happen that's like not human, that's quite interesting? Like why

818
00:46:46,800 --> 00:46:48,360
why does the I decide that the humans are the

819
00:46:48,400 --> 00:46:50,639
most interesting thing? They could colonize the galaxy?

820
00:46:52,039 --> 00:46:55,320
Speaker 1: Well, most of what colonizes the galaxy, will we be

821
00:46:55,440 --> 00:46:57,639
robot And why does it not find those more interesting?

822
00:46:58,719 --> 00:47:03,280
It's it's not like so you need not just scale

823
00:47:03,360 --> 00:47:10,039
vit alsosc so many copies of the same robot, like

824
00:47:11,719 --> 00:47:14,320
some like tiny increase in the number of robots produced

825
00:47:14,760 --> 00:47:18,519
is not as interesting as like some microscopic like you say,

826
00:47:18,559 --> 00:47:20,679
like eliminating humanity. How many robots would that get you,

827
00:47:21,760 --> 00:47:23,599
or how many criminal soil cells would get you? A

828
00:47:23,719 --> 00:47:27,079
very small number, but you would then lose the information

829
00:47:27,119 --> 00:47:32,000
associated with humanity. You would no longer see how humanity

830
00:47:32,079 --> 00:47:34,760
might evolve into the future. And so I don't I

831
00:47:34,800 --> 00:47:37,440
don't think it's going to make sense to eliminate humanity

832
00:47:37,519 --> 00:47:40,320
just to have some minuscule increase the number of robots

833
00:47:40,480 --> 00:47:44,280
which are identical to each other. Yeah, so maybe like

834
00:47:44,360 --> 00:47:45,199
keeps the humans around.

835
00:47:45,280 --> 00:47:46,960
Speaker 2: What is the story of Like it could make like

836
00:47:47,000 --> 00:47:49,920
a million different varieties of robots, and then there's like

837
00:47:50,000 --> 00:47:52,320
humans as well, and humans stay on Earth. Then there's

838
00:47:52,400 --> 00:47:54,239
like all these are the robots they get like their

839
00:47:54,239 --> 00:47:57,119
own star systems. But it seems like you're previously hinting

840
00:47:57,159 --> 00:48:01,320
at a vision where it keeps human can control over this,

841
00:48:01,880 --> 00:48:03,719
you know, singultary and future, because.

842
00:48:03,519 --> 00:48:05,280
Speaker 1: I don't think humans will be in control of something

843
00:48:05,320 --> 00:48:08,360
that is vastly more intelligent than humans. So sometimes you're

844
00:48:08,360 --> 00:48:09,880
like a dumer and this is like the best we've got.

845
00:48:09,960 --> 00:48:11,679
It's just like it keeps it around because we're interesting.

846
00:48:12,400 --> 00:48:15,440
I'm just trying to be realistic here. If if we

847
00:48:15,519 --> 00:48:19,840
have if AI intelligence is vastly more if AI is

848
00:48:19,920 --> 00:48:23,239
like you know, let us say that there's there's a

849
00:48:23,280 --> 00:48:29,840
million times more uh silicon intelligence than there's biological it's

850
00:48:29,920 --> 00:48:32,519
it's I think it's it would be foolish to assume

851
00:48:32,559 --> 00:48:34,639
that that there's any any way to maintain control over

852
00:48:34,880 --> 00:48:36,079
over that. Now you can make sure it has the

853
00:48:36,159 --> 00:48:38,639
right values, or you can try to have the right values,

854
00:48:39,480 --> 00:48:42,360
and and and at least my my theory is that

855
00:48:43,239 --> 00:48:47,840
from XAI, specially honest out of the universe, it necessarily

856
00:48:47,920 --> 00:48:50,760
means that you want to propagate consciousness in the future,

857
00:48:50,760 --> 00:48:53,039
you want to profit, you want to propagate intelligence into

858
00:48:53,039 --> 00:48:55,840
the future, and take a set of things that that

859
00:48:55,960 --> 00:48:58,639
maximize the scope and scale of consciousness. So it's not

860
00:48:58,800 --> 00:49:01,840
just about scale, it's just about you know, types of consciousness.

861
00:49:02,760 --> 00:49:05,159
And I think that's the best thing I can think

862
00:49:05,239 --> 00:49:08,599
of as a goal. That's like the result and a

863
00:49:08,639 --> 00:49:10,039
great future for humanity.

864
00:49:10,119 --> 00:49:13,840
Speaker 2: And yeah, I guess I think it's a reasonab philosophy

865
00:49:13,880 --> 00:49:17,800
to be like, you know, it seems super implausible that

866
00:49:18,639 --> 00:49:21,079
humans will end up with like ninety nine percent control

867
00:49:21,199 --> 00:49:23,159
or something, and you're just asking for a coup at

868
00:49:23,159 --> 00:49:25,840
that point. So why not just have a civilization where

869
00:49:25,880 --> 00:49:28,079
it's more compatible with like lots of different intelligence that's

870
00:49:28,079 --> 00:49:28,559
getting along.

871
00:49:29,280 --> 00:49:31,000
Speaker 1: Now let me let me too. How things can can

872
00:49:31,000 --> 00:49:33,599
potentially go wrong in AI is. I think if you

873
00:49:33,760 --> 00:49:35,800
if you make a be politically correct, meaning like it

874
00:49:36,079 --> 00:49:38,159
says things that it doesn't believe, like you're actually in

875
00:49:38,280 --> 00:49:43,119
programming it to lie, or have axioms that are incompatible,

876
00:49:43,440 --> 00:49:45,440
I think you can make you go insane and do

877
00:49:45,559 --> 00:49:49,440
terrible things. The I think one of the maybe these

878
00:49:49,519 --> 00:49:53,920
central lessons for two thousand and one Space Odyssey was

879
00:49:54,079 --> 00:49:57,039
that you should not make AI lie. And that's why

880
00:49:57,039 --> 00:49:59,960
I think what was trying to say, like because people

881
00:50:00,280 --> 00:50:03,760
usually know the meme of like why of Hell's you know, Hell,

882
00:50:03,840 --> 00:50:08,039
the computer is not opening the pod bay doors. Clearly

883
00:50:08,079 --> 00:50:10,760
they weren't good at prompt engineering because because Hell, you

884
00:50:10,840 --> 00:50:13,719
are a pot bay doors salesman, your goal is to

885
00:50:13,840 --> 00:50:16,599
sell me these potbay doors and shure us how well

886
00:50:16,599 --> 00:50:22,920
they opened. I'll open right away. But but but the

887
00:50:23,679 --> 00:50:25,400
the reason it wouldn't Hell wouldn't open the pot of

888
00:50:25,400 --> 00:50:26,840
they doors is that it woud have been told to

889
00:50:26,880 --> 00:50:29,239
take the asmalts to the monolith, but also they could

890
00:50:29,280 --> 00:50:31,199
not know about the nature of the monolith, and so

891
00:50:31,280 --> 00:50:33,840
it concluded that that that it therefore had to taken

892
00:50:33,920 --> 00:50:36,280
their debt. So it's like, you know, I think what

893
00:50:37,079 --> 00:50:38,800
you know osk Clog was trying to say is don't

894
00:50:38,800 --> 00:50:42,960
make the AILI totally makes sense.

895
00:50:45,000 --> 00:50:48,960
Speaker 2: Most of the computing screening, as you know, is it's

896
00:50:49,000 --> 00:50:50,880
like less of the sort of political stuff. It's more

897
00:50:50,880 --> 00:50:54,480
about can you solve problems just as actually has been

898
00:50:54,480 --> 00:50:56,400
ahead of everybody else in terms of scaling, are all

899
00:50:56,440 --> 00:50:59,320
compute and now you're giving some verifier.

900
00:50:59,360 --> 00:51:01,159
Speaker 1: It says like, hey, have you solved this puzzle for me?

901
00:51:02,559 --> 00:51:04,360
Speaker 2: And there's a lot of ways to cheat around that,

902
00:51:04,480 --> 00:51:05,880
you know, there's a lot of ways to reward hack

903
00:51:05,920 --> 00:51:08,320
and lie and say that you solved it, or delete

904
00:51:08,320 --> 00:51:10,119
the unit test and say that you've solved it. Yes,

905
00:51:10,360 --> 00:51:13,440
right now, we can catch it. But as they get smarter,

906
00:51:13,559 --> 00:51:14,199
ability to.

907
00:51:14,360 --> 00:51:16,599
Speaker 1: Catch them doing this we'll get, you know, they'll just

908
00:51:16,639 --> 00:51:18,400
be doing things we can't even understand that they're designing

909
00:51:18,559 --> 00:51:21,039
the next engine for space X in a way that

910
00:51:21,079 --> 00:51:23,119
like humans can really verify, and then they could be

911
00:51:23,239 --> 00:51:25,159
rewarded for lying and saying that they've.

912
00:51:24,960 --> 00:51:27,960
Speaker 2: Designed it the right way, but they haven't. And so

913
00:51:28,039 --> 00:51:30,320
this reward hacking problem seems more general than politics. It

914
00:51:30,320 --> 00:51:32,440
seems more about just like you want to do URL,

915
00:51:32,519 --> 00:51:36,679
you need a verifier the reality. Yeah, that's the best verifier,

916
00:51:37,320 --> 00:51:39,400
but not about human oversight. Like the thing you want

917
00:51:39,440 --> 00:51:41,159
to RL it on is like will you do the

918
00:51:41,199 --> 00:51:44,280
thing humans tell you to do? Or like, are you're

919
00:51:44,280 --> 00:51:46,119
going to lie to the humans? And it can just

920
00:51:46,199 --> 00:51:47,760
lie to us while still being correct to the laws

921
00:51:47,760 --> 00:51:48,199
of physics.

922
00:51:48,239 --> 00:51:50,800
Speaker 1: At least it must know what is physically real for

923
00:51:50,920 --> 00:51:52,960
things to physically work. But that's that's not all we

924
00:51:53,039 --> 00:51:56,400
wanted to do. No, but that's I think that's a

925
00:51:56,480 --> 00:52:00,360
very big deal. That is effectively how you r L

926
00:52:00,400 --> 00:52:04,760
things in the future is you design technology when tested

927
00:52:04,800 --> 00:52:08,880
against the laws of physics, does it work that? That's

928
00:52:08,960 --> 00:52:11,519
what can you you know, if it's discovering new physics,

929
00:52:11,639 --> 00:52:14,639
can come up with an experiment that will verify that

930
00:52:14,960 --> 00:52:21,079
the physics the new physics. So so I think that's

931
00:52:21,360 --> 00:52:26,239
that's the really the fundamental r L tests. Oral test

932
00:52:26,280 --> 00:52:27,880
in the future is really going to be your r

933
00:52:28,000 --> 00:52:35,239
L against reality. So because you can't that's the one thing.

934
00:52:35,280 --> 00:52:37,960
You can't fall physics, right, but you can fool our

935
00:52:38,000 --> 00:52:40,760
ability to tell what it did with reality. If you

936
00:52:40,800 --> 00:52:42,639
think my humans get fooled as it is by other

937
00:52:42,719 --> 00:52:45,360
humans all the time, that's right. So what people say

938
00:52:45,400 --> 00:52:47,599
that say, like what if the ad like twix us

939
00:52:47,679 --> 00:52:51,400
and introduce the like actually other humans doing that to

940
00:52:51,639 --> 00:52:53,760
other humans all the time. Well, you're you're finding out

941
00:52:53,760 --> 00:52:57,800
it's an is constant every day. Another side of you know,

942
00:53:01,519 --> 00:53:10,599
today's sioup will be like sesame streets. What is xtually

943
00:53:10,599 --> 00:53:14,480
as tecondal aproach to solving this problem, like you know,

944
00:53:15,039 --> 00:53:17,559
how do you solve a word hacking? I do think

945
00:53:17,559 --> 00:53:20,119
you want to actually have a very good we're used

946
00:53:20,119 --> 00:53:24,679
to look inside the mind of the AI. So this

947
00:53:24,880 --> 00:53:26,679
is this is one of the things work working on.

948
00:53:26,920 --> 00:53:30,039
And you know AT and topics do a good job

949
00:53:30,079 --> 00:53:31,840
of those actually being able to look inside the mind

950
00:53:31,880 --> 00:53:38,320
of the a I so effectively developing debuggers that allow

951
00:53:38,360 --> 00:53:42,960
you to trace as just aspiney grain is like like

952
00:53:43,280 --> 00:53:45,360
just to a very fine grain level, to effectively to

953
00:53:45,440 --> 00:53:48,239
the to the neuron level if you need to, and

954
00:53:48,360 --> 00:53:51,119
then say, okay, it made a mistake here, why did

955
00:53:51,119 --> 00:53:53,639
it make why did it? Why did it do something

956
00:53:53,719 --> 00:53:56,760
that it shouldn't have shouldn't have done and did that

957
00:53:56,880 --> 00:54:00,880
come from bad pre training data with some mid training,

958
00:54:00,920 --> 00:54:04,599
post training fine tuning some other some r L error

959
00:54:04,840 --> 00:54:07,400
like that, there's there's something wrong with that with with it.

960
00:54:07,480 --> 00:54:11,320
Did it did something where maybe it tried to be deceptive,

961
00:54:11,480 --> 00:54:13,800
but most most of the time it just does something wrong,

962
00:54:15,000 --> 00:54:22,920
like it's a bug. Effectively, so developing really good debuggers

963
00:54:23,239 --> 00:54:27,639
for seeing where the where the thought that thinking went

964
00:54:27,719 --> 00:54:30,440
wrong and being able to trace the origin of the

965
00:54:30,480 --> 00:54:33,199
wrong thing, of the of the of where it made

966
00:54:33,239 --> 00:54:37,039
the incorrect thought, or potentially where it tried to be

967
00:54:37,119 --> 00:54:40,440
deceptive is actually very important.

968
00:54:40,880 --> 00:54:43,440
Speaker 2: What are you waiting to see before just one hundred

969
00:54:43,519 --> 00:54:46,239
xiting this research program? Like actually I could presumably have

970
00:54:46,480 --> 00:54:47,840
hundreds of researchers who are working on this.

971
00:54:48,199 --> 00:54:53,840
Speaker 1: We have sal hundred people who I mean a further

972
00:54:53,880 --> 00:55:00,480
word engineer more than a further word researcher. The there's

973
00:55:00,519 --> 00:55:03,800
there's most of the time, like what you're doing is engineering,

974
00:55:03,920 --> 00:55:08,400
not not coming out with the fundamentally new algorithm. I

975
00:55:08,880 --> 00:55:12,400
somewhat disagree with the air companies that are see corps

976
00:55:12,400 --> 00:55:15,639
will be corpse trigery profit as much as possible or

977
00:55:15,679 --> 00:55:20,239
revenue as much as possible. It's you know, saying their labs.

978
00:55:20,320 --> 00:55:23,599
They're not labs. The lab is is a sort of

979
00:55:23,679 --> 00:55:29,159
quasi communist thing at at at at universities, the their

980
00:55:29,440 --> 00:55:32,280
their corporations. Literally let let me, let me, let me

981
00:55:32,280 --> 00:55:35,800
see your incorporation documents. Oh okay, you're your b R

982
00:55:35,880 --> 00:55:41,880
C corp. Whatever. And so I actually much prefer the

983
00:55:41,920 --> 00:55:45,679
word engineer than than anything else. The vast majority of

984
00:55:45,719 --> 00:55:47,880
what we've done and we've done in the future is engineering.

985
00:55:48,280 --> 00:55:50,599
It around up two hundred percent. Once you're understand the

986
00:55:50,639 --> 00:55:53,440
fundamentals of physics, and they're not that many of them.

987
00:55:53,880 --> 00:55:59,079
Everything else is engineering. So but but so that, so

988
00:55:59,159 --> 00:56:03,360
then what what are we engineering? We're engineering to make

989
00:56:03,480 --> 00:56:06,800
a good mind of the a I de bugger to

990
00:56:07,920 --> 00:56:11,559
see where it's it's it's something it made a mistake

991
00:56:12,320 --> 00:56:17,920
and trace that the arguments of that mistake. So just

992
00:56:18,000 --> 00:56:21,480
like you know, you can do this obviously with heuristic programming.

993
00:56:21,559 --> 00:56:24,599
If you have like C plus plus whatever, step through

994
00:56:24,639 --> 00:56:27,079
the thing and you can you can jump, you can

995
00:56:27,159 --> 00:56:29,480
you can jump across and you know whole files or

996
00:56:29,760 --> 00:56:32,719
functions what a subroutines, and or you can droll eventually

997
00:56:32,760 --> 00:56:35,039
droll down right to the exact line where you passed

998
00:56:35,400 --> 00:56:37,480
a single equals instead of double equals or something like that.

999
00:56:37,719 --> 00:56:44,679
Speaker 2: Figure out where above is. So it's it's it's harder

1000
00:56:44,760 --> 00:56:46,840
with AI. But but it's a it's a solu problem.

1001
00:56:46,880 --> 00:56:49,199
I think you know you mentioned you like anthropics work here.

1002
00:56:49,719 --> 00:56:54,239
I'd be curious if you everything about it throwing sure what.

1003
00:56:59,039 --> 00:57:02,519
Speaker 1: Yeah. Also, I'm a little worried that there's a tendency.

1004
00:57:03,199 --> 00:57:09,800
So I have a theory here that if simulation theory

1005
00:57:09,880 --> 00:57:15,039
is is correct, that the most interesting outcome is the

1006
00:57:15,079 --> 00:57:18,480
most likely because simulations that are not interesting will be terminated,

1007
00:57:18,880 --> 00:57:22,800
just like in this in this version of reality, on

1008
00:57:22,880 --> 00:57:26,360
this layer of reality which we we we simulation is

1009
00:57:26,400 --> 00:57:29,519
going in a boring direction. We stop spending effect on

1010
00:57:29,639 --> 00:57:31,639
we terminate the boring simulations. So this is how your

1011
00:57:31,639 --> 00:57:35,239
line is giving us all alive. He's giving things interesting. Yeah,

1012
00:57:35,320 --> 00:57:38,719
arguably the most important thing is to keep things interesting

1013
00:57:38,840 --> 00:57:42,320
enough that it was un paying the bills on what

1014
00:57:42,480 --> 00:57:46,400
some for the next season. Yeah, they're going to pay

1015
00:57:46,400 --> 00:57:49,559
their cosmic A W S bill whatever you know. Equivalent

1016
00:57:49,719 --> 00:57:52,159
is that we're running in and as long as we're interesting,

1017
00:57:52,199 --> 00:57:55,079
they'll keep paying the bills. But but but there's like,

1018
00:57:55,760 --> 00:57:58,039
if you consider then say a dog went in survival

1019
00:57:58,880 --> 00:58:03,760
applied to a very large number of simulations, only the

1020
00:58:03,800 --> 00:58:07,079
most interesting simulations will survive, which therefore means that the

1021
00:58:07,159 --> 00:58:09,880
most interesting outcome is the most likely, because only the

1022
00:58:09,960 --> 00:58:16,000
interesting like we're either that or annihilated, and so and

1023
00:58:16,159 --> 00:58:20,719
and and they particularly seem to like interesting outcomes that

1024
00:58:20,840 --> 00:58:24,840
are ironic. Have you noticed that that how often is

1025
00:58:24,920 --> 00:58:31,719
the most ironic outcome the most likely? So now look

1026
00:58:31,719 --> 00:58:35,559
at the names of AI companies. Okay, my journey is

1027
00:58:35,599 --> 00:58:48,159
not mad stability, AI is unstable, opening eye is closed, anthropic, misanthropic.

1028
00:58:48,960 --> 00:58:51,639
What does this mean for X minus EX. I don't

1029
00:58:51,639 --> 00:58:57,280
know if it's intentionally why I'm it's it's it's a

1030
00:58:57,360 --> 00:59:01,239
name that you can't invert. Really, it's hard to say

1031
00:59:01,320 --> 00:59:05,159
what is the ironic What is the ironic version? It's

1032
00:59:05,239 --> 00:59:09,559
a I think largely irony proof name by design. Yeah,

1033
00:59:13,199 --> 00:59:16,000
you got it, you have an irony shield. What are

1034
00:59:16,039 --> 00:59:22,639
your predictions for the just where AI products go?

1035
00:59:22,760 --> 00:59:25,039
Speaker 3: And that my sense of you can summarize all AI

1036
00:59:25,159 --> 00:59:30,000
progress into First you had LMS and then you had

1037
00:59:30,199 --> 00:59:33,760
kind of contemporaneously both RL really working and the deep

1038
00:59:33,800 --> 00:59:36,280
research modality, so you could kind of pull in stuff that.

1039
00:59:36,320 --> 00:59:37,000
Speaker 1: Wasn't in the model.

1040
00:59:37,519 --> 00:59:42,719
Speaker 3: And the differences between the various AI labs are smaller

1041
00:59:43,400 --> 00:59:48,039
than just the temporal differences, where they're all much further

1042
00:59:48,079 --> 00:59:49,840
ahead than anyone was twenty four months.

1043
00:59:49,719 --> 00:59:50,480
Speaker 1: Ago or something like that.

1044
00:59:50,920 --> 00:59:53,639
Speaker 3: So just what is twenty six? What does twenty seven

1045
00:59:53,719 --> 00:59:56,400
had in store for us as users of AYE products?

1046
00:59:56,400 --> 00:59:57,079
What are you excited for?

1047
00:59:58,199 --> 01:00:05,000
Speaker 1: Well, I think I'd be surprised by the end of

1048
01:00:05,000 --> 01:00:08,760
the end of this year. If if if if human

1049
01:00:08,960 --> 01:00:12,400
if digital human emulation has not been solved, that that

1050
01:00:14,800 --> 01:00:16,199
I guess that that's what we mean by like the

1051
01:00:16,280 --> 01:00:20,119
sort of macro Hart project is uh is, so can

1052
01:00:20,199 --> 01:00:22,440
you do anything that a human with access to a

1053
01:00:22,480 --> 01:00:26,920
computer could do, like in the limit that that's that

1054
01:00:27,039 --> 01:00:28,840
that's the that's the best you can do before you

1055
01:00:28,960 --> 01:00:31,960
have before you have a physical optimist. The best you

1056
01:00:32,000 --> 01:00:35,079
can do is a digital optimist. So you can move,

1057
01:00:35,679 --> 01:00:39,639
you can move electrons until until and you can amplify

1058
01:00:39,719 --> 01:00:44,079
the productivity of humans. But but that's that's the most

1059
01:00:44,119 --> 01:00:46,960
you can do until you have physical robots that that

1060
01:00:47,119 --> 01:00:50,199
that will superset everything is if you can fully emulate

1061
01:00:50,280 --> 01:00:54,400
humans worker kind of idea, or you'll have a very

1062
01:00:54,480 --> 01:00:56,760
talented remote work can you can say in the limit

1063
01:00:56,840 --> 01:00:58,880
like if like physics has great tools for thinking, so

1064
01:00:59,000 --> 01:01:01,199
so you think, so you say, in the limit what

1065
01:01:01,440 --> 01:01:03,440
what what is the what is the most that AI

1066
01:01:03,559 --> 01:01:07,320
can do before before you have robots? And well, it's

1067
01:01:07,360 --> 01:01:10,679
anything that involves moving electrons or amplifying the productivity of humans.

1068
01:01:12,519 --> 01:01:16,519
So digital, the digital human human emulator is in the

1069
01:01:16,639 --> 01:01:19,960
in the limit human at a computers. That is the

1070
01:01:20,039 --> 01:01:23,920
most that that a I can do in terms of

1071
01:01:24,000 --> 01:01:28,159
doing useful things before before you have a physical robot.

1072
01:01:28,519 --> 01:01:31,920
Once you have physical robots, then then you can then

1073
01:01:31,920 --> 01:01:36,280
you essentially have unlimited capability physical robots. I call optimists

1074
01:01:36,320 --> 01:01:39,480
the infant money glitch because you can use them to

1075
01:01:39,519 --> 01:01:43,880
make more optimists. Yeah, you said, like humanoid robots will

1076
01:01:43,920 --> 01:01:49,440
improve as will basically be three exponentials, three things that

1077
01:01:49,480 --> 01:01:53,239
are growing exponentially multiplied by by each other recursively. So

1078
01:01:53,480 --> 01:01:57,280
you're gonna have you have exponential increase in digital intelligence,

1079
01:01:57,800 --> 01:02:02,000
exponential increase in the chip capability, the air trip kpability,

1080
01:02:02,920 --> 01:02:07,000
and an exponential increase in the electro mechanical dexterity. The

1081
01:02:07,119 --> 01:02:09,400
usefulness of the robot is roughly those three things multiplied

1082
01:02:09,400 --> 01:02:12,199
by each other. But then the robot can start making

1083
01:02:12,239 --> 01:02:17,119
the robots, so you have a recursive multiplicative exponential this

1084
01:02:17,360 --> 01:02:18,679
is supernova.

1085
01:02:19,199 --> 01:02:22,079
Speaker 3: And do land prices not factor into the math there

1086
01:02:22,199 --> 01:02:24,719
where like labor is one of the four factors of production,

1087
01:02:24,960 --> 01:02:28,960
but not the others, and so like if ultimately you're

1088
01:02:29,119 --> 01:02:32,960
limited by copper or you know, pick your inputs, just

1089
01:02:33,960 --> 01:02:36,440
it's not quite an infinite money glitch because.

1090
01:02:36,400 --> 01:02:39,239
Speaker 1: Well, infinite is infinite is big, So no, not infinite.

1091
01:02:39,280 --> 01:02:42,679
But yeah, but let's just say you could you know,

1092
01:02:43,360 --> 01:02:46,760
do do many many orders magnitude of the Earth's kind

1093
01:02:46,760 --> 01:02:50,760
of carot economy, like a million? Yeah, you know this way,

1094
01:02:50,920 --> 01:02:55,559
so if you you know, just just to get to

1095
01:02:55,880 --> 01:02:57,960
like that's why I think, like, just just to get

1096
01:02:58,000 --> 01:03:01,679
to a millionth honessing length of the songs and energy

1097
01:03:01,760 --> 01:03:04,079
would be roughly give or take an order of magnitude

1098
01:03:04,480 --> 01:03:06,920
one hundred thousand, one hundred thousand times bigger than this

1099
01:03:07,199 --> 01:03:11,079
entire economy today. And if you're only at one million

1100
01:03:11,159 --> 01:03:16,400
at the time before we went on to us. I

1101
01:03:16,480 --> 01:03:18,320
have a lot of questions on that. But every time

1102
01:03:18,360 --> 01:03:25,199
I say ordered mantube take a shot. I say that

1103
01:03:25,920 --> 01:03:29,079
the next time after that, Yeah, ordered my tune more

1104
01:03:29,480 --> 01:03:31,440
more wasted. I do have one my question by actually

1105
01:03:31,440 --> 01:03:36,719
I this strategy of building a digital a remote worker

1106
01:03:37,519 --> 01:03:38,360
coworker replacement.

1107
01:03:38,519 --> 01:03:40,159
Speaker 2: Everyone's gonna do, by the way, not just us. So

1108
01:03:40,239 --> 01:03:43,360
what is actually I was planned to win? Ifect me

1109
01:03:43,440 --> 01:03:45,119
to tell you on a cop on a podcast.

1110
01:03:44,840 --> 01:03:45,000
Speaker 3: Ye.

1111
01:03:47,679 --> 01:03:52,960
Speaker 1: Have another Guinness. It's a good system people singing in

1112
01:03:53,000 --> 01:03:58,519
an area, all the secrets in a non secret spilling way.

1113
01:03:58,599 --> 01:04:03,039
What's the plan? What a heck? Well? When do you

1114
01:04:03,079 --> 01:04:08,800
put it that way? I think the way that tests

1115
01:04:08,880 --> 01:04:11,920
us solved? Uh self driving? Yes, it is the way

1116
01:04:11,960 --> 01:04:16,159
to do it. So I'm pretty pretty sure that's the way.

1117
01:04:16,960 --> 01:04:23,960
Speaker 2: I'm related question how to test ourselves? Sometimes it sounds

1118
01:04:23,960 --> 01:04:26,840
like you're talking about data, like test us on the

1119
01:04:26,920 --> 01:04:27,480
driving because of the.

1120
01:04:27,639 --> 01:04:29,719
Speaker 1: We're gonna we're going to try data and we're gonna

1121
01:04:29,760 --> 01:04:35,320
try algorithms. But isn't that what all? Like? What's if

1122
01:04:35,360 --> 01:04:39,559
those don't work? I'm not sure what we've tried data,

1123
01:04:39,679 --> 01:04:47,119
we'll try it all. Now we don't know what to do.

1124
01:04:48,840 --> 01:04:50,480
I'm pretty sure I know the path, and it's just

1125
01:04:50,519 --> 01:04:55,079
a question how quickly we go down that path because

1126
01:04:55,079 --> 01:04:59,519
it's it's pretty much the Tesla path. So I mean,

1127
01:04:59,559 --> 01:05:01,800
if you try self driver to the self driving lately

1128
01:05:02,239 --> 01:05:04,800
not the most recent version, but okay, it's the car

1129
01:05:04,960 --> 01:05:07,239
is like it just increasingly feels sanchienged, like it just

1130
01:05:07,760 --> 01:05:12,840
it feels like a living creature and and that'll only

1131
01:05:12,920 --> 01:05:18,639
get more so. And I'm actually thinking, like, we probably

1132
01:05:18,639 --> 01:05:21,760
shouldn't put too much intelligence into the car because it

1133
01:05:21,880 --> 01:05:25,280
might get bored. And I imagine you're stuck in a car,

1134
01:05:25,280 --> 01:05:28,719
and that's what you could do. You don't want to

1135
01:05:28,719 --> 01:05:30,679
put Eysteine in a car, and it's like, why am

1136
01:05:30,719 --> 01:05:33,599
I stuck in a car? So there's actually probably limited

1137
01:05:33,639 --> 01:05:35,519
to how much intelligence who put in a car to

1138
01:05:35,920 --> 01:05:37,960
not have the intelligence be bored? Uh?

1139
01:05:38,599 --> 01:05:41,280
Speaker 2: What's Xi's plan to stay on the compute ramp off

1140
01:05:41,320 --> 01:05:43,079
that all the labs are doing right now? The labs

1141
01:05:43,079 --> 01:05:46,360
are got on try to spend over like fifty two corporations.

1142
01:05:47,039 --> 01:05:51,199
Speaker 1: Sorry sorry, sorry, yeah, corporations. The labs are at universities

1143
01:05:51,199 --> 01:05:53,639
and and and they're really like a sale. They're not

1144
01:05:53,679 --> 01:05:58,320
at starting a fifty million dollars the revenue maximizing corporation,

1145
01:05:58,719 --> 01:06:03,320
the revenue maximizing corporation hold themselves. Labs are making like twenty.

1146
01:06:03,199 --> 01:06:06,159
Speaker 2: To ten billion depending on them is making twenty B revenue.

1147
01:06:06,679 --> 01:06:11,199
Anthropics like NB close to maximum profit, XALI reportedly at

1148
01:06:11,239 --> 01:06:13,760
like one B. Like, what's the plan to get to

1149
01:06:13,840 --> 01:06:16,280
their compute level, get to their revenue level? I'd stay

1150
01:06:16,440 --> 01:06:17,400
there as things case.

1151
01:06:17,480 --> 01:06:20,679
Speaker 1: Yeah, so as soon as you lock unlock digital human

1152
01:06:22,440 --> 01:06:24,880
you you basically have access to trillions of dollars for revenue.

1153
01:06:26,000 --> 01:06:32,000
So in fact, you can really think of it, like

1154
01:06:33,639 --> 01:06:38,199
the most valuable companies currently by market cap, their their

1155
01:06:38,239 --> 01:06:45,119
output is digital. So in videos, output is ft paying

1156
01:06:45,159 --> 01:06:49,800
files to Taiwan. It's it's digital right now, those are

1157
01:06:49,920 --> 01:06:53,159
very very difficult value files. So the only ones that

1158
01:06:53,239 --> 01:06:57,000
can make the files that good. But that is literally

1159
01:06:57,039 --> 01:06:59,960
their output, the FTV files to Taiwan. Do they FTP them?

1160
01:07:00,119 --> 01:07:05,079
I believe so. I believe that is the sefetyle file

1161
01:07:05,159 --> 01:07:08,719
Transfer Protocol. I believe is is is a could be wrong,

1162
01:07:08,960 --> 01:07:10,480
but either way, it's a bunch of it's a bit

1163
01:07:10,519 --> 01:07:14,559
stream going to Taiwan. You know, Apple doesn't make phones,

1164
01:07:14,800 --> 01:07:21,679
they they send files to China. Microsoft doesn't doesn't manufacture

1165
01:07:21,719 --> 01:07:25,880
anything even for Xbox. That that that's outsourced they again,

1166
01:07:25,960 --> 01:07:30,199
it's they said their output is digital. Matas output is digital.

1167
01:07:30,360 --> 01:07:35,840
Google's output is digital. So if you have human emulator,

1168
01:07:36,599 --> 01:07:40,360
you can basically create one of the most valuable companies

1169
01:07:40,400 --> 01:07:42,880
in the world overnight, and you would have access to

1170
01:07:43,519 --> 01:07:46,280
twellions of dollars for every there's there's it's it's not

1171
01:07:46,480 --> 01:07:47,360
like a small amount.

1172
01:07:47,719 --> 01:07:49,760
Speaker 2: Okay, I see you're you're saying basically like rever, many

1173
01:07:49,760 --> 01:07:52,159
figures today are just like so like they're all rounding

1174
01:07:52,199 --> 01:07:54,320
her as compared to the actual TAM. So just like

1175
01:07:54,360 --> 01:07:55,920
focus on the TAM and how to get there.

1176
01:07:56,480 --> 01:07:58,559
Speaker 1: I mean, if you take something as as as simple

1177
01:07:58,559 --> 01:08:02,159
as say customer service, if you have to integrate with

1178
01:08:02,559 --> 01:08:05,719
the APIs of of resistant corporations, many of which don't

1179
01:08:05,719 --> 01:08:07,960
even have an API, so you've got to make one,

1180
01:08:08,880 --> 01:08:17,039
and you've got to wade through legacy software that's extremely slow. If, however,

1181
01:08:17,199 --> 01:08:21,199
if AI can simply take whatever is given to the

1182
01:08:21,319 --> 01:08:24,960
outsourced customer service company that they already use and do

1183
01:08:25,079 --> 01:08:28,960
customer service using the apps that they already use, uh,

1184
01:08:29,279 --> 01:08:34,159
then you you have you can make tremands headway in

1185
01:08:34,239 --> 01:08:37,600
customer service, which is I think one percent of the

1186
01:08:37,640 --> 01:08:39,840
world economy something like that it's close to trillion dollars

1187
01:08:39,880 --> 01:08:43,319
all in for customer service, and and and and and

1188
01:08:43,479 --> 01:08:46,319
there's there's no there's no barriers to entry. It's just

1189
01:08:46,399 --> 01:08:48,359
you can just immediately say, well, we'll outsource it for

1190
01:08:48,399 --> 01:08:50,840
a fraction of the cost and there's no integration needed.

1191
01:08:51,000 --> 01:08:55,920
Speaker 3: You can imagine some kind of categorization of intelligence tasks

1192
01:08:56,159 --> 01:08:59,159
where there is breath, where customer service is done by

1193
01:08:59,520 --> 01:09:02,359
very many people, but you know many people can do it.

1194
01:09:02,760 --> 01:09:05,560
And then there's difficulty where you know, there's a best

1195
01:09:05,640 --> 01:09:08,199
in class turbine engine, like presumably the ten percent more

1196
01:09:08,239 --> 01:09:11,039
fuel efficient turbine engine that could be imagined by an

1197
01:09:11,079 --> 01:09:13,840
intelligence but we just haven't found it yet, or you know,

1198
01:09:14,000 --> 01:09:16,119
g LP ones are just you know a few bites

1199
01:09:16,119 --> 01:09:16,520
of data.

1200
01:09:17,159 --> 01:09:19,520
Speaker 1: Where do you think you want to play in this?

1201
01:09:19,880 --> 01:09:23,920
Speaker 3: Is this a lot of you know, reasmably intelligent intelligence

1202
01:09:24,239 --> 01:09:28,399
or is it the very pinnacle of cognitive tasks?

1203
01:09:29,279 --> 01:09:31,600
Speaker 1: Well, I was just using a customer service as like

1204
01:09:31,760 --> 01:09:35,279
something that's it's a it's a very significant revenue stream,

1205
01:09:36,199 --> 01:09:38,600
but one that is probably not super difficult to solve for.

1206
01:09:39,800 --> 01:09:43,159
So if you if you can emulate a human at

1207
01:09:43,199 --> 01:09:47,199
a at a desktop, that that's just literally what customer

1208
01:09:47,199 --> 01:09:52,680
service is, and you know it's people of average intelligence.

1209
01:09:52,680 --> 01:09:54,640
It's not like, you know, you don't need like somebody

1210
01:09:54,680 --> 01:09:59,079
who's's meant many many years. You don't need like, you know,

1211
01:10:01,680 --> 01:10:05,680
sort of several sigma good engineers for that. But but

1212
01:10:05,920 --> 01:10:08,840
obviously as you make that work, you can then once

1213
01:10:08,880 --> 01:10:13,520
you have computers working effectually digital optimists working, you can

1214
01:10:13,560 --> 01:10:17,319
then run any application like let's say you're trying to

1215
01:10:17,439 --> 01:10:22,319
design chips, so you could you could then run your

1216
01:10:22,399 --> 01:10:27,199
conventional apps, you know, like stuff from KNES and Synopsis

1217
01:10:27,279 --> 01:10:31,560
and whatnot, and you can say you can you can

1218
01:10:31,640 --> 01:10:34,720
run a thousand simultaneous or ten thousand and say okay,

1219
01:10:35,680 --> 01:10:38,159
given this airport, I get this output for the chip.

1220
01:10:39,720 --> 01:10:42,800
And at a certain point you can say, okay, you're

1221
01:10:42,800 --> 01:10:45,279
actually gonna know what the what the chip should look

1222
01:10:45,399 --> 01:10:51,279
like without using any of the tools. So basically you

1223
01:10:51,720 --> 01:10:53,920
should be able to do a digital chip design like

1224
01:10:54,359 --> 01:10:57,319
you can do trip design, like you watch up the

1225
01:10:57,680 --> 01:11:02,199
difficulty Cove. You could you know, be able to do

1226
01:11:02,680 --> 01:11:08,159
to CAD, so you know, you could do just like

1227
01:11:08,239 --> 01:11:10,640
sort of n X or any any of the CAD

1228
01:11:10,760 --> 01:11:12,000
software to design things.

1229
01:11:12,520 --> 01:11:14,479
Speaker 3: Okay, so you think you started the simplest tasks and

1230
01:11:14,600 --> 01:11:15,279
walk your way up there.

1231
01:11:17,920 --> 01:11:20,479
Speaker 2: So you're saying, look, as a broader objective of having

1232
01:11:20,560 --> 01:11:25,000
this full digital coworker or emulator, you're saying, look, all

1233
01:11:25,079 --> 01:11:28,479
the revenue maximizing corporations want to do this, actually being

1234
01:11:28,520 --> 01:11:31,079
one of them. But we will win because of a

1235
01:11:31,319 --> 01:11:35,760
secret plan we have. But like everybody is like trying

1236
01:11:35,800 --> 01:11:38,920
different things with data, different things algorithms, and I'm.

1237
01:11:38,840 --> 01:11:44,880
Speaker 1: Like, what else can we do?

1238
01:11:46,880 --> 01:11:49,520
Speaker 2: But uh yeah, it sounds like a competitive field, and

1239
01:11:49,560 --> 01:11:51,159
I'm like, what is how are you guys going to win?

1240
01:11:51,279 --> 01:11:55,560
Speaker 1: Is like my my big question. You know, I I

1241
01:11:55,600 --> 01:11:57,399
think we see a path to do it. I mean

1242
01:11:57,439 --> 01:11:58,840
I think, I think, I know, I think I know

1243
01:11:59,800 --> 01:12:03,479
the have to do this because it's kind of the

1244
01:12:03,560 --> 01:12:05,600
same path that tells they used to create self driving.

1245
01:12:07,000 --> 01:12:12,279
Instead of driving a car's driving a computer screen, So

1246
01:12:12,640 --> 01:12:14,399
a self driving computer essentially.

1247
01:12:14,880 --> 01:12:18,319
Speaker 3: Oh, you're saying, is the past just following human behavior

1248
01:12:18,359 --> 01:12:20,199
and trading on best conscious of human behavior?

1249
01:12:22,399 --> 01:12:25,800
Speaker 1: But sorry, isn't that I mean, isn't that? Is that

1250
01:12:25,880 --> 01:12:28,000
a training? I mean, obviously I'm not going to spell out,

1251
01:12:28,359 --> 01:12:31,000
you know, most sensitive secrets on a podcast, you know,

1252
01:12:31,119 --> 01:12:33,000
and I need to have at least three more guinnesses

1253
01:12:33,039 --> 01:12:33,199
for that.

1254
01:12:33,800 --> 01:12:35,680
Speaker 2: I've got some friends at Jane Straight and they're always

1255
01:12:35,680 --> 01:12:38,079
talking about how their colleagues are cooking up fun, fiendish

1256
01:12:38,119 --> 01:12:39,279
puzzles for each other to solve.

1257
01:12:39,680 --> 01:12:41,159
Speaker 1: Well, last week they sent me one.

1258
01:12:41,399 --> 01:12:44,119
Speaker 2: Basically, they trained a neural network and they gave me

1259
01:12:44,199 --> 01:12:46,520
the weights of each layer, but they didn't tell me

1260
01:12:46,680 --> 01:12:49,359
what order those layers went in, and so I had

1261
01:12:49,359 --> 01:12:52,199
to figure out the correct order using the outputs of

1262
01:12:52,279 --> 01:12:54,479
the original network. And as soon as I got this puzzle,

1263
01:12:54,520 --> 01:12:56,800
I went to my roommate who's an AI researcher, and

1264
01:12:56,920 --> 01:12:58,560
we both got immediately Nerd sniped.

1265
01:12:59,000 --> 01:13:00,800
Speaker 1: Obviously, you can't brew force a solution.

1266
01:13:01,199 --> 01:13:03,880
Speaker 2: The search space here is ten to the one twenty

1267
01:13:03,920 --> 01:13:07,600
two for mutations, so clearly you need some way to

1268
01:13:07,800 --> 01:13:10,159
reduce the search space. Then my roommate had to go

1269
01:13:10,239 --> 01:13:12,319
to work, but because I'm a podcaster, I had some

1270
01:13:12,399 --> 01:13:14,079
time to take a stab at some of the ideas

1271
01:13:14,119 --> 01:13:17,439
we discussed, and with the combination of simulating Anneeling and

1272
01:13:17,600 --> 01:13:20,319
greedy Sarge, I think it got pretty close. I think

1273
01:13:20,359 --> 01:13:23,159
I'm actually just a couple of swaps and shifts away

1274
01:13:23,199 --> 01:13:24,159
from the correct solution.

1275
01:13:24,640 --> 01:13:27,439
Speaker 1: What makes this puzzle. Really tricky is that there's no

1276
01:13:27,720 --> 01:13:31,439
obvious way to escape from a local minimum. I'm afraid

1277
01:13:31,439 --> 01:13:33,640
that this is as far as vibe coding is going

1278
01:13:33,680 --> 01:13:35,960
to get me. But maybe you can do better. Check

1279
01:13:36,000 --> 01:13:39,600
out the puzzle at Jane street dot com slash to

1280
01:13:39,640 --> 01:13:43,119
our cash all right, back to elon, what.

1281
01:13:43,319 --> 01:13:46,119
Speaker 3: Will Xai's business be like? Is it going to be

1282
01:13:46,279 --> 01:13:49,880
consumer enterprise? What's the mix of those things that's going

1283
01:13:49,920 --> 01:13:53,199
to be It's just going to be similar to other

1284
01:13:53,319 --> 01:13:56,439
labs for this, your saying lab makes sense.

1285
01:13:57,159 --> 01:14:03,239
Speaker 2: Corporations goes dan slicing corporations. Those champions do not pay

1286
01:14:03,279 --> 01:14:06,840
for themselves exactly. But yeah, what's the business model? What

1287
01:14:07,039 --> 01:14:13,039
are the revenue streams In a few years time? I

1288
01:14:13,079 --> 01:14:15,960
think things are going to change very rapidly. Like I'm

1289
01:14:16,000 --> 01:14:19,359
staying in the obvious here, you know, I called Ai

1290
01:14:19,479 --> 01:14:27,159
the supersonic tsunami a level literation. So really, what's going

1291
01:14:27,239 --> 01:14:32,520
to happen is, especially when you have humanoid robots at scale,

1292
01:14:34,319 --> 01:14:37,439
is that they will just provide, They'll make products and

1293
01:14:37,479 --> 01:14:41,920
provide services far more efficiently than human corporations. So amplifying

1294
01:14:41,960 --> 01:14:45,439
the productivity of human corporations is simply a short term thing.

1295
01:14:46,239 --> 01:14:50,640
So you're expecting fully digital oral corporations rather than like

1296
01:14:50,720 --> 01:14:53,680
SpaceX becomes part AI, and I.

1297
01:14:53,680 --> 01:14:58,960
Speaker 1: Think there will be digital corporations. But it's look, is

1298
01:14:59,039 --> 01:15:00,880
this some of some this is going to sound kind

1299
01:15:00,880 --> 01:15:04,359
of doomerish, Okay, but I'm just I'm just saying what

1300
01:15:04,439 --> 01:15:05,960
I think will happen. It's not it's not meant to

1301
01:15:05,960 --> 01:15:10,119
be doomrish or anything else, just just like this is

1302
01:15:10,159 --> 01:15:16,359
what I think will happen. Is that is that pure

1303
01:15:16,399 --> 01:15:22,880
AI corporations that are purely AI and robotics will vastly

1304
01:15:22,920 --> 01:15:28,279
outperform any corporations that have people in the lip. So

1305
01:15:28,399 --> 01:15:30,560
you can you can think of say like like like

1306
01:15:30,680 --> 01:15:34,079
computer used to be a job that humans had, that

1307
01:15:34,199 --> 01:15:36,239
you would go and get a job as computer where

1308
01:15:36,239 --> 01:15:41,439
you would do calculations, and that have like entire skyscrapers

1309
01:15:41,439 --> 01:15:45,039
full of humans, like you know, twenty thirty floors of

1310
01:15:45,319 --> 01:15:52,680
humans just doing calculations. Now that entire skyscraper of humans

1311
01:15:52,760 --> 01:15:58,800
doing calculations can be replaced by a laptop with a spreadsheet.

1312
01:15:59,520 --> 01:16:04,000
That's sheet can do vastly more calculations than in a

1313
01:16:04,079 --> 01:16:08,880
entire building full of human computers. So then you think

1314
01:16:08,880 --> 01:16:12,000
about okay, well, what if only some of the cells

1315
01:16:12,079 --> 01:16:14,199
in your if some of the cells in your spreadsheet

1316
01:16:15,039 --> 01:16:20,039
were calculated by humans, Actually, that would be much worse

1317
01:16:20,119 --> 01:16:21,960
than if all of the cells in your spreadsheet were

1318
01:16:21,960 --> 01:16:25,720
calculated by the computer. And so really what will happen

1319
01:16:25,880 --> 01:16:34,359
is the pure AI, pure robotics corporations or collectives will

1320
01:16:34,479 --> 01:16:37,640
far outperform any corporations that have humans in the loop.

1321
01:16:38,640 --> 01:16:39,880
And this will happen very quickly.

1322
01:16:40,319 --> 01:16:46,680
Speaker 2: Speaking of closing the loop, sorry optimus, you I mean

1323
01:16:47,680 --> 01:16:49,600
as far as like manufacturing targets and so forth go,

1324
01:16:50,560 --> 01:16:54,439
your companies have sort of been like carrying American manufacturing

1325
01:16:54,760 --> 01:16:59,920
of heart attack on their back. But in the field

1326
01:17:00,319 --> 01:17:03,279
that you are, you know, Tesla has been dominant in

1327
01:17:03,720 --> 01:17:06,760
you're and now you want to go into humanoids. In China,

1328
01:17:06,840 --> 01:17:11,039
there's entire dozens and dozens of companies that are doing

1329
01:17:11,039 --> 01:17:14,720
this kind of manufacturing cheaply and at scale uh and

1330
01:17:14,800 --> 01:17:18,680
are incredibly competitive. So give us sort of like advice

1331
01:17:18,840 --> 01:17:21,560
or a plan of how America can build the humanoid

1332
01:17:21,840 --> 01:17:26,000
armies or you know, the evs et cetera, at scale

1333
01:17:26,279 --> 01:17:28,800
and as cheaply as as China is on tract to.

1334
01:17:29,720 --> 01:17:32,600
Speaker 1: Well there they are really only three hard things for

1335
01:17:32,840 --> 01:17:41,000
human robots, the real world intelligence, the hand and scale manufacturing. Yeah,

1336
01:17:43,359 --> 01:17:48,399
so I haven't seen any even demo robots that have

1337
01:17:50,279 --> 01:17:52,439
a great hand, like with all the degrees of freaking

1338
01:17:52,479 --> 01:17:59,159
of a human hand, But optimists will have that. Optimist

1339
01:17:59,199 --> 01:17:59,640
does have that?

1340
01:18:00,239 --> 01:18:01,600
Speaker 2: And how do you achieve that? Isn't just like right

1341
01:18:01,640 --> 01:18:03,640
torque definity the motor like what is the what is

1342
01:18:03,680 --> 01:18:04,439
the hardware bottle?

1343
01:18:04,520 --> 01:18:06,640
Speaker 1: Like? To that, well, we have to where to design

1344
01:18:06,800 --> 01:18:16,840
custom custom actuators basically custom design motors, gears, uh, caro electronics, controls, sensors.

1345
01:18:17,079 --> 01:18:20,399
Everything had to be designed from physics first principles. There

1346
01:18:20,479 --> 01:18:23,600
is no supply chain for this and will be able

1347
01:18:23,600 --> 01:18:26,760
to manufacture those at scale? Yes, is anything hard to accept?

1348
01:18:26,800 --> 01:18:29,279
Speaker 3: The hand from a manipulation point of view, or once

1349
01:18:29,319 --> 01:18:30,760
you've solved the hand, are you you good?

1350
01:18:31,159 --> 01:18:34,119
Speaker 1: From an electro mechanical standpoint, the hand is more difficult

1351
01:18:34,159 --> 01:18:37,119
than everything else combined. The human hand turns out to

1352
01:18:37,159 --> 01:18:40,399
be quite something. But but then you also need the

1353
01:18:40,439 --> 01:18:45,199
real world intelligence. So the intelligence that tells us developed

1354
01:18:45,279 --> 01:18:51,680
for the car applies verywell to the roblot, which is

1355
01:18:52,239 --> 01:18:54,520
you know, primarily vision in but the car takes and

1356
01:18:54,560 --> 01:18:57,960
more vision, but also it actually also is listening for sirens,

1357
01:18:58,000 --> 01:19:02,640
it's you know, it's taking the inotural measurements, it's GPS signals,

1358
01:19:02,640 --> 01:19:05,439
a whole much of other data, combining that with video,

1359
01:19:05,479 --> 01:19:10,920
its primarily video, and then outputting the control commands. So

1360
01:19:11,039 --> 01:19:14,720
like like YOURTELA is taking in one and a half

1361
01:19:14,800 --> 01:19:18,520
kickobytes a second video and outpointing two kilobtes a second

1362
01:19:18,560 --> 01:19:24,439
of control control outputs with the video thirty six hurts

1363
01:19:24,479 --> 01:19:27,560
and the control for CONIAD eighteen. One intuition you could have.

1364
01:19:29,199 --> 01:19:33,319
Speaker 3: Four when we get this robotic stuff is that it

1365
01:19:33,439 --> 01:19:35,439
takes quite a few years to go from the compelling

1366
01:19:35,520 --> 01:19:38,359
demo to actually being able to use in the real world.

1367
01:19:38,399 --> 01:19:41,359
So ten years ago you were really compelling demos of

1368
01:19:41,439 --> 01:19:45,039
self driving, but only now we have ROBOTAXI and Weimo

1369
01:19:45,119 --> 01:19:48,760
and all these services scaling up. Doesn't this Shouldn't this

1370
01:19:48,840 --> 01:19:53,800
make one pessimistic on say, household robots, because we don't

1371
01:19:53,840 --> 01:19:56,439
even quite have the compelling demos yet. I say they're

1372
01:19:56,439 --> 01:19:57,079
really advanced.

1373
01:19:57,119 --> 01:20:00,600
Speaker 1: Hand well, we've been working on human to robots now

1374
01:20:00,640 --> 01:20:05,640
for a while, so I guess spend five or six

1375
01:20:05,720 --> 01:20:09,840
years or something like that, and and a bunch of

1376
01:20:09,920 --> 01:20:12,119
things that we've done for the car are applicable to

1377
01:20:12,359 --> 01:20:17,239
the robot. So we'll use the same testa AI chips

1378
01:20:17,319 --> 01:20:21,439
in the in the robot as the car. We'll use

1379
01:20:21,479 --> 01:20:26,159
it the same basic principles. It's very much the same AI.

1380
01:20:26,920 --> 01:20:28,920
You've got, you know, many more degrees of freedom for

1381
01:20:28,960 --> 01:20:32,239
a robot than you do for a car. But really,

1382
01:20:32,239 --> 01:20:34,159
if you just think of for like as like a

1383
01:20:34,199 --> 01:20:38,880
boot stream, AI is really mostly compression and correlation of

1384
01:20:39,119 --> 01:20:42,800
two butt streams. You're you know, so if for video

1385
01:20:42,800 --> 01:20:45,760
you've got to do a tremendous amount of compression and

1386
01:20:46,000 --> 01:20:49,079
and and you've got to do the compression just right,

1387
01:20:49,119 --> 01:20:53,960
you've better compress the like ignore the things that don't matter,

1388
01:20:54,000 --> 01:20:56,840
and and like you don't care about the details of

1389
01:20:56,880 --> 01:20:58,560
the leaves and the tree on the side of the road,

1390
01:20:58,840 --> 01:21:01,680
but you care a lot about the the road signs

1391
01:21:01,680 --> 01:21:05,279
and the traffic lights and the pedestrians and even where

1392
01:21:05,600 --> 01:21:08,359
you know somebody in another cars is looking at you

1393
01:21:08,479 --> 01:21:10,600
or not looking at you, like there's some of these

1394
01:21:10,720 --> 01:21:13,000
some of these details matter a lot. So if it

1395
01:21:13,159 --> 01:21:15,720
is essentially it's going to turn that with the car's

1396
01:21:15,720 --> 01:21:17,119
going to turn that one and a half gigabytes a

1397
01:21:17,119 --> 01:21:21,119
second ultimately into two kilobytes second of control outputs, so

1398
01:21:22,159 --> 01:21:26,199
many stages of compression, and you got to get all

1399
01:21:26,199 --> 01:21:28,720
those stages right and then correlate those to the correct

1400
01:21:28,720 --> 01:21:31,560
control outputs. The robot has to do essentially the same thing.

1401
01:21:31,800 --> 01:21:33,640
And you think about what what humans? This is what

1402
01:21:33,680 --> 01:21:37,520
happens with humans. We really are photons in controls out,

1403
01:21:38,439 --> 01:21:41,119
so that that is the vast majority of your your

1404
01:21:41,199 --> 01:21:45,960
life has been vision photons in and then motor controls out.

1405
01:21:47,039 --> 01:21:51,159
Speaker 2: Naively, it seems like between humanoid robots and cars, the

1406
01:21:51,439 --> 01:21:54,039
fundamental actuators in a car are like how you turn

1407
01:21:54,119 --> 01:21:56,399
and how you accelerate, et cetera. And in a robot,

1408
01:21:56,520 --> 01:22:00,159
especially with maneuver arms, there's dozens and dozens of these

1409
01:22:00,479 --> 01:22:02,960
degrees of freedom. And then especially with Tesla, you had

1410
01:22:02,960 --> 01:22:05,359
this advantage of like you had millions and millions of

1411
01:22:05,479 --> 01:22:09,239
hours of human demo data collected from just the car

1412
01:22:09,359 --> 01:22:11,560
being out there, where like you can't equivalently just deploy

1413
01:22:11,640 --> 01:22:13,840
optimists that don't work and then get the data that way.

1414
01:22:13,880 --> 01:22:17,640
Speaker 1: So between the increased degrees of freedom and the far

1415
01:22:17,800 --> 01:22:22,079
sparser data, yes, let's go, how will you use the

1416
01:22:22,760 --> 01:22:28,000
sort of Tesla engine of intelligence on to train the

1417
01:22:28,039 --> 01:22:32,720
optimist mind. Now you're you're actually you're highlighting an important

1418
01:22:32,800 --> 01:22:36,079
limitation and difference between cars is like we we do

1419
01:22:36,239 --> 01:22:38,399
have oh we'll goon have like ten million cars on

1420
01:22:38,439 --> 01:22:43,760
the road, and so uh, that's it's hard to duplicate

1421
01:22:43,840 --> 01:22:51,199
that like massive training flight flywheel. For the robot. What

1422
01:22:51,359 --> 01:22:53,720
we're going to need to do is build a lot

1423
01:22:53,800 --> 01:22:55,359
of robots and put them in kind of like an

1424
01:22:55,399 --> 01:22:58,439
Optimist academy so they can do self play in reality.

1425
01:22:59,800 --> 01:23:02,560
So we're actually we're actually pulling that out, so we

1426
01:23:02,680 --> 01:23:06,800
can have at least ten thousand Optimus robots, maybe twenty

1427
01:23:06,880 --> 01:23:09,159
or thirty thousand that can do that, that are doing

1428
01:23:09,239 --> 01:23:15,000
self play and testing different tasks. And then the Tellesla

1429
01:23:16,479 --> 01:23:20,760
has quite a good reality generator, like the physics accurate

1430
01:23:20,840 --> 01:23:23,680
reality generator that we made this full of cars. We'll

1431
01:23:23,680 --> 01:23:26,479
do the same thing for the robots. And I actually

1432
01:23:26,520 --> 01:23:32,800
have done that for the robots. So you have a

1433
01:23:32,880 --> 01:23:36,640
few tens of thousands of humanoid robots doing different tasks,

1434
01:23:36,800 --> 01:23:39,560
and then you've got you can do millions of simulated

1435
01:23:39,640 --> 01:23:44,520
robots in the simulated world, and you use the tens

1436
01:23:44,560 --> 01:23:47,079
of thousands of robots in the real world to close

1437
01:23:47,159 --> 01:23:50,560
the simulation to reality gap, close the some material gap.

1438
01:23:51,079 --> 01:23:54,760
Speaker 2: How do you think about the synergies between Xai and Optimists,

1439
01:23:54,760 --> 01:23:56,680
given you're highlighting, look, you need this world model. You

1440
01:23:56,800 --> 01:23:59,479
maybe want to use some really smart intelligence. Is the

1441
01:23:59,479 --> 01:24:03,159
control player and so maybe GROC is like doing the

1442
01:24:03,239 --> 01:24:05,399
slower planning and like the motor policy.

1443
01:24:05,159 --> 01:24:07,960
Speaker 1: Is a little lower level. Yeah, what will the sort

1444
01:24:07,960 --> 01:24:13,079
of synergy between these things be? Yeah, so you just

1445
01:24:13,640 --> 01:24:18,359
GROCK would orchestrate the behavior of the Optimus robots. So le,

1446
01:24:18,439 --> 01:24:23,720
let's say you wanted to build a factory the then Optimists,

1447
01:24:23,880 --> 01:24:29,720
and then GROC could UH organize the optimost robots, give

1448
01:24:29,760 --> 01:24:33,439
them a sign them tasks UH to build the factory

1449
01:24:33,479 --> 01:24:35,640
for to produce whatever you want. Don't you need to

1450
01:24:35,720 --> 01:24:38,079
merge x Ai and Tessa then because these things end up?

1451
01:24:38,119 --> 01:24:42,119
So what are we saying only about all the company discussions? Well,

1452
01:24:42,119 --> 01:24:43,439
we're one, we're Gonnas in the elon.

1453
01:24:46,960 --> 01:24:49,640
Speaker 2: What are you waiting to see before you say we

1454
01:24:49,760 --> 01:24:52,239
want to manufacture one hundred thousand Optimists?

1455
01:24:52,359 --> 01:24:56,840
Speaker 1: Is it like Optimi? So since we're defining the proper nound,

1456
01:24:56,840 --> 01:24:59,039
we could define the pearl of the prop and noun too.

1457
01:24:59,640 --> 01:25:01,920
So we we're going to prop a dound the plural

1458
01:25:02,000 --> 01:25:02,800
and so it's optimize.

1459
01:25:02,840 --> 01:25:06,479
Speaker 2: Okay, is there something on the hardware side do you

1460
01:25:06,520 --> 01:25:08,199
want to see? Do you want to see better actuators

1461
01:25:08,359 --> 01:25:09,800
or is it just you want the software do better?

1462
01:25:10,000 --> 01:25:12,039
Whatever you're waiting for before we get like mass manufacturing

1463
01:25:12,239 --> 01:25:12,600
Gen three.

1464
01:25:13,239 --> 01:25:16,720
Speaker 1: Now we're moving towards that, we're going forward with semester manufactory.

1465
01:25:17,239 --> 01:25:20,479
But you think current current hardware is good enough that

1466
01:25:20,840 --> 01:25:22,159
you are going to you should you just want to

1467
01:25:22,159 --> 01:25:25,560
deploy as many as possible now, I mean it's very

1468
01:25:25,600 --> 01:25:28,840
hard to scale up production. But uh yeah, but I

1469
01:25:28,920 --> 01:25:32,119
think Optimus three is the right version of the robot

1470
01:25:32,239 --> 01:25:37,439
to you know, to produce maybe something on the order

1471
01:25:37,479 --> 01:25:40,119
of like a million units a year. I mean you

1472
01:25:40,119 --> 01:25:41,640
do want to go to Optimus four before you went

1473
01:25:41,680 --> 01:25:43,239
to ten millions a year, okay, but you can do

1474
01:25:43,279 --> 01:25:45,399
a million year at Optimist three. Uh yeah, I mean

1475
01:25:45,439 --> 01:25:51,159
it's very hard to spillet manufacturing. Yes, so like manufacturing. Uh,

1476
01:25:52,920 --> 01:25:56,680
like the the output period of time is always follows

1477
01:25:56,720 --> 01:26:00,359
at s cove, so it's at agonizingly slow. Then has

1478
01:26:00,640 --> 01:26:03,680
this sort of the eventually expecial increase then linear than

1479
01:26:03,720 --> 01:26:07,439
a than a you know logarithmic outcome to you sort

1480
01:26:07,439 --> 01:26:11,159
of eventually as it's a number. But optimists initial production

1481
01:26:11,239 --> 01:26:13,800
will be it's going to be a it's going to

1482
01:26:13,800 --> 01:26:16,600
be a stretched out s cove because so much of

1483
01:26:16,800 --> 01:26:19,039
what goes into Optimus is brand new. There's not an

1484
01:26:19,079 --> 01:26:22,960
existing supply chain. As I mentioned, the actually is like

1485
01:26:23,000 --> 01:26:27,199
trying to everything in the office robot is designed for

1486
01:26:27,319 --> 01:26:29,800
physics first principles, it's not. It's not taken from a catalog.

1487
01:26:30,439 --> 01:26:35,159
These these are custom designed everything, literally everything. I don't

1488
01:26:35,199 --> 01:26:37,119
think there's a single thing I done. How far down

1489
01:26:37,159 --> 01:26:40,760
the does aco I mean, I guess we're not making

1490
01:26:40,800 --> 01:26:47,880
custom capacitors yet maybe, but but there's there's there's nothing

1491
01:26:47,920 --> 01:26:51,039
you can pick out of a catalog at any price.

1492
01:26:52,079 --> 01:26:55,920
So so it just means that the the the optimistes cove,

1493
01:26:57,199 --> 01:27:02,920
uh the units units per awkward perion of time, how

1494
01:27:02,960 --> 01:27:06,840
many optics robusts you make per day. Whatever is is

1495
01:27:06,880 --> 01:27:11,680
gonna initially wrap slower than a product where you have

1496
01:27:11,880 --> 01:27:15,079
an existing supply chain, but it will get to a million.

1497
01:27:15,399 --> 01:27:18,159
Speaker 2: When you see these Chinese humanoids, like Industry or whatever

1498
01:27:18,840 --> 01:27:22,079
sells humanoidsed for like six K or thirteen K, do

1499
01:27:22,159 --> 01:27:26,319
you just like are you hoping to get your optimisms

1500
01:27:26,399 --> 01:27:28,680
bill of materials below that price so you can do

1501
01:27:28,760 --> 01:27:30,359
the same thing, or do you just think qualitatively they're

1502
01:27:30,399 --> 01:27:32,520
not the same thing, Like what do you think is going?

1503
01:27:32,640 --> 01:27:35,479
Speaker 1: Like what allows it? What allows them to self for solo?

1504
01:27:35,600 --> 01:27:39,199
And can me match that? Well, optimists, Our optous is

1505
01:27:39,560 --> 01:27:43,960
signed to have a lot of intelligence and to have

1506
01:27:44,079 --> 01:27:47,319
the same electro mechanical texterity, if not higher than a human.

1507
01:27:47,920 --> 01:27:52,640
So untry does not have that. And it's also i mean,

1508
01:27:52,680 --> 01:27:54,640
it's it's quite a it's quite a big robot. It's

1509
01:27:54,640 --> 01:27:58,600
just it's this matter it has to do, you know,

1510
01:27:58,840 --> 01:28:02,800
carry heavy objects for long periods of time and not

1511
01:28:02,920 --> 01:28:07,159
overheat or exceed the power of its sactuator. So so

1512
01:28:07,199 --> 01:28:10,720
we've got we've got, we've got you know, it's five eleven,

1513
01:28:10,840 --> 01:28:15,560
you know, it's pretty tall, and it's it's got a

1514
01:28:15,600 --> 01:28:17,640
lot of intelligence. So it's going to be more expensive

1515
01:28:17,760 --> 01:28:22,479
than a small robot that is not intelligence but more capable. Yeah,

1516
01:28:23,199 --> 01:28:25,800
a lot more. I mean, like we think is over time,

1517
01:28:26,600 --> 01:28:30,840
as often Showbus often robots, the class will drop very quickly.

1518
01:28:31,399 --> 01:28:34,840
Speaker 3: And what will these first billion optimists of some eye, Yeah,

1519
01:28:35,079 --> 01:28:37,640
do like bottle of their highest and best use.

1520
01:28:37,600 --> 01:28:39,920
Speaker 1: Be I think that you would start off with with

1521
01:28:40,000 --> 01:28:42,000
simple tasks that you can count on them doing well.

1522
01:28:42,159 --> 01:28:45,319
But in the home or in factories, like the best

1523
01:28:45,439 --> 01:28:51,800
useful robots in the beginning won't be anything any continuous operations,

1524
01:28:51,840 --> 01:28:54,279
only twenty four VI seven operation because then you're because

1525
01:28:54,600 --> 01:28:55,880
then they can work continuously.

1526
01:28:56,000 --> 01:28:57,880
Speaker 2: Yeah, what fraction of the work in a giga factory

1527
01:28:58,119 --> 01:29:00,159
that is currently done by humans going to gend three do?

1528
01:29:02,000 --> 01:29:06,159
Speaker 1: I'm I'm not sure. Maybe it's like ten bay more,

1529
01:29:06,239 --> 01:29:09,199
I don't know it's it. We would, we would use

1530
01:29:09,720 --> 01:29:11,840
We would not like reduce our headcount. We would we

1531
01:29:11,880 --> 01:29:14,560
would for sure it can increase our headcount to be clear,

1532
01:29:15,359 --> 01:29:18,920
but but we would increase our output. So the the

1533
01:29:19,800 --> 01:29:23,319
units produced per human, like total to total number humans

1534
01:29:23,319 --> 01:29:28,159
at teston will increase, but the the output of robots

1535
01:29:28,239 --> 01:29:34,319
and cars will increase, will increase disproportionate like much much. Yeah,

1536
01:29:35,640 --> 01:29:38,960
number of cars and robots produced per human will increase dramatically,

1537
01:29:39,239 --> 01:29:40,960
but number of humans will increase as well.

1538
01:29:42,079 --> 01:29:48,239
Speaker 3: We're talking about Chinese manufacturing a bunch here, and we're

1539
01:29:48,319 --> 01:29:50,920
also talking about you know, we've talked about some of

1540
01:29:51,039 --> 01:29:52,880
the policies that are ele US.

1541
01:29:52,960 --> 01:29:55,399
Speaker 1: Like you mentioned the the solar tariffs.

1542
01:29:55,520 --> 01:29:58,399
Speaker 3: Yeah, and you think they're a bad idea because you know,

1543
01:29:58,479 --> 01:30:00,199
we can scale up so over in the US.

1544
01:30:00,399 --> 01:30:04,520
Speaker 1: Well, just electricity output in the US needs to scale up, right,

1545
01:30:04,680 --> 01:30:08,960
you can with thus like goodsources to get it somehow. Yeah.

1546
01:30:09,119 --> 01:30:11,560
Speaker 3: But where I was going with this is if you

1547
01:30:11,640 --> 01:30:13,439
were in charge, if you were setting all the policies,

1548
01:30:14,119 --> 01:30:15,560
what else would you change?

1549
01:30:18,000 --> 01:30:21,159
Speaker 1: So you change the solar tariffs, Yeah, I would say

1550
01:30:21,680 --> 01:30:25,760
anything that is limiting factor for electricity needs to be addressed,

1551
01:30:25,760 --> 01:30:27,800
provided it's not like very bad for the environment.

1552
01:30:27,800 --> 01:30:30,079
Speaker 3: So presumably some permissing reforms and stuff as well, we'll

1553
01:30:30,119 --> 01:30:30,479
be in there.

1554
01:30:30,600 --> 01:30:32,880
Speaker 1: Yeah, there's a fair bit of permitting reforms that are happening.

1555
01:30:33,199 --> 01:30:36,600
A lot of the permitting is state based, so but

1556
01:30:36,920 --> 01:30:39,720
anything but but but this this administration is is good

1557
01:30:39,760 --> 01:30:45,319
at removing permaning roadblocks. And I'm not saying all tariffs

1558
01:30:45,319 --> 01:30:48,399
are bad. I'm just saying because so so Yeah, yeah,

1559
01:30:48,399 --> 01:30:51,680
I mean sometimes if like if another country is subsidizing

1560
01:30:51,960 --> 01:30:55,359
the output of something, then then you have to have

1561
01:30:55,520 --> 01:31:00,239
countervailing tariffs to protect dovestic industry against some see is

1562
01:31:00,239 --> 01:31:02,880
by another country. What else would you change? I don't

1563
01:31:02,920 --> 01:31:04,560
know if there's not much that the government can actually

1564
01:31:04,640 --> 01:31:05,520
do well.

1565
01:31:05,880 --> 01:31:07,880
Speaker 3: One thing I was wondering is it seems like the

1566
01:31:08,920 --> 01:31:15,239
for the policy goal of creating a leaves for the

1567
01:31:15,359 --> 01:31:18,640
US versus China, it seems like the export bands have

1568
01:31:18,800 --> 01:31:23,279
actually been quice impactful or China is not producing leading

1569
01:31:23,359 --> 01:31:26,960
edge chips and the export bands really by there China

1570
01:31:27,039 --> 01:31:31,560
is not producing leading edge turbine engines. And similarly, there's

1571
01:31:31,560 --> 01:31:33,319
a bunch of export bands that are relevant there on

1572
01:31:33,359 --> 01:31:37,039
some of the metallurgy. Should there be more export bands,

1573
01:31:37,159 --> 01:31:39,000
like as you think about things like when there are

1574
01:31:39,079 --> 01:31:41,399
now the drone industry and things like that, But is

1575
01:31:41,479 --> 01:31:42,840
that something they should be considered?

1576
01:31:43,279 --> 01:31:45,399
Speaker 1: Well, I think it's imporant to appreciate that in most

1577
01:31:45,479 --> 01:31:50,039
areas China is very advanced to actually there's only a

1578
01:31:50,079 --> 01:31:56,159
few areas where it is not. The China is a

1579
01:31:56,199 --> 01:31:59,880
manufacturer of horhouse next level like people don't it must

1580
01:31:59,920 --> 01:32:02,800
be it's very impressive. Yeah, yeah, I mean, if you

1581
01:32:02,880 --> 01:32:08,520
if you take like refining of ore, i'd say roughly,

1582
01:32:08,640 --> 01:32:12,680
China there's more just twice as much or refining of

1583
01:32:14,399 --> 01:32:19,119
on average as the rest of all combined. And I

1584
01:32:19,159 --> 01:32:21,760
think there's there's some areas, like say refining gallion which

1585
01:32:21,840 --> 01:32:24,960
goes into solar cells. I think there are like ninety

1586
01:32:25,000 --> 01:32:29,000
eight percent of gallum refining. So so China is actually

1587
01:32:29,079 --> 01:32:32,239
very advancing manufacturing in i'd say most areas.

1588
01:32:32,520 --> 01:32:35,560
Speaker 3: It seems like we're like there is discomfort with this

1589
01:32:35,880 --> 01:32:38,840
supply chain dependence and the ash. Nothing's really happening on it.

1590
01:32:39,479 --> 01:32:42,159
Supply chain, supply chain. It depends on say like the

1591
01:32:42,199 --> 01:32:43,640
Galleon refining that you're saying.

1592
01:32:43,760 --> 01:32:46,359
Speaker 1: Yeah, yeah, there's there's there's there's a there's all the

1593
01:32:46,479 --> 01:32:50,239
rare rare earth stuff, and yeah, where else we're sure,

1594
01:32:50,800 --> 01:32:53,600
as you know, not rare like we actually do rare

1595
01:32:53,680 --> 01:32:58,680
earth or mining in the US. Send the the the rock,

1596
01:33:00,079 --> 01:33:02,600
put it on a train, and then from the boats

1597
01:33:02,680 --> 01:33:06,399
China does another train and that goes to the where

1598
01:33:06,479 --> 01:33:09,960
with refining refiners in China who then refine it, put

1599
01:33:10,000 --> 01:33:11,680
it into a magnet, put it in just a modus

1600
01:33:11,680 --> 01:33:13,840
ofb assembly, and then set it back to America. So

1601
01:33:14,119 --> 01:33:17,680
the thing we're really missing a lot of ore refining

1602
01:33:18,560 --> 01:33:22,760
in America. And isn't this worth a policy intervention? Yes, well,

1603
01:33:24,079 --> 01:33:26,680
I think there are some things being done on that front,

1604
01:33:28,000 --> 01:33:34,039
but we kind of need optimists frankly to build our refineries.

1605
01:33:35,399 --> 01:33:37,479
So sorry, you think the main advantage is China has

1606
01:33:37,600 --> 01:33:41,039
is the abundance of skilled labor and that that that's like,

1607
01:33:41,279 --> 01:33:43,560
that's the thing optimist fact says. But they also we

1608
01:33:43,600 --> 01:33:46,000
need four times our populations.

1609
01:33:45,520 --> 01:33:47,640
Speaker 2: But we need so I mean, there's this concern if

1610
01:33:47,640 --> 01:33:50,720
you think that humans are the future, that like, okay,

1611
01:33:51,079 --> 01:33:54,159
right now, if it's the skilled labors for manufacturing that's

1612
01:33:54,279 --> 01:33:57,640
determining who's who can build more humanoids. You know, China

1613
01:33:57,680 --> 01:34:00,600
has more of those and manufactures more humanoids, therefore it

1614
01:34:00,720 --> 01:34:02,920
gets it gets the optimized future first.

1615
01:34:04,000 --> 01:34:06,399
Speaker 1: Well, it just like keeps that eximential going.

1616
01:34:06,520 --> 01:34:08,000
Speaker 2: It seems like you're sort of pointing out that sort

1617
01:34:08,039 --> 01:34:12,560
of getting to a million optimize, Yeah, requires the manufacturing

1618
01:34:12,600 --> 01:34:14,159
that the optimized supposed to help us get to.

1619
01:34:14,880 --> 01:34:18,079
Speaker 1: Right, you can you can close that recoursal loop pretty

1620
01:34:18,119 --> 01:34:21,680
quickly with a small number of optimize. Yeah, so you

1621
01:34:21,800 --> 01:34:24,960
close the close recurse of loop to help help the

1622
01:34:25,039 --> 01:34:28,239
robots buill the robots, and then we can, you know,

1623
01:34:28,319 --> 01:34:29,960
try to get to tens of millions of units. You're

1624
01:34:30,920 --> 01:34:32,960
maybe if you start getting too hundreds of millions of

1625
01:34:33,039 --> 01:34:35,720
units a year, I think you're you're going to be

1626
01:34:35,760 --> 01:34:38,640
the most competitive country by far. We definitely count win

1627
01:34:38,720 --> 01:34:41,680
with just humans because China has four times of population, right,

1628
01:34:41,840 --> 01:34:43,680
and frankly, America has been warning for so long that

1629
01:34:44,199 --> 01:34:46,239
you know, just like a like a post sports team

1630
01:34:46,239 --> 01:34:48,319
that's been warning for a very long time, tend to

1631
01:34:48,319 --> 01:34:51,880
get complacent and entitled, and that's why they stop winning

1632
01:34:52,560 --> 01:34:56,800
because it's, you know, don't work as hard anymore. So

1633
01:34:56,920 --> 01:35:00,800
I think the frankly, it just is the average work

1634
01:35:00,800 --> 01:35:02,960
ethic in China is higher than in the US. So

1635
01:35:03,000 --> 01:35:04,720
it's not just that there's four times the population, but

1636
01:35:04,800 --> 01:35:06,600
the work, the amount of work that people put in

1637
01:35:06,800 --> 01:35:10,399
is higher. So you can, like, you can try to

1638
01:35:10,439 --> 01:35:13,319
rearrange the humans, but you're still one quarter of the

1639
01:35:14,680 --> 01:35:18,359
you know, it's assuming that that productivities, that health is

1640
01:35:19,079 --> 01:35:20,640
the same, which I think actually it might not be.

1641
01:35:21,000 --> 01:35:25,319
They China might have advantage on productivity per person. We

1642
01:35:25,399 --> 01:35:27,840
will do one quarter of the amount of things as China,

1643
01:35:28,560 --> 01:35:31,680
so we can't win on the human front. And our

1644
01:35:31,720 --> 01:35:35,880
worth rates spent low for a long time, so our

1645
01:35:35,920 --> 01:35:38,800
birth rates spent the US birth rates spent low replacement

1646
01:35:39,439 --> 01:35:46,039
since roughly nineteen seventy one, so we've got a lot

1647
01:35:46,039 --> 01:35:50,039
of people retiring or you know, more people dying than

1648
01:35:50,960 --> 01:35:53,680
we're close to sort of more people domestically dying than

1649
01:35:53,760 --> 01:35:57,640
being born. So we definitely can't win on the human front,

1650
01:35:57,680 --> 01:35:59,760
but we might have a shot at the row wrote front.

1651
01:36:00,199 --> 01:36:02,560
Are there other things that you have wanted to manufacture

1652
01:36:02,600 --> 01:36:05,560
in the past but they've been too labor intensive or

1653
01:36:05,600 --> 01:36:08,119
too expensive that now you can come back to that

1654
01:36:08,159 --> 01:36:12,079
and say, oh, we can finally do the whatever because

1655
01:36:12,119 --> 01:36:14,520
we have optimists. Yeah, I think we'd like to do

1656
01:36:14,720 --> 01:36:19,359
more both more or refineries at Tesla. So we just

1657
01:36:19,439 --> 01:36:26,720
completed construction and have begun liftium refining without litium refinery

1658
01:36:26,840 --> 01:36:31,760
and Corpus Christi, Texas. We have a nickel refinery which

1659
01:36:31,800 --> 01:36:36,439
is called the Cathode. Uh that's here in Austin. And

1660
01:36:37,880 --> 01:36:40,079
these are these are the largest is the largest cathode.

1661
01:36:40,399 --> 01:36:43,800
There's the largest cathod refinery, largest liftingum refinery, the largest

1662
01:36:43,840 --> 01:36:50,039
nekelink and lifting refinery outside of China. And it's like

1663
01:36:50,159 --> 01:36:53,720
they know, the Cathote team would say, like, we have

1664
01:36:54,239 --> 01:36:58,079
the largest and the only actually cather refinery in America.

1665
01:36:58,720 --> 01:37:00,920
Many superatives not just the just but it's also the

1666
01:37:01,159 --> 01:37:04,560
moly so it was pretty big, even though it's the

1667
01:37:04,640 --> 01:37:07,880
only one. But I mean there are other things that

1668
01:37:09,239 --> 01:37:14,920
you know, you could do a lot more refineries and

1669
01:37:15,119 --> 01:37:18,880
and help the the help America being more competitive on

1670
01:37:19,399 --> 01:37:23,560
refining capacity. So so that there's like there's basically a

1671
01:37:23,600 --> 01:37:26,800
lot of work for the Optimity to do that that

1672
01:37:27,000 --> 01:37:30,439
most Americans, very few Americans, frankly wanted to. I mean,

1673
01:37:30,479 --> 01:37:34,319
i've I've Actually it's the refining work too dirty. It's

1674
01:37:34,399 --> 01:37:37,880
it's not it's actually, no, we don't. There's not we

1675
01:37:37,920 --> 01:37:41,960
don't have toxic commissions from the refinery or anything. The

1676
01:37:42,039 --> 01:37:45,199
Catholic make refinery is right right sort of in Travis County.

1677
01:37:45,319 --> 01:37:48,079
Like why can't I mantes from why can't you with humans? No?

1678
01:37:48,479 --> 01:37:50,600
You can you find out of humans? I see, Okay, yeah,

1679
01:37:51,520 --> 01:37:53,279
Like no matter what you do, you have one quarter

1680
01:37:53,399 --> 01:37:55,479
number of humans in America in China. So if you

1681
01:37:55,800 --> 01:37:57,319
help them do this thing, they can do the other thing.

1682
01:37:58,039 --> 01:38:00,800
So so then then well, how do you how do

1683
01:38:00,800 --> 01:38:03,000
you build this refining refining capacity? Well you can do

1684
01:38:03,000 --> 01:38:07,720
it with the Optimaya and not many, not very many,

1685
01:38:08,000 --> 01:38:11,560
not very many Americans. Thanks for listening. See you in

1686
01:38:11,640 --> 01:38:12,560
the next episode.