1
00:00:09,960 --> 00:00:13,240
Welcome back to Cerebral Valley.
I'm here with Max Child and 

2
00:00:13,240 --> 00:00:16,720
James Woolsterman, my friends 
and Co founders of Volley. 

3
00:00:17,080 --> 00:00:20,920
This episode we are talking 
about chips and big tech. 

4
00:00:21,000 --> 00:00:25,000
My interview, after a 
conversation with Max and James 

5
00:00:25,000 --> 00:00:30,280
is with Chris Miller, the author
of Chip War, The fight for the 

6
00:00:30,280 --> 00:00:32,159
World's most critical 
technology. 

7
00:00:32,159 --> 00:00:35,120
He is an expert. 
We talk a lot about NVIDIA. 

8
00:00:36,120 --> 00:00:39,960
Yeah, so stick around for that. 
Max James welcome back. 

9
00:00:40,120 --> 00:00:42,640
Nice to be here. 
Eric, Glad to be here. 

10
00:00:42,840 --> 00:00:45,000
The starting point question that
I wanted to frame things up with

11
00:00:45,000 --> 00:00:49,440
is, do you think GPU capacity, 
the quality of these graphics 

12
00:00:49,440 --> 00:00:53,520
processing units and what NVIDIA
is putting out is the main 

13
00:00:53,520 --> 00:00:57,720
driver for the sort of 
generative AI revolution we're 

14
00:00:57,720 --> 00:01:00,400
seeing right now? 
What do you mostly attribute it 

15
00:01:00,520 --> 00:01:02,840
to? 
The research papers, The new 

16
00:01:02,840 --> 00:01:04,720
approaches. 
How? 

17
00:01:04,879 --> 00:01:05,720
How do you? 
How do you? 

18
00:01:05,720 --> 00:01:06,680
Who do you give? 
Credit. 

19
00:01:07,840 --> 00:01:10,520
I guess I give more credit to 
the research papers. 

20
00:01:11,920 --> 00:01:16,360
Not that I don't discount what 
NVIDIA has done here. 

21
00:01:16,360 --> 00:01:19,520
I just, I think they were in the
right place at the right time 

22
00:01:19,520 --> 00:01:23,680
with the right technology to 
accelerate what was possible 

23
00:01:23,840 --> 00:01:27,040
through due to the research. 
But yeah, the research was the 

24
00:01:27,040 --> 00:01:29,200
breakthrough. 
I do think there was some 

25
00:01:29,200 --> 00:01:31,320
intentionality. 
I mean, it's always hard to tell

26
00:01:31,320 --> 00:01:35,400
with AI when there's so much. 
It's a hypey thing to talk 

27
00:01:35,400 --> 00:01:37,560
about. 
And this is ACEO that also made 

28
00:01:37,560 --> 00:01:39,200
a bunch of money off crypto 
mining. 

29
00:01:39,200 --> 00:01:42,160
So happy to dive into wild 
speculation. 

30
00:01:42,160 --> 00:01:44,160
Clearly I don't know. 
Max, you have a take here. 

31
00:01:45,200 --> 00:01:46,520
I mean I think I agree with 
James. 

32
00:01:46,520 --> 00:01:50,000
I just think that, you know even
four or five years ago we were 

33
00:01:50,000 --> 00:01:52,800
already starting to see the 
beginnings of this with GPT 2 

34
00:01:52,800 --> 00:01:55,560
and you know earlier versions of
the the foundational models, 

35
00:01:55,560 --> 00:01:58,080
right. 
Also in areas outside of large 

36
00:01:58,080 --> 00:02:01,880
language models you saw big 
progress in, you know our hobby 

37
00:02:01,880 --> 00:02:04,200
horses of speech synthesis, 
speech recognition and other 

38
00:02:04,200 --> 00:02:06,480
areas. 
So I think that even with five 

39
00:02:06,480 --> 00:02:10,000
year old GP US, you still could 
pretty to do pretty cool stuff 

40
00:02:10,000 --> 00:02:12,320
with the new research that was 
outright whereas. 

41
00:02:12,720 --> 00:02:15,600
If the research never comes out,
I don't think that you're making

42
00:02:15,600 --> 00:02:18,720
this kind of progress in, in any
of these kind of large large 

43
00:02:18,720 --> 00:02:21,920
language model areas that we or 
or any of these areas of the 

44
00:02:21,920 --> 00:02:24,120
stack that we've seen so far. 
So right. 

45
00:02:24,440 --> 00:02:28,360
I mean, chips, even though 
there's potentially exponential 

46
00:02:28,360 --> 00:02:32,200
improvement, it's still like, 
you know, on a trajectory, 

47
00:02:32,320 --> 00:02:35,520
whereas a paper sort of comes 
out of nowhere, gives people a 

48
00:02:35,520 --> 00:02:40,080
new approach and then sort of 
revolutionizes what's possible. 

49
00:02:40,560 --> 00:02:42,400
I mean, I think a different way 
to think about it would be like,

50
00:02:42,440 --> 00:02:46,320
OK, GPT has gotten like 1000 
times better in the last three 

51
00:02:46,320 --> 00:02:49,040
years and like the chips haven't
gotten 1000 times better in the 

52
00:02:49,040 --> 00:02:51,120
last three years. 
So like, what's driving that? 

53
00:02:51,120 --> 00:02:53,680
Like, it's probably the software
and then the people like 

54
00:02:54,000 --> 00:02:55,800
figuring out how to make GPT, 
right? 

55
00:02:55,800 --> 00:02:59,480
So yeah, I just think it's 
clearly based on the ideas in 

56
00:02:59,480 --> 00:03:01,640
the software. 
The chips are great, no doubt, 

57
00:03:01,640 --> 00:03:06,040
but it's not the key driver. 
I I do think it so NVIDIA has 

58
00:03:06,040 --> 00:03:08,480
clearly gotten rich off this 
stuff. 

59
00:03:08,480 --> 00:03:11,560
It is now a $1.1 trillion 
company. 

60
00:03:11,920 --> 00:03:21,560
In one year the stock is up 258%
and in five years it's up 564%. 

61
00:03:21,720 --> 00:03:27,280
So this, this has been a great 
run for the company. 

62
00:03:27,280 --> 00:03:32,720
I mean what do you, what do you 
guys make of NVIDIA investing in

63
00:03:32,720 --> 00:03:36,880
all these companies? 
I mean similarly if it's into, 

64
00:03:37,120 --> 00:03:41,520
you know Amazon, Google, like 
all these companies are making 

65
00:03:41,520 --> 00:03:43,960
investments. 
The money that they're spending 

66
00:03:43,960 --> 00:03:45,840
obviously comes back to their 
business, right. 

67
00:03:45,840 --> 00:03:50,000
NVIDIA wants AI to be super 
active, so puts out money that 

68
00:03:50,000 --> 00:03:52,880
gets spent on NVIDIA chips. 
Yeah. 

69
00:03:52,880 --> 00:03:56,360
What do you make of, I mean it's
round, it's round tripping in 

70
00:03:56,360 --> 00:04:00,880
the OR it's you know spending 
money to make money, but it's it

71
00:04:00,880 --> 00:04:03,760
doesn't seem as sort of negative
as in some cases. 

72
00:04:04,520 --> 00:04:07,360
I mean, isn't the investment 
NVIDIA is putting out into the 

73
00:04:07,360 --> 00:04:09,720
ecosystem like a rounding error 
compared to their revenue or 

74
00:04:09,720 --> 00:04:12,760
compared to their profits? 
I mean, like, I I mean I think 

75
00:04:12,760 --> 00:04:14,880
that they're making such 
ludicrous quantities of money 

76
00:04:14,880 --> 00:04:18,680
that kicking back like 1 to 2% 
into the startup community that 

77
00:04:18,680 --> 00:04:21,880
might invest in NVIDIA chips in 
the future is, is obviously just

78
00:04:21,880 --> 00:04:23,160
a very good business practice, 
right. 

79
00:04:23,360 --> 00:04:24,680
Right. 
Well, it seems like they're 

80
00:04:24,680 --> 00:04:26,880
literally playing kingmaker, 
right. 

81
00:04:26,880 --> 00:04:30,520
I mean like companies like Core 
Weave and who are sort of chip 

82
00:04:30,840 --> 00:04:35,320
doing chip adjacent stuff depend
on their access to H1 Hundreds 

83
00:04:35,320 --> 00:04:37,880
and a lot of these foundation 
models seem like they're 

84
00:04:37,880 --> 00:04:42,960
fundraising on the premise that 
hey we have access like this is 

85
00:04:43,120 --> 00:04:47,520
differentiated partially because
we have the deal to get GPU 

86
00:04:47,520 --> 00:04:51,160
access when other people don't. 
And so there's this hope that 

87
00:04:51,560 --> 00:04:54,720
you know, you're you have a Moat
in that you have access to to 

88
00:04:54,720 --> 00:04:58,680
chips that other people don't. 
The scarcity today of the best 

89
00:04:58,680 --> 00:05:02,720
in class chip, the H 100 is. 
Has to be a temporary situation.

90
00:05:02,720 --> 00:05:05,320
I just like the economics of 
like making more of these chips 

91
00:05:05,320 --> 00:05:07,600
is a very good idea for NVIDIA. 
So they're going to do it. 

92
00:05:07,600 --> 00:05:12,280
But they they clearly want to 
build like a Amazon Web Services

93
00:05:12,280 --> 00:05:15,840
type competitor. 
And so having eyes and ears in a

94
00:05:15,840 --> 00:05:18,320
bunch of startups and seeing 
maybe who they could acquire to 

95
00:05:18,320 --> 00:05:20,960
do that is smart. 
I do think the interesting 

96
00:05:20,960 --> 00:05:24,840
question again is like Intel 
basically had like 2 decades of 

97
00:05:24,840 --> 00:05:29,080
like total monopolistic 
dominance of of PC chips, right?

98
00:05:29,080 --> 00:05:31,320
And. 
The reason basically was that 

99
00:05:31,320 --> 00:05:35,320
it's like really, really, really
hard to build fabrication plants

100
00:05:35,320 --> 00:05:39,040
to to build CP US. 
And it took a long time for like

101
00:05:39,040 --> 00:05:42,200
TSMC and other people to sort of
catch up on that front and then 

102
00:05:42,200 --> 00:05:43,920
eventually kind of surpass them 
in many ways. 

103
00:05:43,920 --> 00:05:47,880
But the question I guess for GP 
US I think is like, is it that 

104
00:05:47,880 --> 00:05:51,680
technically hard to make GP US 
that like whatever the next 

105
00:05:51,680 --> 00:05:54,320
version of the H 100 is? 
You know that they're going to 

106
00:05:54,320 --> 00:05:57,080
be on to that before anyone's 
even caught up to the H 100, and

107
00:05:57,080 --> 00:05:59,480
so they'll have this 
insurmountable 2 two to 

108
00:05:59,480 --> 00:06:02,280
three-year advantage where if 
you want the best AIGPUS, it's 

109
00:06:02,280 --> 00:06:04,240
always going to be NVIDIA for 
the next 20 years. 

110
00:06:04,360 --> 00:06:06,640
Because that seems like a 
plausible case to me. 

111
00:06:06,760 --> 00:06:09,840
Do you think Microsoft was smart
to invest 10 billion in open AI 

112
00:06:09,840 --> 00:06:13,800
given it's clearly going to 
distract the company as has been

113
00:06:13,800 --> 00:06:15,840
reported? 
I think my friend in the Wall 

114
00:06:15,840 --> 00:06:18,680
Street Journal reported as much 
that they directed sort of their

115
00:06:18,680 --> 00:06:21,040
attention to open AI away from 
Microsoft. 

116
00:06:21,840 --> 00:06:26,080
And and part of what you're 
saying is they they are giving 

117
00:06:26,080 --> 00:06:29,440
up access to their own GPUs to 
Open AI. 

118
00:06:29,920 --> 00:06:34,720
Is that accurate to to say so? 
Yeah, I mean, I think it was. 

119
00:06:34,880 --> 00:06:36,560
I think it was a good decision 
and. 

120
00:06:36,600 --> 00:06:38,320
I agree. 
Definitely a good decision. 

121
00:06:39,200 --> 00:06:45,200
I I can't imagine they would be,
as you know critical to where we

122
00:06:45,200 --> 00:06:48,400
are in the in the AI era right 
now had they not made that 

123
00:06:48,400 --> 00:06:53,120
decision they. 
Have you know early access to 

124
00:06:53,320 --> 00:06:56,720
all the technology that Open AI 
is creating. 

125
00:06:56,720 --> 00:07:02,080
They got GPT 4 into Bing before 
it was in public in you know in 

126
00:07:02,080 --> 00:07:06,080
front of customers for in front 
of Open AIS customers and seems 

127
00:07:06,080 --> 00:07:11,360
to be continuing with what they 
are trying to do with Office and

128
00:07:11,560 --> 00:07:13,160
Windows, you know the Office 
suite. 

129
00:07:14,120 --> 00:07:18,840
So I think. 
Just that early access to Open 

130
00:07:18,840 --> 00:07:21,000
AI products seems worth the 
investment. 

131
00:07:21,400 --> 00:07:24,320
I think, I think the sort of 
meta question you're asking that

132
00:07:24,320 --> 00:07:25,920
I think is really interesting is
like. 

133
00:07:26,480 --> 00:07:29,560
Is big tech going to win the AI 
race or whatever, right? 

134
00:07:29,640 --> 00:07:32,240
Like is like, I mean we always 
used to talk about Fang right? 

135
00:07:32,240 --> 00:07:34,960
Like Facebook, Amazon, Apple 
used to be Netflix and Google 

136
00:07:34,960 --> 00:07:36,320
right? 
You could replace the N and Fang

137
00:07:36,320 --> 00:07:37,640
with NVIDIA and it would be 
perfect. 

138
00:07:37,640 --> 00:07:40,160
So Fang OK and like, goodbye 
streaming. 

139
00:07:40,280 --> 00:07:41,040
Hello. 
Chips. 

140
00:07:41,080 --> 00:07:42,240
Yeah, Goodbye streaming. 
Hello. 

141
00:07:42,240 --> 00:07:42,840
Chips. 
Right. 

142
00:07:43,360 --> 00:07:45,520
And I love Netflix, So they 
never got over like a couple 100

143
00:07:45,520 --> 00:07:47,640
billion, which is where you 
said, you know, NVIDIA is a 

144
00:07:47,640 --> 00:07:51,520
trillion now. 
So pathetic. 200 Three $100 

145
00:07:51,520 --> 00:07:53,680
billion company. 
Yeah. 

146
00:07:53,800 --> 00:07:56,200
Is fan going to win AI? 
Like, I think that's a really 

147
00:07:56,200 --> 00:07:59,200
interesting question. 
And I think like I'm kind of 

148
00:07:59,200 --> 00:08:01,960
leaning, yes, I guess to your 
point, like it seems like 

149
00:08:02,200 --> 00:08:06,320
Microsoft's been able to grab, 
you know, Open AI, which is, per

150
00:08:06,320 --> 00:08:08,880
our last podcast, still with the
most valuable asset in the 

151
00:08:08,880 --> 00:08:12,160
entire industry, right. 
It seems like Amazon's getting 

152
00:08:12,160 --> 00:08:15,200
very cozy with Anthropic, in 
case that's actually the other 

153
00:08:15,200 --> 00:08:17,320
big asset in the foundation 
model category. 

154
00:08:17,840 --> 00:08:21,560
Facebook is absolutely like 
ripping on open source models. 

155
00:08:21,560 --> 00:08:24,800
You know, Llama. 
They're going to put like LLMS 

156
00:08:24,800 --> 00:08:27,200
inside their ad products so you 
can have automated ads on 

157
00:08:27,200 --> 00:08:29,760
Facebook and Instagram and all 
that stuff, you know? 

158
00:08:29,840 --> 00:08:34,919
They're also putting, they're 
also putting LLMS into consumer 

159
00:08:35,080 --> 00:08:36,799
chat. 
You know, products, right? 

160
00:08:36,799 --> 00:08:39,919
You can chat with celebrities, 
and I'm sure they're doing a lot

161
00:08:39,919 --> 00:08:42,200
with image generation. 
Yeah, exactly. 

162
00:08:42,200 --> 00:08:44,640
And then Google seems to be like
fully ripping on, just like 

163
00:08:44,640 --> 00:08:46,840
dumping large language models 
into every single Google 

164
00:08:46,840 --> 00:08:47,920
product. 
Like Google. 

165
00:08:47,920 --> 00:08:51,640
You know, Google Docs, Google 
Photos, I don't know if that is 

166
00:08:51,640 --> 00:08:53,480
the conventional wisdom. 
And so on. 

167
00:08:53,480 --> 00:08:55,520
OK. 
Google ripping for Google. 

168
00:08:55,520 --> 00:08:58,240
For the record, you have to like
grade them on the Google curve, 

169
00:08:58,280 --> 00:09:01,320
which is they're in a competent 
company coming out with new 

170
00:09:01,440 --> 00:09:03,160
ideas. 
Given that Google has not come 

171
00:09:03,160 --> 00:09:06,840
out with a new idea in 15 years,
they are doing a tremendous job 

172
00:09:06,840 --> 00:09:10,200
putting AI products into their 
existing, into their existing 

173
00:09:10,200 --> 00:09:13,480
products, right? 
And so, yeah, for them, the 

174
00:09:13,480 --> 00:09:16,120
number of launches this year has
been tremendous, right, Graded 

175
00:09:16,120 --> 00:09:18,880
on the Google curve, right. 
And then Apple has basically 

176
00:09:18,880 --> 00:09:21,240
done nothing. 
But Apple always does things 

177
00:09:21,240 --> 00:09:23,240
like 2 years late and then 
hopefully they get it right. 

178
00:09:23,240 --> 00:09:26,920
So the, the great dream of Apple
as I understand it is just that,

179
00:09:27,440 --> 00:09:29,760
you know, we get to a point 
where a lot of these small 

180
00:09:29,760 --> 00:09:34,240
models can be run local, local 
and that they have these great 

181
00:09:34,880 --> 00:09:38,040
M1 and M2 whatever generation 
we're on now chips. 

182
00:09:38,400 --> 00:09:41,400
And that even though they're not
GP, they're sort of combined 

183
00:09:41,400 --> 00:09:43,680
chips, right. 
So there is some graphical 

184
00:09:44,080 --> 00:09:46,800
component, right. 
And so that those chips would be

185
00:09:47,200 --> 00:09:50,560
sort of pretty capable at 
handling sort of small local 

186
00:09:50,560 --> 00:09:52,040
models? 
Totally. 

187
00:09:52,240 --> 00:09:54,400
And I mean. 
Apple's argument would be we're 

188
00:09:54,400 --> 00:09:57,920
using transformer based models 
in photo recognition stuff, 

189
00:09:57,920 --> 00:09:59,880
we're using it in speech 
recognition stuff, we're using 

190
00:09:59,880 --> 00:10:02,120
it in auto correct. 
They just put it in the iPhone, 

191
00:10:02,120 --> 00:10:04,160
right. 
They're like their their take is

192
00:10:04,160 --> 00:10:06,800
always like well we will do it 
when there's a real customer 

193
00:10:06,800 --> 00:10:09,000
need and this tool can really 
provide value. 

194
00:10:09,000 --> 00:10:10,640
So I think like Jury is still 
out there. 

195
00:10:10,920 --> 00:10:13,600
I also think with Apple Vision 
Pro it'll be really interesting 

196
00:10:13,600 --> 00:10:18,040
to see what kind of generative 
AI and and also just you know 

197
00:10:18,040 --> 00:10:20,440
the AI driven tool set that 
launches with Apple Vision Pro. 

198
00:10:20,440 --> 00:10:23,680
So I think like. 
It's a really interesting 

199
00:10:23,680 --> 00:10:26,720
question as to whether or not 
like this is whole AI revolution

200
00:10:26,720 --> 00:10:28,960
is just going to entrench the 
power of the big 5 tech 

201
00:10:28,960 --> 00:10:31,960
companies because I think like 
they look pretty good right now.

202
00:10:32,000 --> 00:10:33,560
You didn't. 
You didn't mention Amazon, 

203
00:10:33,560 --> 00:10:36,040
right? 
I I said Amazon was like cozying

204
00:10:36,040 --> 00:10:38,600
up to anthropic. 
And also, I mean, you can 

205
00:10:38,680 --> 00:10:40,600
elaborate on the other stuff 
they're doing, which is quite 

206
00:10:40,720 --> 00:10:42,880
copious. 
On on Alexa specifically. 

207
00:10:42,920 --> 00:10:45,680
Well, and also AWS, right? 
Bedrock and everything, right? 

208
00:10:45,680 --> 00:10:48,760
Yeah, right. 
Well, it feels like Amazon, I 

209
00:10:48,760 --> 00:10:53,640
mean Microsoft is clearly ahead 
like they with Azure and AWS for

210
00:10:53,640 --> 00:10:56,160
those two companies I think are 
the most important. 

211
00:10:56,160 --> 00:10:59,640
I get you guys really care about
Alexa and stuff could happen. 

212
00:10:59,640 --> 00:11:03,240
But for the businesses today, 
like Amazon Web Services is so 

213
00:11:03,240 --> 00:11:06,000
essential. 
And like Microsoft can go to 

214
00:11:06,000 --> 00:11:10,760
people and say like Azure has a 
direct relationship with Open 

215
00:11:10,760 --> 00:11:15,480
AI, whereas Amazon, it feels 
like they've been pretty slow on

216
00:11:15,480 --> 00:11:16,760
this. 
But like like you said, they 

217
00:11:16,760 --> 00:11:19,800
have Bedrock and they're trying 
to come up with these sort of 

218
00:11:19,800 --> 00:11:21,280
partnerships, but. 
Yeah. 

219
00:11:21,400 --> 00:11:23,560
Well, their strategy is they're 
going to put everyone else's 

220
00:11:23,560 --> 00:11:25,480
model on AWS, right. 
So they're going to be the 

221
00:11:25,480 --> 00:11:28,320
middleman, like they're going to
sell you other access to other 

222
00:11:28,320 --> 00:11:30,480
people's models. 
So the anthropic models and I 

223
00:11:30,480 --> 00:11:33,000
think to the Llama models and to
whoever other models they can 

224
00:11:33,000 --> 00:11:34,520
get on there, they're going to 
get everything on there. 

225
00:11:34,800 --> 00:11:38,560
And the only like maybe they 
will, maybe they won't end up 

226
00:11:38,560 --> 00:11:41,040
with open AIS models on there. 
I think that's an interesting 

227
00:11:41,080 --> 00:11:43,600
ongoing business discussion. 
But like, in the end, they're 

228
00:11:43,600 --> 00:11:46,680
going to try to be. 
You know the supermarket for, 

229
00:11:46,680 --> 00:11:49,480
for, for models in the cloud, 
right as they have been in the 

230
00:11:49,640 --> 00:11:52,400
the the last generation of all 
the different web tools they 

231
00:11:52,400 --> 00:11:53,600
sell. 
So like. 

232
00:11:53,600 --> 00:11:58,480
But maybe maybe Azure is not 
allowing Open AI to put that 

233
00:11:58,480 --> 00:12:01,080
model on AWS essentially right? 
Yeah. 

234
00:12:01,480 --> 00:12:05,320
So is Open AI so good that that 
will cripple Amazon? 

235
00:12:06,320 --> 00:12:10,280
But it is interesting that I 
imagine most developers that who

236
00:12:10,280 --> 00:12:14,200
are using GPT models are. 
Doing that directly through the 

237
00:12:14,240 --> 00:12:17,840
open AIAPIS. 
Yeah, probably. 

238
00:12:17,840 --> 00:12:21,240
And there is, you know, there's 
this world now that was already 

239
00:12:21,240 --> 00:12:23,760
trending this way of big 
companies wanting to have like a

240
00:12:23,760 --> 00:12:28,280
foot in Azure and AWS so that 
you can sort of get the best of 

241
00:12:28,280 --> 00:12:31,760
both worlds. 
But that would seem to be good 

242
00:12:31,760 --> 00:12:34,760
for Microsoft given that they 
were not the first place player.

243
00:12:34,760 --> 00:12:38,080
So even saying OK this further 
creates this world where you 

244
00:12:38,080 --> 00:12:43,600
want hybrid cloud or you want to
dual app clouds is is sort of 

245
00:12:43,600 --> 00:12:46,840
good for Microsoft. 
I I can't believe I'm talking 

246
00:12:46,920 --> 00:12:51,320
cloud computing on the podcast. 
I feel like even even even as 

247
00:12:51,360 --> 00:12:53,520
long as we're willing to go it's
funny to be in that room. 

248
00:12:55,040 --> 00:12:57,040
OK. 
Facebook, I think part, you know

249
00:12:57,040 --> 00:13:01,080
Facebook is releasing Llama 
which is sort of their, they're 

250
00:13:01,080 --> 00:13:04,720
sort of the king public company 
and open source models. 

251
00:13:05,320 --> 00:13:10,360
Their their their top AI person 
is Jan Lacun who's sort of he's 

252
00:13:10,360 --> 00:13:14,840
been at this forever and has had
a lot of spicy and open takes. 

253
00:13:15,400 --> 00:13:18,440
I mean I guess the the core 
question is just like what do 

254
00:13:18,440 --> 00:13:22,680
you make of Facebook which is 
sort of a closed social network 

255
00:13:23,000 --> 00:13:28,360
being so pro like let's just 
give, let's give foundation 

256
00:13:28,360 --> 00:13:30,800
models away, make it easy, let's
commoditize them. 

257
00:13:31,000 --> 00:13:33,800
Let's just like give away. 
What do you make of the Llama 

258
00:13:33,800 --> 00:13:34,640
strategy? 
Like what's? 

259
00:13:34,920 --> 00:13:38,000
What's the point of giving away 
Foundation models and making 

260
00:13:38,000 --> 00:13:40,240
that such a core piece of your 
approach? 

261
00:13:41,200 --> 00:13:43,280
I mean, I love it. 
I think it's good. 

262
00:13:44,040 --> 00:13:48,840
Yeah, I think it's great. 
I think it's really, it's really

263
00:13:48,840 --> 00:13:56,520
probably primarily about 
developer relations and creating

264
00:13:56,520 --> 00:14:00,240
goodwill in the open source and 
the developer communities. 

265
00:14:00,560 --> 00:14:06,600
That for them is more important 
overall to the like longevity of

266
00:14:06,600 --> 00:14:11,440
their business than. 
Owning you know, or monetizing 

267
00:14:12,200 --> 00:14:17,960
these APIs, these models through
APIs which would be a just an 

268
00:14:17,960 --> 00:14:22,960
entirely new business for them. 
So I don't know, that's that's 

269
00:14:22,960 --> 00:14:23,960
my take. 
I don't know. 

270
00:14:23,960 --> 00:14:25,280
Do you guys have a different 
perspective? 

271
00:14:26,280 --> 00:14:29,000
I just think that I think that 
there's basically 2 strategies 

272
00:14:29,000 --> 00:14:31,400
here with like open and closed 
source software, right? 

273
00:14:31,400 --> 00:14:34,200
One is like closed source where 
you keep it private. 

274
00:14:34,360 --> 00:14:35,720
You're like we have the best 
stuff. 

275
00:14:36,160 --> 00:14:39,600
And people will have to come to 
us even though they won't know 

276
00:14:39,640 --> 00:14:41,840
the sort of how the internal 
workings of the software are 

277
00:14:41,840 --> 00:14:43,640
because it's the best, right. 
And that's pretty much the open 

278
00:14:43,640 --> 00:14:46,240
the eye strategy, right, like 
and some of the other foundation

279
00:14:46,240 --> 00:14:47,960
model companies like Anthropic 
whatever. 

280
00:14:48,400 --> 00:14:51,080
And then Facebook is like, OK, 
somebody already took the closed

281
00:14:51,080 --> 00:14:52,400
source. 
We're the best strategy. 

282
00:14:52,600 --> 00:14:55,840
We're going to take the the, you
know, opposite strategy which 

283
00:14:55,840 --> 00:14:57,760
is. 
We're going to open source all 

284
00:14:57,760 --> 00:15:00,280
these models, right? 
And we're going to hopefully 

285
00:15:00,280 --> 00:15:02,800
build developer communities 
around these models so that 

286
00:15:02,960 --> 00:15:05,800
people can make them better. 
They can, you know, fine tune 

287
00:15:05,800 --> 00:15:07,760
them. 
They can like, you know, help 

288
00:15:07,760 --> 00:15:11,800
optimize them over time. 
And then we will have like the, 

289
00:15:12,400 --> 00:15:15,240
you know, the. 
Thousands or, you know, 10s of 

290
00:15:15,240 --> 00:15:18,680
thousands of developers who are 
working on top of these models, 

291
00:15:18,840 --> 00:15:22,840
like build and improve them and 
build a network of of plugins 

292
00:15:22,840 --> 00:15:25,120
and, you know, optimize them and
all this good stuff. 

293
00:15:25,120 --> 00:15:27,640
And so we're going to get value 
out of the developer community 

294
00:15:27,640 --> 00:15:30,640
and and therefore the models 
that our business Facebook is 

295
00:15:30,640 --> 00:15:32,600
built on top of will will get 
better and better that way, 

296
00:15:32,600 --> 00:15:34,400
right. 
And so I think it's like it may 

297
00:15:34,400 --> 00:15:37,240
not work, but I think it's 
better than just trying to. 

298
00:15:38,240 --> 00:15:40,560
It's differentiated. 
Yeah, It's differentiated, 

299
00:15:40,560 --> 00:15:41,360
right. 
Yeah. 

300
00:15:41,520 --> 00:15:44,000
And you know, there are lots of 
great startups that have been 

301
00:15:44,000 --> 00:15:45,800
built. 
You know where it's like we're 

302
00:15:45,800 --> 00:15:48,040
going to give away a foundation,
we're going to give away, sorry,

303
00:15:48,040 --> 00:15:50,880
an open source project and then 
you could build a business 

304
00:15:50,880 --> 00:15:56,320
around if it's super successful.
Also, for Facebook, they 

305
00:15:56,640 --> 00:16:00,360
probably benefit a lot from 
hiring great AI talent in the 

306
00:16:00,360 --> 00:16:01,760
long run. 
And and if. 

307
00:16:02,400 --> 00:16:06,760
If that alone just gets them, 
you know marginally, you know 

308
00:16:07,200 --> 00:16:09,120
percentage. 
I know Reels is getting better. 

309
00:16:09,120 --> 00:16:11,920
You know, I I feel like the. 
The reels is good. 

310
00:16:11,920 --> 00:16:15,280
Reels is a good algorithm. 
When if Facebook and Google nail

311
00:16:15,520 --> 00:16:19,920
AI generated ad advertisements, 
right, or add just even the 

312
00:16:19,920 --> 00:16:23,560
words in the advertisements, the
copyright like that, you have to

313
00:16:23,560 --> 00:16:26,480
imagine that's a huge tailwind 
for the business, right? 

314
00:16:26,480 --> 00:16:29,680
Like like. 
You know, personalized ads that 

315
00:16:29,680 --> 00:16:32,360
are created by AI in real time 
as you scroll to Instagram 

316
00:16:32,360 --> 00:16:34,360
feeder or the Facebook feed or 
whatever. 

317
00:16:34,360 --> 00:16:36,000
Like, you have to imagine 
that's. 

318
00:16:36,600 --> 00:16:40,600
Yeah, yeah, you have to imagine 
that's a crazy good business for

319
00:16:40,600 --> 00:16:42,480
them, right? 
Or improvement to their business

320
00:16:42,480 --> 00:16:44,720
and lowers the friction to 
actually buying the ads. 

321
00:16:45,040 --> 00:16:48,280
It's kind of interesting if you 
think of it that way, because 

322
00:16:48,280 --> 00:16:52,720
what you're saying Max is AI 
just improving and getting 

323
00:16:52,720 --> 00:16:56,560
better faster. 
Is, is is what's best for 

324
00:16:56,880 --> 00:17:01,800
Facebook overall, almost like if
we had a breakthrough in, you 

325
00:17:01,800 --> 00:17:04,359
know, energy production of 
energy or clean energy or 

326
00:17:04,359 --> 00:17:06,319
something that would be great 
for Facebook too, probably 

327
00:17:06,319 --> 00:17:09,240
because they would, you know, 
have a lot, we could run their 

328
00:17:09,240 --> 00:17:12,200
data centers a lot cheaper and 
you know all these things, 

329
00:17:12,200 --> 00:17:14,640
right? 
So it's almost just like better 

330
00:17:14,640 --> 00:17:18,200
for them to encourage the 
advancement of this industry 

331
00:17:18,200 --> 00:17:20,760
faster cause a lot of the gains 
will accrue to them. 

332
00:17:21,839 --> 00:17:24,800
We we jumped into this with 
business strategy which I am 

333
00:17:24,800 --> 00:17:27,319
enjoying. 
But to me what's interesting 

334
00:17:27,319 --> 00:17:32,600
about Facebook's positioning is 
that you know Google, Microsoft 

335
00:17:32,600 --> 00:17:36,880
with Open AI to some degree 
they're playing the responsible 

336
00:17:36,880 --> 00:17:39,880
steward right. 
It's we are going to make sure 

337
00:17:40,120 --> 00:17:44,120
our language model behaves 
appropriately, doesn't get off 

338
00:17:44,120 --> 00:17:46,360
the rails. 
We're our brands are at stake 

339
00:17:46,360 --> 00:17:48,400
and we need to really guard 
them. 

340
00:17:48,800 --> 00:17:51,960
You know, Mustafa Suleiman at 
Inflection, who's going to be 

341
00:17:51,960 --> 00:17:55,280
speaking at the conference has 
sort of staked out and you know,

342
00:17:55,280 --> 00:17:56,600
we're going to interrogate this 
more. 

343
00:17:56,600 --> 00:17:58,720
And so I don't want to 
caricature it before he speaks. 

344
00:17:58,720 --> 00:18:02,320
But like a somewhat open source,
skeptical position, right? 

345
00:18:02,320 --> 00:18:06,520
It it undermines the idea that 
we need to really be thoughtful 

346
00:18:06,520 --> 00:18:10,720
and protective about who gets 
access to what and what these 

347
00:18:10,720 --> 00:18:14,280
systems can do. 
And so Facebook is playing sort 

348
00:18:14,280 --> 00:18:19,240
of AI, don't know flamethrower 
fire thrower type role here 

349
00:18:19,240 --> 00:18:23,440
where they're throwing fuel to 
the fire on open source in an 

350
00:18:23,440 --> 00:18:26,040
area where a lot of the other 
big tech companies would say the

351
00:18:26,040 --> 00:18:29,040
thing we can do is like do this 
really well and then keep it 

352
00:18:29,040 --> 00:18:32,760
really closed off so that, you 
know, we have basically we're 

353
00:18:32,760 --> 00:18:36,520
almost as functioning as a state
ourselves where we're really 

354
00:18:36,520 --> 00:18:38,720
being thoughtful. 
What do What do you take of 

355
00:18:39,000 --> 00:18:42,080
Facebook, the big company 
playing this role of 

356
00:18:42,520 --> 00:18:48,040
democratizing? 
I do worry, I guess for them if.

357
00:18:48,640 --> 00:18:51,480
Someone uses Llama for nefarious
purposes like. 

358
00:18:51,480 --> 00:18:53,200
Essentially, it hasn't. 
It's a brand, but. 

359
00:18:53,840 --> 00:18:57,480
Yeah, I think that could happen.
I think that this could be like 

360
00:18:57,480 --> 00:19:01,280
a long running issue now for 
them going forward the next few 

361
00:19:01,280 --> 00:19:04,880
years. 
If for if, if, as we've talked 

362
00:19:04,880 --> 00:19:07,440
about in previous episodes, 
these models can be used for 

363
00:19:07,440 --> 00:19:10,160
nefarious, really is laying the 
groundwork for reporters here. 

364
00:19:10,160 --> 00:19:12,840
You're like, OK, people should 
be held accountable for what 

365
00:19:12,840 --> 00:19:18,880
their foundation models, their 
open source project does. 

366
00:19:19,240 --> 00:19:20,160
I don't. 
I don't. 

367
00:19:20,760 --> 00:19:24,960
Know if I I don't know if I 
personally agree that I'm saying

368
00:19:24,960 --> 00:19:27,320
that will be the. 
That could be a narrative for 

369
00:19:27,320 --> 00:19:32,960
sure, and it just depends on how
how much harm can be done with 

370
00:19:32,960 --> 00:19:35,160
open source models, and I think 
we don't know yet. 

371
00:19:35,640 --> 00:19:38,000
I mean. 
Facebook's like, look, our brand

372
00:19:38,000 --> 00:19:41,560
for trust and safety is already 
so bad that there's nothing we 

373
00:19:41,560 --> 00:19:43,080
can do that could possibly make 
it worse. 

374
00:19:43,080 --> 00:19:46,640
So why don't we just go for it 
On the business front, they're 

375
00:19:46,640 --> 00:19:49,560
like Cambridge Analytica. 
Wait till they see what people 

376
00:19:49,560 --> 00:19:52,240
will do with this model. 
Well, Cambridge Analytica is a 

377
00:19:52,240 --> 00:19:56,920
great similar analogy because 
they thought they were. 

378
00:19:57,400 --> 00:20:00,480
At the time they were opening up
the Thorngren graph and they 

379
00:20:00,480 --> 00:20:04,440
were, you know, giving data to, 
you know, other developers and 

380
00:20:04,440 --> 00:20:06,120
companies. 
And they were, and that was, you

381
00:20:06,120 --> 00:20:10,600
know, encouraged almost at the 
time, right, by developers. 

382
00:20:10,920 --> 00:20:13,040
And it blew back on them 
entirely. 

383
00:20:13,040 --> 00:20:15,320
And it wasn't even good for 
their business to be doing that,

384
00:20:15,560 --> 00:20:19,000
to be giving, leaking that data.
So it is there is kind of a 

385
00:20:19,000 --> 00:20:22,120
similar thing occurring here 
potentially. 

386
00:20:23,480 --> 00:20:26,560
Facebook's willing to be wild. 
Well the narrative kind of can 

387
00:20:26,560 --> 00:20:30,400
change on you right that with at
the beginning of social 

388
00:20:30,400 --> 00:20:33,960
networking and and technology 
this was encouraged that we 

389
00:20:33,960 --> 00:20:37,520
would want you know more more 
data sharing or something with 

390
00:20:37,520 --> 00:20:41,200
developers and that and then 
Fast forward 5-10 years and it's

391
00:20:41,200 --> 00:20:44,160
not have we all staked out 
points of view. 

392
00:20:44,160 --> 00:20:47,200
I think we're all on the same 
page on do we believe in open 

393
00:20:47,200 --> 00:20:50,480
source like data bricks, super 
pro open source, like a lot of 

394
00:20:50,480 --> 00:20:56,120
the, you know, obviously hugging
face clam at CV1, super pro, 

395
00:20:56,120 --> 00:20:58,080
open source. 
Like, I feel like generally 

396
00:20:58,440 --> 00:21:01,280
we've heard from a lot of pro 
open source people. 

397
00:21:01,720 --> 00:21:04,520
Do either of you have 
reservations? 

398
00:21:04,520 --> 00:21:08,280
Are you fully in the camp of 
like it's an arms race and just 

399
00:21:08,280 --> 00:21:10,840
like get the best stuff out and 
let good people sort of take 

400
00:21:10,840 --> 00:21:15,080
advantage of? 
It my take is that open source 

401
00:21:15,080 --> 00:21:19,120
right now is great for the 
community, for developers, it's 

402
00:21:19,120 --> 00:21:23,880
awesome to be able to. 
You know, edit weights or use 

403
00:21:23,880 --> 00:21:27,080
smaller models, run, run in or 
your own inference, all these 

404
00:21:27,080 --> 00:21:33,000
things fine tune, right? 
And that just might not be the 

405
00:21:33,000 --> 00:21:35,280
case going forward. 
I don't know, at some point 

406
00:21:35,280 --> 00:21:37,840
maybe I will change my mind and 
be like, no, the models are too 

407
00:21:37,840 --> 00:21:40,240
powerful. 
Like I don't want these open 

408
00:21:40,240 --> 00:21:45,560
source models in the hands of 
nefarious actors or you know, 

409
00:21:45,800 --> 00:21:48,960
China or something. 
So I I guess my take is it's 

410
00:21:48,960 --> 00:21:51,520
great right now. 
I don't know, but it's hard to 

411
00:21:51,560 --> 00:21:53,360
pull. 
You know, that's sort of the 

412
00:21:53,680 --> 00:21:56,720
once they they have the models 
and it's gone too far, it seems 

413
00:21:56,720 --> 00:22:00,560
hard, hard to unwind the clock 
or whatever. 

414
00:22:01,640 --> 00:22:04,960
Yeah, I guess I I just believe 
the nefarious actors in this 

415
00:22:04,960 --> 00:22:09,120
scenario are unstoppable. 
And as we discussed in The AI 

416
00:22:09,120 --> 00:22:12,040
Kills Us All episode that 
somebody's going to figure out 

417
00:22:12,040 --> 00:22:15,480
how to you know, some bad, bad 
actor is going to get control of

418
00:22:15,480 --> 00:22:18,040
a a really powerful model 
regardless of whether or not it 

419
00:22:18,040 --> 00:22:22,040
was open source or not so. 
Yeah, I mean give the, so you're

420
00:22:22,200 --> 00:22:25,480
you're very pro give the pro 
open source give the give the 

421
00:22:25,480 --> 00:22:27,160
fruits of the labor away to the 
community. 

422
00:22:27,160 --> 00:22:29,840
I mean why do open source? 
Why do we think? 

423
00:22:30,640 --> 00:22:34,760
Why do we think Open AI changed 
its tune so dramatically on 

424
00:22:34,760 --> 00:22:35,320
this? 
Question. 

425
00:22:35,320 --> 00:22:38,600
Awesome business strategy. 
Just like, just protect your 

426
00:22:38,640 --> 00:22:40,680
business. 
Good for the front runner, 

427
00:22:40,680 --> 00:22:43,120
right? 
Like like like regulate it. 

428
00:22:43,120 --> 00:22:45,640
Like let tell. 
Make sure you have to spend a 

429
00:22:45,640 --> 00:22:49,120
lot of money, like following 
rules like that's good for big 

430
00:22:49,120 --> 00:22:50,840
companies. 
Because once they realized how 

431
00:22:50,880 --> 00:22:54,000
good the stuff they were cooking
was, why give it away? 

432
00:22:54,000 --> 00:22:56,800
I mean, open source is the 
strategy when you're not the 

433
00:22:56,800 --> 00:22:59,160
leader. 
Like not, you know, you never, 

434
00:22:59,240 --> 00:23:01,800
never give away the the high 
quality stuff, yeah. 

435
00:23:02,360 --> 00:23:06,280
Or or the strategy that makes 
sense for meta because they have

436
00:23:06,280 --> 00:23:09,560
this these huge other businesses
right like you would. 

437
00:23:09,720 --> 00:23:13,680
It doesn't really make sense. 
To give away models as a start 

438
00:23:13,680 --> 00:23:17,560
up I guess I I mean maybe maybe 
that's what stability is kind of

439
00:23:17,560 --> 00:23:20,520
doing or mid journey but you 
said they're kind of like 

440
00:23:21,480 --> 00:23:25,800
becoming more close source. 
Imagine a world where like these

441
00:23:25,880 --> 00:23:31,920
algorithms are near AGI like and
then they are locked down. 

442
00:23:31,920 --> 00:23:33,240
You know that's the ultimate 
elite. 

443
00:23:33,240 --> 00:23:37,120
So like a a small group of 
companies gets to control like 

444
00:23:37,520 --> 00:23:41,960
this sort of Infinity resource 
like it seems that seems insane.

445
00:23:41,960 --> 00:23:44,960
You know what I mean? 
I feel like that would become 

446
00:23:44,960 --> 00:23:48,680
sort of one of the great freedom
questions of our time, that just

447
00:23:48,880 --> 00:23:52,040
in any way restricting the set 
of people who have limited to 

448
00:23:52,440 --> 00:23:55,360
access to like a God like 
reasoning and information 

449
00:23:55,360 --> 00:23:56,840
system. 
I feel like we're we're 

450
00:23:56,840 --> 00:24:00,920
backtracking on Oh yeah kills us
all episode but but I mean I do 

451
00:24:01,000 --> 00:24:03,960
think open source was key to 
Facebook and that's how we got 

452
00:24:03,960 --> 00:24:06,840
down this path. 
But reasonable that we've gone 

453
00:24:06,880 --> 00:24:10,160
far from that though you've been
the one that makes the point 

454
00:24:10,160 --> 00:24:14,240
that like often these 
discussions undercount the AGI 

455
00:24:14,240 --> 00:24:16,600
potential and how disruptive 
that would be to any of the 

456
00:24:16,640 --> 00:24:19,320
other arguments that we're. 
Having I total, I totally agree.

457
00:24:19,320 --> 00:24:23,080
And that's where I would 
question your assumption that 

458
00:24:23,080 --> 00:24:25,880
everyone should have AGI. 
Like Without knowing how 

459
00:24:25,880 --> 00:24:29,320
dangerous that could be for the 
world, that seems like pretty 

460
00:24:29,320 --> 00:24:33,200
presumptuous to just assume that
that it should not be regulated 

461
00:24:33,200 --> 00:24:37,280
by a handful of companies. 
Yeah, I mean it's a lot just. 

462
00:24:37,520 --> 00:24:41,160
Give away the God like powers. 
Everyone deserves God like 

463
00:24:41,160 --> 00:24:42,920
powers. 
I mean, everyone deserves what 

464
00:24:42,920 --> 00:24:45,800
can go around. 
Yeah, I mean they they 

465
00:24:46,920 --> 00:24:49,840
presumably they will have some 
counter, you know 

466
00:24:50,040 --> 00:24:52,600
countervailing. 
Presumably, sure, yeah. 

467
00:24:52,640 --> 00:24:57,560
I guess if that's the case, you 
know, but, oh, I guess here's an

468
00:24:57,560 --> 00:24:59,120
easy way to make sure we haven't
missed anybody. 

469
00:24:59,360 --> 00:25:03,480
Let's Let's do a collective 
ranking of who we think benefits

470
00:25:03,480 --> 00:25:09,440
the most from AI in terms of 
market cap movements experienced

471
00:25:09,440 --> 00:25:13,840
and to be experienced. 
This is I'm putting, I'm putting

472
00:25:13,840 --> 00:25:15,480
in no we're doing it together. 
I'm putting. 

473
00:25:15,480 --> 00:25:17,120
I think NVIDIA should be our 
number one. 

474
00:25:18,040 --> 00:25:21,040
I mean, Microsoft would be the 
close one there, I think. 

475
00:25:21,720 --> 00:25:25,680
I mean, Microsoft stock isn't 
actually up anywhere nearly. 

476
00:25:25,680 --> 00:25:29,240
So, so NVIDIA then probably, I 
mean to your point I guess, but 

477
00:25:29,320 --> 00:25:30,880
I guess the question. 
What about Google? 

478
00:25:30,880 --> 00:25:34,800
I I would put Google pretty high
here, Microsoft. 

479
00:25:34,800 --> 00:25:40,800
Depends on what time frame. 34% 
over the year and 187% / 5 

480
00:25:40,960 --> 00:25:44,960
years? 
Google you're putting Google. 

481
00:25:45,960 --> 00:25:47,960
I don't know. 
What to me? 

482
00:25:47,960 --> 00:25:51,720
What time rules in though like 
tech becoming more important and

483
00:25:51,720 --> 00:25:54,760
Google being all over it is good
for tech is good for Google. 

484
00:25:54,760 --> 00:25:58,960
But like Google's relative 
dominance, which was so powerful

485
00:26:00,000 --> 00:26:03,840
in search, feels threatened by 
this, so I I just. 

486
00:26:04,360 --> 00:26:10,120
No, I think that's worth talking
about because I just feel pretty

487
00:26:10,680 --> 00:26:17,200
confident that that whole search
impact of AI has like is like 

488
00:26:17,280 --> 00:26:22,320
very, very early and Google will
have plenty of time to 

489
00:26:22,640 --> 00:26:27,080
potentially continue dominance 
with you know using LMS and and 

490
00:26:27,120 --> 00:26:32,280
AI and multimodal models in in 
in their search products, so. 

491
00:26:33,440 --> 00:26:37,400
They have, they have the the 
front page right that everyone 

492
00:26:37,400 --> 00:26:41,640
goes to for search and they have
really powerful AI internally. 

493
00:26:42,960 --> 00:26:45,360
It just hasn't played out yet. 
But I don't. 

494
00:26:45,360 --> 00:26:46,800
I believe that they will get 
there. 

495
00:26:47,760 --> 00:26:50,000
All right. 
So I mean you're you want Google

496
00:26:50,080 --> 00:26:53,000
first it just. 
Depends what we're talking 

497
00:26:53,000 --> 00:26:55,840
about. 
Like are we talking about which 

498
00:26:55,920 --> 00:26:58,560
of the biggest companies will 
have the highest market caps 

499
00:26:58,560 --> 00:27:02,880
from AII mean or just like 
relatively like you know? 

500
00:27:03,360 --> 00:27:05,280
Is their position improved by 
AI? 

501
00:27:05,720 --> 00:27:10,160
How much the percentage of their
position is improved by AI? 

502
00:27:10,360 --> 00:27:15,480
That's what you're asking. 
Yeah, I mean NVIDIA. 

503
00:27:16,200 --> 00:27:23,360
Yeah, NVIDIA for sure. 
Microsoft second agree. 

504
00:27:24,040 --> 00:27:27,120
Facebook Third I would go what? 
Yeah, I don't. 

505
00:27:27,160 --> 00:27:29,600
I don't agree, because 
algorithms are core to what they

506
00:27:29,600 --> 00:27:31,120
do in their discovery. 
Mechanism. 

507
00:27:31,120 --> 00:27:34,880
I think if you actually, if you 
oh, because of algorithms are 

508
00:27:34,880 --> 00:27:37,160
core to what they do, OK. 
I just think if you believe 

509
00:27:37,160 --> 00:27:41,440
there can be a 50% improvement 
in advertising at scale through 

510
00:27:41,560 --> 00:27:45,840
AI generation, then Facebook 
benefits more from that than 

511
00:27:45,840 --> 00:27:48,000
anyone except Google and and 
they might think more than 

512
00:27:48,000 --> 00:27:49,920
Google. 
I have just put Google before. 

513
00:27:50,080 --> 00:27:52,880
OK, I mean, that's fine. 
It's, but it's a totally. 

514
00:27:52,880 --> 00:27:56,840
Agree with Eric, but but. 
It's just about search versus 

515
00:27:56,840 --> 00:27:59,200
the Instagram feed. 
Which one gets more benefits out

516
00:27:59,200 --> 00:28:03,560
of AI like, I think that's the 
real question here, like or 

517
00:28:03,560 --> 00:28:06,640
search versus the Instagram feed
plus the Facebook feed, right? 

518
00:28:06,640 --> 00:28:09,720
And I guess it also has YouTube 
which is competing in like 

519
00:28:09,720 --> 00:28:11,640
shorts. 
Yeah. 

520
00:28:11,640 --> 00:28:13,920
But it's going to take a while 
to do video generation, I guess.

521
00:28:13,920 --> 00:28:16,880
I don't know, like it's sort of 
about the latter of how quickly 

522
00:28:17,040 --> 00:28:19,680
generated personalized ads kind 
of get scaled up. 

523
00:28:19,680 --> 00:28:20,920
So it's. 
Yeah. 

524
00:28:21,480 --> 00:28:24,680
Facebook, you're mostly talking 
about their ad products and. 

525
00:28:24,960 --> 00:28:27,760
Ranking algorithms. 
But there's also an argument 

526
00:28:27,760 --> 00:28:31,480
around on the content generation
piece itself that as we've 

527
00:28:31,480 --> 00:28:35,520
talked about before, there's, 
you know, ability to create more

528
00:28:35,800 --> 00:28:38,400
compelling content and social 
media posts and I. 

529
00:28:38,600 --> 00:28:41,080
I really think the only way you 
can make the case for Google is 

530
00:28:41,080 --> 00:28:43,120
that you would say the ads on 
Google are going to get much 

531
00:28:43,120 --> 00:28:44,360
better, right? 
I mean like. 

532
00:28:44,480 --> 00:28:47,800
Well, what about search itself? 
You think search is going to get

533
00:28:47,800 --> 00:28:50,240
so much better I. 
Think search itself is going to 

534
00:28:50,240 --> 00:28:52,400
get so dramatically better? 
Yes. 

535
00:28:52,400 --> 00:28:55,440
Like I think, I don't know, I 
mean time spent in search you 

536
00:28:55,440 --> 00:28:57,840
would think that go up if 
they're chat type products. 

537
00:28:57,840 --> 00:29:01,560
Well, well, there is an argument
that time spent goes down, 

538
00:29:01,560 --> 00:29:02,880
right? 
Because you're just getting your

539
00:29:02,880 --> 00:29:05,960
answer faster at an end. 
But it's you're not. 

540
00:29:05,960 --> 00:29:09,000
It's no longer a link. 
It's like less of a link system 

541
00:29:09,000 --> 00:29:12,320
and more like in the search, 
which is where Google ads are. 

542
00:29:12,880 --> 00:29:15,520
Well, it just depends. 
If it's more, it becomes more 

543
00:29:15,520 --> 00:29:18,760
like a ChatGPT model where 
you're chatting with Bard or 

544
00:29:18,800 --> 00:29:20,640
instead of searching. 
I guess that's where I'm coming 

545
00:29:20,640 --> 00:29:23,000
from. 
I I see me and myself doing 

546
00:29:23,000 --> 00:29:25,200
that. 
Like almost more than. 

547
00:29:25,200 --> 00:29:27,720
I searched Google today in the 
future, and I think they're the 

548
00:29:27,720 --> 00:29:33,240
best positioned to be that like 
true assistant that I use every 

549
00:29:33,240 --> 00:29:38,040
day to talk to and instead of 
searching by typing into AI mean

550
00:29:38,200 --> 00:29:40,760
I was down on Google before. 
But like Google versus Facebook,

551
00:29:40,760 --> 00:29:42,960
I'm getting heated up just in 
the sense that, like, if you 

552
00:29:42,960 --> 00:29:46,800
think of every product that 
Google offers like if Gmail was 

553
00:29:46,800 --> 00:29:49,920
amazing from AI and like the 
prompts were great and there was

554
00:29:49,920 --> 00:29:52,760
writing like could become an 
even more insane product. 

555
00:29:52,880 --> 00:29:56,000
Obviously Like Google cloud 
services. 

556
00:29:56,000 --> 00:29:59,120
Wait, sorry, it's Google. 
Cloud platform, global cloud 

557
00:29:59,200 --> 00:30:03,240
platform is key. 
If they can improve their like 

558
00:30:03,520 --> 00:30:06,480
search, YouTube, like they're 
just so many pieces of their 

559
00:30:06,480 --> 00:30:10,640
business where they have 
audience and they could put this

560
00:30:10,640 --> 00:30:13,200
to work. 
The only question we all have 

561
00:30:13,200 --> 00:30:17,840
probably is execution on all of 
these areas, but the opportunity

562
00:30:17,840 --> 00:30:21,120
to me seems massive. 
I mean, I guess I'm just going 

563
00:30:21,120 --> 00:30:23,400
based on the fact that Google 
makes no money on anything 

564
00:30:23,400 --> 00:30:25,480
except search and the other 
stuff is great, but it's really 

565
00:30:25,480 --> 00:30:28,600
just a data vacuum for for 
search ads, like they really 

566
00:30:28,600 --> 00:30:30,360
don't make any money. 
I mean, they make one on YouTube

567
00:30:30,360 --> 00:30:31,720
now, so that would be the other 
big thing. 

568
00:30:31,880 --> 00:30:35,560
But like search ads is the whole
ball game if you're just purely 

569
00:30:35,560 --> 00:30:37,640
talking about the business of 
Google, pretty much. 

570
00:30:37,960 --> 00:30:40,520
And so you have to believe that 
search ads are going to get way 

571
00:30:40,520 --> 00:30:43,600
better and or that people are 
going to spend way more time in 

572
00:30:43,600 --> 00:30:45,680
search, which again is an 
execution question about like 

573
00:30:45,680 --> 00:30:48,240
Bard versus all the other, you 
know, large language models and 

574
00:30:48,240 --> 00:30:50,200
chat models. 
So like I buy the case, I just 

575
00:30:50,200 --> 00:30:52,280
don't think it's as clear cut as
you guys do because I just think

576
00:30:52,280 --> 00:30:55,200
Facebook like Instagram ads and 
Facebook ads getting better 

577
00:30:55,200 --> 00:30:57,680
would like to immediately dump 
right into the business. 

578
00:30:57,880 --> 00:31:01,720
Which and what you're saying is 
that there's more of an obvious 

579
00:31:02,240 --> 00:31:06,400
trajectory for. 
The Instagram ads and feed to 

580
00:31:06,400 --> 00:31:09,440
get better from AI than there is
for search ads to get better 

581
00:31:09,440 --> 00:31:10,800
from AI. 
Yeah, yeah. 

582
00:31:10,800 --> 00:31:13,520
I mean, it's close. 
They're, you know, they both are

583
00:31:13,520 --> 00:31:15,240
going to get a lot better. 
So, Yep. 

584
00:31:15,240 --> 00:31:17,000
But it's about which one's 
bigger and what's the bigger 

585
00:31:17,000 --> 00:31:23,200
impact, I guess, Yeah, notably, 
notably like, so yeah, we, we, 

586
00:31:23,520 --> 00:31:27,280
it's Facebook or I don't know, I
Facebook wouldn't necessarily be

587
00:31:27,280 --> 00:31:28,360
my next. 
Either. 

588
00:31:28,400 --> 00:31:29,320
Oh, really? 
Interesting. 

589
00:31:29,320 --> 00:31:30,560
I'm. 
I'm willing to. 

590
00:31:30,800 --> 00:31:32,400
I mean, I. 
Just feel I would bet it forth 

591
00:31:32,400 --> 00:31:35,640
at least, but what? 
What's next? 

592
00:31:35,640 --> 00:31:38,680
We have Apple and Amazon, you 
know, like are now towards the 

593
00:31:38,680 --> 00:31:40,320
bottom of our at least tech 
list. 

594
00:31:40,320 --> 00:31:44,800
Obviously tech being better in 
some ways lifts all boats, but I

595
00:31:44,800 --> 00:31:49,720
don't know what do you make of 
those two and like what it means

596
00:31:49,720 --> 00:31:52,480
for them. 
I would put Amazon next. 

597
00:31:52,480 --> 00:31:58,000
I think they have an obvious 
role to play with AWS and they 

598
00:31:58,000 --> 00:32:00,160
have made a lot of great 
partnerships there with 

599
00:32:00,160 --> 00:32:03,600
Anthropic already. 
They are probably working 

600
00:32:03,600 --> 00:32:09,280
internally on models themselves.
They are the number one cloud 

601
00:32:09,280 --> 00:32:12,920
platform. 
They will benefit in their other

602
00:32:12,920 --> 00:32:18,040
core businesses through AI 
potentially, you know Amazon.com

603
00:32:19,240 --> 00:32:21,520
and ads there, which is a 
increasing part of their 

604
00:32:21,520 --> 00:32:26,920
business and not to ignore it, 
you're allowed to talk about it.

605
00:32:27,120 --> 00:32:30,800
Yeah, Alexa. 
Which is critical to our 

606
00:32:30,800 --> 00:32:36,520
business. 
They have an end point in 30% of

607
00:32:36,560 --> 00:32:39,800
US homes that is an AI 
assistant. 

608
00:32:40,520 --> 00:32:44,480
It historically hasn't been 
amazing at talking to you 

609
00:32:44,680 --> 00:32:48,640
beyond, you know, controlling 
smart home features and and 

610
00:32:48,640 --> 00:32:50,960
telling you the weather and 
playing volley games. 

611
00:32:51,280 --> 00:32:56,720
But it's it's completely 
possible that they will be able 

612
00:32:56,720 --> 00:33:00,320
to. 
AD LLMS and and AI experiences 

613
00:33:00,320 --> 00:33:04,200
as they've already demoed that 
would transform that device in a

614
00:33:04,200 --> 00:33:08,640
way that would be you know 
essential would would would be a

615
00:33:08,640 --> 00:33:10,440
game changer for the Alexa 
business. 

616
00:33:10,920 --> 00:33:15,320
So I think they're the furthest 
ahead in that area on like 

617
00:33:15,320 --> 00:33:19,360
potentially integrating AI with,
you know, your average day 

618
00:33:19,640 --> 00:33:23,760
through a Voice Assistant. 
And Apple, I mean Apple, I mean 

619
00:33:23,760 --> 00:33:27,240
the argument Apple gets away 
with everything in that we can 

620
00:33:27,240 --> 00:33:30,520
put them in like basically last 
place and it's still like, well,

621
00:33:30,520 --> 00:33:32,280
they move slowly and maybe 
they'll figure something out. 

622
00:33:32,280 --> 00:33:35,480
And like, I don't know, it's 
definitely their wheelhouse. 

623
00:33:35,480 --> 00:33:38,360
It's like it's not, it's a lot 
of data. 

624
00:33:38,360 --> 00:33:41,600
It's not about privacy. 
It's sort of not a device. 

625
00:33:41,600 --> 00:33:44,720
I mean, I guess, well, I mean, 
if we really want to criticize 

626
00:33:44,720 --> 00:33:48,480
Apple, we saw this report that, 
you know, maybe Sam Altman and 

627
00:33:48,480 --> 00:33:54,240
Johnny I've and Masa son 
Masayoshi, son of SoftBank are 

628
00:33:54,400 --> 00:33:57,880
getting the team together weird.
The weirdest team headline you 

629
00:33:57,880 --> 00:34:02,120
could imagine to build a phone 
or some device that's sort of 

630
00:34:02,120 --> 00:34:05,000
got AI at its core, that that's 
a threat. 

631
00:34:05,000 --> 00:34:09,000
I mean, I could buy, you know, a
next generation sort of AI first

632
00:34:09,000 --> 00:34:11,679
device. 
I I I just think Apple's Moat 

633
00:34:11,679 --> 00:34:14,800
has always been how unbelievably
difficult it is to build 

634
00:34:14,800 --> 00:34:17,159
hardware at scale, right? 
Especially like high margin 

635
00:34:17,159 --> 00:34:19,000
hardware. 
I mean, there's almost no other 

636
00:34:19,000 --> 00:34:21,000
large hardware businesses in the
industry. 

637
00:34:21,199 --> 00:34:26,600
And even if Johnny is on board 
and Masa is on board, you still 

638
00:34:26,600 --> 00:34:28,480
need like Masa's got a great 
history. 

639
00:34:29,080 --> 00:34:32,159
You still need, yeah, you still 
need about three to five years 

640
00:34:32,159 --> 00:34:35,400
and 30 or $5 billion to build 
even like a small scale, high 

641
00:34:35,400 --> 00:34:37,960
quality hardware device. 
And just like I think that's 

642
00:34:37,960 --> 00:34:42,520
just a long way away and 
anywhere where you know Johnny 

643
00:34:42,520 --> 00:34:45,199
and and. 
Sam are playing like Tim Cook 

644
00:34:45,199 --> 00:34:47,520
will pay some attention to that 
and make sure that they're not 

645
00:34:47,560 --> 00:34:49,080
completely caught off guard, 
right. 

646
00:34:49,080 --> 00:34:51,280
You know, they all talk to the 
same people manufacturing the 

647
00:34:51,280 --> 00:34:52,880
stuff in China, so but would 
you? 

648
00:34:52,880 --> 00:34:56,159
Agree though that like Apple, 
while they're not threatened 

649
00:34:56,159 --> 00:35:00,360
that much, like relatively, they
are probably lowly ranked here. 

650
00:35:00,360 --> 00:35:01,760
Yeah, I think. 
I I think. 

651
00:35:01,760 --> 00:35:04,760
They're the most, most protected
on the downside in some case, in

652
00:35:04,760 --> 00:35:07,440
some cases you could argue and 
then also they don't have huge 

653
00:35:07,440 --> 00:35:09,160
outside this is all bullshit. 
They're fine. 

654
00:35:09,160 --> 00:35:12,520
And if it's good, you know, it 
doesn't feel like they're going 

655
00:35:12,520 --> 00:35:13,760
to disrupt it, yeah. 
Right. 

656
00:35:14,320 --> 00:35:17,280
I would like to throw another 
company in the mix here. 

657
00:35:17,800 --> 00:35:19,400
What do you guys think about 
Tesla? 

658
00:35:20,840 --> 00:35:22,200
Ah, yeah. 
Self driving. 

659
00:35:22,400 --> 00:35:26,760
The old self driving and the 
general xai suite. 

660
00:35:26,760 --> 00:35:30,840
Now, you know, it's like he's 
somehow allowed to like, merge 

661
00:35:31,040 --> 00:35:34,800
AI research, like it seems like 
across all his companies. 

662
00:35:36,480 --> 00:35:39,600
I mean, I just think that I 
think I'd be much longer way MO 

663
00:35:39,600 --> 00:35:41,160
and cruise than than Tesla or 
whatever. 

664
00:35:41,160 --> 00:35:44,400
I mean like, I just think that. 
Tesla's fundamentally built 

665
00:35:44,400 --> 00:35:47,360
around a hardware business model
where you sell cars and like if 

666
00:35:47,360 --> 00:35:49,760
you believe we're entering a 
world where all every car is 

667
00:35:49,760 --> 00:35:51,480
self driving and it can take you
anywhere. 

668
00:35:51,480 --> 00:35:54,120
Like why wouldn't you be long? 
Like, yeah, Uber or Lyft or 

669
00:35:54,120 --> 00:35:56,320
Waymo or Cruise or somebody 
who's built around this model of

670
00:35:56,320 --> 00:35:59,120
like providing cars as a service
rather than cars as like a 

671
00:35:59,360 --> 00:36:02,000
jewelry object that sits in your
garage 99% of the time. 

672
00:36:02,000 --> 00:36:04,440
Like, I just don't like. 
I think Tesla's a great product,

673
00:36:04,480 --> 00:36:07,440
but is it an AI for self driving
first product? 

674
00:36:07,440 --> 00:36:09,760
I always thought. 
The idea that it was gonna turn 

675
00:36:09,760 --> 00:36:13,120
into a self driving taxi fleet 
was like pretty absurd, but. 

676
00:36:13,640 --> 00:36:19,360
I think that's a good point. 
I think Elon, his hype or 

677
00:36:19,360 --> 00:36:24,400
marketing spin on this is that 
the AI that they've built to. 

678
00:36:24,760 --> 00:36:29,200
Successfully get self driving. 
Working at Tesla is so valuable 

679
00:36:29,200 --> 00:36:32,120
in future endeavors, right? 
Like robots, robotics. 

680
00:36:32,480 --> 00:36:35,360
But didn't they literally admit 
they gave up on all the AI 

681
00:36:35,360 --> 00:36:38,240
they've been building for? 
Until a year or two ago, they 

682
00:36:38,240 --> 00:36:41,000
had been using the last 
generation of deep learning 

683
00:36:41,000 --> 00:36:42,800
tools or whatever. 
And then at two years ago, there

684
00:36:42,800 --> 00:36:44,520
was a. 
Like recently there was a story 

685
00:36:44,520 --> 00:36:46,280
that came out that basically 
said, yeah we had to redo the 

686
00:36:46,280 --> 00:36:49,320
whole thing once we realized 
like transformer models, large 

687
00:36:49,320 --> 00:36:50,440
language models are better, 
right. 

688
00:36:50,600 --> 00:36:53,720
So like they they've been saying
that for like 7 years that like 

689
00:36:53,720 --> 00:36:56,440
we're way ahead and everyone on 
building a self driving models, 

690
00:36:56,440 --> 00:36:59,280
we have gazillion trillion miles
of data blah, blah, blah. 

691
00:36:59,280 --> 00:37:01,520
Which is fine is true. 
But like they had to throw it 

692
00:37:01,520 --> 00:37:03,000
all in the trash can like 18 
months ago. 

693
00:37:03,000 --> 00:37:06,960
So like I I don't find that like
a super compelling like case 

694
00:37:06,960 --> 00:37:08,920
that they have some Moat around 
that sort of stuff. 

695
00:37:08,920 --> 00:37:11,160
But you know, who knows? 
I mean, they do have. 

696
00:37:11,280 --> 00:37:13,880
They don't have a lot of trust 
with like regulators. 

697
00:37:13,880 --> 00:37:16,960
Like, I don't. 
I feel like, you know, Waymo and

698
00:37:16,960 --> 00:37:21,760
Cruz have just tried to be as 
responsible as possible. 

699
00:37:21,760 --> 00:37:25,480
You know, I mean, Waymo has been
very conservative in deployment.

700
00:37:25,480 --> 00:37:29,200
And you know, Cruz is like 
General Motors, like a company 

701
00:37:29,200 --> 00:37:32,320
that I feel like every American 
politician is rooting for, 

702
00:37:32,320 --> 00:37:35,120
right. 
I mean, Tesla, it seems like now

703
00:37:35,120 --> 00:37:39,280
has been able to position itself
as very close to the Republican 

704
00:37:39,280 --> 00:37:42,880
Party, but it feels like the 
administrative state and 

705
00:37:42,880 --> 00:37:46,320
Democrats are going to be, like,
extremely skeptical of them just

706
00:37:46,320 --> 00:37:51,000
throwing cars on the road. 
I mean, that's sort of cynical 

707
00:37:51,000 --> 00:37:53,920
even about, I guess, the 
politicians I like. 

708
00:37:53,920 --> 00:37:58,840
But I yeah, I think I agree with
Max's point of technology. 

709
00:37:58,840 --> 00:38:04,480
I also think having a reputation
for being a responsible, sane 

710
00:38:04,480 --> 00:38:07,160
actor is important when you're 
doing something as dangerous as 

711
00:38:07,600 --> 00:38:10,120
deploying the first self driving
cars. 

712
00:38:11,040 --> 00:38:12,680
I think it's a fair, fair 
assessment. 

713
00:38:12,760 --> 00:38:14,760
I mean we'll see if they pivot 
the attack, they're going to be 

714
00:38:14,760 --> 00:38:18,880
a stock premium for it. 
I'm sure like I guess they pivot

715
00:38:18,920 --> 00:38:21,640
the you couldn't you can't I 
don't know how you can ever 

716
00:38:21,640 --> 00:38:24,160
measure this because it's like 
well the stock can stay inflated

717
00:38:24,160 --> 00:38:26,920
for like forever So if they if 
they pivot. 

718
00:38:26,920 --> 00:38:30,280
The entire company to robo taxis
and they start selling $5 Tesla 

719
00:38:30,280 --> 00:38:32,800
rides all around America. 
I will totally change my tune on

720
00:38:32,800 --> 00:38:35,520
this, but you pretty much have 
to embrace a disruptive business

721
00:38:35,520 --> 00:38:38,360
model in the next three to five 
years, which is historically 

722
00:38:38,360 --> 00:38:41,360
been very, very difficult for 
any type of company because 

723
00:38:41,360 --> 00:38:43,320
you're kind of saying all this 
money we're making a day. 

724
00:38:43,800 --> 00:38:45,840
That's going to be 0 in five 
years and we're just going to 

725
00:38:45,840 --> 00:38:48,600
have to jump on the new thing 
and ride it out, which is really

726
00:38:48,600 --> 00:38:51,760
tough. 
I to be clear, I don't think the

727
00:38:51,760 --> 00:38:55,440
self driving deployment cycle 
will be anywhere near as fast as

728
00:38:55,440 --> 00:38:57,760
that would suggest. 
So I don't think it's going to 

729
00:38:57,760 --> 00:39:00,360
destroy their like existing car 
business. 

730
00:39:00,600 --> 00:39:03,600
But I agree with the idea it's 
hard. 

731
00:39:03,800 --> 00:39:06,400
The the classic start up 
situation which you're 

732
00:39:06,400 --> 00:39:09,120
articulating is that like people
don't disrupt their own good 

733
00:39:09,480 --> 00:39:12,720
businesses and so Tesla's in 
sort of the worst situation for 

734
00:39:12,720 --> 00:39:16,520
that. 
Just a counterpoint to that, do 

735
00:39:16,520 --> 00:39:20,320
we believe that other companies 
are going to be able to sell 

736
00:39:20,320 --> 00:39:23,840
fully self driving vehicles 
anytime soon? 

737
00:39:23,960 --> 00:39:27,040
It doesn't doesn't seem like it,
at least besides the three we've

738
00:39:27,040 --> 00:39:28,960
talked about. 
No, I I mean those other 

739
00:39:28,960 --> 00:39:32,880
companies seem to be only 
focused on the robo taxi market 

740
00:39:32,880 --> 00:39:37,360
as opposed to getting consumer 
vehicles that can fully self 

741
00:39:37,360 --> 00:39:40,120
drive themselves into the world 
like if. 

742
00:39:40,640 --> 00:39:43,000
Tesla is the only game in town 
there, and I'm not sure if 

743
00:39:43,000 --> 00:39:46,400
that's the case, but right now 
it seems to be the only one that

744
00:39:46,400 --> 00:39:50,880
is going in that direction. 
That could end up being a very, 

745
00:39:50,880 --> 00:39:55,680
very compelling offering if the 
only car you can buy that will 

746
00:39:55,680 --> 00:39:59,520
drive itself anywhere is a test 
big if we'll see. 

747
00:39:59,520 --> 00:40:02,920
I feel like Waymo and crews are 
going to deploy city by city 

748
00:40:02,920 --> 00:40:06,880
geofence where they know they 
can do it with no deaths, you 

749
00:40:06,880 --> 00:40:08,000
know? 
Yeah. 

750
00:40:08,720 --> 00:40:11,040
Anyway. 
Yeah, are there any other, I 

751
00:40:11,120 --> 00:40:13,080
mean this is great. 
I I'm glad you brought up Tesla.

752
00:40:13,080 --> 00:40:17,480
Like what Any more edge case 
companies we haven't considered.

753
00:40:18,040 --> 00:40:19,960
I mean this is part of the 
problem I think right now and 

754
00:40:19,960 --> 00:40:23,720
why I think a data bricks IPO 
and the Mosaic deal will be 

755
00:40:23,720 --> 00:40:26,560
welcomed. 
It's like there's a public 

756
00:40:26,560 --> 00:40:31,240
market desperation for a way to 
bet on AI, which is why I think 

757
00:40:31,360 --> 00:40:36,000
NVIDIA has done so well because,
you know, these big tech 

758
00:40:36,000 --> 00:40:38,160
companies are already very 
highly valued. 

759
00:40:39,200 --> 00:40:40,400
You know, the startups are 
private. 

760
00:40:41,920 --> 00:40:44,280
Cool, great. 
Thanks for coming on. 

761
00:40:44,280 --> 00:40:48,280
And now we're going to Chris 
Miller with Chip War, who's a 

762
00:40:48,280 --> 00:40:53,160
real genuine expert in the wonky
world of chips. 

763
00:40:53,160 --> 00:40:58,400
And so we go deep in NVIDIA sort
of their history and you know, 

764
00:40:58,560 --> 00:41:03,040
the, the big geopolitical 
question around the race with 

765
00:41:03,040 --> 00:41:04,440
China. 
So give it a listen. 

766
00:41:09,520 --> 00:41:12,600
Chris Miller, author of Chipwar.
Thank you for coming on the 

767
00:41:12,600 --> 00:41:14,160
show. 
Thank you for having me. 

768
00:41:14,360 --> 00:41:19,160
I feel like so much of what sort
of the regular person or even 

769
00:41:19,160 --> 00:41:22,120
you know, I don't know, the 
average startup founder, venture

770
00:41:22,120 --> 00:41:26,400
capitalist in Silicon Valley 
thinks of this artificial 

771
00:41:26,400 --> 00:41:30,760
intelligence phenomenon starts 
and sometimes ends with chat GPE

772
00:41:30,760 --> 00:41:33,120
3 and it's certainly it's very 
much the like. 

773
00:41:33,120 --> 00:41:34,920
What bots can we play around 
with? 

774
00:41:34,920 --> 00:41:39,920
Like how am I talking to some 
sort of AI but you know in the 

775
00:41:39,920 --> 00:41:43,080
background doing reporting on 
some of these conversations 

776
00:41:43,600 --> 00:41:48,280
NVIDIA and like H1 hundreds and 
like weird chips come up over 

777
00:41:48,280 --> 00:41:51,400
and over again and sort of 
people's ability to get access 

778
00:41:51,400 --> 00:41:53,520
to those chips. 
So I really wanted to have you 

779
00:41:53,520 --> 00:41:56,840
on just to sort of like 
interrogate that and understand 

780
00:41:57,240 --> 00:41:59,720
sort of the technology behind 
it. 

781
00:41:59,960 --> 00:42:04,360
Can we just start off like, when
when does NVIDIA start to matter

782
00:42:04,360 --> 00:42:06,160
as a company? 
Or just give us like, the brief 

783
00:42:06,160 --> 00:42:09,640
history of this company, because
it was like a weird, weird sort 

784
00:42:09,640 --> 00:42:12,720
of gaming chip making company, 
right for a long time. 

785
00:42:13,360 --> 00:42:15,760
Yeah, that that's right. 
NVIDIA was founded over 3 

786
00:42:15,760 --> 00:42:20,080
decades ago, but its earliest 
couple of years was all in 

787
00:42:20,080 --> 00:42:24,760
graphics, and graphics was most 
important for games, computer 

788
00:42:24,760 --> 00:42:28,040
games, video games. 
Which, you know, think back two 

789
00:42:28,040 --> 00:42:30,880
decades, they had the most 
complex graphic demands, figures

790
00:42:30,880 --> 00:42:34,400
moving across the scene, 
pictures changing very rapidly. 

791
00:42:34,400 --> 00:42:37,200
And there were special types of 
chips that were produced by 

792
00:42:37,200 --> 00:42:39,760
companies like NVIDIA for 
processing graphics. 

793
00:42:39,760 --> 00:42:43,320
And if you think back, not so 
long ago, people would buy 

794
00:42:43,320 --> 00:42:46,080
specific computers to have the 
best graphics capability. 

795
00:42:46,080 --> 00:42:49,040
Now we sort of take it for 
granted, but for a long time 

796
00:42:49,040 --> 00:42:50,840
that was a real differentiating 
factor. 

797
00:42:50,960 --> 00:42:55,720
Can you explain sort of the 
difference between AGPUA 

798
00:42:55,720 --> 00:42:58,600
Graphics Processing Unit and 
ACPU? 

799
00:42:59,160 --> 00:43:01,920
So it's ACPU, which is the the 
workhorse of traditional 

800
00:43:01,920 --> 00:43:04,480
computing and computers and data
centers. 

801
00:43:04,800 --> 00:43:07,640
They're very good at doing many 
different types of things, but 

802
00:43:07,640 --> 00:43:10,640
they do every computation 
serially, 1 after the other. 

803
00:43:11,280 --> 00:43:14,040
And for most use cases that's 
what you want because you're 

804
00:43:14,040 --> 00:43:16,600
undertaking different types of 
calculations. 

805
00:43:16,600 --> 00:43:21,320
But for for training AI systems,
you want to undertake the same 

806
00:43:21,320 --> 00:43:24,680
type of calculation repeatedly. 
And so parallel processing, 

807
00:43:24,680 --> 00:43:28,440
which is what a GPU does, is 
capable of doing multiple things

808
00:43:28,440 --> 00:43:31,920
in parallel at once, which is 
why they're vastly faster for 

809
00:43:32,120 --> 00:43:34,680
this type of calculation than a 
CPU is. 

810
00:43:35,480 --> 00:43:37,720
But why? 
Why does AI? 

811
00:43:37,960 --> 00:43:42,280
Why do AI tests need to be 
parallel while like a normal 

812
00:43:42,280 --> 00:43:45,880
computer task is sequenced? 
Well, they don't have to be. 

813
00:43:45,880 --> 00:43:48,400
It's just really inefficient, 
right? 

814
00:43:49,200 --> 00:43:51,760
You know, I think the key trend,
there's been great research 

815
00:43:51,760 --> 00:43:55,600
looking at the amount of data on
which cutting edge AI systems 

816
00:43:55,600 --> 00:43:57,960
are trained and and what you 
find, if you look at this for 

817
00:43:57,960 --> 00:44:00,920
the last like 10 years or so and
chart it, the amount of data 

818
00:44:00,920 --> 00:44:04,680
used in cutting edge systems is 
doubling every six to nine 

819
00:44:04,680 --> 00:44:06,560
months. 
There's extraordinary demands 

820
00:44:06,560 --> 00:44:09,720
for putting as much data as 
possible into training. 

821
00:44:10,480 --> 00:44:13,560
And in order to make sense of 
all this data, to train your 

822
00:44:13,560 --> 00:44:16,960
model to learn from the data, it
has to undertake lots of 

823
00:44:16,960 --> 00:44:19,720
calculations. 
And so the the computing demands

824
00:44:19,720 --> 00:44:23,320
of training have shot upwards 
not just exponentially, but 

825
00:44:23,320 --> 00:44:27,360
exponentially, at at a rate 
that's faster than Moore's Law. 

826
00:44:27,480 --> 00:44:30,320
And that's why having the right 
ship for training has become 

827
00:44:30,400 --> 00:44:33,680
very important. 
Did NVIDIA see this coming? 

828
00:44:33,680 --> 00:44:37,120
Like, you know, I their stock is
up year to date. 

829
00:44:37,120 --> 00:44:41,000
I think it's some like 200%. 
I I feel like Jensen Huang, the 

830
00:44:41,000 --> 00:44:42,760
CEO, you hear about them all the
time. 

831
00:44:42,760 --> 00:44:46,440
Like was this a happy accident? 
It's like we're we're making 

832
00:44:46,440 --> 00:44:48,720
great video games and all of a 
sudden people want to buy our 

833
00:44:48,720 --> 00:44:52,480
chips for the hottest thing in 
the world or did they have some 

834
00:44:52,480 --> 00:44:55,160
foresight here? 
How much was this a plan on the 

835
00:44:55,160 --> 00:44:57,880
part of NVIDIA? 
You know, it certainly was a a 

836
00:44:57,880 --> 00:45:00,080
bit of an accident and if you 
talk to the company they'll 

837
00:45:00,080 --> 00:45:03,320
they'll they'll literally admit 
that when they founded NVIDIA. 

838
00:45:03,800 --> 00:45:05,880
They had no idea that AI would 
be an application. 

839
00:45:06,200 --> 00:45:10,080
But over a decade ago, they 
began to realize that there were

840
00:45:10,080 --> 00:45:12,960
a a bunch of PhD students and 
researchers at places like 

841
00:45:12,960 --> 00:45:15,880
Stanford and Berkeley who were 
using their chips for things 

842
00:45:15,880 --> 00:45:20,320
other than gaming for complex 
calculations. 

843
00:45:20,320 --> 00:45:24,400
And that NVIDIA realized that if
they tried to build out chips 

844
00:45:24,400 --> 00:45:28,320
around this application and no 
less importantly, build out a 

845
00:45:28,320 --> 00:45:31,880
software ecosystem around it, 
that they could. 

846
00:45:32,200 --> 00:45:33,960
Provide the chips that AI would 
require. 

847
00:45:33,960 --> 00:45:36,720
And so they've been investing 
very heavily for now, over a 

848
00:45:36,720 --> 00:45:38,160
decade. 
And for a long time that 

849
00:45:38,160 --> 00:45:42,440
investment seemed actually quite
foolish because most of their 

850
00:45:42,440 --> 00:45:46,040
money was still in in in 
graphics and in gaming. 

851
00:45:46,080 --> 00:45:49,280
It's only quite recently that a 
data center in AI has become 

852
00:45:49,280 --> 00:45:52,360
their their primary. 
Well, you just, people talk 

853
00:45:52,360 --> 00:45:56,120
about artificial intelligence as
a way to boost their stock price

854
00:45:56,120 --> 00:45:58,840
or to, you know, seem like a 
more futuristic company than you

855
00:45:58,880 --> 00:46:01,880
are for NVIDIA. 
It's like, OK, you're a gaming 

856
00:46:01,880 --> 00:46:04,600
company guy. 
Like, chill out and now you know

857
00:46:04,600 --> 00:46:07,720
it's been totally validated. 
That sort of all this talk of 

858
00:46:07,720 --> 00:46:13,080
artificial intelligence wasn't 
total wild hype, hypemanship or 

859
00:46:13,080 --> 00:46:14,600
whatever. 
Well, and in between though, 

860
00:46:14,600 --> 00:46:16,120
they were also a a Bitcoin 
mining. 

861
00:46:16,120 --> 00:46:19,200
Yeah, I know, yeah, exactly. 
They did have the crypto. 

862
00:46:19,240 --> 00:46:22,600
They're definitely good at 
seizing the moment, but they 

863
00:46:22,600 --> 00:46:24,680
were they were being used for 
crypto mining, right? 

864
00:46:24,680 --> 00:46:26,200
Is that what it was? 
That's right. 

865
00:46:27,360 --> 00:46:30,880
And like, yeah, how do you see 
Jensen the CEO, like he he's one

866
00:46:30,880 --> 00:46:34,160
of the Co founders writer. 
What's sort of his role at the 

867
00:46:34,160 --> 00:46:37,880
founding and how do you see him 
as sort of ACEO today? 

868
00:46:39,160 --> 00:46:41,480
He was one of the three Co 
founders of a video they they 

869
00:46:41,480 --> 00:46:46,520
met in a Denny's in San Jose to 
devise the business plan that in

870
00:46:46,520 --> 00:46:51,520
the the 1990s and and Jensen I I
think deserves a lot of credit 

871
00:46:51,520 --> 00:46:56,160
for realizing. 
A decade ago, that AI could be a

872
00:46:56,160 --> 00:46:58,560
real growth driver. 
At the time, people thought it 

873
00:46:58,560 --> 00:47:01,120
was a really irresponsible thing
to do plow all this money into 

874
00:47:01,120 --> 00:47:03,560
building a software ecosystem 
around chips. 

875
00:47:03,560 --> 00:47:06,840
No one else did that. 
Most companies just made chips. 

876
00:47:07,560 --> 00:47:13,000
And if you go back to the debate
in 20-15 around that time 

877
00:47:13,000 --> 00:47:15,280
period, you can find lots of 
equity analysts saying, you 

878
00:47:15,280 --> 00:47:17,400
know, what's this company doing 
spending all this money on 

879
00:47:17,400 --> 00:47:18,920
software rather than selling 
chips? 

880
00:47:18,920 --> 00:47:21,800
It's a waste of money And and 
worse, they were giving the 

881
00:47:21,800 --> 00:47:24,800
software away for free. 
But it turned out to be a 

882
00:47:24,800 --> 00:47:27,680
brilliant move because everyone 
started using the the CUDA 

883
00:47:27,720 --> 00:47:32,840
ecosystem, which is is the the 
software layer on top of NVIDIA 

884
00:47:32,840 --> 00:47:35,160
chips. 
And that's put it at the center 

885
00:47:35,160 --> 00:47:38,720
of the the AI world. 
A key piece of the chip world is

886
00:47:38,720 --> 00:47:42,240
this sort of like are you 
designing chips, are you the 

887
00:47:42,240 --> 00:47:44,000
fab? 
Can you just sort of explain 

888
00:47:44,000 --> 00:47:46,960
that and where NVIDIA sort of 
fits in? 

889
00:47:46,960 --> 00:47:49,280
Like, are they? 
Are they actually we say build 

890
00:47:49,280 --> 00:47:50,800
the chips. 
Are they like building the 

891
00:47:50,800 --> 00:47:53,120
chips? 
Yeah that that's a that's a key 

892
00:47:53,120 --> 00:47:54,440
distinction. 
They're not actually building 

893
00:47:54,440 --> 00:47:57,600
the chips. 
They only do chip design from 

894
00:47:57,600 --> 00:47:58,720
day one. 
They were, they were what's 

895
00:47:58,720 --> 00:48:01,120
called a fabulous chip company. 
They didn't have a fab. 

896
00:48:01,400 --> 00:48:06,160
They only did the design and and
chip design is in many ways like

897
00:48:06,160 --> 00:48:08,520
a a software design type of 
business. 

898
00:48:08,520 --> 00:48:11,720
You you basically write the code
of the design and then you 

899
00:48:11,720 --> 00:48:15,240
e-mail it off to the the factory
or a fab and semiconductor 

900
00:48:15,240 --> 00:48:18,440
parlance and and the fab 
actually does the manufacturing 

901
00:48:18,960 --> 00:48:21,880
and so in the case of NVIDIA. 
Their most important 

902
00:48:23,720 --> 00:48:26,920
manufacturing partner has been 
TSMC, the Taiwan Semiconductor 

903
00:48:26,920 --> 00:48:29,520
Manufacturing Company, which 
manufacturers most of their 

904
00:48:29,520 --> 00:48:31,520
chips. 
And so today for all of Nvidia's

905
00:48:31,520 --> 00:48:35,600
key AI training chips, they're 
all manufactured by one company 

906
00:48:36,040 --> 00:48:39,200
in Taiwan. 
And do they they make all 

907
00:48:39,200 --> 00:48:43,680
basically all NVIDIA chips or? 
They make, they currently make 

908
00:48:43,680 --> 00:48:46,800
all of the key AI chips that 
NVIDIA produces and they. 

909
00:48:47,560 --> 00:48:51,760
Been the most important producer
of video chips from day one, and

910
00:48:51,760 --> 00:48:55,280
is NVIDIA like, how far ahead is
NVIDIA right now in this sort of

911
00:48:55,480 --> 00:48:58,080
AI chip? 
Are they singular? 

912
00:48:58,080 --> 00:49:00,400
Like, yeah, where where are they
today? 

913
00:49:01,200 --> 00:49:04,520
Well, in terms of market share, 
Nvidia's far, far ahead. 

914
00:49:04,520 --> 00:49:09,840
So it's estimated that up to 90%
of cutting edge AI systems are 

915
00:49:09,840 --> 00:49:12,280
trained on NVIDIA chips. 
And you can debate what's the 

916
00:49:12,280 --> 00:49:15,440
definition of cutting edge, but 
Nvidia's market position is. 

917
00:49:15,920 --> 00:49:19,560
That is extraordinary right now 
and that's reflected in in in 

918
00:49:19,560 --> 00:49:21,880
their stock price. 
I I think if you asked are they 

919
00:49:21,880 --> 00:49:25,800
technologically ahead of of 
competitors, whether AMD, which 

920
00:49:25,800 --> 00:49:29,200
is another chip maker or Google,
which has a chip called the TPU,

921
00:49:29,200 --> 00:49:31,960
which is also used for AI 
training, that that's a harder 

922
00:49:31,960 --> 00:49:34,040
question to answer. 
Exactly. 

923
00:49:34,040 --> 00:49:37,000
Yeah, exactly, exactly. 
And that's a hard question to 

924
00:49:37,000 --> 00:49:40,640
answer because there's both, you
know, technical specifications 

925
00:49:40,640 --> 00:49:43,080
you could look at, but there's 
also ease of use and so if 

926
00:49:43,080 --> 00:49:45,920
you've built a system. 
Trains it on NVIDIA if you 

927
00:49:45,920 --> 00:49:49,080
yourself as a engineer have 
gotten comfortable with 

928
00:49:49,080 --> 00:49:50,840
invidious products and with 
CUDA. 

929
00:49:51,000 --> 00:49:54,240
There's also a a a switching 
cost of moving to somebody else,

930
00:49:54,680 --> 00:49:56,920
and so Nvidia's both got their 
own capabilities. 

931
00:49:56,920 --> 00:49:59,000
They also got the first mover 
advantage. 

932
00:49:59,000 --> 00:50:02,280
Everyone's used to their 
products, but but they're really

933
00:50:02,280 --> 00:50:05,040
scarce right now, right? 
I mean, my sense talking to 

934
00:50:05,040 --> 00:50:09,000
startups and investors, I mean, 
you hear Nvidia's like in every 

935
00:50:09,000 --> 00:50:12,720
venture round now because as 
part of the funding, they can 

936
00:50:12,760 --> 00:50:15,600
sort of promise people their 
chips, which then they make 

937
00:50:15,600 --> 00:50:17,800
money back from the startups. 
It's like a great world. 

938
00:50:17,800 --> 00:50:20,880
They're playing kingmaker. 
Like what? 

939
00:50:20,880 --> 00:50:24,960
What's your sense of how scarce 
their artificial intelligence 

940
00:50:24,960 --> 00:50:28,240
chips are? 
Right now they're scarce. 

941
00:50:28,240 --> 00:50:31,280
I hear the same anecdotes that 
you're hearing and NVIDIA and 

942
00:50:31,280 --> 00:50:35,560
TSMC, the manufacturer partner 
have have both said it's going 

943
00:50:35,560 --> 00:50:38,760
to be a year or so until supply 
normalizes. 

944
00:50:39,280 --> 00:50:42,480
And the problem is actually not 
that TSMC can't produce enough 

945
00:50:42,480 --> 00:50:46,040
chips, but they can't package 
them because they use a a very 

946
00:50:46,040 --> 00:50:49,840
complex, a new packaging 
technology to put the chips 

947
00:50:49,840 --> 00:50:52,760
right next to the specialized. 
I assume you don't mean in a 

948
00:50:52,760 --> 00:50:54,920
box. 
Yeah, you're like, you place the

949
00:50:54,920 --> 00:50:56,520
chip on something, right? 
Yeah. 

950
00:50:56,720 --> 00:50:59,360
That's right, yeah. 
It's a Traditionally chips were 

951
00:50:59,360 --> 00:51:00,840
packaged in. 
That'd be funny though, if it's 

952
00:51:00,880 --> 00:51:02,720
like. 
Oh, the Apple cardboard box is 

953
00:51:02,720 --> 00:51:06,080
like holding them up at the end 
anyway, so go ahead. 

954
00:51:06,480 --> 00:51:08,400
Yeah. 
So, so packaging traditionally 

955
00:51:08,400 --> 00:51:10,040
was a really boring part of the 
chip business. 

956
00:51:10,040 --> 00:51:13,240
You take a PIC a chip, you put 
it in a a plastic or a ceramic 

957
00:51:13,240 --> 00:51:14,800
package and that was that was 
done. 

958
00:51:14,800 --> 00:51:18,320
But now it's it's increasingly 
important because you need to 

959
00:51:18,320 --> 00:51:21,880
get your your processor and your
memory as close as possible with

960
00:51:21,880 --> 00:51:23,680
as fast and interconnect as 
possible. 

961
00:51:24,800 --> 00:51:29,920
And so companies like NVIDIA are
leaning more heavily on having 

962
00:51:29,920 --> 00:51:33,120
the right packaging capabilities
to provide the most advanced. 

963
00:51:34,440 --> 00:51:36,600
Outcomes. 
And so it's actually TSM, CS 

964
00:51:36,600 --> 00:51:39,520
packaging capabilities that are 
the limiting factor right now 

965
00:51:39,520 --> 00:51:45,640
for producing more NVIDIA GPUs. 
And and what is an H100 or like 

966
00:51:45,640 --> 00:51:48,440
what what, what are, how 
different are these chips from 

967
00:51:48,440 --> 00:51:51,760
the the GPUs that everybody got 
excited about in the 1st place? 

968
00:51:51,760 --> 00:51:55,520
Like are they server chips Are 
explaining, you know, what is an

969
00:51:55,520 --> 00:51:58,080
H100? 
But it's just the newest version

970
00:51:58,080 --> 00:52:02,920
of of AGPU that NVIDIA offers 
specifically for AI training 

971
00:52:02,920 --> 00:52:04,640
and. 
The reason people get excited 

972
00:52:04,640 --> 00:52:07,800
about new versions of chips is 
because the rate of change is so

973
00:52:07,800 --> 00:52:10,600
fast that a new version is 
always not just incrementally 

974
00:52:10,600 --> 00:52:13,520
better, but a lot better than 
the old version. 

975
00:52:14,200 --> 00:52:17,400
You know, it's it's it's not 
like the iPhone 15, which is a 

976
00:52:17,400 --> 00:52:20,160
bit better than the iPhone 14. 
But we've stopped noticing and 

977
00:52:20,760 --> 00:52:25,520
I'm like, this is not enough 
necessarily to get me from 13 

978
00:52:25,640 --> 00:52:27,800
Pro or whatever. 
Yeah, yeah. 

979
00:52:27,800 --> 00:52:31,440
The chips are different because 
of the the Moore's Law dynamic, 

980
00:52:31,440 --> 00:52:34,160
which is, you know, says you 
roughly double your processing 

981
00:52:34,160 --> 00:52:37,400
power every two years. 
And that's not a perfect, a 

982
00:52:37,400 --> 00:52:40,760
perfect metric, but ballpark 
that tells you just how rapidly 

983
00:52:40,920 --> 00:52:43,560
the technology improves. 
And the H100 is the most 

984
00:52:43,560 --> 00:52:45,640
advanced chip that NVIDIA 
offers. 

985
00:52:46,080 --> 00:52:49,600
So yeah, to put Moore's Law is 
still alive today. 

986
00:52:50,440 --> 00:52:52,200
Moores law is actually not a 
law. 

987
00:52:52,200 --> 00:52:55,360
It's a prediction that was set 
out by Gordon Moore, who founded

988
00:52:55,360 --> 00:52:59,240
Intel in 1965. 
He predicted that the number of 

989
00:52:59,240 --> 00:53:03,480
transistors per chip would 
double every year or two, which 

990
00:53:03,480 --> 00:53:06,600
means that the computing power 
of chips would double every year

991
00:53:06,600 --> 00:53:09,080
or two. 
And that's been and he said it 

992
00:53:09,080 --> 00:53:11,920
would be a decade, but then it's
been true for a long time, 

993
00:53:11,920 --> 00:53:14,600
right? 
That's right, I'm just stealing 

994
00:53:14,600 --> 00:53:17,680
from your book. 
Of course it makes me sound 

995
00:53:17,680 --> 00:53:20,640
smarter when you get to read the
book in advance, so you know all

996
00:53:20,640 --> 00:53:21,680
the answers. 
Anyway, go ahead. 

997
00:53:22,320 --> 00:53:24,520
But yeah, that's right. 
So Gordon Moore thought, would 

998
00:53:24,520 --> 00:53:27,320
last for a decade, and here we 
are half a century later. 

999
00:53:27,680 --> 00:53:31,160
And Moore's law is is still 
basically holding up the the 

1000
00:53:31,160 --> 00:53:34,400
rate of change increases and 
decreases at certain times. 

1001
00:53:34,400 --> 00:53:38,280
But it's basically true that if 
you wait 2 years, a new chip 

1002
00:53:38,280 --> 00:53:40,840
will have twice as many 
transistors and therefore twice 

1003
00:53:40,840 --> 00:53:44,960
as much computing power. 
I mean, this is sort of a hard 

1004
00:53:44,960 --> 00:53:47,840
question, so I don't know if you
there's an easy answer, but 

1005
00:53:48,240 --> 00:53:52,040
yeah, how much credit would you 
give then the improvement of 

1006
00:53:52,520 --> 00:53:56,640
GPUs for what's going on in 
artificial intelligence right 

1007
00:53:56,640 --> 00:54:00,840
now relative to, you know, 
attention is all you need or you

1008
00:54:00,920 --> 00:54:04,640
know, the papers in terms of the
approaches to actually use these

1009
00:54:04,640 --> 00:54:08,480
chips to to run, run the 
software? 

1010
00:54:09,480 --> 00:54:11,680
Well, you know, I think it's 
certainly it's the case that 

1011
00:54:11,760 --> 00:54:14,840
that there's there's multiple 
factors driving innovation. 

1012
00:54:14,840 --> 00:54:17,200
But I just the way I like to 
think about it is, you know 

1013
00:54:17,240 --> 00:54:19,240
what's improved most over the 
last decade? 

1014
00:54:19,240 --> 00:54:23,920
Is it the case that that 
software engineers are 16 times 

1015
00:54:23,920 --> 00:54:28,400
smarter or is it the case that? 
Algorithms are 16 times better 

1016
00:54:28,720 --> 00:54:32,680
or is it the case that because 
of Moore's Law 2 to the fifth, 

1017
00:54:32,680 --> 00:54:36,000
we've got, you know, a vast 
increase in the number of of 

1018
00:54:36,000 --> 00:54:38,520
transistors? 
You know, I think it's it's 

1019
00:54:38,520 --> 00:54:41,200
primarily computing power that's
that's driving us. 

1020
00:54:41,800 --> 00:54:44,080
That's what you want. 
The chips guy comes on, he says 

1021
00:54:44,080 --> 00:54:45,680
it's the chips. 
Yeah. 

1022
00:54:46,720 --> 00:54:49,960
A key theme, you know, obviously
in the book and sort of for 

1023
00:54:49,960 --> 00:54:52,640
anyone thinking about the 
situation is, you know, the 

1024
00:54:52,640 --> 00:54:59,040
rivalry with China and and sort 
of the the delicacy of the 

1025
00:54:59,040 --> 00:55:03,560
global supply chains and you 
know, the, there's the chip 

1026
00:55:03,560 --> 00:55:05,720
rivalry with China. 
But also in artificial 

1027
00:55:05,720 --> 00:55:08,560
intelligence, you know, if we're
worried that this is the 

1028
00:55:08,560 --> 00:55:12,760
potential sort of path to some 
sort of, you know, generalized 

1029
00:55:12,760 --> 00:55:16,920
intelligence, there's also the 
real question of whether the US 

1030
00:55:16,920 --> 00:55:20,120
gets there first. 
Yeah. 

1031
00:55:20,120 --> 00:55:22,320
So I mean, there's so much to 
hit at. 

1032
00:55:22,320 --> 00:55:25,600
I mean, I guess the first 
question for me is, is the 

1033
00:55:25,600 --> 00:55:29,040
United States getting most of 
Nvidia's chips or like how much?

1034
00:55:29,600 --> 00:55:31,320
How many of those chips are 
going to China? 

1035
00:55:31,320 --> 00:55:35,120
Is China getting access to those
right now while they're scarce? 

1036
00:55:36,400 --> 00:55:41,000
Well, the the US last year 
imposed new rules that said 

1037
00:55:41,000 --> 00:55:44,720
NVIDIA can't transfer its most 
advanced GPU to China. 

1038
00:55:44,720 --> 00:55:47,720
And not just NVIDIA. 
Any company producing chips 

1039
00:55:47,720 --> 00:55:51,160
above a certain threshold can't 
transfer them to China, so H1 

1040
00:55:51,160 --> 00:55:55,120
hundreds are illegal to. 
To transfer to China, China's 

1041
00:55:55,120 --> 00:55:58,720
not getting any of those. 
But but NVIDIA has design chips 

1042
00:55:58,720 --> 00:56:01,320
that go basically right up to 
the threshold called A8 

1043
00:56:01,320 --> 00:56:03,960
hundreds. 
And Chinese firms are reportedly

1044
00:56:03,960 --> 00:56:06,080
buying very large volumes of 
those. 

1045
00:56:06,440 --> 00:56:09,120
So China's getting a lot of 
GPUs, but they're less advanced 

1046
00:56:09,120 --> 00:56:12,320
than what AUS firm can buy. 
What would their next best 

1047
00:56:12,440 --> 00:56:15,880
option be? 
Well, the next best option is 

1048
00:56:15,880 --> 00:56:19,640
Nvidia's second best chip. 
Oh, man, OK, right. 

1049
00:56:20,680 --> 00:56:24,920
And China tried to build its own
fab, right? 

1050
00:56:24,920 --> 00:56:28,400
A competitor of TSMC. 
What's, what's the status of 

1051
00:56:28,400 --> 00:56:31,080
that and how much? 
You know it's fun, especially in

1052
00:56:31,120 --> 00:56:33,560
the US and given the market 
returns, like to talk about 

1053
00:56:33,560 --> 00:56:37,160
NVIDIA, but like how much does 
this all hang on TSMC? 

1054
00:56:38,200 --> 00:56:40,520
So China has been investing very
heavily in its own chip 

1055
00:56:40,520 --> 00:56:42,320
industry. 
It's got both companies doing 

1056
00:56:42,320 --> 00:56:45,240
the design side and the 
manufacturing side. 

1057
00:56:46,000 --> 00:56:50,200
There there are a number of 
Chinese GPU designers that seem 

1058
00:56:50,200 --> 00:56:54,480
to have pretty competitive 
products, although even in China

1059
00:56:54,480 --> 00:56:57,320
Chinese firms prefer to buy 
invidious chips, so they're 

1060
00:56:57,320 --> 00:57:01,120
they're not as good it seems. 
But the big challenge in China 

1061
00:57:01,120 --> 00:57:04,600
is the manufacturing side 
because TSMC, the Taiwanese 

1062
00:57:04,600 --> 00:57:09,560
firm, has been around five years
ahead of the leading Chinese 

1063
00:57:09,560 --> 00:57:12,000
firm SMEC. 
For a very long time, for at 

1064
00:57:12,000 --> 00:57:13,840
least a decade, there's been a 
five year gap. 

1065
00:57:13,840 --> 00:57:16,840
So both companies regular 
regularly improve the 

1066
00:57:16,840 --> 00:57:20,480
manufacturing processes, but the
rate of change in both is 

1067
00:57:20,480 --> 00:57:23,000
roughly the same. 
So TSMC is always half a decade 

1068
00:57:23,000 --> 00:57:25,280
ahead, and that holds to up the 
Today. 

1069
00:57:25,280 --> 00:57:26,520
And what about the United 
States? 

1070
00:57:26,520 --> 00:57:29,280
Like it hadn't Intel been trying
to become a fab? 

1071
00:57:29,280 --> 00:57:32,600
Like what's what's the state of 
our ability to actually 

1072
00:57:32,720 --> 00:57:35,800
manufacture these chips if you 
know, there's a war in Taiwan, 

1073
00:57:35,800 --> 00:57:38,400
God forbid? 
Well, yeah, that that's that's 

1074
00:57:38,400 --> 00:57:40,520
the dilemma. 
So Intel does make its own 

1075
00:57:40,520 --> 00:57:45,360
chips, but around five years ago
it faced some severe problems 

1076
00:57:45,360 --> 00:57:47,360
with its manufacturing 
operations. 

1077
00:57:47,640 --> 00:57:51,360
And so for its most cutting edge
chips today, Intel now turns to 

1078
00:57:51,360 --> 00:57:55,760
TSMC for the manufacturing. 
Now Intel right now is is trying

1079
00:57:55,760 --> 00:58:00,400
to reformulate its manufacturing
processes and it hopes that by 

1080
00:58:00,560 --> 00:58:03,800
they're saying by 2025 they're 
going to be producing. 

1081
00:58:04,160 --> 00:58:06,560
Have their most advanced chips 
back in house again, and that 

1082
00:58:06,560 --> 00:58:09,360
those chips will be as capable 
as a TSMC made chip. 

1083
00:58:09,360 --> 00:58:12,120
But right now, Intel's most 
advanced chips are actually 

1084
00:58:12,120 --> 00:58:16,640
manufactured by TSMC. 
And then there's like a key, the

1085
00:58:16,640 --> 00:58:22,120
people who make the machines 
that help T that allow TSMC to 

1086
00:58:22,120 --> 00:58:24,960
do what they do, There's like 
one company of those or what's 

1087
00:58:24,960 --> 00:58:27,400
that company? 
Yeah, that that's right. 

1088
00:58:27,400 --> 00:58:30,880
So TSMC, they know how to use 
the machines to make chips, but 

1089
00:58:30,880 --> 00:58:33,800
the machines themselves are 
produced by a handful of other 

1090
00:58:33,800 --> 00:58:36,840
companies, a couple in 
California, a couple in Japan, 

1091
00:58:36,840 --> 00:58:39,960
and then one in the Netherlands 
called ASML, which produces the 

1092
00:58:39,960 --> 00:58:42,840
most complex of these chip 
making tools. 

1093
00:58:43,840 --> 00:58:48,200
So is quantity. 
Everything here with the H1? 

1094
00:58:48,200 --> 00:58:51,960
Hundreds like, I mean, you know,
I'm at, you know, Microsoft has 

1095
00:58:51,960 --> 00:58:55,160
this cloud computing offering. 
You know, Google has its, Amazon

1096
00:58:55,160 --> 00:58:57,160
obviously has Amazon Web 
Services. 

1097
00:58:57,480 --> 00:59:01,080
How much are those services 
trying to stockpile AI chips? 

1098
00:59:01,320 --> 00:59:05,040
And you know, then these 
startups, whether it's, you 

1099
00:59:05,040 --> 00:59:08,360
know, anthropic or open AI, how 
much is their game? 

1100
00:59:08,360 --> 00:59:11,280
Just like we're going to win if 
we assemble the most H1 

1101
00:59:11,280 --> 00:59:15,560
hundreds, Is quantity sort of 
the game right now in terms of 

1102
00:59:15,640 --> 00:59:19,400
getting access to these chips? 
So long as there are shortages, 

1103
00:59:19,400 --> 00:59:21,840
I think Quantity's a key part of
the game. 

1104
00:59:22,280 --> 00:59:25,960
And and what you see is that 
NVIDIA now realizes it's got a 

1105
00:59:25,960 --> 00:59:29,160
lot of influence over the future
of the cloud computing market. 

1106
00:59:29,520 --> 00:59:33,280
And so NVIDIA has been, I I 
think, pretty actively trying to

1107
00:59:33,320 --> 00:59:39,440
build up other cloud computing 
firms by giving them exactly 

1108
00:59:39,600 --> 00:59:43,160
exactly giving them access to 
GPUs and NVIDIA. 

1109
00:59:43,640 --> 00:59:46,720
I think quite rationally wants a
more fragmented cloud computing 

1110
00:59:46,720 --> 00:59:49,320
market because that's a market 
in which it has more market 

1111
00:59:49,320 --> 00:59:53,120
power relative to Microsoft, to 
Amazon and to Google. 

1112
00:59:54,840 --> 00:59:57,560
So so we alluded to this, but 
sort of can you walk me through 

1113
00:59:57,560 --> 01:00:03,480
the tech giants in terms of 
credible competitors to NVIDIA? 

1114
01:00:03,480 --> 01:00:05,840
I mean we talked, I guess start 
with Google, it sounds like 

1115
01:00:05,840 --> 01:00:07,840
maybe they're in the lead. 
I want to know about Intel. 

1116
01:00:07,840 --> 01:00:10,640
And then is there anybody else 
who's who's relevant or even 

1117
01:00:10,640 --> 01:00:12,320
close? 
Yeah. 

1118
01:00:12,320 --> 01:00:16,800
So Google has a a in house 
designed chip. 

1119
01:00:16,800 --> 01:00:18,680
They've got a big design team 
that has designed a chip called 

1120
01:00:18,680 --> 01:00:22,720
ATPU which stands for tensor 
processing unit but fairly 

1121
01:00:22,720 --> 01:00:28,080
similar to AGPU that is used for
for AI applications and in 

1122
01:00:28,320 --> 01:00:31,800
Google has a vast cloud 
computing business as well as 

1123
01:00:31,800 --> 01:00:34,760
its own, its its own vast data 
center. 

1124
01:00:34,760 --> 01:00:38,560
So lots of experience running 
very complex computing 

1125
01:00:39,160 --> 01:00:44,760
operations but empirically we 
know that companies prefer to 

1126
01:00:44,760 --> 01:00:46,720
use NVIDIA chips at least they 
do right now. 

1127
01:00:46,720 --> 01:00:50,360
You know that could change as 
Google's chips get better as 

1128
01:00:50,880 --> 01:00:53,040
scarcity drives up the price of 
GPUs. 

1129
01:00:53,040 --> 01:00:55,240
But that hasn't been the the 
trend, the preference has been 

1130
01:00:55,240 --> 01:00:57,960
for NVIDIA. 
Then AMD is the other company 

1131
01:00:57,960 --> 01:01:02,480
that that is a competitor that 
in producing GPUs it's a a chip 

1132
01:01:02,480 --> 01:01:05,560
design firm. 
They also manufacture with TSMC 

1133
01:01:06,160 --> 01:01:08,880
that in. 
They're an American firm. 

1134
01:01:08,880 --> 01:01:12,360
That's right. 
Based in Texas and and they 

1135
01:01:12,360 --> 01:01:14,440
manufacture fairly competitive 
GPUs. 

1136
01:01:14,880 --> 01:01:18,120
But the key difference with 
between their chips and Nvidia's

1137
01:01:18,120 --> 01:01:21,320
is that NVIDIA has the the 
ecosystem around it, the CUDA 

1138
01:01:21,360 --> 01:01:24,560
ecosystem. 
And so once you're bought into 

1139
01:01:24,560 --> 01:01:28,240
that ecosystem, right now the 
switching costs are substantial 

1140
01:01:28,240 --> 01:01:30,760
enough where you just prefer to 
stick with NVIDIA if you can get

1141
01:01:30,760 --> 01:01:34,080
access to enough chips. 
Is Intel relevant at all? 

1142
01:01:34,920 --> 01:01:39,720
Intel is is trying hard to 
produce competitive chips and I 

1143
01:01:39,720 --> 01:01:41,800
think over the next couple of 
years we'll see it roll out a 

1144
01:01:41,800 --> 01:01:44,040
series of new products that are 
designed to compete. 

1145
01:01:44,040 --> 01:01:45,360
But right now it's a small 
player. 

1146
01:01:46,360 --> 01:01:49,080
In my understanding like in 
cloud computing, like part of 

1147
01:01:49,080 --> 01:01:53,200
the beauty of like an AWS is 
that you can sort of you know, 

1148
01:01:53,560 --> 01:01:56,680
it's flexible, it's elastic, we 
can sort of have some people do 

1149
01:01:56,680 --> 01:01:59,280
the computing, then you sell it 
to another customer at a 

1150
01:01:59,280 --> 01:02:02,640
different time. 
Why hasn't that been the case 

1151
01:02:02,640 --> 01:02:05,200
with this artificial 
intelligence computing, I mean? 

1152
01:02:05,560 --> 01:02:08,560
I hear people talk about it like
you're you're running full bore,

1153
01:02:08,560 --> 01:02:10,160
like these data centers are 
super hot. 

1154
01:02:10,160 --> 01:02:14,000
Like what's what's so different 
about AI processing? 

1155
01:02:14,800 --> 01:02:18,240
I think the the, the key 
difference is that the demand 

1156
01:02:18,240 --> 01:02:23,320
for compute power in AI training
is so large that there's just a 

1157
01:02:23,320 --> 01:02:25,160
deficit of it. 
There's just a deficit. 

1158
01:02:25,160 --> 01:02:28,480
And so the the systems that 
firms like Open AI are trying to

1159
01:02:28,480 --> 01:02:34,360
train are so vast that they they
can't share their infrastructure

1160
01:02:34,360 --> 01:02:36,440
because they need all of their 
infrastructure and they need 

1161
01:02:36,440 --> 01:02:38,520
even more of it than they 
actually can get access to. 

1162
01:02:38,520 --> 01:02:42,240
And so the cloud computing 
business has been a business, as

1163
01:02:42,240 --> 01:02:46,280
you say, about engineering more 
efficient systems via sharing. 

1164
01:02:46,920 --> 01:02:50,040
But in AI training, no one wants
to share because everyone needs 

1165
01:02:50,040 --> 01:02:51,760
more compute than they can 
actually get access to. 

1166
01:02:53,000 --> 01:02:54,960
Right. 
And then there is also this sort

1167
01:02:54,960 --> 01:02:58,520
of perceived almost like 0 sum, 
like you want to deny your 

1168
01:02:58,520 --> 01:03:02,880
competitor access. 
One company we haven't talked 

1169
01:03:02,880 --> 01:03:07,320
about, you know, that we talked 
about in the laptop world for 

1170
01:03:07,320 --> 01:03:10,600
chips is Apple. 
Like what's what's the status of

1171
01:03:10,600 --> 01:03:12,680
Apple on this? 
Sometimes people talk about 

1172
01:03:12,680 --> 01:03:15,240
Apple and artificial 
intelligence on like local 

1173
01:03:15,240 --> 01:03:17,720
processing. 
Is that a possibility or how do 

1174
01:03:17,720 --> 01:03:21,680
you see Apple as being relevant 
or not in this race? 

1175
01:03:23,120 --> 01:03:26,000
Yeah, you know, thus far we've 
been talking about AI training 

1176
01:03:26,000 --> 01:03:29,120
and and AI training happens 
almost exclusively in big data 

1177
01:03:29,120 --> 01:03:31,280
centers. 
But there's a big question about

1178
01:03:31,520 --> 01:03:34,000
what inference will look like in
the future. 

1179
01:03:34,960 --> 01:03:37,600
Some people think that inference
will mostly happen in data 

1180
01:03:37,600 --> 01:03:41,480
centers, that you're gonna ask a
question of ChatGPT, you'll send

1181
01:03:41,480 --> 01:03:44,000
it back the Internet to the data
center, tips to the data center.

1182
01:03:44,000 --> 01:03:45,480
We'll think about it and give 
you an answer back. 

1183
01:03:46,440 --> 01:03:48,120
And that is efficient in some 
ways. 

1184
01:03:48,120 --> 01:03:50,400
You have all the compute, all 
the chips in one big data 

1185
01:03:50,400 --> 01:03:52,320
center. 
And so you can start to set up 

1186
01:03:52,320 --> 01:03:54,640
some of the sharing and 
efficiencies that that we just 

1187
01:03:54,640 --> 01:03:56,000
discussed. 
But the downside is you have 

1188
01:03:56,000 --> 01:03:58,360
latency issues if you're sending
all the data back and forth and 

1189
01:03:58,360 --> 01:04:00,360
there's costs associated with 
moving data. 

1190
01:04:00,680 --> 01:04:04,200
And so an alternative paradigm 
is doing more of your inference 

1191
01:04:04,240 --> 01:04:07,320
on the edge of networks in your 
phone, in your car, for 

1192
01:04:07,320 --> 01:04:11,160
autonomous driving systems, in 
in in your PC. 

1193
01:04:11,480 --> 01:04:14,320
And that's where companies like 
Apple could start to play a much

1194
01:04:14,560 --> 01:04:17,480
bigger role. 
Right now, I think the the 

1195
01:04:17,480 --> 01:04:21,840
market for inference on the edge
is still very much in flux. 

1196
01:04:21,840 --> 01:04:25,000
We're in the process of seeing 
many new types of chips rolled 

1197
01:04:25,000 --> 01:04:29,040
out precisely for that purpose. 
So, so it's just sort of depends

1198
01:04:29,040 --> 01:04:34,040
on where it's one of these like 
sort of personal computer versus

1199
01:04:34,040 --> 01:04:38,240
server level questions back in 
the 90s where it's like OK, is 

1200
01:04:38,240 --> 01:04:39,920
computing going to continue to 
happen? 

1201
01:04:40,280 --> 01:04:42,280
In the cloud? 
Or is there some model where 

1202
01:04:42,280 --> 01:04:45,720
local computers are going to do 
and you're saying that if it is 

1203
01:04:45,720 --> 01:04:48,640
local, then maybe Apple's in a 
better position because they 

1204
01:04:48,640 --> 01:04:51,880
have these great chips for for 
individual computers. 

1205
01:04:51,880 --> 01:04:54,000
Is is that right? 
Yeah, that that, that's right. 

1206
01:04:54,000 --> 01:04:56,480
And there's I think a pretty 
straightforward cost equation 

1207
01:04:56,480 --> 01:05:01,120
of, you know, if the cost of 
data transfer is high, you'll 

1208
01:05:01,120 --> 01:05:03,120
try to do as much as your 
processing can on the edge. 

1209
01:05:03,120 --> 01:05:06,600
If the cost of the transfer is 
low and the benefits of having 

1210
01:05:07,000 --> 01:05:10,760
big centralized facilities are 
are substantial in terms of the 

1211
01:05:10,760 --> 01:05:12,880
efficiencies you can reap, then 
you'll have a much more 

1212
01:05:12,880 --> 01:05:16,480
centralized inference process. 
And it it might vary even 

1213
01:05:16,480 --> 01:05:20,280
between different use cases. 
In a car, for example, if you've

1214
01:05:20,280 --> 01:05:25,960
got a pretty autonomous car, 
your willingness to to depend on

1215
01:05:25,960 --> 01:05:28,920
a data center to tell you to 
turn left or to turn right, it's

1216
01:05:28,920 --> 01:05:30,680
going to be pretty limited. 
I would, I would, I would 

1217
01:05:30,680 --> 01:05:32,360
hypothesize. 
Whereas for your phone maybe. 

1218
01:05:32,360 --> 01:05:35,240
You want to make sure no matter 
what you can get the answer but 

1219
01:05:35,240 --> 01:05:41,360
you know if it's for mid journey
I can wait till exactly you 

1220
01:05:41,360 --> 01:05:46,640
know, I mean we we talked about 
NVIDIA earlier on in terms of oh

1221
01:05:46,640 --> 01:05:49,600
there was a brief period where 
they were essential to crypto 

1222
01:05:49,600 --> 01:05:54,000
mining and I and one of the key 
storylines I think think for the

1223
01:05:54,000 --> 01:05:56,400
non crypto world that was just 
watching it. 

1224
01:05:56,880 --> 01:06:02,640
Was all the energy consumption 
around those crypto mines is? 

1225
01:06:02,640 --> 01:06:05,880
Is that true for AII haven't 
heard that same environmental 

1226
01:06:05,880 --> 01:06:08,160
story get picked up in this 
case. 

1227
01:06:08,160 --> 01:06:12,160
But are are we destroying the 
environment and our pursuit of 

1228
01:06:12,160 --> 01:06:16,000
artificial intelligence? 
Why, I think we we haven't heard

1229
01:06:16,000 --> 01:06:18,640
much about it just because 
there's such an extraordinary 

1230
01:06:18,640 --> 01:06:23,040
demand for computing that 
companies haven't gotten around 

1231
01:06:23,040 --> 01:06:24,480
to thinking about the 
consequences. 

1232
01:06:24,480 --> 01:06:28,080
But the short answer is, is yes,
it's extraordinarily energy 

1233
01:06:28,080 --> 01:06:32,400
intensive and and we're going to
get more efficient at it as as 

1234
01:06:32,400 --> 01:06:36,120
time passes. 
But right now there are, you 

1235
01:06:36,120 --> 01:06:39,960
know, such demands for energy in
data centers that there are some

1236
01:06:39,960 --> 01:06:42,360
places in the US where it's 
difficult to build a data center

1237
01:06:42,360 --> 01:06:45,480
because there's no spare 
electricity capacity for the 

1238
01:06:45,480 --> 01:06:47,920
data center. 
It's it's already a gating 

1239
01:06:47,920 --> 01:06:49,880
factor. 
Have you seen any? 

1240
01:06:49,880 --> 01:06:52,360
Yes, it's about energy use or 
anybody trying. 

1241
01:06:53,400 --> 01:06:56,200
Well it's it's, it's tricky 
because everything depends on 

1242
01:06:56,200 --> 01:06:59,000
how much more efficient you 
think things will get over time.

1243
01:06:59,000 --> 01:07:01,440
So you've got to assume we're 
going to be a lot more efficient

1244
01:07:01,880 --> 01:07:05,120
making the same compute 
available at lower energy demand

1245
01:07:05,120 --> 01:07:08,720
in 10 years time. 
But at what curve it's it's very

1246
01:07:08,720 --> 01:07:10,680
difficult to say. 
Yeah. 

1247
01:07:10,680 --> 01:07:14,400
And to be clear, I think what 
what is the point of energy if 

1248
01:07:14,400 --> 01:07:17,120
not to sort of push the cutting 
edge of like computing? 

1249
01:07:17,120 --> 01:07:21,600
It seems like a very valuable 
use to me, but it yeah, it'd be 

1250
01:07:21,600 --> 01:07:23,480
interesting. 
I mean, you obviously want to 

1251
01:07:23,480 --> 01:07:26,480
make sure companies aren't being
sort of cavalier about it. 

1252
01:07:26,480 --> 01:07:30,320
And yeah, well, there's 
certainly not cavalier because 

1253
01:07:30,320 --> 01:07:32,640
if you look at the cost of 
running a data center, 'cause I 

1254
01:07:32,640 --> 01:07:34,640
don't know what it is, yeah, 
critical cost. 

1255
01:07:35,640 --> 01:07:38,120
Well, I mean, I don't know if 
you're Open AI, you raised $10 

1256
01:07:38,120 --> 01:07:42,360
billion, it's you can become 
sort of irresponsible. 

1257
01:07:42,360 --> 01:07:45,080
You know there isn't the same 
market pressure if these 

1258
01:07:45,080 --> 01:07:47,760
companies are able to raise what
are basically like research 

1259
01:07:47,760 --> 01:07:49,880
grants to to figure this all 
out. 

1260
01:07:51,080 --> 01:07:53,480
Can you, I mean, we sort of 
jumped into artificial 

1261
01:07:53,480 --> 01:07:57,760
intelligence, but I mean can the
broad theme of the book, the 

1262
01:07:57,760 --> 01:08:02,720
idea that sort of that we often 
think of global infrastructure 

1263
01:08:02,720 --> 01:08:05,400
and the sort of weaknesses and 
vulnerabilities in terms of like

1264
01:08:05,400 --> 01:08:07,160
oil, right. 
I mean it's like oh, did we go 

1265
01:08:07,160 --> 01:08:12,160
to war in Iraq over oil? 
What what is sort of the real 

1266
01:08:12,480 --> 01:08:16,160
sort of what are the weaknesses 
with chips and what what why do 

1267
01:08:16,160 --> 01:08:19,200
you think sort of this chip's 
arms race is? 

1268
01:08:19,880 --> 01:08:22,800
So important when we're thinking
about sort of geopolitics. 

1269
01:08:24,160 --> 01:08:27,319
Well, the reason I, I, I wrote 
chipwork was when I realized 

1270
01:08:27,319 --> 01:08:31,160
first, that we're surrounded by 
thousands and thousands of chips

1271
01:08:31,160 --> 01:08:34,080
over the course of our daily 
lives, Not just our phones and 

1272
01:08:34,080 --> 01:08:37,120
our PCs, but it's cars and 
dishwashers and coffee makers 

1273
01:08:37,120 --> 01:08:39,279
too. 
Second, when I began to 

1274
01:08:39,319 --> 01:08:42,359
understand how critical they are
for AI. 

1275
01:08:42,359 --> 01:08:45,040
And then third, when I realized 
how concentrated the production 

1276
01:08:45,040 --> 01:08:47,800
is in just a tiny number of 
countries and really just a 

1277
01:08:47,800 --> 01:08:52,520
handful of companies, Taiwan 
being the the, the, the, the 

1278
01:08:52,520 --> 01:08:56,640
most surprising example of that.
And so when you look at the the 

1279
01:08:56,840 --> 01:09:01,240
US, China race to develop more 
advanced AI systems and in tech 

1280
01:09:01,240 --> 01:09:04,279
in general, then realize that 
both China and the US are 

1281
01:09:04,279 --> 01:09:07,920
dependent to a shocking degree 
on ships made in Taiwan to power

1282
01:09:07,920 --> 01:09:11,439
their most advanced AI systems. 
It's an extraordinary situation 

1283
01:09:11,439 --> 01:09:14,279
that we've all found ourselves 
in and it's a very dangerous one

1284
01:09:14,600 --> 01:09:17,200
given the concentration and 
given the dependency. 

1285
01:09:18,240 --> 01:09:21,800
What would be the consequences 
of China having a chip edge 

1286
01:09:22,000 --> 01:09:24,240
broadly and an artificial 
intelligence? 

1287
01:09:25,439 --> 01:09:29,080
Well, just like we're preventing
China from accessing our most 

1288
01:09:29,080 --> 01:09:32,720
advanced ships, I think one has 
to assume that any country with 

1289
01:09:32,720 --> 01:09:34,520
that position would do the same 
to us. 

1290
01:09:35,279 --> 01:09:38,800
And right now, Chinese firms 
that are trying to develop AI 

1291
01:09:38,800 --> 01:09:41,399
systems are doing so at least 
severe disadvantage because 

1292
01:09:41,399 --> 01:09:45,600
they've got much less access to 
high quality hardware than does 

1293
01:09:45,600 --> 01:09:50,080
Open AI or or Google. 
And that's I think the only way 

1294
01:09:50,080 --> 01:09:53,600
to accelerate into the future as
countries, including the US 

1295
01:09:53,600 --> 01:09:56,600
become more restrictive in terms
of controlling access to 

1296
01:09:56,920 --> 01:10:00,640
different types of AI hardware. 
And so it's a good thing that 

1297
01:10:00,640 --> 01:10:04,560
we've got most of it designed 
here and most of it manufactured

1298
01:10:04,960 --> 01:10:07,960
in friendly countries, because 
what we're going to see, what 

1299
01:10:07,960 --> 01:10:10,120
we're already seeing, what we'll
keep seeing is more 

1300
01:10:10,120 --> 01:10:12,800
politicization, I think, of the 
hardware that makes AI possible.

1301
01:10:13,880 --> 01:10:16,000
Is the United? 
I mean the United States still 

1302
01:10:16,000 --> 01:10:18,680
seems great. 
I guess that chip design, given 

1303
01:10:18,680 --> 01:10:22,200
we have so many of the leading 
firms, could, do you think we 

1304
01:10:22,200 --> 01:10:24,320
could ever produce a fab or like
what? 

1305
01:10:24,440 --> 01:10:29,920
What makes TSMC so uniquely 
capable at building a fab? 

1306
01:10:29,920 --> 01:10:34,120
Like, we're talking culture, 
money, like, what's what's 

1307
01:10:34,120 --> 01:10:36,040
driving that in your opinion? 
Yeah. 

1308
01:10:36,040 --> 01:10:38,320
You know it's it's it's not it 
can't be a cultural thing 

1309
01:10:38,320 --> 01:10:41,200
because Taiwan hasn't always 
been the center of the chip 

1310
01:10:41,200 --> 01:10:43,480
industry. 
And if you go back four decades 

1311
01:10:43,480 --> 01:10:46,480
ago, it was Japan that was the 
world's biggest producer of of 

1312
01:10:46,480 --> 01:10:50,800
advanced semiconductors. 
So it's primarily I think the 

1313
01:10:50,800 --> 01:10:54,960
result of TSMC business model 
which has let it scale and the 

1314
01:10:54,960 --> 01:10:57,080
scale has let it Dr. 
efficiencies. 

1315
01:10:57,080 --> 01:11:01,480
And so when TSMC was founded in 
1987 at that time almost all 

1316
01:11:01,480 --> 01:11:04,800
companies both designed and 
manufactured chips in house. 

1317
01:11:04,800 --> 01:11:08,320
They did, both sides of the of 
the equation, And the founder of

1318
01:11:08,320 --> 01:11:11,800
TSMC, Morris Chang, realized 
that as it was getting more 

1319
01:11:11,800 --> 01:11:14,040
complex to manufacture, 
companies would prefer to 

1320
01:11:14,040 --> 01:11:16,520
outsource it. 
And so he decided that he wanted

1321
01:11:16,520 --> 01:11:18,600
to be sort of like Gutenberg was
for books. 

1322
01:11:18,600 --> 01:11:20,640
Gutenberg didn't write any 
books, he only printed them. 

1323
01:11:20,960 --> 01:11:23,120
Morris Chang and TSMC, they 
didn't design any chips. 

1324
01:11:23,120 --> 01:11:25,920
They only manufactured them. 
And what that meant is that he 

1325
01:11:25,920 --> 01:11:29,280
could produce chips not just for
one company, but for lots of 

1326
01:11:29,280 --> 01:11:32,040
companies. 
For AMD, for Apple, for 

1327
01:11:32,040 --> 01:11:34,800
Qualcomm, for NVIDIA. 
And so today, TSMC is the 

1328
01:11:34,800 --> 01:11:37,640
world's largest chip maker. 
And precisely because it's the 

1329
01:11:37,640 --> 01:11:40,880
largest, it's also the most 
advanced because it can hone, it

1330
01:11:40,880 --> 01:11:43,560
processes over every single 
silicon wafer that it 

1331
01:11:43,560 --> 01:11:47,240
manufacturers. 
How can you can you make any 

1332
01:11:47,240 --> 01:11:51,240
predictions about how much you 
think AI chips are going to 

1333
01:11:51,240 --> 01:11:53,880
improve? 
Or like what do you see sort of 

1334
01:11:53,880 --> 01:11:57,680
the next couple years looking 
like in terms of like do 

1335
01:11:57,680 --> 01:12:01,840
availability problems resolve 
and and how much more do you 

1336
01:12:01,840 --> 01:12:05,240
think these chips will improve 
over the next couple years? 

1337
01:12:06,680 --> 01:12:10,520
Yeah, on the the availability 
front, NVIDIA and TSMC has said 

1338
01:12:10,520 --> 01:12:14,640
they expect it to be a year or 
so until the shortages get 

1339
01:12:14,640 --> 01:12:16,680
resolved, but I think they're 
going to get resolved. 

1340
01:12:17,120 --> 01:12:20,800
Prices are such that there's a 
very strong incentive to solve 

1341
01:12:20,800 --> 01:12:23,040
the availability problem and 
sell more. 

1342
01:12:23,280 --> 01:12:27,160
Sometimes people think they want
to keep a shortage to keep sort 

1343
01:12:27,160 --> 01:12:29,080
of demand high. 
You you think they want to 

1344
01:12:29,080 --> 01:12:31,920
produce as many chips as they 
can get out there to make the 

1345
01:12:31,920 --> 01:12:35,120
money while it's there to be 
made or yeah I think that's 

1346
01:12:35,120 --> 01:12:37,800
right because they're they're it
looks like they're a 

1347
01:12:37,800 --> 01:12:40,240
monopolistic producer. 
They've got 90% of the market 

1348
01:12:40,240 --> 01:12:43,960
but if they can't supply chips 
there are competitors there is 

1349
01:12:43,960 --> 01:12:47,280
Google with TPUS there is AMD. 
And so if there are long run 

1350
01:12:47,280 --> 01:12:50,480
shortages that that are are in 
the supply chain, people will 

1351
01:12:50,480 --> 01:12:52,240
turn to competitors. 
And so they've got a strong 

1352
01:12:52,240 --> 01:12:54,520
incentive to get get supply 
increase. 

1353
01:12:54,520 --> 01:12:57,160
And actually part of their 
strategy is again to put their 

1354
01:12:57,160 --> 01:13:01,560
chips at the center of the AI 
world and make it the the gold 

1355
01:13:01,560 --> 01:13:04,000
standard. 
And the more people that design,

1356
01:13:04,000 --> 01:13:08,240
the more people that train, more
people that study using their 

1357
01:13:08,240 --> 01:13:11,240
system using the Kuda ecosystem,
the more likely NVIDIA is to be 

1358
01:13:11,240 --> 01:13:14,840
at the center going forward. 
And then I guess sort of the 

1359
01:13:14,840 --> 01:13:17,320
next chip they have is hard to 
know or have there been any 

1360
01:13:17,320 --> 01:13:21,000
leaks about what what the future
looks like in terms of these 

1361
01:13:21,000 --> 01:13:24,160
artificial intelligence chips? 
I mean I think the the best 

1362
01:13:24,160 --> 01:13:27,680
guide is is to look at the past 
and and what you find is that 

1363
01:13:28,880 --> 01:13:33,240
that for over the past half 
decade they've released a series

1364
01:13:33,240 --> 01:13:37,600
of better and better chips that 
you know ballpark Moore's law 

1365
01:13:37,600 --> 01:13:40,560
gives you a pretty good guide as
to where things should be in a 

1366
01:13:40,560 --> 01:13:44,520
couple of years time bigger, 
more powerful hopefully somewhat

1367
01:13:44,520 --> 01:13:49,320
more power efficient and and the
other area focuses the the 

1368
01:13:49,320 --> 01:13:52,400
interconnect speed between 
semiconductors. 

1369
01:13:52,400 --> 01:13:54,600
So there's the GPUs themselves, 
there's all the networking 

1370
01:13:54,600 --> 01:13:57,600
equipment that moves data 
between chips. 

1371
01:13:57,600 --> 01:14:01,160
That's also an area of really 
intense focus right now. 

1372
01:14:01,400 --> 01:14:03,680
And we're I think we're gonna 
see a lot of progress for the 

1373
01:14:03,680 --> 01:14:06,800
coming years. 
Jensen I mean, did you interact 

1374
01:14:06,800 --> 01:14:10,760
with him at all for this book or
no, I didn't actually. 

1375
01:14:10,760 --> 01:14:12,720
I I interviewed his his Co 
founder but not not. 

1376
01:14:12,760 --> 01:14:13,440
Interesting. 
Yeah. 

1377
01:14:13,800 --> 01:14:18,360
I mean, just like observe. 
Like it must be crazy to go from

1378
01:14:18,360 --> 01:14:21,960
sort of being seen as this, 
like, I don't know, wild sort of

1379
01:14:21,960 --> 01:14:25,840
futurist to all of it. 
Seeming to come true or the 

1380
01:14:25,840 --> 01:14:29,120
world sort of agreeing with your
futurism. 

1381
01:14:29,120 --> 01:14:32,640
I mean he sort of he he wears 
like leather jackets and dresses

1382
01:14:32,640 --> 01:14:35,760
all black or like what? 
What do you make of him? 

1383
01:14:35,760 --> 01:14:39,040
And yeah, how much has this been
sort of a vindication for him? 

1384
01:14:40,200 --> 01:14:44,560
Well, I think he he sort of sees
himself like like Steve Jobs was

1385
01:14:44,560 --> 01:14:47,160
for the iPhone moment, he is for
the AI moment. 

1386
01:14:47,160 --> 01:14:50,320
And I think that's right. 
I I think he he has played that 

1387
01:14:50,320 --> 01:14:54,720
role because he does with a lot 
of credit for investing in the 

1388
01:14:54,720 --> 01:14:58,480
types of products that that have
made AI possible at a time when 

1389
01:14:58,480 --> 01:15:02,440
most people thought it was 
either crazy or possible but 

1390
01:15:02,440 --> 01:15:07,560
only relevant for niche academic
uses or something that would 

1391
01:15:07,560 --> 01:15:10,600
never be financially viable. 
And look, he's built one of the 

1392
01:15:10,600 --> 01:15:13,120
world's largest tech companies 
around it. 

1393
01:15:13,120 --> 01:15:16,160
And already there are products 
being produced on a regular 

1394
01:15:16,160 --> 01:15:19,400
basis using chips that his 
company has pioneered. 

1395
01:15:20,520 --> 01:15:23,200
I believe I agree with you. 
You can make a fortune. 

1396
01:15:23,400 --> 01:15:26,400
Silicon Valley can love you. 
But I feel like the world will 

1397
01:15:26,400 --> 01:15:29,360
only love you if it's your 
device in their hands or it's 

1398
01:15:29,360 --> 01:15:30,960
your brand. 
You know, it. 

1399
01:15:31,280 --> 01:15:34,480
I, I, it just doesn't seem like 
they have a path for like 

1400
01:15:34,600 --> 01:15:37,960
consumer love, right? 
I mean, do you see them trying 

1401
01:15:37,960 --> 01:15:42,840
to take advantage of this and 
go, you know, offer other 

1402
01:15:42,840 --> 01:15:45,120
products or, you know, they're 
so good at chips, they're 

1403
01:15:45,200 --> 01:15:47,800
they're gonna, they'll win the 
technologist's heart and be 

1404
01:15:47,800 --> 01:15:50,920
happy with their cash mile. 
Yeah. 

1405
01:15:51,520 --> 01:15:54,880
They're so far from the computer
in terms of where they sit in 

1406
01:15:54,880 --> 01:15:58,280
the tech stack that like most 
consumers don't even know they 

1407
01:15:58,280 --> 01:16:02,160
exist. 
Or gamers do is sort of the 

1408
01:16:02,160 --> 01:16:03,040
funny gamers do. 
Right. 

1409
01:16:03,040 --> 01:16:07,120
Yeah. 
How did how did you get into 

1410
01:16:07,120 --> 01:16:09,880
this? 
Like what were you sorry, I 

1411
01:16:09,880 --> 01:16:11,160
don't know as much of your 
background. 

1412
01:16:11,160 --> 01:16:12,720
Yeah. 
Before you wrote this book, were

1413
01:16:12,720 --> 01:16:15,480
you a chip nerd or what? 
No, not at all. 

1414
01:16:15,640 --> 01:16:17,680
Oh, really? 
OK what was your story to the 

1415
01:16:17,680 --> 01:16:20,760
book? 
So I I'm a an economic historian

1416
01:16:20,760 --> 01:16:22,880
by training written a couple of 
books on different aspects of 

1417
01:16:22,880 --> 01:16:26,160
economic history and and decided
to write the book when I came to

1418
01:16:26,160 --> 01:16:30,200
realize how important ships were
and was shocked myself that I'd 

1419
01:16:30,200 --> 01:16:32,680
never paid any attention to this
industry. 

1420
01:16:32,680 --> 01:16:36,000
And so it was the combination of
learning that almost all 

1421
01:16:36,000 --> 01:16:38,680
advanced ships are made in 
Taiwan, which was shocking. 

1422
01:16:39,160 --> 01:16:42,640
Learning that chips are so 
extraordinarily difficult to 

1423
01:16:42,640 --> 01:16:46,640
manufacture and transistors in 
your iPhone, for example, are 

1424
01:16:46,640 --> 01:16:49,080
small in the size of a 
coronavirus manufactured by the 

1425
01:16:49,080 --> 01:16:52,040
billions. 
That was an extraordinary fact 

1426
01:16:52,040 --> 01:16:54,280
that got me into it. 
And then and then finally 

1427
01:16:54,280 --> 01:16:57,560
looking at the ramifications for
AII, like most people have 

1428
01:16:57,560 --> 01:17:00,840
thought about AI as a as a 
question of of of either 

1429
01:17:00,840 --> 01:17:03,120
software or of data. 
Data is the new oil I'd read a 

1430
01:17:03,120 --> 01:17:05,880
million times, but it turns out 
that's not exactly right. 

1431
01:17:06,800 --> 01:17:10,840
Did you go dump all your money 
into NVIDIA or I should have? 

1432
01:17:11,440 --> 01:17:12,840
Yeah. 
Did you see the stock? 

1433
01:17:13,080 --> 01:17:15,480
I mean I I know. 
Well you're if you're an 

1434
01:17:15,480 --> 01:17:19,560
economic historian like do you 
have a view on whether, OK, this

1435
01:17:19,560 --> 01:17:21,560
is a great company but it's 
overvalued. 

1436
01:17:21,560 --> 01:17:24,520
I mean you know, yeah it's do 
you, I mean you shouldn't turn 

1437
01:17:24,520 --> 01:17:26,480
to me for for stock market 
advice. 

1438
01:17:26,520 --> 01:17:30,800
But I mean I think the, the key 
question about NVIDIA is, is 

1439
01:17:30,800 --> 01:17:33,200
will it. 
I, I think it's clear it will 

1440
01:17:33,200 --> 01:17:36,160
continue to play a dominant role
in AI training. 

1441
01:17:36,920 --> 01:17:38,800
Maybe it'll lose market share, 
maybe it'll win market share 

1442
01:17:38,800 --> 01:17:40,200
with Google, but it'll play a 
central role. 

1443
01:17:40,200 --> 01:17:44,960
But in the inference market will
NVIDIA be as central there or 

1444
01:17:44,960 --> 01:17:46,720
not? 
That's I think the key question 

1445
01:17:46,720 --> 01:17:48,880
that we're gonna find the answer
to over the next couple of 

1446
01:17:48,880 --> 01:17:51,560
years. 
And do you think China is close 

1447
01:17:51,560 --> 01:17:53,960
to catching up? 
You know, I I think China's is 

1448
01:17:53,960 --> 01:17:58,120
not that far behind, but China 
has been not that far behind for

1449
01:17:58,120 --> 01:18:01,120
some time. 
And because the rate of advance 

1450
01:18:01,120 --> 01:18:04,440
is so rapid in terms of design 
in the US and manufacturing in 

1451
01:18:04,440 --> 01:18:07,440
Taiwan, it's very, very hard to 
catch up. 

1452
01:18:08,160 --> 01:18:12,400
And so long as that companies 
like NVIDIA and TSMC keep racing

1453
01:18:12,400 --> 01:18:15,080
forward to catching up is just 
going to be extraordinarily 

1454
01:18:15,080 --> 01:18:16,800
challenging for China and no 
matter how much money they pour 

1455
01:18:16,800 --> 01:18:19,320
into it. 
And are Japan and South Korea 

1456
01:18:19,320 --> 01:18:22,960
relevant to this conversation? 
Well, it's that's an interesting

1457
01:18:22,960 --> 01:18:26,240
question. 
Japan is relevant because if you

1458
01:18:26,240 --> 01:18:29,480
look inside of a chip making 
facility, there is a ton of 

1459
01:18:29,480 --> 01:18:34,520
Japanese machine tools and 
there's also a wide variety of 

1460
01:18:34,600 --> 01:18:37,240
ultra specialized chemicals made
in Japan. 

1461
01:18:37,240 --> 01:18:40,160
So a lot of the chemicals used 
in chip making because of the 

1462
01:18:40,480 --> 01:18:46,400
precision required have to be 
purified to the 99.99999% level.

1463
01:18:46,560 --> 01:18:48,240
A lot of those are only made in 
Japan. 

1464
01:18:49,600 --> 01:18:53,240
I wanted to go back and I know 
this is like a complicated 

1465
01:18:53,240 --> 01:18:58,480
question, but like what are the 
tasks that are being thrown at 

1466
01:18:58,480 --> 01:19:01,440
these chips, right? 
Or you talked about sort of 

1467
01:19:01,440 --> 01:19:04,760
matrix computing or? 
Is there like 1 task over and 

1468
01:19:04,760 --> 01:19:07,880
over again that these chips are 
being asked to do or it's a lot 

1469
01:19:07,880 --> 01:19:10,960
of different things or how? 
How are the foundation models 

1470
01:19:11,240 --> 01:19:15,040
sort of translating to the chips
as best you can explain it? 

1471
01:19:16,320 --> 01:19:18,840
I I guess the way I would 
explain it is you know if you're

1472
01:19:18,840 --> 01:19:24,000
trying to train a a a chat bot 
like ChatGPT, you train it by 

1473
01:19:24,000 --> 01:19:27,760
taking all of the language data 
in a a vast data set like 

1474
01:19:27,760 --> 01:19:30,640
Wikipedia. 
You essentially read Wikipedia 

1475
01:19:30,640 --> 01:19:33,120
and identify patterns, and so 
and. 

1476
01:19:33,520 --> 01:19:35,800
That's and that's like inference
that you've been taught. 

1477
01:19:36,160 --> 01:19:37,520
That's training. 
That's training. 

1478
01:19:37,520 --> 01:19:39,880
So training is is identifying 
the pattern. 

1479
01:19:39,880 --> 01:19:48,080
So you know the what's the what?
What is the next word I'm going 

1480
01:19:48,080 --> 01:19:52,240
to say? 
You know, you know what say, and

1481
01:19:52,240 --> 01:19:55,840
Chachi Piti knows that because 
it's identified thousands of 

1482
01:19:55,840 --> 01:19:59,720
other instances where sentences 
like that ended with the word 

1483
01:19:59,920 --> 01:20:02,760
say. 
And it did that because it read 

1484
01:20:02,760 --> 01:20:06,440
lots and lots of books and 
articles and Reddit posts 

1485
01:20:06,440 --> 01:20:08,680
online. 
And then when I you go back and 

1486
01:20:08,680 --> 01:20:11,280
ask a question which is 
inference, it gives you a pretty

1487
01:20:11,280 --> 01:20:14,400
good answer because it has a 
pretty good sense of what the 

1488
01:20:14,400 --> 01:20:18,160
most likely word is going to be.
So you're saying they're they're

1489
01:20:18,160 --> 01:20:23,040
training the chips based on a a 
lot of data and then it makes an

1490
01:20:23,040 --> 01:20:26,800
inference about what it thinks 
the the sort of fill in the 

1491
01:20:26,800 --> 01:20:30,720
blank of of the prompt? 
Basically, yeah, inference is 

1492
01:20:30,720 --> 01:20:33,120
just an educated guess, a 
probabilistic guess. 

1493
01:20:33,760 --> 01:20:37,000
And what's cool about Chachi PT 
is that it's not just fill in 

1494
01:20:37,000 --> 01:20:39,360
one word, it's fill in dozens 
of. 

1495
01:20:39,360 --> 01:20:40,640
Words. 
I know, it's insane. 

1496
01:20:41,640 --> 01:20:44,960
One thing I'm still trying to 
get my head around is, you know,

1497
01:20:44,960 --> 01:20:50,040
like DeepMind, right? 
Like when they trained, you 

1498
01:20:50,040 --> 01:20:55,840
know, computer systems, I guess 
to to compete on chess or Go or 

1499
01:20:55,840 --> 01:20:58,400
Dota. 
Like my sense is that you know 

1500
01:20:58,400 --> 01:21:01,800
they're using. 
Sort of this sort of scale data 

1501
01:21:01,800 --> 01:21:04,880
approach, but they also have 
like specific ways they're 

1502
01:21:04,880 --> 01:21:08,120
teaching it to think about 
specific problems. 

1503
01:21:08,120 --> 01:21:13,560
Or like, do you, are those types
of approaches to AI also using 

1504
01:21:13,560 --> 01:21:17,560
these same chips, or are they 
still sort of invested in the 

1505
01:21:17,560 --> 01:21:21,280
same arms race? 
Yeah, well, the chips just do 

1506
01:21:21,280 --> 01:21:22,960
the math for you. 
And so they'll do what other 

1507
01:21:22,960 --> 01:21:24,480
math? 
Whatever math you ask them to 

1508
01:21:24,480 --> 01:21:26,320
do, Yeah. 
Exactly. 

1509
01:21:27,880 --> 01:21:31,120
And they're not. 
And then the interaction between

1510
01:21:31,120 --> 01:21:34,880
like Nvidia's like programming 
language is super interesting 

1511
01:21:36,040 --> 01:21:38,840
because they're not like the the
foundation models are not being 

1512
01:21:38,840 --> 01:21:40,800
written in that language, right?
But they're they're sort of 

1513
01:21:40,800 --> 01:21:43,440
bridging. 
That's right. 

1514
01:21:43,480 --> 01:21:46,080
Yep. 
And it's it's just about, it's 

1515
01:21:46,080 --> 01:21:48,760
fundamentally allows them to 
optimize what's happening on 

1516
01:21:48,760 --> 01:21:52,320
these chips much more than maybe
they would on like an AMD chip 

1517
01:21:52,320 --> 01:21:55,320
even if it were and it's it's 
it's it's easier, it's more 

1518
01:21:55,320 --> 01:21:58,880
familiar and therefore there's a
big Moat around invidious 

1519
01:21:58,880 --> 01:22:00,080
position. 
Yeah. 

1520
01:22:01,360 --> 01:22:06,200
So I have you like become like 
were you bullish on like AI 

1521
01:22:06,200 --> 01:22:10,080
before writing this book or I'm 
curious how much your view has 

1522
01:22:10,080 --> 01:22:13,680
really changed on, yeah like I 
don't know like generalized 

1523
01:22:13,680 --> 01:22:16,640
intelligence or like the 
imminence of self driving cars 

1524
01:22:16,640 --> 01:22:19,920
or any of that. 
You know I I think that actually

1525
01:22:19,920 --> 01:22:23,480
we've I I'm I'm in the long run,
I'm very bullish but I think in 

1526
01:22:23,560 --> 01:22:26,600
in the the short run the last 12
months there's been all a whole 

1527
01:22:26,600 --> 01:22:30,240
lot of excitement of the 
imminent arrival of AGI which 

1528
01:22:30,440 --> 01:22:33,680
I'm I'm not so sure I see it 
happening in 2023. 

1529
01:22:34,960 --> 01:22:37,200
I think if you look at the 
trends, the trends are all very 

1530
01:22:37,200 --> 01:22:39,120
positive. 
AI system is getting better, 

1531
01:22:39,120 --> 01:22:44,920
more capable, but I think there 
there there was due to ChatGPT a

1532
01:22:44,920 --> 01:22:47,280
popular sense that we're going 
to get some extraordinary 

1533
01:22:47,280 --> 01:22:48,760
products in our hands right 
away. 

1534
01:22:48,760 --> 01:22:51,640
And actually, it's going to be a
couple of years before we see 

1535
01:22:51,640 --> 01:22:54,800
AI, the current wave of AI 
products deployed in a big way 

1536
01:22:55,200 --> 01:22:57,600
that meaningfully impacts 
economy or society. 

1537
01:22:57,760 --> 01:23:00,840
Yeah, I think we're all watching
to see if this hype wave is 

1538
01:23:00,840 --> 01:23:03,720
about to end or not. 
I mean there's there's a world 

1539
01:23:03,720 --> 01:23:06,720
where this is sort of peak 
NVIDIA, right? 

1540
01:23:06,720 --> 01:23:10,160
Like we have this sort of mania 
around AI, everybody's running 

1541
01:23:10,160 --> 01:23:15,880
towards it and then it turns 
out, you know, chat GPD doesn't 

1542
01:23:15,880 --> 01:23:18,200
improve that much and self 
driving cars still have a lot of

1543
01:23:18,200 --> 01:23:21,520
edge cases. 
And you know, the chips don't 

1544
01:23:21,520 --> 01:23:23,160
solve everything. 
Even if they do get better. 

1545
01:23:23,160 --> 01:23:24,520
Like you're saying, like, I 
don't know. 

1546
01:23:24,520 --> 01:23:27,880
How likely do you think that is?
Or do you have a sense of, yeah,

1547
01:23:27,880 --> 01:23:31,200
whether better years are ahead 
or this is sort of a peak moment

1548
01:23:31,200 --> 01:23:34,200
for the company. 
Well it's it's been a very, very

1549
01:23:34,200 --> 01:23:36,080
good year for the company. 
So I wouldn't be surprised if 

1550
01:23:36,080 --> 01:23:38,520
it's a it's it's at least a 
localized peak. 

1551
01:23:38,520 --> 01:23:43,160
But I I think you know for for 
AI applications most people 

1552
01:23:43,160 --> 01:23:46,400
think of what's the consumer 
application for AI, you know 

1553
01:23:46,400 --> 01:23:50,560
what's what's the iPhone of AI. 
But it seems to me that for the 

1554
01:23:50,560 --> 01:23:53,800
next couple of years actually 
most of the application and most

1555
01:23:53,800 --> 01:23:56,920
of the money in AI will be in 
enterprises because enterprises 

1556
01:23:56,920 --> 01:23:59,840
that have vast data, they've got
the desire to use it efficiently

1557
01:23:59,840 --> 01:24:02,160
to monetize it. 
And so that's where I think 

1558
01:24:02,160 --> 01:24:04,920
we're actually already seeing a 
lot of the investment and a lot 

1559
01:24:04,920 --> 01:24:07,640
of the early products are 
happening in pretty boring 

1560
01:24:07,640 --> 01:24:10,240
places inside of enterprises. 
But that that's probably OK 

1561
01:24:10,240 --> 01:24:14,000
that's that's what's going to 
drive I think the the, the 

1562
01:24:14,000 --> 01:24:15,800
building of effective products 
in the long run. 

1563
01:24:17,160 --> 01:24:19,840
Enterprises, you know, 
businesses just translate it for

1564
01:24:19,840 --> 01:24:21,720
like the regular person, you 
know, I used to work at 

1565
01:24:21,720 --> 01:24:23,560
Bloomberg. 
They've actually come out, I 

1566
01:24:23,560 --> 01:24:26,520
think with their own foundation 
model to show how they can train

1567
01:24:26,840 --> 01:24:29,320
financial data. 
And you you have a good example 

1568
01:24:29,320 --> 01:24:33,920
of where you think corporations 
will put this approach and data 

1569
01:24:33,920 --> 01:24:37,400
to use. 
If you imagine a company like 

1570
01:24:37,400 --> 01:24:41,800
Walmart, they have to everyday 
decide what price to put all of 

1571
01:24:41,800 --> 01:24:45,120
their products, and they decide 
pricing based on a whole variety

1572
01:24:45,120 --> 01:24:46,720
of different factors. 
What their competitors are 

1573
01:24:46,720 --> 01:24:48,680
pricing? 
The availability of products. 

1574
01:24:48,720 --> 01:24:53,040
Very complex process and that 
seems like a perfect use case 

1575
01:24:53,040 --> 01:24:57,200
for trying to use more advanced 
AI algorithms to set better 

1576
01:24:57,200 --> 01:24:59,040
prices more rapidly. 
Boring is it? 

1577
01:24:59,440 --> 01:25:01,760
Kmart this blue light discounts.
What's Walmart? 

1578
01:25:01,760 --> 01:25:03,840
I forget. 
But yeah, who gets the discount?

1579
01:25:03,880 --> 01:25:07,480
You know where, right? 
A lot of products to keep track 

1580
01:25:07,480 --> 01:25:09,800
of. 
Have you like? 

1581
01:25:10,880 --> 01:25:14,240
It feels so absurd to ask this, 
but like, what the chip business

1582
01:25:14,240 --> 01:25:18,800
looks like in a world of AGI, 
generalized intelligence? 

1583
01:25:18,800 --> 01:25:20,960
Or like, have you gained that 
out at all? 

1584
01:25:20,960 --> 01:25:24,240
I mean, there are people who 
think it could happen, you know,

1585
01:25:24,720 --> 01:25:28,240
five years from now or less. 
Like, do you have a view on it 

1586
01:25:28,360 --> 01:25:30,160
first of all and then second? 
Yeah, what? 

1587
01:25:30,160 --> 01:25:33,000
What would it mean for the chip 
world? 

1588
01:25:34,160 --> 01:25:37,440
Well I I guess my sense is that 
we're we're we're a long way 

1589
01:25:37,440 --> 01:25:38,640
away. 
I'm I'm still waiting for 

1590
01:25:38,640 --> 01:25:41,160
ChatGPT to accurately complete 
all my senses. 

1591
01:25:41,160 --> 01:25:45,120
You know still at a 90% rate and
the other 10% are pretty bad. 

1592
01:25:46,120 --> 01:25:50,000
I think it's gonna require 
tremendous advances in 

1593
01:25:50,000 --> 01:25:54,080
semiconductors to make possible 
the tremendous increase in 

1594
01:25:54,080 --> 01:25:56,560
computing that more advanced AI 
systems will need. 

1595
01:25:56,560 --> 01:25:57,760
I mean that that's been the 
trend. 

1596
01:25:57,760 --> 01:26:00,720
The trend has been AI systems 
only advance. 

1597
01:26:00,720 --> 01:26:03,880
We apply more computing to them,
and so if we want systems to be 

1598
01:26:03,880 --> 01:26:07,320
twice as good as they are today,
we're gonna need something not 

1599
01:26:07,320 --> 01:26:09,720
too far off from twice as much 
compute to make it possible. 

1600
01:26:10,760 --> 01:26:14,480
Are there, you know there was 
this I I covered Righetti and 

1601
01:26:14,480 --> 01:26:16,360
sort of the quantum computing 
world. 

1602
01:26:17,560 --> 01:26:21,320
Do you have much optimism there?
Do you spend time on them in in 

1603
01:26:21,320 --> 01:26:23,960
the book? 
You know, I not not in the book.

1604
01:26:25,000 --> 01:26:29,400
I my sense in in, in, in 
speaking to people in the 

1605
01:26:29,400 --> 01:26:32,880
quantum world and then looking 
also at how in history new 

1606
01:26:32,880 --> 01:26:36,040
computing technologies have 
disseminated is that actually 

1607
01:26:36,440 --> 01:26:40,120
even revolutionary technologies,
they're implemented slowly. 

1608
01:26:40,680 --> 01:26:43,280
And so suppose we get to the 
next couple of years, the first 

1609
01:26:43,280 --> 01:26:45,000
practical use case of quantum 
computing. 

1610
01:26:45,400 --> 01:26:49,080
It's going to be a years long 
process of beginning to 

1611
01:26:49,080 --> 01:26:51,880
implement that in all sorts of 
different computing use cases. 

1612
01:26:51,880 --> 01:26:54,000
And so I think we shouldn't 
expect to have, you know, a 

1613
01:26:54,000 --> 01:26:55,840
quantum powered iPhone anytime 
soon. 

1614
01:26:56,640 --> 01:27:00,880
It's it's insane to think like 
on the one hand we have. 

1615
01:27:01,480 --> 01:27:04,800
We have public companies that 
people are sort of speculating 

1616
01:27:04,800 --> 01:27:08,320
and betting on they yeah produce
quantum computers but they 

1617
01:27:08,320 --> 01:27:13,400
really do nothing practical. 
I I you can go with I you know I

1618
01:27:13,400 --> 01:27:15,720
swing back and forth. 
On the one hand it's like great 

1619
01:27:15,720 --> 01:27:19,520
it's amazing that our system 
will invest in such bleeding 

1620
01:27:19,520 --> 01:27:21,280
edge technology and give it a 
chance. 

1621
01:27:21,560 --> 01:27:23,800
On the other hand for the 
shareholders, you know like 

1622
01:27:24,120 --> 01:27:26,720
everybody could decide to give 
up on the effort with like 

1623
01:27:26,720 --> 01:27:29,560
higher interest rates, it's, I 
don't know, it's a crazy 

1624
01:27:29,560 --> 01:27:33,600
function of. 
The global economy, well, I I 

1625
01:27:33,600 --> 01:27:35,440
think it's interesting that a 
lot of the companies that are 

1626
01:27:35,440 --> 01:27:38,640
investing the heaviest are also 
big cloud computing operators. 

1627
01:27:39,280 --> 01:27:42,720
And that's I think because they 
believe that quantum will be 

1628
01:27:42,720 --> 01:27:46,640
actually most useful in a 
context where it's closely 

1629
01:27:46,640 --> 01:27:49,800
interlinked with huge volumes of
classical computing. 

1630
01:27:50,200 --> 01:27:53,040
And so actually, we're going to 
need more advanced silicon chips

1631
01:27:53,040 --> 01:27:55,360
to make quantum practically 
applicable. 

1632
01:27:55,840 --> 01:27:58,480
And do you know, is there any 
sense whether Quantum and the 

1633
01:27:58,800 --> 01:28:02,960
GPU sync up, or where they fit 
in that story? 

1634
01:28:03,400 --> 01:28:06,080
There There are many different 
paradigms for how you 

1635
01:28:06,440 --> 01:28:11,040
specifically sync classical and 
quantum computing, but right now

1636
01:28:11,040 --> 01:28:15,280
nobody knows which of the many 
paradigms will win, if any of 

1637
01:28:15,280 --> 01:28:18,240
them. 
To wrap it up, I mean, what do 

1638
01:28:18,240 --> 01:28:21,560
you think the regular person 
should take from all this? 

1639
01:28:21,560 --> 01:28:26,320
Like it feels very far. 
From their lives like they they 

1640
01:28:26,320 --> 01:28:28,600
maybe don't even know what 
NVIDIA is. 

1641
01:28:28,600 --> 01:28:32,120
Like they tried ChatGPT, their 
kids using it for homework. 

1642
01:28:32,120 --> 01:28:35,360
It's not in their life yet. 
Like you know Silicon Valley 

1643
01:28:35,360 --> 01:28:38,080
loves to get itself worked up 
about things that don't 

1644
01:28:38,200 --> 01:28:40,520
sometimes translate. 
Don't like what what do you 

1645
01:28:40,520 --> 01:28:44,000
think the lesson or the the key 
thing for like the regular 

1646
01:28:44,000 --> 01:28:47,320
person right now is in terms of 
what's happening with these 

1647
01:28:47,320 --> 01:28:49,360
chips. 
I think if you went back to 

1648
01:28:49,360 --> 01:28:53,320
1965, which is when Gordon Moore
first set out the the phrase 

1649
01:28:53,320 --> 01:28:57,520
Moore's Law and you asked what's
the impact of of of Moore's Law,

1650
01:28:57,520 --> 01:28:59,960
the average person, the the 
answer in the short run was 

1651
01:29:00,080 --> 01:29:03,040
approximately nothing. 
But the answer in the long run 

1652
01:29:03,040 --> 01:29:06,680
is that it totally transformed 
society, economy, technology, 

1653
01:29:06,680 --> 01:29:10,480
everything, because we put 
computing and therefore 

1654
01:29:10,480 --> 01:29:13,480
semiconductors into basically 
every product that we rely on. 

1655
01:29:13,480 --> 01:29:16,960
And I think we should expect the
same to be true for AI. 

1656
01:29:16,960 --> 01:29:19,080
You know, what does it mean for 
me tomorrow? 

1657
01:29:19,080 --> 01:29:21,520
Probably not much. 
And what is it going to mean in 

1658
01:29:21,520 --> 01:29:25,120
in 10 years and in 20 years when
every aspect of human life is 

1659
01:29:25,120 --> 01:29:27,680
being impacted by it? 
Well, it'll be transformative in

1660
01:29:28,080 --> 01:29:30,400
all sorts of ways, most of which
we probably can't even imagine 

1661
01:29:30,400 --> 01:29:32,640
today. 
For the startup entrepreneur, 

1662
01:29:32,640 --> 01:29:36,240
where do you think there's 
opportunity to build a business?

1663
01:29:36,240 --> 01:29:38,720
You know they, I mean maybe you 
think they should go out and 

1664
01:29:38,720 --> 01:29:42,520
start the next NVIDIA like it's 
a heavy lift, but like where 

1665
01:29:42,520 --> 01:29:46,480
where do you think really with 
the progress that we've made in 

1666
01:29:46,480 --> 01:29:50,200
these chips that there's real 
business opportunities still? 

1667
01:29:51,320 --> 01:29:54,160
Well, I think if you've got the 
idea of the next NVIDIA, you 

1668
01:29:54,160 --> 01:30:00,240
should absolutely go do it. 
But I I think you're, you're 

1669
01:30:00,240 --> 01:30:03,200
right that, you know, companies 
like NVIDIA, they they create 

1670
01:30:03,200 --> 01:30:06,680
the infrastructure on which many
different types of systems can 

1671
01:30:06,680 --> 01:30:09,240
be built. 
And and if you go back to, you 

1672
01:30:09,240 --> 01:30:11,840
know, the smartphone for 
example, smartphones were 

1673
01:30:11,840 --> 01:30:14,160
themselves a platform owners 
could build lots of different 

1674
01:30:14,160 --> 01:30:16,000
things. 
And I think we're still in the 

1675
01:30:16,000 --> 01:30:18,640
Super early stages. 
We're basically in stage zero in

1676
01:30:18,640 --> 01:30:21,400
terms of figuring out what are 
the ways you can create products

1677
01:30:21,400 --> 01:30:24,640
out of generative AI systems. 
Right now, there are hardly any 

1678
01:30:24,640 --> 01:30:27,280
companies that make money 
selling their AI systems. 

1679
01:30:27,400 --> 01:30:29,760
And of course, the challenge for
entrepreneurs is, you know, 

1680
01:30:30,160 --> 01:30:33,840
people who built early iPhone 
apps and Facebook apps, you 

1681
01:30:33,840 --> 01:30:36,720
know, they don't all pan out. 
Sometimes you come too early. 

1682
01:30:36,720 --> 01:30:40,160
You can be right that a 
technology is transformational 

1683
01:30:40,160 --> 01:30:42,880
and get the timing wrong. 
And I guess that's. 

1684
01:30:43,720 --> 01:30:47,360
The sort of confluence of luck 
and insight in, in the business 

1685
01:30:47,360 --> 01:30:49,760
world, it's a tough one to 
predict. 

1686
01:30:50,960 --> 01:30:53,320
Chris, this was awesome. 
Thank you so much for coming on 

1687
01:30:53,320 --> 01:30:54,160
the show. 
It's great. 

1688
01:30:54,280 --> 01:30:56,360
Great to talk to you. 
Thanks for having me. 

1689
01:30:56,600 --> 01:30:59,760
That's our episode on chips and 
big tech. 

1690
01:30:59,760 --> 01:31:03,000
Thanks for listening. 
I'm Eric Newcomer, your host, 

1691
01:31:03,000 --> 01:31:06,000
author of newcomer. 
Thanks so much to Max Child and 

1692
01:31:06,000 --> 01:31:08,880
James Wilsterman. 
Co founder Zavali and my long 

1693
01:31:08,880 --> 01:31:12,120
time friends shout out to Scott 
Brody, our producer Riley 

1694
01:31:12,120 --> 01:31:15,560
Kinsella, my Chief of Staff 
Gabby Caliendo at Volley who's 

1695
01:31:15,560 --> 01:31:19,000
helping organize the conference 
and playing a big role behind 

1696
01:31:19,000 --> 01:31:21,240
the scenes. 
Thank you to young Chomsky for 

1697
01:31:21,240 --> 01:31:24,120
the theme music. 
Please like, comment, subscribe 

1698
01:31:24,120 --> 01:31:29,160
on YouTube, give me a review on 
Apple podcast and subscribe to 

1699
01:31:29,160 --> 01:31:33,440
the sub stack newcomer.co. 
Go try out Volley on your Alexa 

1700
01:31:34,080 --> 01:31:35,840
play song quiz maybe we're going
to have. 

1701
01:31:36,200 --> 01:31:39,160
Couple more episodes before 
Cerebral Valley. 

1702
01:31:39,400 --> 01:31:41,760
Cerebral Valley is on November 
15th. 

1703
01:31:42,280 --> 01:31:46,040
We will publish our 
conversations to YouTube. 

1704
01:31:46,040 --> 01:31:47,440
We will probably publish them 
all. 

1705
01:31:47,440 --> 01:31:50,080
That's what we did last time. 
You can see the conversations 

1706
01:31:50,080 --> 01:31:53,920
from Cerebral Valley One, which 
is on March 30th. 

1707
01:31:54,360 --> 01:31:58,840
And then yeah, we'll play some 
of our favourites probably. 

1708
01:31:58,840 --> 01:32:02,520
We did a sort of distilled 
version in our podcast feed, so 

1709
01:32:02,520 --> 01:32:04,080
follow the podcast feeds for 
those. 

1710
01:32:04,560 --> 01:32:07,040
And yeah, I'll be covering it in
the newcomer newsletter. 

1711
01:32:07,240 --> 01:32:09,600
So newcomer.co, Thanks so much.