1
00:00:03,420 --> 00:00:06,560
Josh:
One of the most important technologies in the world that is happening as we

2
00:00:06,560 --> 00:00:11,160
Josh:
speak every day is the rise of autonomy, and particularly around autonomous robots.

3
00:00:11,380 --> 00:00:14,440
Josh:
Robots can be many things. Robots can be humanoids, they can be cars.

4
00:00:14,660 --> 00:00:17,820
Josh:
And today we're going to talk about both, because there's one company that is

5
00:00:17,820 --> 00:00:21,140
Josh:
at the frontier of both of those areas, and that's Tesla.

6
00:00:21,320 --> 00:00:27,000
Josh:
Tesla has the most unbelievable set of autopilot software that I think exists in the world.

7
00:00:27,140 --> 00:00:30,740
Josh:
I've been using it personally for eight years now. And it's been amazing to

8
00:00:30,740 --> 00:00:31,560
Josh:
see how good it's gotten.

9
00:00:31,820 --> 00:00:34,920
Josh:
And EJS, now there's, for the first time ever, we have the secrets.

10
00:00:35,120 --> 00:00:39,460
Josh:
The secret sauce that shares exactly how they've been able to get autonomy this

11
00:00:39,460 --> 00:00:40,420
Josh:
powerful, this impressive.

12
00:00:40,660 --> 00:00:44,820
Josh:
And there's now very clearly a world in which I can imagine waking up in the

13
00:00:44,820 --> 00:00:47,380
Josh:
morning, getting ready to go to work, stepping outside, and there's a cyber

14
00:00:47,380 --> 00:00:50,820
Josh:
cab waiting for me outside that will just take me wherever I want for a fraction

15
00:00:50,820 --> 00:00:53,640
Josh:
of the cost that it takes for a normal driver.

16
00:00:53,840 --> 00:00:57,820
Josh:
And I think this is an incredibly powerful unlock and to see a behind the scenes

17
00:00:57,820 --> 00:01:02,040
Josh:
of this is awesome so the entire episode today is behind the scenes of the most

18
00:01:02,040 --> 00:01:05,140
Josh:
impressive new front-end tier technology that exists.

19
00:01:05,140 --> 00:01:10,960
Ejaaz:
I think what i'm most excited about today josh is the fact that i've always

20
00:01:10,960 --> 00:01:15,380
Ejaaz:
thought tesla ai and robotics is so cool but i i just don't know how any of

21
00:01:15,380 --> 00:01:19,920
Ejaaz:
this works and they've refused to tell us and finally they've they've spilt their secrets today

22
00:01:20,620 --> 00:01:24,420
Ejaaz:
to quickly paint some context for the listeners here, up until yesterday,

23
00:01:24,420 --> 00:01:29,460
Ejaaz:
we only thought of Tesla AI as something called a neural network.

24
00:01:29,580 --> 00:01:33,340
Ejaaz:
That's their secret source. And a neural network can be thought of as a software

25
00:01:33,340 --> 00:01:37,260
Ejaaz:
program that is designed to function like the human brain.

26
00:01:37,520 --> 00:01:41,580
Ejaaz:
So it takes in information and it discovers patterns, trends,

27
00:01:41,760 --> 00:01:43,420
Ejaaz:
and it can also sometimes make predictions.

28
00:01:43,820 --> 00:01:47,560
Ejaaz:
Now, this contrasts directly to some of Tesla's competitors,

29
00:01:47,560 --> 00:01:50,740
Ejaaz:
which do self-driving and robotics in a very different way.

30
00:01:50,900 --> 00:01:54,200
Ejaaz:
They take more modular and sensor-driven approaches, right?

31
00:01:54,480 --> 00:01:59,820
Ejaaz:
The reason why Tesla's neural network is so special is they have an end-to-end

32
00:01:59,820 --> 00:02:04,200
Ejaaz:
neural network, which means that they feed a bunch of raw data from one side

33
00:02:04,200 --> 00:02:07,360
Ejaaz:
and out comes the output, which is an action.

34
00:02:07,460 --> 00:02:11,620
Ejaaz:
In this case of Tesla cars, it would be driving, steering, and acceleration.

35
00:02:12,080 --> 00:02:15,080
Ejaaz:
And they took this approach for a few different ways.

36
00:02:15,180 --> 00:02:21,180
Ejaaz:
The most important being, it's really hard Josh to codify what human values

37
00:02:21,180 --> 00:02:25,080
Ejaaz:
are and what I mean by that is let's say in this example that you're seeing

38
00:02:25,080 --> 00:02:30,740
Ejaaz:
on your screen right now you are driving your car and there's a massive puddle on your lane but

39
00:02:30,740 --> 00:02:32,060
Josh:
You see that you.

40
00:02:32,060 --> 00:02:36,780
Ejaaz:
Could potentially drive into the oncoming lane to skirt around it now for humans

41
00:02:36,780 --> 00:02:39,760
Ejaaz:
it's really easy to do that right it's like okay maybe I should just go through

42
00:02:39,760 --> 00:02:45,360
Ejaaz:
it because there's no cars coming but for a machine to do that it requires a

43
00:02:45,360 --> 00:02:47,140
Ejaaz:
lot of effort. It's hard to hard code.

44
00:02:47,280 --> 00:02:50,540
Ejaaz:
So that's one special thing around the neural network. But Josh,

45
00:02:50,720 --> 00:02:52,180
Ejaaz:
I want to jump into the secrets.

46
00:02:52,420 --> 00:02:54,360
Ejaaz:
Can you lead us with the first one?

47
00:02:54,910 --> 00:02:57,490
Josh:
Well, what you mentioned is really important, the end-to-end stuff.

48
00:02:57,710 --> 00:02:59,750
Josh:
And I want to walk through a little experiment.

49
00:02:59,890 --> 00:03:02,670
Josh:
So when you kick a soccer ball, I think this is an experience everyone's kind

50
00:03:02,670 --> 00:03:05,770
Josh:
of went through, right? What do you do when you kick a soccer ball?

51
00:03:05,970 --> 00:03:11,870
Ejaaz:
Yeah, I see the soccer ball coming towards me. I kind of prepare my legs ready to kind of kick.

52
00:03:12,010 --> 00:03:14,390
Ejaaz:
I'm right-footed, so I'm kicking with my right foot.

53
00:03:14,750 --> 00:03:19,390
Ejaaz:
And then I guess the rest is kind of intuitive, Josh. I just kind of run up to it and kick it.

54
00:03:19,390 --> 00:03:23,430
Josh:
Yeah, yeah. And I think that's exactly the point is when you kick a soccer ball,

55
00:03:23,490 --> 00:03:24,930
Josh:
this is something a lot of people have experienced.

56
00:03:25,090 --> 00:03:27,950
Josh:
You're not actually thinking about all the parts of kicking a soccer ball.

57
00:03:28,010 --> 00:03:31,490
Josh:
You're not thinking of where it is on the ground, where your ankle is,

58
00:03:31,570 --> 00:03:34,050
Josh:
where your knee is, where your leg is, the positioning, how hard you're going

59
00:03:34,050 --> 00:03:35,290
Josh:
to kick it. It just feels very intuitive.

60
00:03:35,730 --> 00:03:40,390
Josh:
And with a lot of other car companies, they're hard coding these intuitions as code.

61
00:03:40,650 --> 00:03:43,730
Josh:
So it does have to think about each section. It does have to calculate each

62
00:03:43,730 --> 00:03:46,510
Josh:
section. And what's different about Tesla and what we learned from this article,

63
00:03:46,550 --> 00:03:49,990
Josh:
this is from Ashok, who is the person who's in charge of Tesla AI,

64
00:03:50,290 --> 00:03:52,630
Josh:
is that they use this thing called end-to-end neural networks.

65
00:03:52,670 --> 00:03:56,610
Josh:
And what does that mean? In like a fun, simple way, it's basically the intuition

66
00:03:56,610 --> 00:03:59,670
Josh:
that you just described with kicking a soccer ball, the AI model,

67
00:03:59,850 --> 00:04:01,690
Josh:
the chip on a car is able to emulate that.

68
00:04:01,810 --> 00:04:05,170
Josh:
So instead of making these minute decisions all the way through a fixed decision

69
00:04:05,170 --> 00:04:08,490
Josh:
tree, they're able to take a ton of data and use these things that we've learned

70
00:04:08,490 --> 00:04:13,350
Josh:
over time, which are gradients and weights, and basically move the gradients

71
00:04:13,350 --> 00:04:16,210
Josh:
and weights throughout the decision process to reach an end goal.

72
00:04:16,390 --> 00:04:20,370
Josh:
So if the end goal is to kick a soccer ball, there's a very clear stated end goal.

73
00:04:20,510 --> 00:04:24,570
Josh:
And the neural network's job is to figure out the full sweep of gradients as

74
00:04:24,570 --> 00:04:26,310
Josh:
it goes across to get to that end goal.

75
00:04:26,410 --> 00:04:29,390
Josh:
And it uses a bunch of this training data that they collect in order to get there.

76
00:04:29,510 --> 00:04:32,530
Josh:
So this is this remarkable technology that breakthrough that they have.

77
00:04:32,610 --> 00:04:34,150
Josh:
And they have some really interesting examples here.

78
00:04:34,310 --> 00:04:37,570
Josh:
So in the case of the ducks, like we're looking at an example on the screen

79
00:04:37,570 --> 00:04:39,810
Josh:
right now, there's ducks standing in the middle of the road.

80
00:04:40,400 --> 00:04:44,060
Josh:
When you're coding an AI system, when you're coding a car, you're not hard coding

81
00:04:44,060 --> 00:04:46,000
Josh:
in, if you see ducks, do this.

82
00:04:46,220 --> 00:04:49,360
Josh:
What the car is understanding intuitively is like, okay, there's an obstacle

83
00:04:49,360 --> 00:04:51,660
Josh:
here and they are ducks. They're not moving.

84
00:04:52,060 --> 00:04:56,540
Josh:
The interesting thing is the example above is the car recognizes that the ducks

85
00:04:56,540 --> 00:04:57,760
Josh:
are actually moving across the road.

86
00:04:57,840 --> 00:05:00,520
Josh:
So it knows to wait and then it could pass once they've moved.

87
00:05:00,620 --> 00:05:02,700
Josh:
But the second one, it notices they're just kind of chilling.

88
00:05:03,020 --> 00:05:06,660
Josh:
The ducks aren't going anywhere. And what does it do? It understands that intuitively

89
00:05:06,660 --> 00:05:10,260
Josh:
and it is able to back up and then move around them. And that's the difference

90
00:05:10,260 --> 00:05:13,840
Josh:
in how Tesla does it versus some other companies is they're not hard coding

91
00:05:13,840 --> 00:05:15,060
Josh:
a series of fixed parameters.

92
00:05:15,280 --> 00:05:17,580
Josh:
They are doing it all entirely through these neural networks.

93
00:05:17,720 --> 00:05:22,640
Ejaaz:
If we move on to secret number one, Josh, it kind of explains how they're able

94
00:05:22,640 --> 00:05:24,900
Ejaaz:
to achieve this at a pretty high level, right?

95
00:05:25,420 --> 00:05:30,080
Ejaaz:
So it's titled The Curse of Dimensionality. And what it basically describes

96
00:05:30,080 --> 00:05:35,880
Ejaaz:
is you can imagine for a car to self-drive, it requires a ton of data.

97
00:05:35,880 --> 00:05:39,260
Ejaaz:
I think Tesla, the average car, has about seven cameras.

98
00:05:39,300 --> 00:05:43,740
Ejaaz:
It ingests a ton of audio data, a ton of navigation GPS data,

99
00:05:44,020 --> 00:05:46,260
Ejaaz:
and kinematics. So speed is tracking your speed.

100
00:05:46,820 --> 00:05:50,680
Ejaaz:
And so all this data is roughly equivalent to 2 billion tokens.

101
00:05:51,020 --> 00:05:54,500
Ejaaz:
And if you think about it, it needs to run through this end-to-end neural network

102
00:05:54,500 --> 00:05:58,620
Ejaaz:
that you just described, Josh, and it needs to output pretty much two tokens.

103
00:05:58,920 --> 00:06:03,480
Ejaaz:
One token, which determines which way the car should steer, and the other token

104
00:06:03,480 --> 00:06:07,520
Ejaaz:
determining how fast should that car be at that point? Should it decelerate

105
00:06:07,520 --> 00:06:08,900
Ejaaz:
or should it accelerate?

106
00:06:09,220 --> 00:06:13,500
Ejaaz:
And you can imagine this is an incredibly nuanced and complex process.

107
00:06:13,500 --> 00:06:18,820
Ejaaz:
And the way that the Tesla neural engine or the neural network is designed is

108
00:06:18,820 --> 00:06:23,340
Ejaaz:
it has really special data lanes that process this data in a very nuanced way

109
00:06:23,340 --> 00:06:28,160
Ejaaz:
to understand what exactly it needs to map onto when it comes to steering and acceleration.

110
00:06:28,160 --> 00:06:33,020
Ejaaz:
Now, you might think that's pretty cool, but Tesla's secret source when it comes

111
00:06:33,020 --> 00:06:36,700
Ejaaz:
to this particular component is the driving data, right, Josh?

112
00:06:37,200 --> 00:06:41,780
Ejaaz:
So they get access to all the camera data, audio data, GPS data that I just

113
00:06:41,780 --> 00:06:45,500
Ejaaz:
mentioned from their entire fleet of Tesla cars.

114
00:06:45,620 --> 00:06:50,300
Ejaaz:
So the equivalent of data that they get every day is something crazy like 500

115
00:06:50,300 --> 00:06:52,120
Ejaaz:
years worth of driving data.

116
00:06:52,320 --> 00:06:55,920
Ejaaz:
Now, you can imagine if it processes this amount of rich data,

117
00:06:56,060 --> 00:06:59,940
Ejaaz:
and not all of that data is important, right? It's kind of like the same kind of standard things.

118
00:07:00,480 --> 00:07:02,600
Josh:
Over those years of data.

119
00:07:02,800 --> 00:07:07,780
Ejaaz:
You get access to the one or two random nuanced incidents which feed in and

120
00:07:07,780 --> 00:07:11,380
Ejaaz:
improve the collective intelligence of the entire Tesla fleet.

121
00:07:11,540 --> 00:07:14,580
Ejaaz:
So whether you're on the other side of the world driving a Tesla or you're in

122
00:07:14,580 --> 00:07:17,820
Ejaaz:
the local neighborhood, you still benefit from the same types of improvements.

123
00:07:18,300 --> 00:07:20,960
Josh:
I want to talk a little bit about the scale because you mentioned 2 billion

124
00:07:20,960 --> 00:07:25,160
Josh:
inputs and it's kind of difficult to comprehend what 2 billion actually means.

125
00:07:25,160 --> 00:07:31,360
Josh:
And as a good example, I want you to imagine your phone processing every TikTok

126
00:07:31,360 --> 00:07:35,640
Josh:
that exists on the platform every single second in order to determine the next

127
00:07:35,640 --> 00:07:37,480
Josh:
turn. That is two billion inputs.

128
00:07:37,660 --> 00:07:41,200
Josh:
It is an astronomical amount of data. You're basically, you take the whole TikTok

129
00:07:41,200 --> 00:07:45,900
Josh:
catalog every second in order to make every decision and you distill that entire

130
00:07:45,900 --> 00:07:48,060
Josh:
data set into two single points.

131
00:07:48,160 --> 00:07:52,140
Josh:
And it's just, it's a remarkable amount of compression and then a remarkable

132
00:07:52,140 --> 00:07:55,360
Josh:
amount of precision to make the right decision over and over and over again,

133
00:07:55,480 --> 00:07:57,380
Josh:
and then adjust and calculate as things change.

134
00:07:58,190 --> 00:08:01,110
Josh:
The way that they do this, they're not doing this raw. They're not actually

135
00:08:01,110 --> 00:08:02,270
Josh:
ingesting all this data.

136
00:08:02,570 --> 00:08:06,690
Josh:
They have this data curation process that they use in order to help them kind

137
00:08:06,690 --> 00:08:08,770
Josh:
of figure out what is important and what is just noise.

138
00:08:09,070 --> 00:08:11,610
Josh:
And what they do, and we have a great example on screen here,

139
00:08:11,790 --> 00:08:13,530
Josh:
is they pick the juiciest clips.

140
00:08:13,670 --> 00:08:17,450
Josh:
It's like kind of curating like a viral playlist and they use it to train the

141
00:08:17,450 --> 00:08:18,610
Josh:
AI on these weird scenarios.

142
00:08:18,750 --> 00:08:22,430
Josh:
So we're seeing on the screen, there's someone rolling through an intersection of wheelchair.

143
00:08:22,630 --> 00:08:26,750
Josh:
It's actually very funny to see and scary to see what types of things happen.

144
00:08:26,750 --> 00:08:29,550
Josh:
I mean, this is crazy. Two cars crashing right in front of you,

145
00:08:30,010 --> 00:08:31,610
Josh:
driving on a snow blind street.

146
00:08:31,950 --> 00:08:33,670
Josh:
There's kids that are running out in the middle of the road.

147
00:08:33,950 --> 00:08:38,350
Josh:
There's these tremendous amount of edge cases that are really difficult to understand.

148
00:08:38,550 --> 00:08:43,010
Josh:
And because of the 500 years of driving data every single day that they ingest,

149
00:08:43,130 --> 00:08:45,670
Josh:
they're able to analyze and to kind of sift through.

150
00:08:45,810 --> 00:08:48,490
Josh:
And they've come up with systems to curate the most viral clips,

151
00:08:48,810 --> 00:08:54,630
Josh:
not viral, but the clips with the most implications of safety that are kind of the weird edge cases.

152
00:08:54,870 --> 00:08:57,130
Josh:
And then we have this example here. Do you want to walk through the chart that

153
00:08:57,130 --> 00:09:01,150
Josh:
we're seeing, because it's really fascinating how the car can kind of see it before the human does.

154
00:09:01,810 --> 00:09:06,410
Ejaaz:
Yeah. So what's interesting is when I first watched this clip and for those who are listening,

155
00:09:06,750 --> 00:09:12,670
Ejaaz:
it is a car driving on a very rainy evening on the highway and a car in front

156
00:09:12,670 --> 00:09:17,630
Ejaaz:
of it kind of crashes out and goes and starts to spin and kind of enter its own lane.

157
00:09:17,970 --> 00:09:21,410
Ejaaz:
When I first watched this video, Josh, I didn't even notice the car spinning

158
00:09:21,410 --> 00:09:24,130
Ejaaz:
out because it happens so far away.

159
00:09:24,230 --> 00:09:29,010
Ejaaz:
And so what's effective about this particular video is, given everything that

160
00:09:29,010 --> 00:09:33,870
Ejaaz:
you just described, the Tesla self-driving software and machinery is able to

161
00:09:33,870 --> 00:09:38,950
Ejaaz:
detect things that you necessarily as a human aren't able to do this.

162
00:09:39,090 --> 00:09:42,470
Ejaaz:
This graph specifically, Josh, can you explain what I'm looking at here?

163
00:09:43,120 --> 00:09:46,460
Josh:
Yeah, so this is the gradient. This is the weighted decision tree in real time.

164
00:09:46,600 --> 00:09:49,380
Josh:
So you could kind of see every single frame that it receives,

165
00:09:49,680 --> 00:09:52,780
Josh:
the chart moves, and then you could actually see the point in which it realizes

166
00:09:52,780 --> 00:09:54,660
Josh:
there's a threat and it adjusts very quickly.

167
00:09:54,820 --> 00:09:57,940
Josh:
So what you're seeing here is the real time visual representation of what the

168
00:09:57,940 --> 00:10:00,920
Josh:
brain sees. And we're going to get into this a little bit later where you can

169
00:10:00,920 --> 00:10:02,340
Josh:
actually communicate with this system.

170
00:10:02,520 --> 00:10:04,940
Josh:
You could talk to it just like it's a large language model. It's pretty insane.

171
00:10:05,080 --> 00:10:07,180
Josh:
But I want to move on to the next section because this is my favorite.

172
00:10:07,540 --> 00:10:10,400
Josh:
When I saw this, it just really blew my mind on how they

173
00:10:10,400 --> 00:10:13,600
Josh:
were able to basically emulate real world

174
00:10:13,600 --> 00:10:16,480
Josh:
driving scenarios and each as I want to start this section with an

175
00:10:16,480 --> 00:10:18,540
Josh:
example that they showed if you don't mind scrolling down and sharing the one

176
00:10:18,540 --> 00:10:23,640
Josh:
of the the fake screen so after these splats there's one a little bit later

177
00:10:23,640 --> 00:10:28,800
Josh:
and basically it shows a driving further down even sorry the like next section

178
00:10:28,800 --> 00:10:34,980
Josh:
then we'll go right back up oh sure sure yeah this one yeah yeah so this example

179
00:10:34,980 --> 00:10:36,340
Josh:
that we're looking at on the screen.

180
00:10:36,460 --> 00:10:39,860
Josh:
This looks like a standard traditional driving setup.

181
00:10:40,100 --> 00:10:44,180
Josh:
So the car has, what is that, seven cameras and each one of them ingest data.

182
00:10:44,660 --> 00:10:47,520
Josh:
The thing with this EJAS is what you're seeing on screen is not real.

183
00:10:47,680 --> 00:10:52,500
Josh:
That is a 100% virtual representation of this real world.

184
00:10:52,620 --> 00:10:55,880
Josh:
And it's unbelievable because it looks so good. And as I'm watching this,

185
00:10:55,980 --> 00:11:00,480
Josh:
I'm like, man, I hope GTA 6 looks like this because the quality,

186
00:11:00,680 --> 00:11:04,420
Josh:
the fidelity of this artificially generated world is indistinguishable from

187
00:11:04,420 --> 00:11:05,760
Josh:
real life, the entire thing.

188
00:11:06,060 --> 00:11:09,160
Josh:
And the reason they're able to do this is by ingesting all this data.

189
00:11:09,300 --> 00:11:12,760
Josh:
So now that you've seen how impressive it gets, this is kind of how they build

190
00:11:12,760 --> 00:11:14,700
Josh:
it. So we can go back up to the Gaussian splatting examples.

191
00:11:15,060 --> 00:11:19,160
Josh:
And Gaussian splats are kind of a fancy way of saying, as the car drives through,

192
00:11:19,560 --> 00:11:21,980
Josh:
you could imagine the cameras as scanners.

193
00:11:22,160 --> 00:11:26,900
Josh:
So if you flipped a camera into a scanner, it maps this 3D world and creates a world.

194
00:11:27,040 --> 00:11:31,040
Josh:
And then they're actually able to move around and navigate the 3D world they

195
00:11:31,040 --> 00:11:33,160
Josh:
create using just the cameras on your car.

196
00:11:33,280 --> 00:11:35,980
Josh:
And I want to reiterate that every Tesla you see on the road,

197
00:11:36,180 --> 00:11:40,580
Josh:
regardless of when it was made, is capable of collecting this data and creating

198
00:11:40,580 --> 00:11:42,660
Josh:
these 3D models that you see on the screen. So...

199
00:11:43,120 --> 00:11:45,320
Josh:
The interesting thing here is that top bar is what the car sees.

200
00:11:45,580 --> 00:11:48,060
Josh:
The bottom bar is what the car is generated to see.

201
00:11:48,180 --> 00:11:51,280
Josh:
And what it's able to do as a result is it's able to kind of get a better understanding

202
00:11:51,280 --> 00:11:55,360
Josh:
of the world around it and make much better decisions that in turn make it much

203
00:11:55,360 --> 00:11:56,840
Josh:
safer than a human driver does.

204
00:11:56,960 --> 00:12:01,800
Ejaaz:
This just looks like a computer game, Josh. Like one of those massive MMORPGs

205
00:12:01,800 --> 00:12:05,700
Ejaaz:
that kind of generates the world as I navigate and move through it as I interact

206
00:12:05,700 --> 00:12:06,640
Ejaaz:
with different characters.

207
00:12:06,980 --> 00:12:11,880
Ejaaz:
This is kind of that, but for self-driving specifically. And why I think this is so cool,

208
00:12:12,000 --> 00:12:15,540
Ejaaz:
and these are kind of like widely known as world simulators,

209
00:12:15,540 --> 00:12:20,880
Ejaaz:
it's like an AI model that generates simulated realities, is that this data

210
00:12:20,880 --> 00:12:25,960
Ejaaz:
can be modified in so many different ways and so many different scenarios to

211
00:12:25,960 --> 00:12:29,620
Ejaaz:
train the car for experiences or accidents that it hasn't even,

212
00:12:29,720 --> 00:12:32,120
Ejaaz:
that hasn't even encountered just yet.

213
00:12:32,120 --> 00:12:36,260
Ejaaz:
And this is really cool because I think one major constraint that a lot of AI

214
00:12:36,260 --> 00:12:40,680
Ejaaz:
models and self-driving models come up against is sometimes there's not enough

215
00:12:40,680 --> 00:12:44,000
Ejaaz:
data to account for every single different type of scenario.

216
00:12:44,420 --> 00:12:48,700
Ejaaz:
So a way to kind of address that is to create something known as synthetic data.

217
00:12:49,140 --> 00:12:53,780
Ejaaz:
World simulators is one step towards being able to do that super effectively

218
00:12:53,780 --> 00:12:59,300
Ejaaz:
whilst bending this simulated reality to how the actual world works,

219
00:12:59,400 --> 00:13:03,060
Ejaaz:
right? Physics is super important, but hard to translate into an AI model.

220
00:13:03,260 --> 00:13:07,700
Ejaaz:
And so seeing something like this at scale for a product, a car,

221
00:13:07,860 --> 00:13:13,660
Ejaaz:
that is used by almost every human on the world is just so amazing to see.

222
00:13:13,820 --> 00:13:16,320
Josh:
And the answer to the question, well, why hasn't everybody done this?

223
00:13:16,400 --> 00:13:19,440
Josh:
Is because to generate these world models generally takes tens of seconds to do.

224
00:13:19,840 --> 00:13:24,320
Josh:
Tesla's figured out a way to do it in 0.2 seconds. So it's a remarkable efficiency

225
00:13:24,320 --> 00:13:26,480
Josh:
improvement that allows them to actually do this.

226
00:13:26,560 --> 00:13:28,220
Josh:
It's not like the rest of the world doesn't want to do this.

227
00:13:28,220 --> 00:13:31,040
Josh:
Is that technically speaking, it's just very, very difficult to do.

228
00:13:31,180 --> 00:13:35,200
Josh:
And the next example they shared was one of my favorite ones because it really just created.

229
00:13:35,890 --> 00:13:39,190
Josh:
It made it feel very familiar where you can actually talk to these models like

230
00:13:39,190 --> 00:13:39,870
Josh:
they're a language model.

231
00:13:40,190 --> 00:13:43,890
Josh:
Yeah. And the example above where you could just say, well, why are you not turning left?

232
00:13:44,210 --> 00:13:48,130
Josh:
And it will explain to you, well, there's a detour sign. And why shouldn't you

233
00:13:48,130 --> 00:13:50,450
Josh:
turn right? Well, because the detour sign is pointing to the left.

234
00:13:50,730 --> 00:13:55,350
Josh:
And it really, you start to get a sense the same way yesterday in our episode

235
00:13:55,350 --> 00:13:59,190
Josh:
yesterday, where you can see the behind the scenes of how the model thinks when it trades.

236
00:13:59,430 --> 00:14:03,290
Josh:
You can now see the behind the scenes of the brain and you could start to understand

237
00:14:03,290 --> 00:14:05,110
Josh:
how it works, why it works, how it's reasoning.

238
00:14:05,390 --> 00:14:09,070
Josh:
And the results from this is pretty fascinating. It's not only is it interpreting

239
00:14:09,070 --> 00:14:12,330
Josh:
inputs like where the lines on the road are, but it's also able to read signs.

240
00:14:12,390 --> 00:14:16,310
Josh:
They have an example where you're able to see a human who's like kind of giving

241
00:14:16,310 --> 00:14:18,450
Josh:
you a high five, like saying, wait one second, I'm about to pull out.

242
00:14:18,550 --> 00:14:19,910
Josh:
And then the car recognizes that and stops.

243
00:14:20,150 --> 00:14:22,830
Josh:
So there's these like unbelievable improvements that they have.

244
00:14:22,910 --> 00:14:27,570
Josh:
And this section I want to get into next is because they can reevaluate these

245
00:14:27,570 --> 00:14:30,350
Josh:
new decision trees on existing historical models.

246
00:14:30,510 --> 00:14:34,330
Josh:
So my car, I've had a few near collision experiences that have been a little

247
00:14:34,330 --> 00:14:35,950
Josh:
scary, but they've been narrowly avoided.

248
00:14:36,170 --> 00:14:39,510
Josh:
What they can do is they can actually take the exact camera inputs from the

249
00:14:39,510 --> 00:14:42,790
Josh:
car and emulate if the collision had actually happened.

250
00:14:42,990 --> 00:14:46,850
Josh:
And then they could run these new tests on it and see how the new models would

251
00:14:46,850 --> 00:14:47,670
Josh:
compare to the old models.

252
00:14:47,830 --> 00:14:51,750
Josh:
So in the case that you narrowly miss an accident, well, you could test it on

253
00:14:51,750 --> 00:14:54,490
Josh:
a new model and see if it does better. And in the first example, it does.

254
00:14:54,650 --> 00:14:56,950
Josh:
And it actually moves away faster than the others.

255
00:14:57,150 --> 00:15:01,270
Josh:
The second example that they have here is that you can create artificial examples.

256
00:15:01,610 --> 00:15:05,010
Josh:
So you can take a car, remove it, place it into this virtual world,

257
00:15:05,050 --> 00:15:08,470
Josh:
but it looks like the real world. It emulates a real world scenario. And it just.

258
00:15:09,060 --> 00:15:12,640
Josh:
As I'm looking at this, Ejas, to your point, it all feels like a video game.

259
00:15:12,820 --> 00:15:16,080
Josh:
And it's a really high fidelity video game where they can take things from reality.

260
00:15:16,540 --> 00:15:20,780
Josh:
They can distort them. They could create fake realities. And as I was scrolling

261
00:15:20,780 --> 00:15:24,220
Josh:
through this post, I started to lose track of what was real and what wasn't

262
00:15:24,220 --> 00:15:26,500
Josh:
because it all looks so real to me.

263
00:15:26,620 --> 00:15:29,020
Josh:
And to the video game point, which you might be able to share,

264
00:15:29,200 --> 00:15:32,480
Josh:
is that they actually allow you to play it as if it was a video game.

265
00:15:32,580 --> 00:15:36,720
Josh:
You can drive through these virtual worlds without actually needing a Tesla vehicle.

266
00:15:36,720 --> 00:15:41,460
Ejaaz:
Yeah, so what I have here is the Tesla's Neural World Simulator,

267
00:15:41,460 --> 00:15:45,220
Ejaaz:
where you have someone that is in basically a driver's seat,

268
00:15:45,400 --> 00:15:47,320
Ejaaz:
but it's one of those video gaming driving setups.

269
00:15:47,520 --> 00:15:50,040
Ejaaz:
And they are driving through what looks

270
00:15:50,040 --> 00:15:54,220
Ejaaz:
like a pretty pleasant suburban neighborhood on a sunny blue sky day.

271
00:15:54,440 --> 00:15:58,080
Ejaaz:
And it looks really real, Josh. It looks like something that would be recorded

272
00:15:58,080 --> 00:16:02,080
Ejaaz:
from Tesla's seven cameras, except that none of it is real.

273
00:16:02,220 --> 00:16:08,160
Ejaaz:
He is navigating through roads. He's skirting around cars. He's narrowly avoiding collisions.

274
00:16:08,560 --> 00:16:11,900
Ejaaz:
And every single perspective and animal that you see from the three different

275
00:16:11,900 --> 00:16:15,920
Ejaaz:
cameras on the screen here is completely and utterly simulated.

276
00:16:16,280 --> 00:16:19,260
Josh:
The most remarkable part is that all of this amazing stuff that we've just talked

277
00:16:19,260 --> 00:16:23,280
Josh:
about for the last 20 minutes, it's actually cross compatible with the next

278
00:16:23,280 --> 00:16:25,820
Josh:
most important form of autonomy, which is robots.

279
00:16:26,140 --> 00:16:29,560
Josh:
Now, everyone knows Tesla's making Optimus. They signal plans to make hundreds

280
00:16:29,560 --> 00:16:30,820
Josh:
of thousands of these by next year.

281
00:16:31,080 --> 00:16:34,680
Josh:
And the problem with training robots for a lot of other companies is that they

282
00:16:34,680 --> 00:16:36,200
Josh:
don't have the data, they don't have the neural models.

283
00:16:36,320 --> 00:16:39,660
Josh:
Well, all of the progress and all of the data that's been made previously through

284
00:16:39,660 --> 00:16:46,400
Josh:
Tesla is cross-compatible directly with the robot team and Optimus as a humanoid robot.

285
00:16:46,560 --> 00:16:51,740
Josh:
And that is one of the most impressive things because as the program gets better

286
00:16:51,740 --> 00:16:56,100
Josh:
through AI's autopilot stack, it improves dramatically through Optimus.

287
00:16:56,200 --> 00:16:57,600
Josh:
And what you're able to see is,

288
00:16:58,190 --> 00:17:01,790
Josh:
A lot of, like you mentioned, Ejaz, the goldmine is the digital data because

289
00:17:01,790 --> 00:17:02,890
Josh:
you just want more data to train.

290
00:17:03,330 --> 00:17:04,410
Josh:
Optimus gets better. And that

291
00:17:04,410 --> 00:17:09,390
Josh:
moves us on to the price of Tesla and the second order effects of Tesla.

292
00:17:09,590 --> 00:17:13,410
Josh:
Because now that we have humanoid robots that are learning quickly,

293
00:17:13,550 --> 00:17:16,990
Josh:
now that we have cars that are able to drive themselves, well, there's two things.

294
00:17:17,130 --> 00:17:19,950
Josh:
One of them is being the chip that unifies the two.

295
00:17:20,110 --> 00:17:24,970
Josh:
The other is the second order effects of what happens when this gets rolled out across the world.

296
00:17:25,330 --> 00:17:28,450
Josh:
And he says, maybe you want to tee that up for us, because this is a very bullish

297
00:17:28,450 --> 00:17:30,050
Josh:
scenario that we're guiding towards.

298
00:17:30,410 --> 00:17:35,650
Ejaaz:
Okay, so this is the most exciting part for me for this entire episode,

299
00:17:35,650 --> 00:17:39,710
Ejaaz:
because as you mentioned, this data and these neural networks aren't just super

300
00:17:39,710 --> 00:17:41,610
Ejaaz:
valuable for the Tesla cars.

301
00:17:41,830 --> 00:17:47,050
Ejaaz:
It's for the robots and pretty much any other kind of robotic machine that they create in the future.

302
00:17:47,250 --> 00:17:50,290
Ejaaz:
And the beautiful thing about this is that it's self-recursive.

303
00:17:50,390 --> 00:17:53,950
Ejaaz:
So whatever is learned from all the camera information and audio information

304
00:17:53,950 --> 00:17:56,510
Ejaaz:
that's pulled from the cars can feed into the robots,

305
00:17:56,510 --> 00:17:59,210
Ejaaz:
which is like kind of what we're seeing in the demo on our screen here with

306
00:17:59,210 --> 00:18:04,250
Ejaaz:
this Optimus robot navigating through what seems to be a manufacturing site, right?

307
00:18:04,910 --> 00:18:09,130
Ejaaz:
This is incredibly bullish for Tesla, the stock, in my opinion,

308
00:18:09,470 --> 00:18:13,710
Ejaaz:
because it takes it from, well, it's currently breaching or sitting under its

309
00:18:13,710 --> 00:18:16,390
Ejaaz:
all-time high, right, Josh? What is that market cap right now?

310
00:18:16,490 --> 00:18:19,830
Josh:
We're just under an all-time high, which puts it right around $1.5 trillion.

311
00:18:19,830 --> 00:18:25,010
Ejaaz:
Okay, so $1.5 trillion in today's age seems pretty small.

312
00:18:25,310 --> 00:18:29,150
Ejaaz:
You just had Microsoft and Apple today cross $4 trillion market cap.

313
00:18:29,290 --> 00:18:33,750
Ejaaz:
If you compare that to Tesla, and if you factor in the fact that these humanoid

314
00:18:33,750 --> 00:18:38,870
Ejaaz:
robots are largely going to replace or work in conjunction with a large swathe

315
00:18:38,870 --> 00:18:41,290
Ejaaz:
of the human manual labor force,

316
00:18:41,430 --> 00:18:46,850
Ejaaz:
that prices this up at least up until a $10 trillion company as this scales out.

317
00:18:47,110 --> 00:18:49,910
Ejaaz:
Josh, I have a feeling you're probably similarly bullish when it

318
00:18:49,910 --> 00:18:53,790
Josh:
Comes to this. Obviously, I share your sentiment. I have been maximally bullish

319
00:18:53,790 --> 00:18:56,370
Josh:
on Tesla for over a decade now. It's about,

320
00:18:56,970 --> 00:18:58,930
Josh:
12 years. Did your dad.

321
00:18:58,930 --> 00:19:02,030
Ejaaz:
Buy your Tesla stock for you at the start? You asked him to?

322
00:19:02,330 --> 00:19:05,970
Josh:
Yeah, I was too young to have my own brokerage account. So we were very early

323
00:19:05,970 --> 00:19:09,090
Josh:
shares of Tesla and continue to be maximally bullish on it.

324
00:19:09,330 --> 00:19:12,910
Josh:
And we're actually, I'm going to be recording a bull thesis episode about Tesla

325
00:19:12,910 --> 00:19:15,490
Josh:
because I'm so bullish on it. So if you're interested in that, let me know.

326
00:19:15,590 --> 00:19:19,250
Josh:
But I'm going to pull some notes from that to use here, just to kind of outline

327
00:19:19,250 --> 00:19:20,770
Josh:
the humanoid robotic opportunity.

328
00:19:20,910 --> 00:19:24,110
Josh:
Because EJ, as you said, $10 trillion, which is an outrageous market cap,

329
00:19:24,110 --> 00:19:27,310
Josh:
considering NVIDIA is the largest company in the world sitting at four trillion.

330
00:19:27,590 --> 00:19:30,870
Josh:
So that's a long way to go. And NVIDIA is on top of the world.

331
00:19:30,990 --> 00:19:36,430
Josh:
But if you think of humanoids as labor, right, you have kind of four billion

332
00:19:36,430 --> 00:19:37,530
Josh:
people in the labor market.

333
00:19:37,650 --> 00:19:40,090
Josh:
And this becomes a global trend. This is not just for the United States.

334
00:19:40,450 --> 00:19:44,650
Josh:
And if the average wage, which is what it is right now, is about $10,000 per

335
00:19:44,650 --> 00:19:47,270
Josh:
year, that's a $40 trillion market size.

336
00:19:47,650 --> 00:19:52,430
Josh:
So the labor opportunity is $40 trillion, assuming we don't have any productivity

337
00:19:52,430 --> 00:19:56,610
Josh:
unlocks that generate brand new opportunities, that generate more use cases for labor.

338
00:19:56,750 --> 00:19:58,910
Josh:
So that's just given the current state of the world today.

339
00:19:59,170 --> 00:20:04,010
Josh:
So if one humanoid at $5 an hour can replace two humans working at $25 an hour,

340
00:20:04,430 --> 00:20:10,310
Josh:
the value per humanoid becomes $200,000 per robot, which is pretty high given

341
00:20:10,310 --> 00:20:14,710
Josh:
that the costs are projected to be around $20,000 to $30,000 once it's all said and done.

342
00:20:15,070 --> 00:20:17,910
Josh:
The US labor market, there's 160 million people.

343
00:20:18,050 --> 00:20:23,530
Josh:
So if just 1% is substituted by humanoid robots, that is greater than $300 billion in value.

344
00:20:24,460 --> 00:20:27,260
Josh:
That's a lot of revenue. That is a tremendous amount of revenue.

345
00:20:27,420 --> 00:20:31,040
Josh:
And then you get to a point where you're starting to offset significant percentages of GDP.

346
00:20:31,400 --> 00:20:35,060
Josh:
So in the 1950s, the US manufacturing share of GDP, it was 30%.

347
00:20:35,060 --> 00:20:40,200
Josh:
Today it sits at 10%. And if this goes further, we'll have a total reliance on foreign entities.

348
00:20:40,340 --> 00:20:43,940
Josh:
So there's all the incentives in the world to bring robots into the United States.

349
00:20:44,060 --> 00:20:47,160
Josh:
So we don't continue this trend of decreasing our manufacturing capabilities.

350
00:20:47,240 --> 00:20:50,600
Josh:
There's a lot of headwinds and a lot of trends that all converge on the humanoid

351
00:20:50,600 --> 00:20:53,040
Josh:
robot opportunity. It's just a matter of making these.

352
00:20:53,160 --> 00:20:58,320
Josh:
And it's possible because of this new software stack and also because of this

353
00:20:58,320 --> 00:21:00,260
Josh:
new chip, which is the AI5 chip.

354
00:21:00,280 --> 00:21:06,400
Josh:
And the AI5 chip is the brand new golden child of Tesla. And it is going to

355
00:21:06,400 --> 00:21:08,260
Josh:
be cross compatible between both robots and,

356
00:21:08,870 --> 00:21:12,230
Josh:
cyber cabs. But you just maybe you want to walk us through exactly why this is interesting.

357
00:21:12,570 --> 00:21:19,150
Ejaaz:
Yeah. So the way I think about this is this is Tesla's bold attempt to replace the GPU.

358
00:21:19,470 --> 00:21:23,890
Ejaaz:
And as we've spoken about many times on this show before, Nvidia kind of rules the kingdom.

359
00:21:24,070 --> 00:21:27,850
Ejaaz:
We mentioned that they are sitting at a $4 trillion or above a well above a

360
00:21:27,850 --> 00:21:30,710
Ejaaz:
$4 trillion market cap. They are the kings of the roost.

361
00:21:30,990 --> 00:21:34,710
Ejaaz:
And the reason why is because they provide the hardware that kind of fuels all

362
00:21:34,710 --> 00:21:35,430
Ejaaz:
these different things.

363
00:21:35,650 --> 00:21:40,350
Ejaaz:
Now, what Tesla identified is whilst all these GPUs that they've been using

364
00:21:40,350 --> 00:21:45,890
Ejaaz:
are really helpful, they're not specifically designed to fit certain niche use

365
00:21:45,890 --> 00:21:48,970
Ejaaz:
cases when it comes to a range of different things that they're involved in, right?

366
00:21:49,330 --> 00:21:52,730
Ejaaz:
Cars, humanoid robots, and an array of different things.

367
00:21:52,870 --> 00:21:57,830
Ejaaz:
And now they've released their AI5 chip, which is basically their brand new

368
00:21:57,830 --> 00:22:01,030
Ejaaz:
chip, which is going to be used across all their different robots.

369
00:22:01,170 --> 00:22:04,550
Ejaaz:
So it's going to be used in cars, on humanoids, and the like.

370
00:22:04,830 --> 00:22:07,290
Ejaaz:
And the coolest part about this, Josh, we were speaking about this before the

371
00:22:07,290 --> 00:22:13,250
Ejaaz:
show, is it improves this whole GPU experience for them by 40 times.

372
00:22:13,410 --> 00:22:15,710
Ejaaz:
But can you help me unpack as to why exactly?

373
00:22:15,890 --> 00:22:19,170
Ejaaz:
Is this like a sizing thing? Can they add more compute? How does this work?

374
00:22:19,530 --> 00:22:23,870
Josh:
Okay, so first thing, AI5 isn't out just yet. It's coming. They have completed the spec.

375
00:22:24,010 --> 00:22:27,030
Josh:
Elon's been working on it. He said on the most recent earnings call that it

376
00:22:27,030 --> 00:22:30,310
Josh:
has been his number one focus for weeks and weeks and weeks on end,

377
00:22:30,410 --> 00:22:32,130
Josh:
which is very high signal that it means a lot.

378
00:22:32,290 --> 00:22:34,990
Josh:
So it is coming soon. They're working on tooling and they're working to roll

379
00:22:34,990 --> 00:22:39,050
Josh:
this out, I assume, in companion with the Optimus robot that is probably coming

380
00:22:39,050 --> 00:22:41,090
Josh:
next year. You mentioned it's 40 times better.

381
00:22:41,450 --> 00:22:44,550
Josh:
Why is it 40 times better? And why do companies make their own chips?

382
00:22:44,670 --> 00:22:46,790
Josh:
I think this is an important question because a lot of people don't know.

383
00:22:47,290 --> 00:22:50,910
Josh:
Well, NVIDIA makes awesome GPUs. Why would I go through all the R&D budgeting

384
00:22:50,910 --> 00:22:52,410
Josh:
costs and pain in the ass because...

385
00:22:53,210 --> 00:22:57,050
Josh:
To make my own chip? And the answer is because vertical integration allows you

386
00:22:57,050 --> 00:22:59,150
Josh:
to be hyper customized in what you're able to do.

387
00:22:59,430 --> 00:23:03,710
Josh:
So what Tesla has done is they, it's funny, they do this with everything,

388
00:23:03,830 --> 00:23:06,210
Josh:
but they kind of, they looked at the chip through first principles.

389
00:23:06,410 --> 00:23:08,750
Josh:
They looked at all the different modules that sit on this chip.

390
00:23:08,990 --> 00:23:13,170
Josh:
You could think one of them processes graphics, one of them processes images, one is processing math.

391
00:23:13,490 --> 00:23:16,370
Josh:
The reason why all of these GPUs from other companies need to have all of these

392
00:23:16,370 --> 00:23:18,090
Josh:
is because they need to satisfy their customers.

393
00:23:18,290 --> 00:23:22,110
Josh:
They need to be able to be diverse in the types of computing they can do.

394
00:23:22,290 --> 00:23:27,510
Josh:
In the narrow band of use cases that Tesla has, they're able to reconsider this and optimize for it.

395
00:23:27,610 --> 00:23:31,650
Josh:
So for example, there's this image signal processor that sits on a chip and

396
00:23:31,650 --> 00:23:34,950
Josh:
it's meant to what it says. It processes image signals that come in.

397
00:23:35,090 --> 00:23:38,250
Josh:
What Tesla has done is they're not actually processing images.

398
00:23:38,250 --> 00:23:40,710
Josh:
They're processing photons and photons can be binary.

399
00:23:40,870 --> 00:23:43,030
Josh:
They could be expressed in code. So there's this.

400
00:23:43,580 --> 00:23:46,960
Josh:
Big chip that sits on a larger chip, they're able to completely remove that

401
00:23:46,960 --> 00:23:50,300
Josh:
image processing chip because they said, actually, we don't need to look at images ever.

402
00:23:50,460 --> 00:23:54,480
Josh:
We're just doing photons in, photons out, baby. And that unlocks X percent of

403
00:23:54,480 --> 00:23:58,820
Josh:
this board to add more compute power to the specific type of compute you need.

404
00:23:58,940 --> 00:24:03,280
Josh:
So for the first time ever, you're getting these chips that don't actually look

405
00:24:03,280 --> 00:24:04,300
Josh:
like traditional chips.

406
00:24:04,480 --> 00:24:07,360
Josh:
They're built very different because of the narrow band use case that's required.

407
00:24:07,640 --> 00:24:11,260
Josh:
And that allows them to not only be much more efficient in terms of compute

408
00:24:11,260 --> 00:24:15,260
Josh:
per watt, but also cost per watt, and also the cross compatibility across all these devices.

409
00:24:15,440 --> 00:24:18,160
Josh:
So a lot of companies, they have, like if you think of Apple,

410
00:24:18,240 --> 00:24:21,920
Josh:
they have the M series chip for the computers and the iPhones,

411
00:24:22,220 --> 00:24:27,240
Josh:
whereas NVIDIA has 12 different GPUs for mobile devices, for power,

412
00:24:27,400 --> 00:24:29,040
Josh:
general computers, for data centers.

413
00:24:29,320 --> 00:24:32,620
Josh:
It's this really remarkable unlock that we're going to start to see roll out

414
00:24:32,620 --> 00:24:36,300
Josh:
next year in both that enables both the CyberCab and the humanoid robot.

415
00:24:36,300 --> 00:24:42,400
Ejaaz:
There's an increasing trend of these new age AI tech companies that once they

416
00:24:42,400 --> 00:24:46,820
Ejaaz:
reach escape velocity for a bunch of consumer and enterprise facing products,

417
00:24:47,180 --> 00:24:51,720
Ejaaz:
they start to vertically integrate with a part of which includes creating their

418
00:24:51,720 --> 00:24:54,060
Ejaaz:
own custom design GPUs and chips.

419
00:24:54,060 --> 00:24:58,220
Ejaaz:
The most recent example I can think of aside from Tesla is OpenAI,

420
00:24:58,260 --> 00:25:00,980
Ejaaz:
who announced that a partnership with Broadgate,

421
00:25:01,180 --> 00:25:05,820
Ejaaz:
that they're going to be developing their own custom GPUs to fuel certain niche

422
00:25:05,820 --> 00:25:10,620
Ejaaz:
use cases that their future GPT-6 models and ahead will utilize.

423
00:25:10,900 --> 00:25:16,060
Ejaaz:
They haven't quite revealed what those chips are going to be facilitating exactly.

424
00:25:16,300 --> 00:25:21,040
Ejaaz:
But what we do know is that they're using the AI model itself to help them design

425
00:25:21,040 --> 00:25:25,120
Ejaaz:
this chip. So this thing around AI5 is the most Elon thing ever,

426
00:25:25,160 --> 00:25:28,960
Ejaaz:
because we've seen what he's done when he's taken a hammer to data centers.

427
00:25:29,100 --> 00:25:33,200
Ejaaz:
And we're seeing now what he's what he's done by creating the probably the most

428
00:25:33,200 --> 00:25:36,740
Ejaaz:
valuable resource going forwards for tech companies at the GPU layer.

429
00:25:36,840 --> 00:25:38,880
Ejaaz:
So I don't know. I'm excited about this, Josh.

430
00:25:39,060 --> 00:25:41,640
Ejaaz:
It makes me unfathomably bullish.

431
00:25:41,860 --> 00:25:46,820
Ejaaz:
My earlier $10 trillion estimate is probably too conservative after what we've just discussed.

432
00:25:47,610 --> 00:25:50,290
Josh:
Well, with Elon's new pay package, there is a direct incentive alignment.

433
00:25:50,450 --> 00:25:54,770
Josh:
One thing on the Broadcom partnership with OpenAI, the difference there is that

434
00:25:54,770 --> 00:25:57,610
Josh:
Broadcom exists and Tesla is a single entity.

435
00:25:58,130 --> 00:26:04,010
Josh:
So OpenAI doesn't really have the resources in order to create their own chips in-house.

436
00:26:04,010 --> 00:26:07,690
Josh:
And I think that's a really big difference because when there is that physical

437
00:26:07,690 --> 00:26:11,250
Josh:
gap between different companies when you're designing these chips,

438
00:26:11,390 --> 00:26:14,570
Josh:
it makes it a little bit more difficult to do that really hardcore,

439
00:26:14,570 --> 00:26:18,650
Josh:
like cost-cutting vertical integration that Tesla has. Tesla's doing this.

440
00:26:18,770 --> 00:26:21,230
Josh:
They're making their own ship in-house. They're designing it in-house.

441
00:26:21,610 --> 00:26:25,910
Josh:
OpenEye is outsourcing that responsibility. And that's where you'll maybe start to see discrepancy.

442
00:26:26,030 --> 00:26:30,790
Josh:
So I am hopeful that they will do great, but I still suspect Tesla will do better.

443
00:26:31,070 --> 00:26:35,810
Josh:
And Tesla also has manufacturing prowess. So yeah, I think if we walk away with

444
00:26:35,810 --> 00:26:38,530
Josh:
anything from this episode is that both of us share the sentiment that we are

445
00:26:38,530 --> 00:26:41,650
Josh:
unfathomably bullish for an assortment of reasons. And this is just one of them.

446
00:26:41,910 --> 00:26:44,010
Josh:
The Tesla bookcase will be coming soon, I promise.

447
00:26:44,450 --> 00:26:49,090
Josh:
And there's a lot more to the company, but this is autonomy. This is autopilot.

448
00:26:49,250 --> 00:26:52,830
Josh:
This is the secrets of Tesla finally unveiled for the world.

449
00:26:52,990 --> 00:26:56,190
Josh:
And I imagine the rest of the world, granted, they've probably been trying to

450
00:26:56,190 --> 00:27:00,150
Josh:
emulate this. It's not really much of a secret, but we'll have a very difficult time in doing so.

451
00:27:00,330 --> 00:27:03,670
Ejaaz:
I think that wraps it up for today's episode.

452
00:27:03,930 --> 00:27:07,950
Ejaaz:
We hope you enjoyed this breakdown. We are unfathomably excited and bullish,

453
00:27:07,990 --> 00:27:11,610
Ejaaz:
as I've said multiple times about Tesla, but are you?

454
00:27:11,810 --> 00:27:15,510
Ejaaz:
Let us know in the comments. Are we crazy? is the vision that we're engaging

455
00:27:15,510 --> 00:27:17,730
Ejaaz:
in around Tesla completely insane?

456
00:27:17,930 --> 00:27:20,850
Ejaaz:
Are robots not really a thing in your opinion? Let us know in the comments.

457
00:27:21,010 --> 00:27:23,930
Ejaaz:
We're also going to be releasing one more episode this week,

458
00:27:24,110 --> 00:27:26,950
Ejaaz:
which is going to be the AI Weekly Roundup, which we're going to cover all the

459
00:27:26,950 --> 00:27:29,490
Ejaaz:
hottest topics. There's some crazy stuff that has happened this week.

460
00:27:30,050 --> 00:27:32,570
Ejaaz:
And if there's anything else that we've missed or that you want to hear about,

461
00:27:32,830 --> 00:27:37,650
Ejaaz:
let us know in the comments. DM us. We're always available. And we will see you in the next one.

462
00:27:38,150 --> 00:27:39,230
Josh:
Thanks for watching. See you guys.