1
00:00:00,080 --> 00:00:03,757
Basically, the whole crypto industry
was in the process of moving offshore and

2
00:00:03,840 --> 00:00:07,037
America was being deprived
of this industry of the future.

3
00:00:07,120 --> 00:00:10,077
>> I think the country had a significant
victory with the president signing the

4
00:00:10,160 --> 00:00:13,997
stablecoin bill into law. I think
that the positive consequences of that law

5
00:00:14,080 --> 00:00:15,277
have been even bigger than we thought.

6
00:00:15,360 --> 00:00:18,717
>> The legislation was declared dead
many times. I think that we will get the

7
00:00:18,800 --> 00:00:21,837
clarity act done. This will provide
the regulatory framework for all the other

8
00:00:21,920 --> 00:00:24,317
tokens besides stable coins.
There are a lot of twists and turns in the

9
00:00:24,400 --> 00:00:26,237
legislative process. It's definitely
true that you don't want to see the

10
00:00:26,320 --> 00:00:30,840
sausage getting made. >> [music]

11
00:00:34,060 --> 00:00:36,080
[music]

12
00:00:38,960 --> 00:00:41,277
>> Welcome to Web 3 with A6Z Crypto.

13
00:00:41,360 --> 00:00:45,197
Today's episode is a special
crossover with the A16Z podcast [music]

14
00:00:45,280 --> 00:00:49,997
featuring David Saxs, the Trump
administration's AI and cryptozar in

15
00:00:50,080 --> 00:00:51,437
conversation [music] with A16Z

16
00:00:51,520 --> 00:00:54,397
co-founders Mark Andre and Ben Horowitz,

17
00:00:54,480 --> 00:00:57,357
and general partner Eric Torenberg.

18
00:00:57,440 --> 00:01:01,677
They dig into how the US is approaching
AI and crypto policy, the fight over

19
00:01:01,760 --> 00:01:05,837
open- source software, and what's
at stake in the global tech race with

20
00:01:05,920 --> 00:01:08,077
China. Be sure [snorts] to follow and

21
00:01:08,160 --> 00:01:13,197
subscribe to the A16Z podcast for more
conversations like this one. David,

22
00:01:13,280 --> 00:01:15,437
welcome to the A6Z podcast.
Thanks for joining.

23
00:01:15,520 --> 00:01:19,277
>> Yeah, good to be here. So, David,
you're the AI and cryptosar. Why don't you

24
00:01:19,360 --> 00:01:22,557
first talk about why it makes sense
to have those as a portfolio? What do they

25
00:01:22,640 --> 00:01:25,197
have to do with each other? and then
I'll have you lay out what's the what's

26
00:01:25,280 --> 00:01:28,957
the what's the Trump plan on those
those two categories and how we're doing.

27
00:01:29,040 --> 00:01:30,877
>> Well, there are two technologies that I

28
00:01:30,960 --> 00:01:33,917
guess are relatively new and so there's

29
00:01:34,000 --> 00:01:39,597
a lot of fear of them and uh I think
people don't know that much about them.

30
00:01:39,680 --> 00:01:43,597
They don't really know what to make
of them. I think that from a policy

31
00:01:43,680 --> 00:01:46,237
standpoint and we can talk about
the similarities and differences. The the

32
00:01:46,320 --> 00:01:49,037
approaches are a little different. Um,

33
00:01:49,120 --> 00:01:51,277
I think with crypto, the main thing

34
00:01:51,360 --> 00:01:54,557
that's needed is regulatory certainty.

35
00:01:54,640 --> 00:01:58,477
You know, all the entrepreneurs I've
talked to over the years, they they all

36
00:01:58,560 --> 00:02:01,437
say the same thing, which is just tell
us what the rules are. We're happy to

37
00:02:01,520 --> 00:02:05,037
comply, but people, you know, watch
them won't tell us what they And in fact,

38
00:02:05,120 --> 00:02:06,477
during the Biden years,

39
00:02:06,560 --> 00:02:10,717
>> you had um an SEC chairman who uh took

40
00:02:10,800 --> 00:02:14,717
an approach which I guess has been
called um regulation through uh

41
00:02:14,800 --> 00:02:17,917
enforcement, which basically means you
just get prosecuted. They don't tell you

42
00:02:18,000 --> 00:02:21,437
what the rules are. You just basically
get indicted and then everyone else is

43
00:02:21,520 --> 00:02:24,637
supposed to divine what the rules are um

44
00:02:24,720 --> 00:02:27,197
as you get prosecuted and fined and

45
00:02:27,280 --> 00:02:29,357
imprisoned. So that was the approach for

46
00:02:29,440 --> 00:02:31,677
several years and um you know and and

47
00:02:31,760 --> 00:02:34,397
and there and as a result of that

48
00:02:34,480 --> 00:02:36,397
basically the whole crypto industry was

49
00:02:36,480 --> 00:02:39,437
in the process of moving offshore and uh

50
00:02:39,520 --> 00:02:41,597
and America I think was being deprived

51
00:02:41,680 --> 00:02:45,037
of this industry of the future. And so

52
00:02:45,120 --> 00:02:48,077
President Trump during his campaign and

53
00:02:48,160 --> 00:02:53,037
last year he gave a now famous speech
in Nashville in which he declared that he

54
00:02:53,120 --> 00:02:56,717
would make the United States the crypto
capital of the planet and that he would

55
00:02:56,800 --> 00:02:59,757
fire Gendler. That was like
the big applause line. I think he

56
00:02:59,840 --> 00:03:03,437
>> he's applauded. [laughter] >>
He's talked about how surprised he was

57
00:03:03,520 --> 00:03:08,397
that you know what a big uh you know
ovation he got at that. So he said it

58
00:03:08,480 --> 00:03:11,517
again and you know the crowd erupted
again. But in any event, he promised

59
00:03:11,600 --> 00:03:16,877
basically to provide this clarity so
that um you know the the industry would

60
00:03:16,960 --> 00:03:21,357
um understand what the rules are, be
able to comply. In turn, that gives that

61
00:03:21,440 --> 00:03:23,677
should provide uh greater protection for

62
00:03:23,760 --> 00:03:26,637
consumers and um businesses, everyone

63
00:03:26,720 --> 00:03:31,037
who's part of the ecosystem. Uh and it
makes America more competitive. So I

64
00:03:31,120 --> 00:03:32,957
think that's sort of the the the the

65
00:03:33,040 --> 00:03:36,957
mandate on on crypto is uh it's in a way

66
00:03:37,040 --> 00:03:40,077
it's pro- regulation. is basically we we

67
00:03:40,160 --> 00:03:45,117
want to put in place regulations.
In a in a way, AI is kind of the opposite

68
00:03:45,200 --> 00:03:48,877
where I think the Biden administration
was too heavy-handed. They were starting

69
00:03:48,960 --> 00:03:52,317
to really regulate this area without
even understanding what it was. They

70
00:03:52,400 --> 00:03:57,117
weren't they had no one had really taken
the time to understand uh how AI was

71
00:03:57,200 --> 00:04:00,957
even being used, what the real um
dangers were. there was this intense

72
00:04:01,040 --> 00:04:02,477
fear-mongering

73
00:04:02,560 --> 00:04:05,037
and as a result of that, you know, the

74
00:04:05,120 --> 00:04:09,517
approach of the B administration
was uh they were in the process of of

75
00:04:09,600 --> 00:04:13,277
implementing very heavy-handed
regulations on both the software and

76
00:04:13,360 --> 00:04:15,597
hardware side. And we can drill into

77
00:04:15,680 --> 00:04:17,917
that. I think that with the Trump

78
00:04:18,000 --> 00:04:20,637
administration, um the approach has been

79
00:04:20,720 --> 00:04:25,037
that we want the United States
to win the AI race. Um it's a global

80
00:04:25,120 --> 00:04:29,597
competition. Um, sometimes we mention
the fact that China is our probably our

81
00:04:29,680 --> 00:04:33,117
main competitor in this area. They're
the only other country that has the

82
00:04:33,200 --> 00:04:36,317
technological capability, the talent,

83
00:04:36,400 --> 00:04:39,757
the knowhow, the um expertise to to beat

84
00:04:39,840 --> 00:04:44,317
us in this area. And we want to make
sure the United States wins. Um, and of

85
00:04:44,400 --> 00:04:47,837
course in the US, it's not really
the government that's responsible for

86
00:04:47,920 --> 00:04:51,437
innovation, it's the private sector.
So that means that our companies have to

87
00:04:51,520 --> 00:04:55,357
win. And if you're imposing all sorts
of crazy um burdensome regulation on them,

88
00:04:55,440 --> 00:04:57,197
then that's going to hurt not not help.

89
00:04:57,280 --> 00:05:00,317
So um the president gave a I think very

90
00:05:00,400 --> 00:05:04,957
important AI policy speech a couple
months ago on July 23rd where he

91
00:05:05,040 --> 00:05:09,517
declared in no uncertain terms that we
had to win the AI race and he laid out

92
00:05:09,600 --> 00:05:11,757
several pillars on on how we do that. It

93
00:05:11,840 --> 00:05:16,637
was pro- innovation, pro infrastructure,
which also means uh pro energy and pro

94
00:05:16,720 --> 00:05:20,077
export. And we can drill into all those
things if you want, but that was sort of

95
00:05:20,160 --> 00:05:22,717
the the high line. And so I think that,

96
00:05:22,800 --> 00:05:27,997
you know, again with with AI, um the idea
is kind of like h how do we unleash

97
00:05:28,080 --> 00:05:31,117
innovation? And I think with crypto,
it's been more about how do we create

98
00:05:31,200 --> 00:05:36,637
regulatory certainty? Um but you know,
in terms of my role, like what you know,

99
00:05:36,720 --> 00:05:39,677
why why am I doing both? I mean, I think
the common denominator is just again

100
00:05:39,760 --> 00:05:42,877
these are new technologies. They um you

101
00:05:42,960 --> 00:05:46,877
know they're both um you know obviously
come from from the tech industry which

102
00:05:46,960 --> 00:05:50,877
has a very different culture than
Washington does and I kind of see as my

103
00:05:50,960 --> 00:05:53,437
role to help be a bridge between what's

104
00:05:53,520 --> 00:05:54,637
happening in Silicon Valley and what's

105
00:05:54,720 --> 00:05:58,477
happening in Washington and helping

106
00:05:58,560 --> 00:06:02,317
Washington understand not just
the policy that's needed or or the

107
00:06:02,400 --> 00:06:07,277
innovation that's happening but also
kind of culturally you know what what

108
00:06:07,360 --> 00:06:10,237
makes the tech industry different
and special and and how that needs to be

109
00:06:10,320 --> 00:06:14,957
protected.
um from government doing something

110
00:06:15,040 --> 00:06:18,157
excessively heavy-handed. >> So um
so David, you know, we're going to

111
00:06:18,240 --> 00:06:21,517
talk a lot about AI today, but just
on crypto, I've I've had this interesting

112
00:06:21,600 --> 00:06:24,417
experience this year kind of after the,
you know, election, you know, kind

113
00:06:24,500 --> 00:06:27,117
[snorts] of people adjusted to the
change uh of government. And I've had

114
00:06:27,200 --> 00:06:30,717
this discussion with, you know, a number
of people who, let's say, in in politics

115
00:06:30,800 --> 00:06:34,237
who were previously anti-crypto who have
been, you know, trying to figure out how

116
00:06:34,320 --> 00:06:37,277
to kind of get to a more sensible
position. Um and then also actually

117
00:06:37,360 --> 00:06:39,837
people in the in the financial services
industry who kind of followed it from a

118
00:06:39,920 --> 00:06:42,877
distance um and maybe you know
participated in the you know the various

119
00:06:42,960 --> 00:06:45,917
debanking things without really
understanding what was happening. Um but

120
00:06:46,000 --> 00:06:49,117
the the common denominator has been
they're like you know Mark I didn't

121
00:06:49,200 --> 00:06:51,917
really understand um how bad it was. Um,

122
00:06:52,000 --> 00:06:56,797
I basically thought you guys in tech
were basically just whining a lot. Um,

123
00:06:56,880 --> 00:06:59,757
and you know, pleading as a special
interest, um, and kind of doing the

124
00:06:59,840 --> 00:07:02,957
normal thing. And I figured, you know,
the the, you know, the horror stories

125
00:07:03,040 --> 00:07:06,397
were kind of made up, you know, of people
getting prosecuted and, you know,

126
00:07:06,480 --> 00:07:09,197
entrepreneurs getting their houses
raided by the FBI and, you know, like

127
00:07:09,280 --> 00:07:12,557
the whole paniply of things that,
you know, that happened and like I didn't

128
00:07:12,640 --> 00:07:15,917
really I I now in retrospect now that I
now that I go back and look, I'm like,

129
00:07:16,000 --> 00:07:20,077
"Oh my god, this was actually much worse
than I thought." Um, what do you do you

130
00:07:20,160 --> 00:07:23,277
do you have that experience? as as as
you're in there and kind of as you now

131
00:07:23,360 --> 00:07:26,237
have like a complete view of everything
that happened like do you think people

132
00:07:26,320 --> 00:07:28,237
understand actually how bad it was?

133
00:07:28,320 --> 00:07:31,117
>> I mean I think it's a a great point.
I mean I didn't really know either. You

134
00:07:31,200 --> 00:07:35,917
kind of heard um generally I mean we
knew that there was debanking going on

135
00:07:36,000 --> 00:07:38,877
and and by the way it wasn't just crypto
companies that were being debanked but

136
00:07:38,960 --> 00:07:42,957
their founders were being debanked
personally. So if you were the founder

137
00:07:43,040 --> 00:07:45,837
of a crypto company you couldn't open
a bank account. I mean that's a huge

138
00:07:45,920 --> 00:07:50,397
problem. It's like how do you how do you
transact? How do you make payments? um

139
00:07:50,480 --> 00:07:52,957
how do you pay people?
I mean it basically deprivives you of a

140
00:07:53,040 --> 00:07:56,637
livelihood. It's a very extreme form
of censorship. So that was definitely

141
00:07:56,720 --> 00:08:00,957
happening. And then of course you have
all the the prosecutions uh that the SEC

142
00:08:01,040 --> 00:08:02,877
was behind. So yeah, it was it was

143
00:08:02,960 --> 00:08:05,677
really bad. I remember um back in I

144
00:08:05,760 --> 00:08:08,077
think it was in March uh we had a crypto

145
00:08:08,160 --> 00:08:10,717
summit at the White House and um one of

146
00:08:10,800 --> 00:08:15,437
the attendees said that you know a year
ago uh I would have thought it was more

147
00:08:15,520 --> 00:08:19,917
likely that I'd be in jail than that I'd
be at the White House. Um and so it was,

148
00:08:20,000 --> 00:08:23,357
you know, it was a really big milestone
for for the industry. They had never

149
00:08:23,440 --> 00:08:28,477
ever received any kind of recognition
like that. The idea that this was even a

150
00:08:28,560 --> 00:08:30,237
industry that you would do an event at

151
00:08:30,320 --> 00:08:33,437
the White House. I mean um at a minimum

152
00:08:33,520 --> 00:08:36,157
I think crypto is seen as very day class

153
00:08:36,240 --> 00:08:40,957
A. Um but uh but in any event, yeah,
no, it's been a it's been a huge shift. I

154
00:08:41,040 --> 00:08:43,037
mean, we basically have have um have

155
00:08:43,120 --> 00:08:48,477
stopped that. And um and you know,
it was very unfair because again, these

156
00:08:48,560 --> 00:08:52,317
founders wanted to comply with the rules,
but they weren't told what they

157
00:08:52,400 --> 00:08:56,717
were. Um and that was all part of a
deliberate strategy, I think, to drive

158
00:08:56,800 --> 00:08:58,557
crypto offshore.

159
00:08:58,640 --> 00:09:02,317
>> Yeah. One of the things that um is very

160
00:09:02,400 --> 00:09:05,997
different between crypto and AI
that we've noticed is that on the crypto

161
00:09:06,080 --> 00:09:08,317
front, everybody just wanted rules. um

162
00:09:08,400 --> 00:09:11,437
and the industry was relatively unified.

163
00:09:11,520 --> 00:09:14,797
Whereas in AI, we've seen very like

164
00:09:14,880 --> 00:09:19,677
interesting kind of calls coming
from inside the house with uh certain

165
00:09:19,760 --> 00:09:23,357
companies really going for regulatory
capture. You know, people have early

166
00:09:23,440 --> 00:09:27,597
leads saying let's cut off all new
companies from developing AI and so

167
00:09:27,680 --> 00:09:29,917
forth. Um what do you make of that and

168
00:09:30,000 --> 00:09:32,077
and where do you think that's going?

169
00:09:32,160 --> 00:09:34,637
>> I think it's I think it's a very big

170
00:09:34,720 --> 00:09:39,117
problem. Um I actually recently
criticized uh one of our AI model

171
00:09:39,200 --> 00:09:42,237
companies for engaging in a
regulatory capture strategy.

172
00:09:42,320 --> 00:09:45,357
>> Yes. Um >> it's a very fair
criticism [laughter] by the way.

173
00:09:45,440 --> 00:09:48,397
>> It is very fair
and actually um you know

174
00:09:48,480 --> 00:09:51,117
of course they denied it and then uh

175
00:09:51,200 --> 00:09:52,397
should I tell the story? I mean

176
00:09:52,480 --> 00:09:55,757
>> yes rarely do you
get vindicated on on X

177
00:09:55,840 --> 00:10:00,157
so so thoroughly and completely as I
did on this because [laughter] after this

178
00:10:00,240 --> 00:10:04,077
company was basically anthropic after
they denied it. There was um I mean what

179
00:10:04,160 --> 00:10:07,277
basically happened is that Jack Clark
who's a co-founder and head of policy

180
00:10:07,360 --> 00:10:09,597
for anthropic gave a speech at a a

181
00:10:09,680 --> 00:10:13,917
conference where he said that uh he

182
00:10:14,000 --> 00:10:17,837
compared fear of AI to a child um you

183
00:10:17,920 --> 00:10:21,197
know seeing monsters in the dark or or
or thinking there were monsters in the

184
00:10:21,280 --> 00:10:23,037
dark and but then you turn the lights on

185
00:10:23,120 --> 00:10:25,117
and the monsters are there. I thought

186
00:10:25,200 --> 00:10:28,717
that was such a um ridiculous analogy. I

187
00:10:28,800 --> 00:10:32,957
mean it's basically purile. I mean,
it's it's so childish just to be almost

188
00:10:33,040 --> 00:10:37,757
self-indicting because you're basically
admitting the fear is made up, not real.

189
00:10:37,840 --> 00:10:41,117
>> Uh, in any event, so I said, well, this
is like fear-mongering and part of the

190
00:10:41,200 --> 00:10:44,877
regulatory capture strategy. And of
course, they denied it. But then um a

191
00:10:44,960 --> 00:10:48,237
lawyer who was in the crowd at his
speech said, "Well, yeah, but Jack's not

192
00:10:48,320 --> 00:10:50,477
telling you what he said during the Q&A,

193
00:10:50,560 --> 00:10:55,357
which he basically admitted that
everything that Anthropic was doing was

194
00:10:55,440 --> 00:10:59,677
uh you know, with with like things
like SB53, which is supposedly just

195
00:10:59,760 --> 00:11:03,357
implementing transparency." He said,
"No, that's all that he admitted that

196
00:11:03,440 --> 00:11:08,077
was just a stepping stone to their real
goal, which was to get an approval, a

197
00:11:08,160 --> 00:11:10,237
system of pre-approvals in Washington

198
00:11:10,320 --> 00:11:13,197
before you can release new models." and

199
00:11:13,280 --> 00:11:17,517
he admitted as part of the Q&A that
making people very afraid was part of

200
00:11:17,600 --> 00:11:19,597
their strategy. So again, just as much

201
00:11:19,680 --> 00:11:22,077
of a smoking gun as you could ever get

202
00:11:22,160 --> 00:11:24,957
um in a in a spaton X. But the reason

203
00:11:25,040 --> 00:11:26,717
why I think that [snorts] that approach

204
00:11:26,800 --> 00:11:31,117
is so damaging is that the thing that's

205
00:11:31,200 --> 00:11:35,277
uh really made I think Silicon Valley
special over the past several decades is

206
00:11:35,360 --> 00:11:39,997
permissionless innovation. Right?
It's that two guys in a garage can just

207
00:11:40,080 --> 00:11:44,477
pursue their idea. you know, maybe they
raise some capital from angels or VC

208
00:11:44,560 --> 00:11:48,717
firms. B basically people who are
willing to lose all of their money and

209
00:11:48,800 --> 00:11:51,837
these are people who, you know, are
young founders. They could also be, you

210
00:11:51,920 --> 00:11:55,117
know, it could be the the the dropout,
the future dropout in a dorm room and

211
00:11:55,200 --> 00:11:57,037
they're able just to pursue their idea.

212
00:11:57,120 --> 00:12:00,077
And um and the only reason that I think

213
00:12:00,160 --> 00:12:01,757
has happened in Silicon Valley whereas

214
00:12:01,840 --> 00:12:04,077
you look at in you know industries like

215
00:12:04,160 --> 00:12:09,517
um I don't know like pharma or um you
know healthcare or defense or banking or

216
00:12:09,600 --> 00:12:13,357
these highly regulated industries where
you just don't see a lot of startups is

217
00:12:13,440 --> 00:12:15,837
because they're all heavily regulated
which means you have to go to Washington

218
00:12:15,920 --> 00:12:20,317
to get permission to do things. And the
thing I've seen in Washington is just

219
00:12:20,400 --> 00:12:22,637
that, you know, the approvals get set up

220
00:12:22,720 --> 00:12:25,917
for reasons, but those reasons very

221
00:12:26,000 --> 00:12:30,157
quickly stop mattering. And it just
matters like how good your government

222
00:12:30,240 --> 00:12:33,597
affairs team is at navigating through
the bureaucracy and figuring out how to

223
00:12:33,680 --> 00:12:37,357
get those approvals. And it's not
something that your typical startup

224
00:12:37,440 --> 00:12:40,797
founders are going to be good at. Um,
it's something that big companies get

225
00:12:40,880 --> 00:12:43,117
good at because they've got the resources.
And that's exactly what

226
00:12:43,200 --> 00:12:45,597
regulatory capture means. So the the

227
00:12:45,680 --> 00:12:48,077
whole basis of Silicon Valley success,

228
00:12:48,160 --> 00:12:51,357
the reason why it's really the the crown

229
00:12:51,440 --> 00:12:55,837
jewel of the American economy and the
envy of the rest of the world. We see

230
00:12:55,920 --> 00:12:59,197
all these attempts by all these other
countries to create their own Silicon

231
00:12:59,280 --> 00:13:02,877
Valley. The the reason that's the case
is because of permissionless innovation.

232
00:13:02,960 --> 00:13:05,597
And what is being contemplated and

233
00:13:05,680 --> 00:13:09,997
discussed and implemented with respect
to AI is an approval system for both

234
00:13:10,080 --> 00:13:14,157
software and hardware. Uh and this is
not theoretical. This has already been

235
00:13:14,240 --> 00:13:18,317
happening on the the hardware side.
One of the last things that the Biden

236
00:13:18,400 --> 00:13:22,957
administration did the last week
the Biden administration was um impose the

237
00:13:23,040 --> 00:13:25,037
so-called Biden diffusion rule which

238
00:13:25,120 --> 00:13:29,597
requires that every sale of a GPU on

239
00:13:29,680 --> 00:13:32,477
Earth be licensed by the government

240
00:13:32,560 --> 00:13:34,637
which is to say pre-approved um unless

241
00:13:34,720 --> 00:13:37,197
it fits into some category of exception

242
00:13:37,280 --> 00:13:39,677
but basically the overall idea is that

243
00:13:39,760 --> 00:13:42,077
you know uh compute is now going to be

244
00:13:42,160 --> 00:13:44,877
um a licensed and and pre-approved um

245
00:13:44,960 --> 00:13:47,837
category. uh we rescended that. Um and

246
00:13:47,920 --> 00:13:49,677
then on the the software side, like I

247
00:13:49,760 --> 00:13:52,317
said, I mean the goal very clearly is to

248
00:13:52,400 --> 00:13:55,277
start with these reporting requirements

249
00:13:55,360 --> 00:13:57,917
to the government to the states and then

250
00:13:58,000 --> 00:14:02,397
where that ramps up to is you have
to go to Washington to get permission uh

251
00:14:02,480 --> 00:14:04,797
before you release a new model. And um

252
00:14:04,880 --> 00:14:06,557
you know it's this would drastically

253
00:14:06,640 --> 00:14:09,597
slow down innovation and make America

254
00:14:09,680 --> 00:14:12,237
less competitive. I mean, you know,

255
00:14:12,320 --> 00:14:16,637
these approvals can take months, they
can take years. Uh, when models are when

256
00:14:16,720 --> 00:14:19,597
when a new chip is released every year
and we have licenses that have been

257
00:14:19,680 --> 00:14:23,837
sitting in the hopper for 2 years,
I mean, the requests are obsolete by the

258
00:14:23,920 --> 00:14:27,517
time they finally get approved. And um,
that would be even more true with models

259
00:14:27,600 --> 00:14:31,197
where, you know, the the cycle time is,
you know, like three three or four

260
00:14:31,280 --> 00:14:35,197
months for a new model. I mean,
you know, and and and what exactly is a

261
00:14:35,280 --> 00:14:39,677
bureaucracy in Washington going to know
about uh this technology that, you know,

262
00:14:39,760 --> 00:14:42,957
that they're that they're going to be
in a good position to to approve it. Any

263
00:14:43,040 --> 00:14:45,997
event, but this is what is being
contemplated right now and I think it

264
00:14:46,080 --> 00:14:50,477
would be a disaster for for Silicon
Valley, but also and for innovation, but

265
00:14:50,560 --> 00:14:52,397
therefore for American competitiveness

266
00:14:52,480 --> 00:14:56,317
and I think we will lose the AI race uh

267
00:14:56,400 --> 00:15:00,237
to countries like China if you know if
this is if this is the set of rules that

268
00:15:00,320 --> 00:15:02,877
we have. >> Yeah.
One of the really diabolical

269
00:15:02,960 --> 00:15:05,437
things about their argument is if they

270
00:15:05,520 --> 00:15:10,317
really believe there was a monster, then
why are they buying GPUs like at a rate

271
00:15:10,400 --> 00:15:13,597
faster than anybody? And then the other
thing that we know from being in the

272
00:15:13,680 --> 00:15:16,797
industry is their reputation is the

273
00:15:16,880 --> 00:15:20,637
literally the worst security practices
in the entire industry with respect to

274
00:15:20,720 --> 00:15:23,997
their own code. So if you were building
this monster, the last thing you'd want

275
00:15:24,080 --> 00:15:27,997
to do is like leave a bunch of holes
around for people to hack it. Um, so

276
00:15:28,080 --> 00:15:31,757
they don't believe anything they're
saying. It's like completely made up to

277
00:15:31,840 --> 00:15:35,517
try and maintain their lead
by the really psychotic.

278
00:15:35,600 --> 00:15:37,917
>> I think there is um
I think it's a heady

279
00:15:38,000 --> 00:15:42,637
drug to basically say that um that you

280
00:15:42,720 --> 00:15:47,437
know we're we're creating this you know
new super intelligence that is going to

281
00:15:47,520 --> 00:15:51,117
could destroy humanity but we're
the only ones who are virt virtuous enough

282
00:15:51,200 --> 00:15:54,797
>> to ensure [laughter] >> that
this is done correctly. Right. And

283
00:15:54,880 --> 00:15:58,557
I think that you know >> it's
a good recruiting tool. Yeah. join

284
00:15:58,640 --> 00:16:02,157
the virtuous team. >> Yes.
Um I think that's right.

285
00:16:02,240 --> 00:16:04,557
>> The >> but yeah but I
think that is definitely

286
00:16:04,640 --> 00:16:08,957
um you know I I I think of of all
the companies that particular one has been

287
00:16:09,040 --> 00:16:12,477
the most aggressive uh in terms
of the regulatory capture

288
00:16:12,560 --> 00:16:15,197
and pushing these um for these these

289
00:16:15,280 --> 00:16:20,077
regulations and just I mean let's just
bring it up a level just doesn't have to

290
00:16:20,160 --> 00:16:22,157
be about them. Um there's now something

291
00:16:22,240 --> 00:16:24,397
like 1,200 bills going through state uh

292
00:16:24,480 --> 00:16:27,997
legislatures right now to regulate AI.

293
00:16:28,080 --> 00:16:31,517
Uh 25% of them are in um the the top

294
00:16:31,600 --> 00:16:33,837
four blue states which are California,

295
00:16:33,920 --> 00:16:38,797
New York, Colorado, and Illinois. Uh
over a 100 measures have already passed.

296
00:16:38,880 --> 00:16:43,517
Um I think three of them just got signed
in the last month in California alone.

297
00:16:43,600 --> 00:16:47,677
I'll tell you just let me tell you what
Colorado is actually Colorado, Illinois,

298
00:16:47,760 --> 00:16:52,317
and California have all done some
version of a thing called algorithmic

299
00:16:52,400 --> 00:16:56,557
discrimination. Um,
which I think is it's really

300
00:16:56,640 --> 00:16:58,557
troubling where where it's headed. Um,

301
00:16:58,640 --> 00:17:00,877
what what this concept means is that if

302
00:17:00,960 --> 00:17:04,877
the model produces an output

303
00:17:04,960 --> 00:17:08,717
that has a disparate impact
on a protected group, then that is

304
00:17:08,800 --> 00:17:11,357
algorithmic discrimination. And the list

305
00:17:11,440 --> 00:17:14,317
of protected groups is um is very long.

306
00:17:14,400 --> 00:17:16,237
It's more than just the the the usual

307
00:17:16,320 --> 00:17:18,797
ones. So for example, in Colorado,

308
00:17:18,880 --> 00:17:22,557
they've defined um people who um may not

309
00:17:22,640 --> 00:17:25,837
have English language proficiency as a

310
00:17:25,920 --> 00:17:29,757
protected group. So I guess if the model
says something bad about, you know,

311
00:17:29,840 --> 00:17:33,197
illegal aliens and that would be,
you know, that would basically violate the

312
00:17:33,280 --> 00:17:36,557
law. I don't know exactly how model

313
00:17:36,640 --> 00:17:41,277
companies are even supposed to comply
with this rule. I mean presumably

314
00:17:41,360 --> 00:17:44,157
presumably discrimination is already
illegal. So if you're a business and you

315
00:17:44,240 --> 00:17:46,957
violate the civil rights laws and you
engage in discrimination, you're already

316
00:17:47,040 --> 00:17:48,557
liable for that. You know, there's no

317
00:17:48,640 --> 00:17:51,517
reason, you know, if if you happen to um

318
00:17:51,600 --> 00:17:56,237
you know make that mistake and you use
any kind of tool in the process of doing

319
00:17:56,320 --> 00:17:59,437
it, we don't really need to go after the
tool developer because we can already go

320
00:17:59,520 --> 00:18:02,237
after the business that's made that

321
00:18:02,320 --> 00:18:04,637
decision. But the whole purpose of of

322
00:18:04,720 --> 00:18:07,837
these laws is to get at the tool they're

323
00:18:07,920 --> 00:18:10,797
making. not just the the business that

324
00:18:10,880 --> 00:18:15,597
is using AI liable, they're making the
tool developer liable. And I don't even

325
00:18:15,680 --> 00:18:18,797
know how the tool developer is supposed
to anticipate this because how do you

326
00:18:18,880 --> 00:18:22,297
know all the ways that your tool
is going to be used? How do you

327
00:18:22,380 --> 00:18:26,157
[clears throat] know that um that this
output, you know, especially if the

328
00:18:26,240 --> 00:18:28,957
output is 100% true and accurate and the

329
00:18:29,040 --> 00:18:32,477
model is doing its job? um you know then

330
00:18:32,560 --> 00:18:36,877
how are you supposed to know that that
was that output was used as part of a

331
00:18:36,960 --> 00:18:39,277
decision that had a disperate impact?

332
00:18:39,360 --> 00:18:43,117
Nevertheless, you're liable.
And the only way that I can see for model

333
00:18:43,200 --> 00:18:44,557
developers to even attempt to comply

334
00:18:44,640 --> 00:18:47,837
with this is to build a DEI layer into

335
00:18:47,920 --> 00:18:52,957
their models that tries to anticipate
could this answer have a disparit

336
00:18:53,040 --> 00:18:54,477
impact? And if it does, we either can't

337
00:18:54,560 --> 00:18:56,637
give you the answer. We have to sanitize

338
00:18:56,720 --> 00:19:00,557
or distort the answer. Uh, and you know,

339
00:19:00,640 --> 00:19:04,157
you just take this to its logical
conclusion and we're back to, you know,

340
00:19:04,240 --> 00:19:07,757
uh, woke AI, which by the way
was a major objective of the Biden

341
00:19:07,840 --> 00:19:12,477
administration. Um, that that Biden
executive order on AI that we rescended

342
00:19:12,560 --> 00:19:15,997
as part of the Trump administration had
something like 20 pages of DEI language

343
00:19:16,080 --> 00:19:18,557
in it. They were very
much trying to promote

344
00:19:18,640 --> 00:19:22,157
>> uh, DEI values,
they called it, in models

345
00:19:22,240 --> 00:19:25,197
>> and then we saw what the result of that
was. you know, uh, it was, you know, we

346
00:19:25,280 --> 00:19:28,797
saw the whole black George Washington
thing where history was being rewritten

347
00:19:28,880 --> 00:19:31,437
in real time because somebody built, you

348
00:19:31,520 --> 00:19:34,317
know, a DEI layer into the model. Um,

349
00:19:34,400 --> 00:19:39,117
and, um, and I, you know, and I,
I almost feel like the term woke AI is

350
00:19:39,200 --> 00:19:42,317
insufficient >> to explain
what's going on because it

351
00:19:42,400 --> 00:19:45,357
somehow trivializes it. I mean, what
we're really talking about is Orwellian

352
00:19:45,440 --> 00:19:47,917
AI. You know, we're talking about AI

353
00:19:48,000 --> 00:19:52,477
that um that lies to you, that um that

354
00:19:52,560 --> 00:19:55,517
distorts an an answer that rewrites

355
00:19:55,600 --> 00:19:59,677
history in real time to serve a current
political agenda of the people who are

356
00:19:59,760 --> 00:20:02,157
in power. I mean, it's it's um it's very

357
00:20:02,240 --> 00:20:04,957
Orwellian and um and we were definitely

358
00:20:05,040 --> 00:20:07,437
on that path before President Trump's

359
00:20:07,520 --> 00:20:09,517
election. Uh it was part of the Biden

360
00:20:09,600 --> 00:20:14,717
EO. We saw it happen, you know, in in
the release of that first Gemini model.

361
00:20:14,800 --> 00:20:16,477
uh that was not an accident that you

362
00:20:16,560 --> 00:20:18,797
know that th those distorted outputs

363
00:20:18,880 --> 00:20:22,717
came from somewhere. Um so it was you

364
00:20:22,800 --> 00:20:26,397
know to just to me this is the the

365
00:20:26,480 --> 00:20:30,317
biggest risk of AI actually is it's not

366
00:20:30,400 --> 00:20:33,997
it was not described by James Cameron it
was described by George Orwell you know

367
00:20:34,080 --> 00:20:38,157
it's in my view it's not
the Terminator it's 1984

368
00:20:38,240 --> 00:20:40,877
that um you know that as AI eats the

369
00:20:40,960 --> 00:20:43,037
internet and becomes the main way that

370
00:20:43,120 --> 00:20:46,237
we interact and get our information

371
00:20:46,320 --> 00:20:49,517
online that it'll be used by the people

372
00:20:49,600 --> 00:20:52,317
in power to um to control the

373
00:20:52,400 --> 00:20:54,317
information we receive. Um that it'll

374
00:20:54,400 --> 00:20:56,957
contain an ideological bias that

375
00:20:57,040 --> 00:21:01,917
essentially it'll censor us. All
that trust and safety apparatus that was

376
00:21:02,000 --> 00:21:05,197
created for social media will be ported
over to this new world of AI. Mark, I

377
00:21:05,280 --> 00:21:07,757
know that you've spoken about this quite
a bit. I think you're absolutely right

378
00:21:07,840 --> 00:21:11,917
about that. And then on top of that,
you've got the surveillance um issues

379
00:21:12,000 --> 00:21:15,597
where, you know, AI is going to know
everything about you. Uh it's going to

380
00:21:15,680 --> 00:21:17,837
be your kind of personal assistant. And

381
00:21:17,920 --> 00:21:21,357
so it's kind of the perfect tool for the

382
00:21:21,440 --> 00:21:24,237
government to um monitor and control

383
00:21:24,320 --> 00:21:28,877
you. And to me that is by far
the biggest risk of of AI. And that's the

384
00:21:28,960 --> 00:21:32,717
thing we should be working towards
preventing. And the problem is a lot of

385
00:21:32,800 --> 00:21:35,517
these regulations that are being whipped

386
00:21:35,600 --> 00:21:38,237
up by these um fear-mongering

387
00:21:38,320 --> 00:21:40,077
techniques. They're actually empowering

388
00:21:40,160 --> 00:21:43,277
the government to uh to engage in this

389
00:21:43,360 --> 00:21:45,997
type of control that I think we should

390
00:21:46,080 --> 00:21:50,637
all be very afraid of. actually is Sam
Alman earlier this week said that in

391
00:21:50,720 --> 00:21:52,877
2028 or by 2028 he expects to have

392
00:21:52,960 --> 00:21:55,837
automated researchers. I'm I'm curious

393
00:21:55,920 --> 00:21:59,997
just for your sort of state of of AI
sort of model development or just

394
00:22:00,080 --> 00:22:03,917
progress in general and what do you
think are the implications? Some people

395
00:22:04,000 --> 00:22:06,797
have been sort of you know saying that

396
00:22:06,880 --> 00:22:11,677
you know AGI is two years away sort
of the AI 2027 papers the old Ashen Brener

397
00:22:11,760 --> 00:22:14,717
situational awareness papers.
I'm curious kind of what's your reading of

398
00:22:14,800 --> 00:22:17,757
the of the state of play in terms
of a AI development and what what are the

399
00:22:17,840 --> 00:22:22,157
implications from that? >> So so
my sense is that um people are in

400
00:22:22,240 --> 00:22:25,357
Silicon Valley are kind of pulling
back from the let's call it imminent AGI

401
00:22:25,440 --> 00:22:28,557
narrative. Um I saw Andre Karpathy gave

402
00:22:28,640 --> 00:22:33,837
an interview where all of a sudden he's
>> he's um rewritten this and he says AGI

403
00:22:33,920 --> 00:22:37,917
is at least a decade away. He's
basically saying that um that you know

404
00:22:38,000 --> 00:22:41,357
reinforcement learning has its limits.
I mean it's it's very useful. It's the

405
00:22:41,440 --> 00:22:44,717
main paradigm right now that they're
making a lot of progress with. But he

406
00:22:44,800 --> 00:22:48,077
says that actually the way that
humans learn is not really through

407
00:22:48,160 --> 00:22:51,437
reinforcement. We do something a little
different, which I think is a good thing

408
00:22:51,520 --> 00:22:55,757
because it means that human and AI will
be synergistic, right? I mean that the

409
00:22:55,840 --> 00:22:59,517
AI's um understanding if it's based
on RL will be a little different than the

410
00:22:59,600 --> 00:23:02,717
way that we intuit it and and reason. Uh

411
00:23:02,800 --> 00:23:06,717
but in any event, I I sense more
of a pullback from this imminent AGI

412
00:23:06,800 --> 00:23:10,877
narrative. um you know the idea that AGI
is two years away. Of course it's like

413
00:23:10,960 --> 00:23:14,957
kind of unclear what people mean by AGI
but it's kind of you know was used in

414
00:23:15,040 --> 00:23:18,957
this like scary way that um it's kind
of this super intelligence that would grow

415
00:23:19,040 --> 00:23:23,997
beyond our control. Um I feel like
people are pulling back from that and

416
00:23:24,080 --> 00:23:25,837
understanding that yes we're still

417
00:23:25,920 --> 00:23:28,077
making a lot of progress and the

418
00:23:28,160 --> 00:23:31,597
progress is amazing but at the same time

419
00:23:31,680 --> 00:23:35,277
you know what we mean
by intelligence is multifaceted

420
00:23:35,360 --> 00:23:38,477
and it's not like you know there's
progress being made along some

421
00:23:38,560 --> 00:23:40,557
dimensions but it's not along every

422
00:23:40,640 --> 00:23:42,077
dimension.

423
00:23:42,160 --> 00:23:46,077
Um, and so therefore I think again I

424
00:23:46,160 --> 00:23:49,117
would just I mean I've described
actually the situation we're in right

425
00:23:49,200 --> 00:23:53,357
now as a little bit of a Goldilock
scenario where um you know the extremes

426
00:23:53,440 --> 00:23:58,877
would be you know you kind of have the
the scary Terminator situation imminent

427
00:23:58,960 --> 00:24:02,237
super intelligence that'll grow beyond
our control on the other the other

428
00:24:02,320 --> 00:24:05,277
narrative you hear in the press a lot
is that we're in a big bubble. So in other

429
00:24:05,360 --> 00:24:09,037
words the whole thing is fake >>
and the media is basically pushing both

430
00:24:09,120 --> 00:24:12,797
narratives at the same time. >> Yeah.
Um, but any event, I think that

431
00:24:12,880 --> 00:24:14,237
the truth is is more in the middle.

432
00:24:14,320 --> 00:24:15,677
That's kind of a Goldilock scenario

433
00:24:15,760 --> 00:24:17,757
where um we're seeing a lot of

434
00:24:17,840 --> 00:24:21,437
innovation. I think the um the progress

435
00:24:21,520 --> 00:24:24,557
is impressive. Uh it's I think we're

436
00:24:24,640 --> 00:24:29,517
going to see big productivity gains in
the economy from this. But uh but I like

437
00:24:29,600 --> 00:24:32,557
the observations that Bology made
recently where he said there's a couple

438
00:24:32,640 --> 00:24:34,797
of things that really struck me. One was

439
00:24:34,880 --> 00:24:38,077
um AI is polytheistic, not monotheistic.

440
00:24:38,160 --> 00:24:42,877
Meaning what we're seeing is many
instead of just one all- knowing all

441
00:24:42,960 --> 00:24:47,517
powerful god. What we're seeing
is a bunch of smaller deities more

442
00:24:47,600 --> 00:24:49,677
specialized models

443
00:24:49,760 --> 00:24:52,797
um you know uh you know it's it's not

444
00:24:52,880 --> 00:24:57,037
that sort of um we're not on that
kind of recursive self-improvement

445
00:24:57,120 --> 00:25:00,797
uh track just yet but you know we're
seeing many different kinds of models

446
00:25:00,880 --> 00:25:03,117
make progress in different areas. Um and

447
00:25:03,200 --> 00:25:08,237
then the other the other um one was just
his observation that AI was um was

448
00:25:08,320 --> 00:25:11,677
middle to middle whereas humans are end
to end and therefore the relationship is

449
00:25:11,760 --> 00:25:16,077
is pretty synergistic and um I think
that's right. I mean I think I think all

450
00:25:16,160 --> 00:25:19,997
th those observations are like resonate
with me in terms of where we're at right

451
00:25:20,080 --> 00:25:23,277
now. >> Yeah. And that that that's very
consistent with what we're seeing as

452
00:25:23,360 --> 00:25:26,477
well where um you know ideas that we

453
00:25:26,560 --> 00:25:29,517
thought would for sure get subsumed by

454
00:25:29,600 --> 00:25:33,677
the big models are becoming amazingly

455
00:25:33,760 --> 00:25:36,157
differentiated businesses. Um just

456
00:25:36,240 --> 00:25:40,077
because the fat tail of the universe is

457
00:25:40,160 --> 00:25:43,757
very fat and you need really kind of

458
00:25:43,840 --> 00:25:48,077
specific understanding of certain
scenarios to build an effective model

459
00:25:48,160 --> 00:25:52,077
and that that's just how it's going. you
know, no model has just like figured out

460
00:25:52,160 --> 00:25:55,517
how to do everything. >> Yeah.
I mean, and the models work best

461
00:25:55,600 --> 00:25:58,717
when they have context, you know, and um

462
00:25:58,800 --> 00:26:02,557
the more I mean, we've all seen this,
the more general your prompt, the less

463
00:26:02,640 --> 00:26:04,797
likely it is that you're going to be

464
00:26:04,880 --> 00:26:07,597
able to um you know, get get a great

465
00:26:07,680 --> 00:26:10,717
response. And um I don't know, if you

466
00:26:10,800 --> 00:26:14,157
tell the AI uh you know, uh something

467
00:26:14,240 --> 00:26:19,837
very general like um what business can
I create to make a billion dollars? it

468
00:26:19,920 --> 00:26:23,117
it's not going to give you something
actionable. You know, you have to get

469
00:26:23,200 --> 00:26:27,197
very specific about what you're trying
to do and it has to have access to

470
00:26:27,280 --> 00:26:28,637
relevant data and then it can give you

471
00:26:28,720 --> 00:26:32,397
some specific answers to to um to a

472
00:26:32,480 --> 00:26:34,477
prompt. Um and I think this is, you

473
00:26:34,560 --> 00:26:39,597
know, partly Bali's point, which is,
you know, the the the AI does not come up

474
00:26:39,680 --> 00:26:42,317
with its own objective. You know,
it needs to be prompted. It needs to be

475
00:26:42,400 --> 00:26:44,237
told what to do. We've seen no evidence

476
00:26:44,320 --> 00:26:46,717
that that's at this stage that that's

477
00:26:46,800 --> 00:26:49,517
changing. we're still at step zero in

478
00:26:49,600 --> 00:26:53,277
terms of AI kind of, you know, somehow

479
00:26:53,360 --> 00:26:58,077
coming up with its own objective. And um
uh and as a result of that, you know,

480
00:26:58,160 --> 00:27:00,957
the model has to be prompted and then it
gives you an output and that output has

481
00:27:01,040 --> 00:27:04,797
to be validated. You have to somehow
make sure it's correct because models

482
00:27:04,880 --> 00:27:09,597
can still be wrong. Um and more likely
you have to iterate a few times because

483
00:27:09,680 --> 00:27:13,197
it doesn't give you exactly what you want.
So now you kind of reprompt and

484
00:27:13,280 --> 00:27:14,397
we've all had this experience, right?

485
00:27:14,480 --> 00:27:17,117
you this is why like the chat the chat

486
00:27:17,200 --> 00:27:21,837
um interface is so is so necessary is
because it takes you a few times to kind

487
00:27:21,920 --> 00:27:26,477
of iterate to get to the output that
actually has value for you. Um again you

488
00:27:26,560 --> 00:27:30,237
know the humans are antenn and the the um
the AI is middle to middle. I just

489
00:27:30,320 --> 00:27:33,757
don't you know we haven't seen any
evidence that that fundamental dynamic

490
00:27:33,840 --> 00:27:38,557
is is changing. Um I mean I think you
know we're at the I mean I'd love to

491
00:27:38,640 --> 00:27:39,517
hear what you guys think about this.

492
00:27:39,600 --> 00:27:44,317
this. I mean, we're obviously at the
outset of of agents and um you know, an

493
00:27:44,400 --> 00:27:47,037
agents you can give an objective to and
then they'll be able to take tasks on

494
00:27:47,120 --> 00:27:50,797
your behalf, but I suspect that the
agents will work better as well when

495
00:27:50,880 --> 00:27:52,957
they have a much more narrow context.

496
00:27:53,040 --> 00:27:54,877
They're much less likely to go off the

497
00:27:54,960 --> 00:27:57,357
rails, start going in weird directions.

498
00:27:57,440 --> 00:28:02,317
If you give it like a very broad,
you know, a very broad task, it's just not

499
00:28:02,400 --> 00:28:06,157
likely to completely figure it out
before it needs human intervention. But

500
00:28:06,240 --> 00:28:09,757
if you give it something very narrow
to do, then it's much more likely to be

501
00:28:09,840 --> 00:28:12,957
successful. Um, so, you know, I would

502
00:28:13,040 --> 00:28:16,557
just guess like, okay, just you tell the

503
00:28:16,640 --> 00:28:20,317
AI, um, you know, sell my product. You

504
00:28:20,400 --> 00:28:23,757
know, it's it's very unlikely that it's
just going to figure out like what that

505
00:28:23,840 --> 00:28:26,237
means and how to do that. But if you're

506
00:28:26,320 --> 00:28:30,317
a sales rep and you're using the AI to

507
00:28:30,400 --> 00:28:33,677
help you, there's probably very specific
tasks that you can tell it to do and it

508
00:28:33,760 --> 00:28:37,677
would be much more successful doing that.
So I I just tend to think I mean

509
00:28:37,760 --> 00:28:40,957
this also kind of speaks to the whole
job loss narrative. I just think that

510
00:28:41,040 --> 00:28:43,037
this is going to be a very synergistic

511
00:28:43,120 --> 00:28:45,757
tool for a long time. I don't think it's

512
00:28:45,840 --> 00:28:49,197
going to wipe out um human human jobs. I

513
00:28:49,280 --> 00:28:52,397
don't think the need for human cognition

514
00:28:52,480 --> 00:28:56,877
is going away. Um it's something that you
know we'll we'll all use to kind of

515
00:28:56,960 --> 00:29:01,437
get this big productivity boost at least
for the foreseeable future. I mean, I

516
00:29:01,520 --> 00:29:03,677
don't know. I don't know if anyone can
any of us can predict what's going to

517
00:29:03,760 --> 00:29:07,037
happen beyond five or 10 years, but I
mean, that's just what I'm seeing right

518
00:29:07,120 --> 00:29:11,197
now. I don't know. I'm curious what
what are you guys seeing this front?

519
00:29:11,280 --> 00:29:14,157
>> Yeah, generally consistent with that,
things are improving. So, like on

520
00:29:14,240 --> 00:29:18,397
agents, um the the early agents,
the longer the running task, the more they

521
00:29:18,480 --> 00:29:23,437
would go like completely bananas and off
off the rails. People are working on

522
00:29:23,520 --> 00:29:27,037
that. I do think like
everything's working better

523
00:29:27,120 --> 00:29:30,077
in a context. I at least from what we've

524
00:29:30,160 --> 00:29:33,197
seen that will continue and even you

525
00:29:33,280 --> 00:29:35,357
know to your point on like super smart

526
00:29:35,440 --> 00:29:39,357
models there's like a dozen video models

527
00:29:39,440 --> 00:29:43,037
out there and there's not one that's the

528
00:29:43,120 --> 00:29:46,797
best at everything or even close
to the best at everything. There's like

529
00:29:46,880 --> 00:29:49,517
literally a dozen that are all the best

530
00:29:49,600 --> 00:29:53,517
at one thing. Um, and

531
00:29:53,600 --> 00:29:57,517
which is a little surprising at least
to me because you would think, you know,

532
00:29:57,600 --> 00:29:59,757
just the sheer size of the data would be

533
00:29:59,840 --> 00:30:04,157
an advantage, but even that um hasn't

534
00:30:04,240 --> 00:30:07,597
quite proven out. Uh, you know, it it is

535
00:30:07,680 --> 00:30:09,197
like depending on what you want. Do you

536
00:30:09,280 --> 00:30:12,237
want a meme? Do you want a movie? Do you

537
00:30:12,320 --> 00:30:15,997
want, you know, an ad? Like it's all

538
00:30:16,080 --> 00:30:20,877
very very different. And I think
this gets to your main point which is and

539
00:30:20,960 --> 00:30:23,997
Mark Zuckerberg said something that I
really liked. He's like intelligence is

540
00:30:24,080 --> 00:30:27,037
not life. Um and these things that we

541
00:30:27,120 --> 00:30:28,877
associate

542
00:30:28,960 --> 00:30:33,597
with life um like we have an objective.

543
00:30:33,680 --> 00:30:37,197
Um we have free will. We're sentient. uh

544
00:30:37,280 --> 00:30:42,477
th those just aren't part of a
mathematical model that is you know

545
00:30:42,560 --> 00:30:46,157
searching through a distribution and
figuring out an answer or even you know

546
00:30:46,240 --> 00:30:48,877
a model that you know through a

547
00:30:48,960 --> 00:30:50,717
reinforcement learning technique can

548
00:30:50,800 --> 00:30:54,397
kind of improve its logic. So it's just

549
00:30:54,480 --> 00:30:58,557
like the the the comparison
to humans I think is it just

550
00:30:58,640 --> 00:31:02,637
falls short in a lot of ways is what
we're seeing you know we're just different

551
00:31:02,720 --> 00:31:06,317
>> you know then the models are very good
at things they're better than humans at

552
00:31:06,400 --> 00:31:09,997
things already many things >>
the other thing I'd bring up related to

553
00:31:10,080 --> 00:31:14,557
this which is sort of this which I think
is a little orthogonal but also quite

554
00:31:14,640 --> 00:31:18,637
related which is sort of is is the future
of the world going to be one or a

555
00:31:18,720 --> 00:31:21,837
small number of companies or for that
matter governments or super AIs that

556
00:31:21,920 --> 00:31:25,197
kind of own and control everything. Um,
and sort of all the value rolls up into

557
00:31:25,280 --> 00:31:27,997
a handful of of entities and and you
know, and there you get into this

558
00:31:28,080 --> 00:31:30,237
there's like the hyper capitalist
version of it where a few companies make

559
00:31:30,320 --> 00:31:32,957
all the money or there's like the hyper
communist version of it where you have

560
00:31:33,040 --> 00:31:34,877
total state control or whatever.

561
00:31:34,960 --> 00:31:38,957
>> Um, you know, or is this a technology
that's going to diffuse out um and be

562
00:31:39,040 --> 00:31:42,797
like in everybody's hands um and be
a tool of empowerment and creativity um

563
00:31:42,880 --> 00:31:46,477
and individual effort, you know,
expressiveness and um and um and as a

564
00:31:46,560 --> 00:31:49,357
tool for basically everybody to use.
I think one of the really striking things

565
00:31:49,440 --> 00:31:52,877
about this period of time and you being
in this role is that this this is the

566
00:31:52,960 --> 00:31:57,117
period in which the scenario number
two is I think very clearly playing out

567
00:31:57,200 --> 00:32:01,197
which is that this is the time in which
AI is actually hyperdemocratizing.

568
00:32:01,280 --> 00:32:04,557
I think actually AI is actually
hyperdemocratized. It has spread to more

569
00:32:04,640 --> 00:32:07,917
individuals um uh both in the country
and around the world uh in the shortest

570
00:32:08,000 --> 00:32:10,637
period of time than any new technology
I think in history. you know, we're we're

571
00:32:10,720 --> 00:32:13,837
it's something like 600 million,
you know, users today on on rapidly on the

572
00:32:13,920 --> 00:32:17,037
way to a billion rapidly on the way
to to 5 billion, you know, kind of across

573
00:32:17,120 --> 00:32:20,317
all the all the consumer products. And
then and then the best AIS in the world

574
00:32:20,400 --> 00:32:21,997
are are in the consumer products, right?

575
00:32:22,080 --> 00:32:25,757
Um and so if you use the if you use,
you know, current day Chad GPT or Grock or

576
00:32:25,840 --> 00:32:29,437
any of these things like, you know,
I I I can't spend more money, you know, uh

577
00:32:29,520 --> 00:32:32,637
and uh and and get access to better AI,
it's it's it's in the consumer products.

578
00:32:32,720 --> 00:32:36,477
And so just in practice, what you have
like playing out in real time is this

579
00:32:36,560 --> 00:32:39,757
technology is going to be in everybody's
hands. Um, and everybody is going to be

580
00:32:39,840 --> 00:32:42,237
able to use it to optim, you know,
optimize the things that they do. Have

581
00:32:42,320 --> 00:32:45,757
it be a thought partner. Have it be, you
know, somebody, you know, a a a um, you

582
00:32:45,840 --> 00:32:48,797
know, an assistant for for for building
companies, you know, starting companies

583
00:32:48,880 --> 00:32:52,717
or creating art or, you know, doing all
the things that people want to do. Um,

584
00:32:52,800 --> 00:32:55,837
you know, my wife was just using it
this morning to design a new uh an

585
00:32:55,920 --> 00:32:58,697
entrepreneurship curriculum for our
10-year-old, right? Um, you know,

586
00:32:58,780 --> 00:33:01,037
[laughter] right? Like,
you know, like literally

587
00:33:01,120 --> 00:33:04,077
it's like, "Oh, wow. That's like
a really great idea." Um, and it took her

588
00:33:04,160 --> 00:33:06,877
a couple hours and she has like a full
curriculum for him to be able to start

589
00:33:06,960 --> 00:33:09,437
his first video game company and here's
all the different skills that he needs

590
00:33:09,520 --> 00:33:12,397
to learn and here's all the resources
and and like that's just a level of

591
00:33:12,480 --> 00:33:16,157
capability. I mean to to to you know
to have done that without without these

592
00:33:16,240 --> 00:33:19,517
modern consumer AI tools you know you'd
have to go like hire a you know special

593
00:33:19,600 --> 00:33:21,997
ed education specialist or something
[snorts] you know which is basically

594
00:33:22,080 --> 00:33:23,517
impossible um to do that kind of thing.

595
00:33:23,600 --> 00:33:26,317
So and and you know everybody has these
stories now in their lives among people

596
00:33:26,400 --> 00:33:30,477
they know. So we're we have the I think
we have like a lot of proof that the the

597
00:33:30,560 --> 00:33:32,957
track that this on is that this is going
to be in everybody's hands and in fact

598
00:33:33,040 --> 00:33:35,997
that is going to be a really good thing
and and and I think and I think David I

599
00:33:36,080 --> 00:33:39,357
think you guys are really playing
a key role in making making it happen.

600
00:33:39,440 --> 00:33:42,157
>> I think it's so important that this um

601
00:33:42,240 --> 00:33:44,717
technology remain decentralized because

602
00:33:44,800 --> 00:33:49,277
the kind of the Orwellian concern
is kind of the ultimate centralization

603
00:33:49,360 --> 00:33:52,877
>> and fortunately so far what we're
seeing in the market is that it's hyper

604
00:33:52,960 --> 00:33:55,037
competitive. There's five major model

605
00:33:55,120 --> 00:33:58,237
companies all making huge investments.

606
00:33:58,320 --> 00:34:02,717
Um, and the uh the the benchmark the the

607
00:34:02,800 --> 00:34:07,437
the model performance the evaluations
are relatively clustered and there's a

608
00:34:07,520 --> 00:34:11,437
lot of leaprogging going on. So,
you know, Grock releases a new model, it

609
00:34:11,520 --> 00:34:15,517
leaprogs CH GBT and but then Chad GBT
releases something new, they they leap

610
00:34:15,600 --> 00:34:18,237
frog. So, they're all like very
competitive and close to each other and

611
00:34:18,320 --> 00:34:20,877
I think that's uh that's a good thing.

612
00:34:20,960 --> 00:34:22,717
And it's the opposite of what was

613
00:34:22,800 --> 00:34:25,117
predicted, you know, through this like

614
00:34:25,200 --> 00:34:27,917
imminent AGI story where the the the

615
00:34:28,000 --> 00:34:30,317
sort of the the storytelling there was

616
00:34:30,400 --> 00:34:34,477
that one model would get a lead and then

617
00:34:34,560 --> 00:34:38,317
it would direct its own intelligence
to making itself better and then so

618
00:34:38,400 --> 00:34:40,877
therefore its lead would get bigger
and bigger and you kind of get this

619
00:34:40,960 --> 00:34:44,957
recursive self-improvement and pretty
soon you're off to the the singularity.

620
00:34:45,040 --> 00:34:48,637
And um and and we haven't really seen
that, you know, we haven't seen one

621
00:34:48,720 --> 00:34:53,837
model completely pull away in terms
of capabilities. Um and I think that's a a

622
00:34:53,920 --> 00:34:57,037
good thing. And so Eric to your point
about this narrative about the virtual

623
00:34:57,120 --> 00:34:59,917
AI researcher that that was one variant

624
00:35:00,000 --> 00:35:02,397
of this sort of um imminent AGI

625
00:35:02,480 --> 00:35:06,157
narrative is that the steps would be um

626
00:35:06,240 --> 00:35:11,357
you know models get smarter then models
create virtual AI researcher and then

627
00:35:11,440 --> 00:35:14,397
you get [snorts] a million virtual
AI researchers and then you know it's

628
00:35:14,480 --> 00:35:16,557
singularity and I think just the slight

629
00:35:16,640 --> 00:35:19,757
of hand in that is um what is a virtual

630
00:35:19,840 --> 00:35:21,997
AI researcher right it's like a very

631
00:35:22,080 --> 00:35:24,957
easy thing to say, but like what does

632
00:35:25,040 --> 00:35:30,077
that really mean? And you know,
top toy's point about um you know, AI is

633
00:35:30,160 --> 00:35:34,397
still middle to middle. It doesn't it's
not end to end. So if an AI researcher

634
00:35:34,480 --> 00:35:38,237
is endtoend, there's like things
that has to you things the person has to

635
00:35:38,320 --> 00:35:40,637
figure out. They've got to set their own
objective. They've got to be able to

636
00:35:40,720 --> 00:35:44,077
pivot in ways that AI can't. Um you

637
00:35:44,160 --> 00:35:47,837
know, like is it really is that really
feasible to create a virtual AI

638
00:35:47,920 --> 00:35:52,317
researcher? I think there's like parts
of the job that you know AI could get

639
00:35:52,400 --> 00:35:53,997
really good at or even better than

640
00:35:54,080 --> 00:35:57,117
humans, but probably that tool has to be

641
00:35:57,200 --> 00:36:00,717
used by a human AI researcher. So I

642
00:36:00,800 --> 00:36:03,917
guess the the the the argument I suspect

643
00:36:04,000 --> 00:36:07,357
could be um sort of teological in the

644
00:36:07,440 --> 00:36:10,077
sense that you might need AGI to create

645
00:36:10,160 --> 00:36:14,397
a virtual AI researcher as opposed
to the other way around. And if that's the

646
00:36:14,480 --> 00:36:18,397
case, you don't just, you know, you're
not going to get like singularity. So,

647
00:36:18,480 --> 00:36:22,877
I'm a little bit skeptical of that claim.
Um, you know, we'll see. Sam,

648
00:36:22,960 --> 00:36:27,517
would Sam say he could do it in 2028?
I mean, I guess we'll see. And

649
00:36:27,600 --> 00:36:30,957
>> I I think all those claims tend
to be like recruiting ideas as [laughter]

650
00:36:31,040 --> 00:36:32,877
opposed to actual predictions.

651
00:36:32,960 --> 00:36:36,317
>> He's he's not the first to to mention
that idea. Other other model companies

652
00:36:36,400 --> 00:36:40,717
have been promoting it, but um yeah,
Leopold mentioned that too. um you know,

653
00:36:40,800 --> 00:36:42,637
we'll we'll we'll see. But I suspect

654
00:36:42,720 --> 00:36:45,037
that that's what's wrong with that

655
00:36:45,120 --> 00:36:49,117
argument is that virtual AI researcher

656
00:36:49,200 --> 00:36:52,397
um requires AGI and so the idea that

657
00:36:52,480 --> 00:36:56,877
you're going to get AGI through
a virtual AI researcher is um is

658
00:36:56,960 --> 00:36:59,277
backwards, but we'll see,
you know, we'll see.

659
00:36:59,360 --> 00:37:01,357
>> David, you've also you and the
administration, I think, have also been

660
00:37:01,440 --> 00:37:04,477
very supportive of open source AI, which
I think also dovtales into this in terms

661
00:37:04,560 --> 00:37:06,397
of like the market being very competitive.

662
00:37:06,480 --> 00:37:09,197
>> Yes. Yes. Do you want to spend a moment
on what you guys have been able to do on

663
00:37:09,280 --> 00:37:11,277
that and how you think about it?

664
00:37:11,360 --> 00:37:15,277
>> Yeah, I mean so so just to the you open
source is very important because I just

665
00:37:15,360 --> 00:37:20,077
think it's synonymous with um you know
freedom I mean software freedom. You

666
00:37:20,160 --> 00:37:24,077
basically could run your own models
on your own hardware and retain control

667
00:37:24,160 --> 00:37:28,077
over your own information.
And by the way, this is what enterprises

668
00:37:28,160 --> 00:37:30,237
typically do all the time is, you know,

669
00:37:30,320 --> 00:37:35,677
um about half the global data center
market is, you know, is onrem, meaning

670
00:37:35,760 --> 00:37:38,477
enterprises and governments create their
own data centers. They don't go to the

671
00:37:38,560 --> 00:37:41,997
big clouds. And by the way, I've got
nothing against the hyperscalers, but

672
00:37:42,080 --> 00:37:45,437
just, you know, people like to run their

673
00:37:45,520 --> 00:37:50,077
own um you know, own data centers
and you know, maintain control over their

674
00:37:50,160 --> 00:37:53,757
own data and that kind of thing. And I
think that will be true for consumers to

675
00:37:53,840 --> 00:37:57,677
some degree as well. Um so I do think
it's an important area that we should

676
00:37:57,760 --> 00:37:59,517
you know want to encourage and promote.

677
00:37:59,600 --> 00:38:02,157
The irony right now in the market is

678
00:38:02,240 --> 00:38:06,637
that um is that uh the best open source

679
00:38:06,720 --> 00:38:12,237
models are Chinese. Uh and it's it's
sort of a quirk, right? It's it's the

680
00:38:12,320 --> 00:38:15,277
opposite of what you'd expect. You'd
expect like the American system would

681
00:38:15,360 --> 00:38:19,677
promote open and somehow the Chinese
system would promote closed. That has

682
00:38:19,760 --> 00:38:22,077
kind of um you know ended up being a

683
00:38:22,160 --> 00:38:23,917
little backwards. Um, I think there's

684
00:38:24,000 --> 00:38:26,877
like good reasons for it. Uh, it could

685
00:38:26,960 --> 00:38:31,917
just be it could just be kind of a
historical accident, the fact that the

686
00:38:32,000 --> 00:38:35,117
Deep Seek founder was like very
committed to open source and kind of

687
00:38:35,200 --> 00:38:38,957
just kind of that like got things
started that way or it could be part of

688
00:38:39,040 --> 00:38:42,477
a deliberate strategy. Uh, if you're
China and you're trying to catch up,

689
00:38:42,560 --> 00:38:45,517
open source is a really good way to do
that because you get all the non-aligned

690
00:38:45,600 --> 00:38:49,517
developers to want to help your project
which they can't do if you know with a

691
00:38:49,600 --> 00:38:51,197
closed project. So, it's a great

692
00:38:51,280 --> 00:38:53,677
strategy for catching up. And then also

693
00:38:53,760 --> 00:38:56,557
if you think that your business model

694
00:38:56,640 --> 00:38:58,877
you know as a company or as a country is

695
00:38:58,960 --> 00:39:00,797
let's say scale manufacturing of

696
00:39:00,880 --> 00:39:04,237
hardware then you would want the the

697
00:39:04,320 --> 00:39:07,997
software part to be free or cheap
because it's your compliment right so

698
00:39:08,080 --> 00:39:12,317
you try to commoditize your compliment
and I don't know whether it's by

699
00:39:12,400 --> 00:39:15,837
accident or part of design that that
seems to be what the Chinese strategy

700
00:39:15,920 --> 00:39:17,917
has been I think that the the right

701
00:39:18,000 --> 00:39:23,277
answer for the US in this is to you know
tocourage urge our own open source. I

702
00:39:23,360 --> 00:39:27,597
mean, we I think it'd be a great thing
if we saw more open- source initiatives

703
00:39:27,680 --> 00:39:29,037
get get going. I guess there's one

704
00:39:29,120 --> 00:39:31,917
promising one called um Reflection,

705
00:39:32,000 --> 00:39:36,317
which um was founded by former um you

706
00:39:36,400 --> 00:39:41,277
know, engineers from uh from Google
DeepMind. Uh so I hope we see more open

707
00:39:41,360 --> 00:39:44,957
source innovation in in the west.
But um but look, I think it's it's very

708
00:39:45,040 --> 00:39:49,037
important. is critical and like I said
I in my view it's synonymous with with

709
00:39:49,120 --> 00:39:52,557
freedom and it's definitely not
something we want to suppress. Now just

710
00:39:52,640 --> 00:39:57,117
back to the the the closed ecosystem
for a second. It's true we have five major

711
00:39:57,200 --> 00:40:01,357
competitors there and they're all
spending a lot of money. I do worry a

712
00:40:01,440 --> 00:40:05,357
little bit that at some point some point
in time that the model consolid not the

713
00:40:05,440 --> 00:40:10,157
model the um the market consolidates and
we end up with you know like a monopoly

714
00:40:10,240 --> 00:40:13,757
or duopoly or something like that as
we've seen in other technology markets.

715
00:40:13,840 --> 00:40:16,557
We saw this with search and you know and

716
00:40:16,640 --> 00:40:21,517
so on down the line. And I just think
that it would be good if this market

717
00:40:21,600 --> 00:40:23,357
stayed more competitive than just one or

718
00:40:23,440 --> 00:40:28,157
two winners. Um and um I I don't I don't

719
00:40:28,240 --> 00:40:31,197
really know what to do about that. I'm
just making that observation. And I I do

720
00:40:31,280 --> 00:40:33,277
think that having open source as an

721
00:40:33,360 --> 00:40:36,237
option always then ensures that even if

722
00:40:36,320 --> 00:40:41,517
the market does consolidate that you
do have an alternative and it's an

723
00:40:41,600 --> 00:40:42,877
alternative that's more fully within

724
00:40:42,960 --> 00:40:45,197
your control as opposed to a large

725
00:40:45,280 --> 00:40:49,037
corporation or um you know or or the

726
00:40:49,120 --> 00:40:54,397
deep state you know working with that
corporation as we saw in the Twitter

727
00:40:54,480 --> 00:40:58,637
files that you know the deep state
was working with all these um you know

728
00:40:58,720 --> 00:41:03,837
social media companies um you know
in in implementing much more widespread

729
00:41:03,920 --> 00:41:07,917
censorship than I think we
any of us thought possible. So

730
00:41:08,000 --> 00:41:10,877
uh we've seen evidence in the past
and again in the social networking space

731
00:41:10,960 --> 00:41:13,277
about um about how the government could

732
00:41:13,360 --> 00:41:16,157
get involved in nefarious ways and it

733
00:41:16,240 --> 00:41:18,637
would be good to have alternatives so

734
00:41:18,720 --> 00:41:22,477
that so you know to to prevent that or

735
00:41:22,560 --> 00:41:26,477
to to make it less likely that
that scenario comes about with AI.

736
00:41:26,560 --> 00:41:29,757
>> Yeah. Well, as you know, we we
and others uh um are very aggressively

737
00:41:29,840 --> 00:41:33,517
investing um in new new model companies
of many kinds, including new foundation

738
00:41:33,600 --> 00:41:36,557
model companies, and then also,
you know, as you're probably there are a

739
00:41:36,640 --> 00:41:39,677
whole bunch of new open source efforts
um you know, that are not yet public

740
00:41:39,760 --> 00:41:42,717
that, you know, hopefully will bear
fruit over the next couple years. So, I

741
00:41:42,800 --> 00:41:46,557
think that at least in the in the medium
term, I think we're looking at an an

742
00:41:46,640 --> 00:41:49,117
explosion of model development
as opposed to consolidation and then, you

743
00:41:49,200 --> 00:41:50,717
know, we'll we'll see
what happens from there.

744
00:41:50,800 --> 00:41:53,917
>> Yeah, that's really good to hear. I
mean I think you know if we assess kind of

745
00:41:54,000 --> 00:41:57,917
the the state of the AI race um visv China

746
00:41:58,000 --> 00:42:00,077
>> uh this is the only
area where we appear

747
00:42:00,160 --> 00:42:04,717
to be behind is in this area of open
source models um I I think you know if

748
00:42:04,800 --> 00:42:07,757
you don't care whether it's open
or closed I think we have the lead

749
00:42:07,840 --> 00:42:12,157
>> uh I think our top models uh model
companies are ahead of the top Chinese

750
00:42:12,240 --> 00:42:16,237
companies although they're quite good
but uh but just this narrow area of open

751
00:42:16,320 --> 00:42:19,517
source seems to be where they have an
advantage so it's great to hear that you

752
00:42:19,600 --> 00:42:24,637
guys are seeing >> you know a lot
more um efforts um coming to market.

753
00:42:24,720 --> 00:42:25,997
>> Yeah. Yeah, there's more coming.

754
00:42:26,080 --> 00:42:28,397
>> Yeah. Good. >> Yeah.
Yeah. Definitely more coming.

755
00:42:28,480 --> 00:42:32,797
>> The um Peter Teal quipped, you know,
many years ago that he thought, you

756
00:42:32,880 --> 00:42:36,957
know, crypto would be libertarian
or decentralizing and that uh AI would be

757
00:42:37,040 --> 00:42:39,917
communist or or centralizing.
And I think one thing we've perhaps we've

758
00:42:40,000 --> 00:42:43,277
learned that uh technology isn't
deterministic and that the there are set

759
00:42:43,360 --> 00:42:47,037
of choices that determine whether these
technologies are decentralizing or or

760
00:42:47,120 --> 00:42:51,117
centralizing. And um maybe we could
segue use that as a segue to go deeper

761
00:42:51,200 --> 00:42:54,957
into the the state of the race with China.
Uh maybe David, you could lay out

762
00:42:55,040 --> 00:42:58,637
the sort of the what's most important
to to get right. You've already indicated,

763
00:42:58,720 --> 00:43:02,637
you know, open source is one example.
Um you know, you you alluded earlier to

764
00:43:02,720 --> 00:43:06,797
sort of our strategy as it relates to
chips. Um you know, some people say that

765
00:43:06,880 --> 00:43:09,757
yes, it's it's a good idea to do what
we're doing because it'll you know,

766
00:43:09,840 --> 00:43:12,317
limit domestic semiconductor production.

767
00:43:12,400 --> 00:43:15,357
Other people say, "Oh, well, you know,
some of these companies say chips are

768
00:43:15,440 --> 00:43:18,877
their biggest, you know, limiting
factors and so are we enabling them in

769
00:43:18,960 --> 00:43:22,397
some way?" May why don't you talk about
the sort of state of play and then our

770
00:43:22,480 --> 00:43:23,437
strategy?

771
00:43:23,520 --> 00:43:26,077
>> Yeah. So, you know, when we talk about

772
00:43:26,160 --> 00:43:30,637
winning the AI race, um, sometimes
we say we're in a race against China,

773
00:43:30,720 --> 00:43:33,117
sometimes we just leave it a little bit
more vague because I I don't think we

774
00:43:33,200 --> 00:43:35,917
should become overly obsessed with our

775
00:43:36,000 --> 00:43:40,237
uh competitors or adversaries. I think
whether we win or not will mostly have

776
00:43:40,320 --> 00:43:41,757
to do with the decisions we make about

777
00:43:41,840 --> 00:43:44,477
our own technology ecosystem, not about,

778
00:43:44,560 --> 00:43:47,677
you know, what we do v visav them. And

779
00:43:47,760 --> 00:43:50,877
so the president in in his July 23rd

780
00:43:50,960 --> 00:43:52,877
speech on AI policy, I think mentioned a

781
00:43:52,960 --> 00:43:55,677
few of the key pillars of of, you know,

782
00:43:55,760 --> 00:43:59,997
of of how we win this this AI race.
Um, and by the way, I'm not saying it ever

783
00:44:00,080 --> 00:44:01,917
ends. This it might be an infinite game,

784
00:44:02,000 --> 00:44:04,077
but um but we want to be in the lead at

785
00:44:04,160 --> 00:44:07,357
least. And um and I I I I do think that

786
00:44:07,440 --> 00:44:11,677
um there could be a period of time where
like you know take the internet where I

787
00:44:11,760 --> 00:44:15,517
mean the internet's still going on. It's
happened but we understand that kind of

788
00:44:15,600 --> 00:44:19,437
who the winners are is kind of baked now.
Um so there could be a period of

789
00:44:19,520 --> 00:44:22,797
time in which you know it's kind of baked
who the the winners in AI are. But

790
00:44:22,880 --> 00:44:27,357
um in any event um you know in terms
of how we win this race um you know I

791
00:44:27,440 --> 00:44:28,477
mentioned a few of the key pillars.

792
00:44:28,560 --> 00:44:31,917
Number one is innovation. You know it's
very important to support the private

793
00:44:32,000 --> 00:44:34,717
sector because they're the ones who
do the innovation. We're not going to

794
00:44:34,800 --> 00:44:38,877
regulate our way to beating our our
adversary. We just have to out innovate

795
00:44:38,960 --> 00:44:41,677
them. Um I mentioned I I think right now

796
00:44:41,760 --> 00:44:45,277
the biggest obstacle is the the the um

797
00:44:45,360 --> 00:44:47,757
frenzy of overregulation happening at

798
00:44:47,840 --> 00:44:52,797
the states. I desperately think we need
um a federal a single federal standard.

799
00:44:52,880 --> 00:44:56,237
Uh a patchwork of 50 different
regulatory regimes can be incredibly

800
00:44:56,320 --> 00:45:00,237
burdensome to comply with. I think even
the people who support a lot of this

801
00:45:00,320 --> 00:45:03,837
regulation are now acknowledging that
we're going to need a federal standard.

802
00:45:03,920 --> 00:45:08,237
Uh the problem is that when they talk
about it, what they really want is to

803
00:45:08,320 --> 00:45:11,437
federalize the most ownorous
version of all the state laws

804
00:45:11,520 --> 00:45:15,037
>> and that can't be allowed either.
So, you know, there's going to be a battle

805
00:45:15,120 --> 00:45:17,917
to come. I think as the states um become

806
00:45:18,000 --> 00:45:22,797
more and more unwieldy, you know, as it
becomes more of a trap for startups that

807
00:45:22,880 --> 00:45:26,077
they now have to report into 50
different states at 50 different times

808
00:45:26,160 --> 00:45:29,117
to 50 different agencies about 50
different things, people are going to

809
00:45:29,200 --> 00:45:32,397
realize this is crazy and they're going
to try to federalize it. And then the

810
00:45:32,480 --> 00:45:35,837
question, I think, is whether we get
preeemption heavy or preeemption light.

811
00:45:35,920 --> 00:45:40,237
Um you know, do we do we get a I think
everyone's gonna ultimately be in favor

812
00:45:40,320 --> 00:45:44,237
of a of a single federal standard.
Um because I think one of America's

813
00:45:44,320 --> 00:45:48,557
greatest advantages is that we have
a large national market right not 50

814
00:45:48,640 --> 00:45:52,477
separate state markets like kind
of you know Europe before the EU wasn't

815
00:45:52,560 --> 00:45:55,597
competitive at all on the internet
because it's 30 different regulatory

816
00:45:55,680 --> 00:45:59,197
regimes and so you know if you're
a European startup and even if you won

817
00:45:59,280 --> 00:46:03,677
your country it didn't get you very far
because you still had to like you know

818
00:46:03,760 --> 00:46:06,717
figure out how to compete in 30 other
countries before you could even win

819
00:46:06,800 --> 00:46:10,157
Europe and then meanwhile your American
competitors won the entire American

820
00:46:10,240 --> 00:46:14,237
market and ready to scale scale up
globally. So the fact that we have a

821
00:46:14,320 --> 00:46:17,677
single national market is just
fundamental to our competitiveness and

822
00:46:17,760 --> 00:46:19,517
it's why you know winners in America

823
00:46:19,600 --> 00:46:25,037
then go on to kind of win the whole world.
So we have to preserve that and I

824
00:46:25,120 --> 00:46:27,677
think we will eventually get some
some federal preeemption. I think the

825
00:46:27,760 --> 00:46:29,597
question will just again be whether we

826
00:46:29,680 --> 00:46:32,797
preempt heavy or preempt light. Um

827
00:46:32,880 --> 00:46:36,797
second big area is infrastructure you
know and energy you know we want to help

828
00:46:36,880 --> 00:46:42,237
this this um amazing infrastructure boom
that's happening and the biggest I think

829
00:46:42,320 --> 00:46:46,237
limiting factor there is just going
to be around energy uh I think President

830
00:46:46,320 --> 00:46:49,197
Trump's been incredibly far-sighted
in this I mean he was talking about drill

831
00:46:49,280 --> 00:46:53,277
baby drill many years ago um he
understood that energy is the basis for

832
00:46:53,360 --> 00:46:57,357
everything it's definitely the basis
for this AI boom and we want to basically

833
00:46:57,440 --> 00:46:59,837
get all of these unnecessary regulations

834
00:46:59,920 --> 00:47:01,837
the permitting restrictions, a lot of

835
00:47:01,920 --> 00:47:05,997
the nimism out of the way so that um AI

836
00:47:06,080 --> 00:47:09,517
companies can build data centers and get
power for them. And we can talk about

837
00:47:09,600 --> 00:47:10,797
that more if you want, but I think that

838
00:47:10,880 --> 00:47:15,037
that's a second really huge um uh part

839
00:47:15,120 --> 00:47:18,477
of the of what it's going to take to win
the AI race. And then the third area is

840
00:47:18,560 --> 00:47:21,037
is around exports. And maybe this has

841
00:47:21,120 --> 00:47:24,317
been the most controversial one. Um and

842
00:47:24,400 --> 00:47:28,077
it really speaks to the cultural
divide between Silicon Valley and and

843
00:47:28,160 --> 00:47:31,597
Washington. So I think
all of us in Silicon Valley

844
00:47:31,680 --> 00:47:35,517
understand that the way that you win
a technology race is by building the

845
00:47:35,600 --> 00:47:40,077
biggest ecosystem, right? You get
the most developers building on your your

846
00:47:40,160 --> 00:47:43,837
platform. You get the most apps in your
app store. Everyone just uses you. I

847
00:47:43,920 --> 00:47:48,077
mean, you know, those are the companies
that typically win are the ones that get

848
00:47:48,160 --> 00:47:52,077
all the users, all the developers, and so
on. And so we in Silicon Valley have

849
00:47:52,160 --> 00:47:55,517
a partnership mentality. You know,
we want to just publish the APIs and get

850
00:47:55,600 --> 00:47:59,037
everyone using them. Uh Washington has
a different mentality, right? It's much

851
00:47:59,120 --> 00:48:03,357
more of a command and control. We want
you to get approved. Um, you know, we

852
00:48:03,440 --> 00:48:04,797
kind of want to hoard this technology.

853
00:48:04,880 --> 00:48:07,757
Only America should have it. Um, and and

854
00:48:07,840 --> 00:48:12,077
this this was really fundamental,
I think, to the Biden diffusion rule where

855
00:48:12,160 --> 00:48:15,677
the the point of that rule is to stop
diffusion, right? Diffusion is a is a is

856
00:48:15,760 --> 00:48:18,397
a bad word. >> Um,
but in Silicon Valley, we understand

857
00:48:18,480 --> 00:48:22,797
that diffusion is how you win. [snorts]
>> Um, and now I mean, I don't think we

858
00:48:22,880 --> 00:48:25,757
ever called it diffusion before. That
was a new word for me. We just called it

859
00:48:25,840 --> 00:48:28,557
usage. >> Yeah.
Uh but we understand that like

860
00:48:28,640 --> 00:48:29,997
getting the most users is how you win.

861
00:48:30,080 --> 00:48:32,637
Um so there's like a fundamental culture

862
00:48:32,720 --> 00:48:35,837
clash going on right now. And um you

863
00:48:35,920 --> 00:48:40,397
know the way I kind of parse it
is that what we decide to sell to China is

864
00:48:40,480 --> 00:48:43,837
always going to be comp uh complicated
because you know they're our competitor

865
00:48:43,920 --> 00:48:48,717
and our adversary and there's the whole
potential dual use and so the the the

866
00:48:48,800 --> 00:48:52,797
question of what you sell to China is is
nuanced. Um but what we sell to the rest

867
00:48:52,880 --> 00:48:55,917
of the world that should be an easy
question which is we should want to do

868
00:48:56,000 --> 00:48:59,117
business with the rest of the world.
We should want to have the largest

869
00:48:59,200 --> 00:49:04,157
ecosystem possible. Uh and every country
we exclude from our technology alliance

870
00:49:04,240 --> 00:49:06,637
we're basically driving into the arms
of China and it makes their ecosystem

871
00:49:06,720 --> 00:49:09,197
bigger. And what we saw under the the

872
00:49:09,280 --> 00:49:12,557
the the Biden years is that they were

873
00:49:12,640 --> 00:49:15,517
constantly pushing other countries uh

874
00:49:15,600 --> 00:49:17,677
into the arms of China. Starting with

875
00:49:17,760 --> 00:49:23,277
the Gulf States in um October of 2023,
basically the the Gulf States, you I'm

876
00:49:23,360 --> 00:49:26,477
talking about countries like Saudi
Arabia, UAE, long-standing US allies,

877
00:49:26,560 --> 00:49:30,797
they weren't allowed to buy chips
from the US. In other words, they weren't

878
00:49:30,880 --> 00:49:36,157
allowed to set up data centers
and participate in in AI. And um you know

879
00:49:36,240 --> 00:49:40,237
here we are telling all these countries
that you know AI is fundamental to the

880
00:49:40,320 --> 00:49:43,757
future. It's going to be the basis
of the economy and yet we're excluding you

881
00:49:43,840 --> 00:49:47,917
from participating in the American tech
stack. Well you know it's obvious what

882
00:49:48,000 --> 00:49:51,597
they're going to do. You know the only
play we're giving them is to go to China

883
00:49:51,680 --> 00:49:55,357
and so you know all of these rules
basically just create pent-up demand for

884
00:49:55,440 --> 00:49:58,557
for um for Chinese chips and models and

885
00:49:58,640 --> 00:50:01,197
it creates a Huawei belt and road. And

886
00:50:01,280 --> 00:50:03,837
we are hearing that um that Huawei is

887
00:50:03,920 --> 00:50:08,077
starting to uh proliferate or diffuse um

888
00:50:08,160 --> 00:50:12,397
in the Middle East and in Southeast Asia.
And I just think it's a really

889
00:50:12,480 --> 00:50:14,957
counterproductive strategy.
We're completely shooting ourselves in the

890
00:50:15,040 --> 00:50:18,877
foot. And like the the greatest irony is

891
00:50:18,960 --> 00:50:22,237
that the people who've been pushing this
strategy of driving all these countries

892
00:50:22,320 --> 00:50:24,557
into China's arms have called themselves

893
00:50:24,640 --> 00:50:29,437
China hawks. You know, as if what
they're doing is uh is hurting China.

894
00:50:29,520 --> 00:50:32,957
No, it's like it's it's helping China.
I mean, it's basically just handing them

895
00:50:33,040 --> 00:50:35,997
markets. And our products are better. Uh

896
00:50:36,080 --> 00:50:40,397
but if you don't give these countries
a choice to buy the American tech stack,

897
00:50:40,480 --> 00:50:42,477
obviously they're going to go with the

898
00:50:42,560 --> 00:50:45,037
Chinese tech stack. And um you know,

899
00:50:45,120 --> 00:50:49,117
China China's out there promoting um you

900
00:50:49,200 --> 00:50:53,917
know, uh Deepseek models and Huawei
chips and they're not like ringing their

901
00:50:54,000 --> 00:50:55,757
hands about

902
00:50:55,840 --> 00:50:58,877
you know whether um you know exporting

903
00:50:58,960 --> 00:51:02,957
chips for a data center in UAE is going
to like create the Terminator and you

904
00:51:03,040 --> 00:51:06,957
know all these like ridiculous
narratives that we've invented to you

905
00:51:07,040 --> 00:51:12,237
know reasons we've invented not to sell
American technology to our friends. So,

906
00:51:12,320 --> 00:51:15,757
um, you know, that that has ended up
being, I think, surprisingly maybe the

907
00:51:15,840 --> 00:51:20,797
most controversial part of what we've
advocated for. Um, but, uh, but there

908
00:51:20,880 --> 00:51:23,757
you have it. So, in any event, I'll
stop there. Those are kind of some of the

909
00:51:23,840 --> 00:51:26,797
major pillars of what
we've been advocating.

910
00:51:26,880 --> 00:51:28,077
>> Should we should
we go deeper on sort of

911
00:51:28,160 --> 00:51:31,277
the infrastructure energy point in terms
of what it's really going to take to to

912
00:51:31,360 --> 00:51:34,877
to get enough capacity or what's what's
most important in that second bullet you

913
00:51:34,960 --> 00:51:38,557
were talking about? Yeah. I mean, well,
so I mean, there there are definitely

914
00:51:38,640 --> 00:51:42,157
people who are much more knowledgeable
about energy than I am and are experts

915
00:51:42,240 --> 00:51:46,797
in the space, but here's what I've been
able to kind of divine is um so first of

916
00:51:46,880 --> 00:51:51,037
all, the administration, President Trump
has signed multiple executive orders to

917
00:51:51,120 --> 00:51:52,797
uh to allow for nuclear to make

918
00:51:52,880 --> 00:51:54,877
permitting easier. We've even freed up

919
00:51:54,960 --> 00:51:58,637
federal land uh for for data centers to

920
00:51:58,720 --> 00:52:02,157
hopefully trying to help get around some
of these state and local restrictions.

921
00:52:02,240 --> 00:52:05,197
And obviously the president has made it

922
00:52:05,280 --> 00:52:08,637
a lot easier to um to stand up new um uh

923
00:52:08,720 --> 00:52:11,277
energy projects uh power generation all

924
00:52:11,360 --> 00:52:13,837
that kind of stuff. Um I still think

925
00:52:13,920 --> 00:52:16,877
though that we have um a growing nimi

926
00:52:16,960 --> 00:52:21,677
problem at the state and local level
uh in in the US that that is becoming a

927
00:52:21,760 --> 00:52:26,797
little bit worrisome.
uh and um and uh and if if we don't

928
00:52:26,880 --> 00:52:30,877
figure out a way to to address it, then
it could really slow down uh the build

929
00:52:30,960 --> 00:52:33,517
out of this infrastructure. Um in terms

930
00:52:33,600 --> 00:52:38,397
of of power, so my understanding is that
nuclear is going to take 5 or 10 years.

931
00:52:38,480 --> 00:52:41,197
It's just it's just not something that
we're going to be able to do in the next

932
00:52:41,280 --> 00:52:45,277
2 or 3 years. So in the short term,
it really means that like gas is the way

933
00:52:45,360 --> 00:52:50,317
that these data centers are going to get
um powered. And the the issue with gas

934
00:52:50,400 --> 00:52:54,637
is the shortage there is not I mean
America has plenty of natural gas and um

935
00:52:54,720 --> 00:52:59,357
there's uh it exists in plenty of enough
red states where you could just build

936
00:52:59,440 --> 00:53:03,357
out data centers like close to the
source which be smart. Uh the issue is

937
00:53:03,440 --> 00:53:06,957
there's like a shortage of these um gas
turbines. You know there's only like two

938
00:53:07,040 --> 00:53:09,677
or three companies that make these

939
00:53:09,760 --> 00:53:14,317
things and like there's like a backlog
of two or three years. So I think that's

940
00:53:14,400 --> 00:53:16,397
probably the the the immediate problem

941
00:53:16,480 --> 00:53:21,597
there that needs to get solved. Um uh
however I do think that in the next 2

942
00:53:21,680 --> 00:53:23,037
or 3 years we could get a lot more out

943
00:53:23,120 --> 00:53:25,917
of the grid. So, I've had um you know,

944
00:53:26,000 --> 00:53:29,437
energy executives tell me that uh that

945
00:53:29,520 --> 00:53:33,677
if we could just shed um 40 hours a year

946
00:53:33,760 --> 00:53:38,077
of peak load from the grid to like
backup generators, to diesel, things

947
00:53:38,160 --> 00:53:39,997
like that, you could free up an

948
00:53:40,080 --> 00:53:42,957
additional 80 gawatt of power, which is

949
00:53:43,040 --> 00:53:47,757
a lot because I guess
the way that it works is um

950
00:53:47,840 --> 00:53:49,837
is, you know, the the grid is only used

951
00:53:49,920 --> 00:53:54,877
about 50% uh 50% of the capacity is used
throughout the here uh because they have

952
00:53:54,960 --> 00:53:59,037
to build enough capacity for the the peak
the peak days like the hottest day

953
00:53:59,120 --> 00:54:00,317
in summer, the coldest day in winter.

954
00:54:00,400 --> 00:54:05,597
Those are your peak days. Uh and they
don't want to um commit to a bunch of

955
00:54:05,680 --> 00:54:08,637
the capacity being used and then you
know you find out that you have a really

956
00:54:08,720 --> 00:54:10,557
cold day in winter and people can't get

957
00:54:10,640 --> 00:54:15,917
enough heat for their their homes.
And so um so they can't like overcommit to

958
00:54:16,000 --> 00:54:19,677
you know to say contracts or data
centers, things like that. Um, but if

959
00:54:19,760 --> 00:54:22,557
you could again just like um if you

960
00:54:22,640 --> 00:54:27,437
could shed that like 40 hours
a year of peak load to backup

961
00:54:27,520 --> 00:54:31,437
uh then you would be able to free up 80
80 gigawatts which is a lot. Um and that

962
00:54:31,520 --> 00:54:35,117
would definitely get us through the next
you know 2 or 3 years until the gas

963
00:54:35,200 --> 00:54:38,317
turbine bottleneck's been alleviated.
Um and then eventually you get to to

964
00:54:38,400 --> 00:54:41,917
nuclear. Uh so that would be very good.

965
00:54:42,000 --> 00:54:45,277
I think um the the issue there
is just there's a whole bunch of insane

966
00:54:45,360 --> 00:54:47,597
regulations preventing

967
00:54:47,680 --> 00:54:49,677
uh you know load shedding. So like for

968
00:54:49,760 --> 00:54:53,277
example, you can't use diesel. Um and uh

969
00:54:53,360 --> 00:54:56,717
this you know uh Chris Wright is the
secretary of energy is very good on all

970
00:54:56,800 --> 00:55:00,077
this stuff and I think he's working
on unraveling all of this so we could

971
00:55:00,160 --> 00:55:03,117
actually uh actually do this.
>> It's funny David as you talk about this

972
00:55:03,200 --> 00:55:05,997
stuff it's I I can't help but it's
a little bit like it's a little bit like

973
00:55:06,080 --> 00:55:11,117
the principle is just do
the opposite of the EU. [laughter]

974
00:55:11,200 --> 00:55:14,877
Yeah. Like basically I think everything
we've talked about so far is basically

975
00:55:14,960 --> 00:55:18,077
the opposite of the European approach.

976
00:55:18,160 --> 00:55:21,757
>> Yeah. Well, I mean the Europeans I mean
they have a really different mindset for

977
00:55:21,840 --> 00:55:26,157
all this stuff. when when they
talk about AI leadership, um

978
00:55:26,240 --> 00:55:31,357
>> what they mean is that they're taking
the lead in defining the regulations and

979
00:55:31,440 --> 00:55:34,317
they're all, you know, it's like that's
what they're proud of is that like they

980
00:55:34,400 --> 00:55:37,597
think that's what their comparative
advantage is is that, you know, they get

981
00:55:37,680 --> 00:55:40,557
together in Brussels and figure
out what all the rules should be

982
00:55:40,640 --> 00:55:42,477
>> and that's what they call leadership.

983
00:55:42,560 --> 00:55:46,237
Um, >> the the EU the EU
just announced a uh I

984
00:55:46,320 --> 00:55:48,797
shouldn't be on them too much, but they
just announced a big new growth fund, a

985
00:55:48,880 --> 00:55:53,197
big new public private sector uh tech
growth fund to grow uh EU companies to

986
00:55:53,280 --> 00:55:56,317
scale. And I I just I would literally I
was just like, well, it's it's actually

987
00:55:56,400 --> 00:55:58,477
quite it's almost like a game show
or something. They do everything they can

988
00:55:58,560 --> 00:56:01,517
to strangle them
in their crib [clears throat]

989
00:56:01,600 --> 00:56:04,957
and then if they if they make it if they
make it through like a [laughter] decade

990
00:56:05,040 --> 00:56:08,077
of abuse of small companies, then
they're going to give the money to grow.

991
00:56:08,160 --> 00:56:12,717
Um, >> well, yeah. Well, it's what, you
know, Ronald Reagan had a line about this,

992
00:56:12,800 --> 00:56:16,557
which is if it moves, tax it. If it
keeps moving, regulate it. If it stops

993
00:56:16,640 --> 00:56:18,477
moving, subsidize it. >> Yeah.

994
00:56:18,560 --> 00:56:22,077
>> Uh, the Europeans are definitely
at the subsidized stage. [laughter]

995
00:56:22,160 --> 00:56:25,517
>> Yeah. And I Yeah. I shouldn't be on
them too much, but I just like I've always

996
00:56:25,600 --> 00:56:29,197
been proud to be an American, but
particularly now u because like we just

997
00:56:29,280 --> 00:56:32,317
the fact that we've we've we've been
we we are we it really feels like we're

998
00:56:32,400 --> 00:56:34,797
reentering on core American values
in a lot of the things that we're talking

999
00:56:34,880 --> 00:56:36,317
about, which is just really great.

1000
00:56:36,400 --> 00:56:39,437
>> Yeah. I mean again
it's you know our our

1001
00:56:39,520 --> 00:56:43,677
view is that the first of all we have
to win the I race. We want America to lead

1002
00:56:43,760 --> 00:56:47,437
in this critical area. It's like
fundamental for our economy and our

1003
00:56:47,520 --> 00:56:49,757
national security. How do you do that?

1004
00:56:49,840 --> 00:56:52,237
Well, our companies have to be
successful because they're the ones who

1005
00:56:52,320 --> 00:56:55,517
do the innovation. Again, you're not
going to regulate your way to winning

1006
00:56:55,600 --> 00:56:58,877
the AI race. I'm not saying we don't
need any regulations, but the point is

1007
00:56:58,960 --> 00:57:01,757
just that's not how that's not what's
going to determine whether we're the

1008
00:57:01,840 --> 00:57:06,477
winners or not. David, you recently
tweeted that climate dumerism perhaps is

1009
00:57:06,560 --> 00:57:11,277
giving way to AI dumerism based on you
know Bill Bill Gates recent comments. Um

1010
00:57:11,360 --> 00:57:14,317
what do you mean by this? Do do you mean
it's going to be a major fl you know of

1011
00:57:14,400 --> 00:57:17,437
of of the you know uh US left or what do

1012
00:57:17,520 --> 00:57:19,677
what do you mean by this comment? Well,

1013
00:57:19,760 --> 00:57:22,317
I I think the left needs a central

1014
00:57:22,400 --> 00:57:25,917
organizing catastrophe to justify their

1015
00:57:26,000 --> 00:57:30,157
uh takeover of the economy and to to

1016
00:57:30,240 --> 00:57:34,797
regulate everything and especially
to control the information space. Uh and I

1017
00:57:34,880 --> 00:57:38,477
think that you're seeing that kind
of the allure of the whole climate change

1018
00:57:38,560 --> 00:57:40,957
doomer narrative has kind of faded.

1019
00:57:41,040 --> 00:57:43,437
Maybe it's the fact that they predicted
10 years ago that the whole world would

1020
00:57:43,520 --> 00:57:46,237
be underwater in 10 years and that
hasn't happened. So it's like at a

1021
00:57:46,320 --> 00:57:50,717
certain point you get discredited
by your own catastrophic predictions. I

1022
00:57:50,800 --> 00:57:53,597
suspect that's where we'll be with AI
dumerism in a few years. But in the

1023
00:57:53,680 --> 00:57:56,877
meantime it's a really good narrative
to kind of take the place of the climate

1024
00:57:56,960 --> 00:58:01,357
dumerism. There's actually a lot
of similarities I would say. Um you know

1025
00:58:01,440 --> 00:58:03,357
you've kind of got um there's a lot of

1026
00:58:03,440 --> 00:58:07,197
kind of um pre-existing Hollywood um

1027
00:58:07,280 --> 00:58:10,957
storytelling and pop culture that
supports this idea. You know you've got

1028
00:58:11,040 --> 00:58:13,357
the Terminator movies and the Matrix and

1029
00:58:13,440 --> 00:58:16,397
all this kind of stuff. Uh so people

1030
00:58:16,480 --> 00:58:18,237
have been taught to be afraid of this.

1031
00:58:18,320 --> 00:58:21,677
Um and then you know you you um there

1032
00:58:21,760 --> 00:58:26,077
there's enough kind of pseudocience
behind it you know kind of like um you

1033
00:58:26,160 --> 00:58:30,477
know with uh like you've got all these
contrived studies that like the one

1034
00:58:30,560 --> 00:58:35,277
where they claimed that the AI
researcher got blackmailed by his own AI

1035
00:58:35,360 --> 00:58:37,117
model or whatever. Look, it's very easy

1036
00:58:37,200 --> 00:58:40,877
to steer the model towards um the the

1037
00:58:40,960 --> 00:58:44,317
answer that you want and uh and a lot
of these studies have been very contrived,

1038
00:58:44,400 --> 00:58:49,437
but there's this patina pseudocience to
it. It's certainly technical enough that

1039
00:58:49,520 --> 00:58:53,197
the average person doesn't feel
comfortable saying that this doesn't

1040
00:58:53,280 --> 00:58:56,717
make any sense. I mean, it's more like
you're not an expert. What do you know?

1041
00:58:56,800 --> 00:59:01,197
Uh and even Republican politicians,
I think, um are kind of falling for this.

1042
00:59:01,280 --> 00:59:05,597
So yeah, I mean it's a rarely desirable
nar and of course you know as AI touches

1043
00:59:05,680 --> 00:59:09,037
more and more things, more and more
parts of the economy, every business is

1044
00:59:09,120 --> 00:59:12,877
going to use it to some degree. If you
can regulate AI, then that kind of gives

1045
00:59:12,960 --> 00:59:17,917
you a lot of control over lots of other
things. And like I mentioned, AI is kind

1046
00:59:18,000 --> 00:59:20,077
of eating the internet.
It's like the main way that you're getting

1047
00:59:20,160 --> 00:59:25,117
information. So again, if you can kind
of get your hooks into what the AI is

1048
00:59:25,200 --> 00:59:30,077
showing people, now you can control what
they see and hear and think, uh, which

1049
00:59:30,160 --> 00:59:32,877
dovetales with the whole with the left
censorship, you know, agenda, which

1050
00:59:32,960 --> 00:59:37,677
they've never given up on. Uh, dovetales
with their agenda to brainwash kids, uh,

1051
00:59:37,760 --> 00:59:41,597
which we, you know, which is kind of
the whole woke thing. So, I mean, this is

1052
00:59:41,680 --> 00:59:45,197
this is going to be very desirable
for the left and and this is why I mean,

1053
00:59:45,280 --> 00:59:46,637
look, they're they're already doing

1054
00:59:46,720 --> 00:59:48,237
this. This is not like some prediction

1055
00:59:48,320 --> 00:59:52,317
on my part. basically after um scam,

1056
00:59:52,400 --> 00:59:56,397
bankr run fraud uh did what he did with
FTX and got sent to jail. He was like a

1057
00:59:56,480 --> 01:00:01,517
big effective altruist and he had made
um pandemics like their big cause. They

1058
01:00:01,600 --> 01:00:06,557
needed a new cause and they got behind
this idea of of X risk existential risk.

1059
01:00:06,640 --> 01:00:10,877
The idea being if there's like a 1%
chance of um AI ending the world, then

1060
01:00:10,960 --> 01:00:14,477
we should drop everything and just
focus on that because you do the expected

1061
01:00:14,560 --> 01:00:17,597
value calculation and so if it ends
humanity, then that's the only thing you

1062
01:00:17,680 --> 01:00:22,317
should focus on. even if it's, you know,
very small percentage chance. But they

1063
01:00:22,400 --> 01:00:26,477
really like reorganized behind this with
all and, you know, they've got quite a

1064
01:00:26,560 --> 01:00:28,637
few advocates. And actually, it's it's

1065
01:00:28,720 --> 01:00:30,077
an amazing story about how much

1066
01:00:30,160 --> 01:00:33,837
influence they were able to achieve um

1067
01:00:33,920 --> 01:00:37,757
largely behind the scenes or in the
shadows during the Biden years. They

1068
01:00:37,840 --> 01:00:41,917
basically convinced all of the major
Biden staffers of this um view of this

1069
01:00:42,000 --> 01:00:45,677
like imminent super intelligence
is coming. We should be really afraid of

1070
01:00:45,760 --> 01:00:49,357
it. we need to consolidate control
over it. There should only be, you know,

1071
01:00:49,440 --> 01:00:52,877
ideally two or three companies that have
it. We don't want anyone in the rest of

1072
01:00:52,960 --> 01:00:55,597
the world to get it. And then, you know,
what they said is, you know, once we

1073
01:00:55,680 --> 01:00:59,437
make sure that there's only two or three
American companies, we'll u we'll solve

1074
01:00:59,520 --> 01:01:02,397
the coordination problems. That's what
they consider to be the, you know, the

1075
01:01:02,480 --> 01:01:05,837
free market, uh, we'll solve those
coordination problems for behind those

1076
01:01:05,920 --> 01:01:08,237
um, companies and we'll be able to

1077
01:01:08,320 --> 01:01:13,037
control this whole thing and prevent
the the genie from escaping the bottle. I

1078
01:01:13,120 --> 01:01:17,037
think it was like this totally paranoid
version of what would happen and it's

1079
01:01:17,120 --> 01:01:18,397
already being it's already in the

1080
01:01:18,480 --> 01:01:22,317
process of being refuted. But this this

1081
01:01:22,400 --> 01:01:26,957
vision is fundamentally what animated
the Biden executive order on AI. It's

1082
01:01:27,040 --> 01:01:30,797
what animated the Biden diffusion rule
and Mark and you've talked about how you

1083
01:01:30,880 --> 01:01:34,557
were in a meeting with Biden folks
and they were going to basically ban open

1084
01:01:34,640 --> 01:01:36,637
source and >> they were
going to they're basically

1085
01:01:36,720 --> 01:01:40,157
going to anoint two or three
winners and that was it. Um,

1086
01:01:40,240 --> 01:01:43,437
>> yeah, they told us that. They told
us that explicitly. Um, and uh, yeah, and

1087
01:01:43,520 --> 01:01:45,197
they told us exactly what you just said.

1088
01:01:45,280 --> 01:01:47,677
They told us they're going to ban open
source. And when we challenged them on

1089
01:01:47,760 --> 01:01:50,477
the ability to ban open source because
it's, you know, we're talking about, you

1090
01:01:50,560 --> 01:01:54,237
know, math, you know, like mathematical
algorithms that are taught in textbooks

1091
01:01:54,320 --> 01:01:57,837
and YouTube videos and universities.
Um, you know, they they they said, "Well,

1092
01:01:57,920 --> 01:02:01,117
during the Cold War, we banned entire
areas of physics um, and put them off

1093
01:02:01,200 --> 01:02:05,437
limits and we'll do the same thing
for math if we have to." Um, yeah. And

1094
01:02:05,520 --> 01:02:09,277
that's the um yeah that's
the uh yeah that that was the

1095
01:02:09,360 --> 01:02:12,717
>> and you'll be happy to know that
the guy who actually said that is now an

1096
01:02:12,800 --> 01:02:16,077
anthropic employee it's not

1097
01:02:16,160 --> 01:02:19,837
>> no that's exactly right all all those
and and I mean literally the minute the

1098
01:02:19,920 --> 01:02:23,277
Biden administration was over all
the top Biden AI employees went to go work

1099
01:02:23,360 --> 01:02:26,557
at anthropic which tells you
who they were working with during

1100
01:02:26,640 --> 01:02:31,197
>> the Biden years. Um >> but no
but I mean this was this was very

1101
01:02:31,280 --> 01:02:34,077
much the narrative you
sort of had this imminent

1102
01:02:34,160 --> 01:02:35,757
>> uh super intelligence and then you know

1103
01:02:35,840 --> 01:02:38,557
the one of the refrains you heard um was

1104
01:02:38,640 --> 01:02:43,357
that uh AI is like nuclear weapons
and uh and GPUs are like uranium or

1105
01:02:43,440 --> 01:02:48,077
plutonium or something >> and um
and so and and therefore we need

1106
01:02:48,160 --> 01:02:51,437
like an you know the way the proper
way to to regulate this is with like an

1107
01:02:51,520 --> 01:02:53,917
international atomic energy commission

1108
01:02:54,000 --> 01:02:58,957
>> and so you know again everything would
be sort of centralized and um controlled

1109
01:02:59,040 --> 01:03:02,957
and they would anoint two or three
winners and you know now this I think

1110
01:03:03,040 --> 01:03:08,397
this narrative really started to fall
apart uh with the launch of deepseek which

1111
01:03:08,480 --> 01:03:10,877
really happened in the first I
don't know couple weeks of the Trump

1112
01:03:10,960 --> 01:03:13,037
administration because you know if you

1113
01:03:13,120 --> 01:03:18,317
asked any of these people what they
thought of um of China during this time

1114
01:03:18,400 --> 01:03:21,757
when they were pushing all these
regulations they basically you know and

1115
01:03:21,840 --> 01:03:25,037
specifically well wait if if we shoot
ourselves in the foot by overregulating

1116
01:03:25,120 --> 01:03:28,877
AI you know Won't China just win the AI
race? If you were to ask them that, what

1117
01:03:28,960 --> 01:03:32,637
they what they would have said, they did
say is that China is so far behind us,

1118
01:03:32,720 --> 01:03:36,237
it doesn't matter. And furthermore,
uh, and this was said completely without

1119
01:03:36,320 --> 01:03:40,637
evidence that if we basically slow
down to impose, you know, all these

1120
01:03:40,720 --> 01:03:43,917
supposedly healthy regulations, well,
China will just copy us and do the same

1121
01:03:44,000 --> 01:03:48,717
thing. I think it was an absurdly naive
view. I think that if we shoot ourselves

1122
01:03:48,800 --> 01:03:51,517
in the foot, China will just be like,
"Thank you very much. We'll just

1123
01:03:51,600 --> 01:03:55,597
>> take leadership in this technology."
Uh, why wouldn't we? Um but this is what

1124
01:03:55,680 --> 01:03:57,997
they said and um and you know when when

1125
01:03:58,080 --> 01:04:00,797
when the Biden executive order on AI was

1126
01:04:00,880 --> 01:04:03,917
um crafted there was no discussion

1127
01:04:04,000 --> 01:04:08,397
whatsoever of the China compete.
You know it was just it was just again

1128
01:04:08,480 --> 01:04:12,877
assumed that we were so far ahead that
um that we could basically do anything

1129
01:04:12,960 --> 01:04:16,237
to our companies and it would just be
it wouldn't it wouldn't really affect our

1130
01:04:16,320 --> 01:04:20,237
our um competitiveness.
And I think I think that narrative

1131
01:04:20,320 --> 01:04:24,477
really started to fall apart with um
DeepSeek at the model level. Uh back in

1132
01:04:24,560 --> 01:04:28,877
April, uh Huawei launched a technology
called Cloud Matrix in which they

1133
01:04:28,960 --> 01:04:32,557
compensated for the fact that their
chips individually are not as good as

1134
01:04:32,640 --> 01:04:35,837
Nvidia's chips by networking more of

1135
01:04:35,920 --> 01:04:38,957
them together. So they took 384 of them.

1136
01:04:39,040 --> 01:04:41,357
You use their prowess in networking to

1137
01:04:41,440 --> 01:04:46,237
create you know this rack system cloud
matrix. And it was demonstrated to show

1138
01:04:46,320 --> 01:04:49,277
that, you know, yes, Nvidia chips
are better. They're much more power

1139
01:04:49,360 --> 01:04:52,477
efficient, but at the rack level, at the

1140
01:04:52,560 --> 01:04:55,037
system level, you know, um, Huawei could

1141
01:04:55,120 --> 01:04:59,517
get the job done with these, you know,
Ascend chips and cloud matrix. And so,

1142
01:04:59,600 --> 01:05:02,317
again, I think that showed that, you know,
we're not the only game in town on

1143
01:05:02,400 --> 01:05:05,997
chips, which means that if we don't sell
our chips to, you know, our our friends

1144
01:05:06,080 --> 01:05:11,357
and allies in the Middle East and other
places, then Huawei certainly will. Uh

1145
01:05:11,440 --> 01:05:14,317
so I think it's just been kind of one
revelation after another in which we've

1146
01:05:14,400 --> 01:05:18,157
learned that um that a lot of their
preconceptions and belief were wrong.

1147
01:05:18,240 --> 01:05:21,757
And we've talked about the fact that um
that the that the markets ended up being

1148
01:05:21,840 --> 01:05:25,277
much more decentralized than they ever
could have predicted. And and I'd also

1149
01:05:25,360 --> 01:05:28,237
say one other thing which is they they

1150
01:05:28,320 --> 01:05:33,117
um they also believe that there'd be
you know um imminent catastrophes that

1151
01:05:33,200 --> 01:05:35,677
haven't so this is kind of like
the equivalent to the global warming thing

1152
01:05:35,760 --> 01:05:38,957
where we're all supposed to be
underwater by now. They they were saying

1153
01:05:39,040 --> 01:05:43,677
that models trained on I think I don't
know like 10 to 25 flops or whatever

1154
01:05:43,760 --> 01:05:45,837
were like way too risky. Well, every

1155
01:05:45,920 --> 01:05:51,197
single model now at at the frontier is
trained on that level of compute. And so

1156
01:05:51,280 --> 01:05:54,877
they would have banned us from even
being at the place we're at today if we

1157
01:05:54,960 --> 01:05:59,757
had listened to these people back in 2023.
So just a couple years ago. So

1158
01:05:59,840 --> 01:06:02,717
that's like really important to keep
in mind that their predictions of imminent

1159
01:06:02,800 --> 01:06:04,637
catastrophe have already been refuted.

1160
01:06:04,720 --> 01:06:09,117
And so things are moving in a direction
that um that I think are very different

1161
01:06:09,200 --> 01:06:12,397
than you know what they thought in you
know let's call it the first year after

1162
01:06:12,480 --> 01:06:16,317
the launch of Chachi BBT. >> Right.
So David just to uh come back

1163
01:06:16,400 --> 01:06:20,557
real quick while we still have you on on
crypto. Um so the administration um and

1164
01:06:20,640 --> 01:06:23,597
I think the country had a significant
victory uh earlier this year with the

1165
01:06:23,680 --> 01:06:27,357
president signing the u
stable coin bill into law

1166
01:06:27,440 --> 01:06:31,437
>> uh which was the the genius act. Um,
and I I think that I'll just tell you what

1167
01:06:31,520 --> 01:06:35,117
we see is like the positive consequences
of that law have been even bigger than

1168
01:06:35,200 --> 01:06:39,597
we thought. Um, and I would say that's
both for the stable coin industry and

1169
01:06:39,680 --> 01:06:42,797
you now see actually a lot of a lot
of financial institutions of all kinds

1170
01:06:42,880 --> 01:06:46,317
embracing stable coins um, in a way
that that they weren't before. Um, and you

1171
01:06:46,400 --> 01:06:48,717
know, sort of the the phenomenon
spreading in America, you know, by the

1172
01:06:48,800 --> 01:06:51,757
way, doing, you know, being being in the
lead and doing very well there. Um, but

1173
01:06:51,840 --> 01:06:55,117
but but just also more broadly just as
a signal to the crypto industry that like

1174
01:06:55,200 --> 01:06:58,397
this, you know, this really is a,
you know, this really is a new day. um and

1175
01:06:58,480 --> 01:07:00,877
they're really there really are going
to be regulatory frameworks that that make

1176
01:07:00,960 --> 01:07:03,677
these things possible and and you know
that that are responsible but also make

1177
01:07:03,760 --> 01:07:07,597
this industry really possible to flourish
um and in the US um as you know

1178
01:07:07,680 --> 01:07:11,117
there's a second piece of legislation
um you know being constructed right now

1179
01:07:11,200 --> 01:07:14,797
which is the market structure bill
called the clarity act um which is sort

1180
01:07:14,880 --> 01:07:18,317
of the phase two of the legislative
agenda um and I wondered maybe if you

1181
01:07:18,400 --> 01:07:20,877
could just tell us a little bit about
your view of of the importance of that

1182
01:07:20,960 --> 01:07:24,477
of that bill and and then you know kind
of how do you think that process is going

1183
01:07:24,560 --> 01:07:29,197
>> I think it's extremely important um So,
as you mentioned, we we passed the the

1184
01:07:29,280 --> 01:07:33,117
Genius Act few few months ago, but that
was just for stable coins. Stable coins

1185
01:07:33,200 --> 01:07:35,437
are about 6% of the total market cap in

1186
01:07:35,520 --> 01:07:38,397
terms of tokens. So, 94% are all the

1187
01:07:38,480 --> 01:07:42,637
other types of tokens. And the Clarity
Act would apply to all of that and

1188
01:07:42,720 --> 01:07:44,797
provide the regulatory framework for all

1189
01:07:44,880 --> 01:07:46,557
those other crypto projects and

1190
01:07:46,640 --> 01:07:50,557
companies. Um, you know, if we could be

1191
01:07:50,640 --> 01:07:55,117
sure that, you know, currently we have
a great SEC chairman, Paul Atkins, and if

1192
01:07:55,200 --> 01:07:58,637
we could be sure that Paul Atkins and
a person like Paul Atkins was always at

1193
01:07:58,720 --> 01:08:02,157
the SEC forever, that we wouldn't
necessarily need legislation because

1194
01:08:02,240 --> 01:08:05,117
they're already in the process
of implementing like much better rules and

1195
01:08:05,200 --> 01:08:08,557
providing regulatory clarity. But the
truth is that we don't know for sure.

1196
01:08:08,640 --> 01:08:12,717
And if you're a founder who's trying
to make a decision now about where you're

1197
01:08:12,800 --> 01:08:15,757
going to build your company, you want
to have certainty for 10 years out, 20

1198
01:08:15,840 --> 01:08:20,317
years out. uh you know we want
to encourage long-term projects and so

1199
01:08:20,400 --> 01:08:23,837
again I think it's very
important to pro to canonize

1200
01:08:23,920 --> 01:08:27,757
uh the the the rules that that's first
to provide the clarity and then to make

1201
01:08:27,840 --> 01:08:32,397
sure there's enough stability around
them and sort of canonize those rules in

1202
01:08:32,480 --> 01:08:35,757
legislation. That's the only way that you
provide that long-term stability. I

1203
01:08:35,840 --> 01:08:38,797
think that we will get the clarity act
done. Like you mentioned it passed the

1204
01:08:38,880 --> 01:08:43,197
house with about 300 votes of about
78 Democrats. So it was substantially

1205
01:08:43,280 --> 01:08:47,037
bipartisan. I think it will ultimately
we're it's now going through the Senate.

1206
01:08:47,120 --> 01:08:50,157
Um I think it will ultimately get done.

1207
01:08:50,240 --> 01:08:53,357
Uh we're negotiating with a dozen or so

1208
01:08:53,440 --> 01:08:55,197
Democrat. We have to get to 60 votes. So

1209
01:08:55,280 --> 01:09:00,477
that's the the hard part is under
the filibuster. We got to get to 60. Uh so

1210
01:09:00,560 --> 01:09:03,757
but we're negotiating with about a dozen
Democrats. And I I do think that we will

1211
01:09:03,840 --> 01:09:06,317
ultimately get to to that number. We by

1212
01:09:06,400 --> 01:09:09,277
the way we ended up having 68 uh votes

1213
01:09:09,360 --> 01:09:14,157
in the Senate for genius uh including
18 Democrats. So, I do think that even if

1214
01:09:14,240 --> 01:09:17,357
we just get, you know,
twothirds of the number of

1215
01:09:17,440 --> 01:09:20,557
Democrats that we got for Genius, then,
you know, we'll we'll be we'll be fine

1216
01:09:20,640 --> 01:09:25,357
on on clarity. But I, you know,
this will provide the the regulatory

1217
01:09:25,440 --> 01:09:28,797
framework again for all the other tokens
besides stable coins. And um I think

1218
01:09:28,880 --> 01:09:32,637
it's just a critical
piece of of legislation.

1219
01:09:32,720 --> 01:09:36,957
And uh yeah, this this would ultimately
I think kind of complete the crypto

1220
01:09:37,040 --> 01:09:41,277
agenda where we've kind of you know
moving from you know Biden's war on

1221
01:09:41,360 --> 01:09:46,237
crypto to Trump's crypto capital the
planet. Um and then you know I think the

1222
01:09:46,320 --> 01:09:50,397
industry will have the stability it
needs and can just focus on innovating and

1223
01:09:50,480 --> 01:09:53,037
>> there'll be you know rule updates
and things like that but you know we'll

1224
01:09:53,120 --> 01:09:56,477
fundamentally have the foundation
for the industry in place.

1225
01:09:56,560 --> 01:10:01,677
>> On Genius Act um you know President
Trump really made that bill possible. I

1226
01:10:01,760 --> 01:10:05,437
mean, first of all, it was his election
that completely shifted the conversation

1227
01:10:05,520 --> 01:10:10,237
>> on crypto. We would still be, you know,
if a different result had been reached,

1228
01:10:10,320 --> 01:10:14,237
we would have again like figure out the
SEC. The founders would still be getting

1229
01:10:14,320 --> 01:10:16,957
prosecuted. We wouldn't know what
the rules are. Elizabeth Warren would be

1230
01:10:17,040 --> 01:10:20,237
calling the shots. So, President Trump's
election made everything possible. And

1231
01:10:20,320 --> 01:10:23,277
it's his commitment to the industry and

1232
01:10:23,360 --> 01:10:27,837
he and his commitment to keeping his um
promises during the election that's made

1233
01:10:27,920 --> 01:10:31,917
all of this possible. But also I mean he
got directly involved in making sure the

1234
01:10:32,000 --> 01:10:35,117
Genius Act passed. It was the
legislation was declared dead many

1235
01:10:35,200 --> 01:10:38,237
times. Uh I saw it with my own eyes that

1236
01:10:38,320 --> 01:10:43,677
you know he was able to persuade um
Rickstrrent votes and uh twist arms

1237
01:10:43,760 --> 01:10:45,197
cajul

1238
01:10:45,280 --> 01:10:48,557
uh and um charm and um you know he

1239
01:10:48,640 --> 01:10:53,037
ultimately got it done and um and I
think that clarity will be a similar

1240
01:10:53,120 --> 01:10:58,157
result. people are always prematurely
declaring these things to be dead or

1241
01:10:58,240 --> 01:11:01,517
whatever. Um there are a lot of twists
and turns in the legislative process.

1242
01:11:01,600 --> 01:11:04,157
It's definitely true that you don't want
to see the sausage getting made, but um

1243
01:11:04,240 --> 01:11:06,637
but anyway, I think we're
on a good track right now.

1244
01:11:06,720 --> 01:11:10,957
>> Good. Fantastic. >> Great.
>> Pete Pete Buddhajes went on all in

1245
01:11:11,040 --> 01:11:15,197
recently and you guys talked about the
the left's identity crisis and you he's

1246
01:11:15,280 --> 01:11:19,357
hoping for more of a moderate, you know,
center center left um to to to emerge.

1247
01:11:19,440 --> 01:11:22,877
At the same time, we see Mumani in New
York. I'm I'm curious what you you think

1248
01:11:22,960 --> 01:11:26,077
of what are you seeing in terms of what
is the future in terms of for the for

1249
01:11:26,160 --> 01:11:29,517
the Democratic party in terms of it
is there a more moderate presence or is it

1250
01:11:29,600 --> 01:11:33,197
kind of this Mani style
you know woke populism?

1251
01:11:33,280 --> 01:11:35,197
>> I mean it certainly seems to me that

1252
01:11:35,280 --> 01:11:40,557
Mani and and um I don't know like the woke
socialism seems to be the future of

1253
01:11:40,640 --> 01:11:43,997
the party. I mean that's where all the
energy is and their base. I mean I don't

1254
01:11:44,080 --> 01:11:48,237
want that to be the case. I'd rather
have a rational Democrat party, but um

1255
01:11:48,320 --> 01:11:50,317
but that seems to be where their base

1256
01:11:50,400 --> 01:11:53,197
is, where the energy is. And um you

1257
01:11:53,280 --> 01:11:56,077
don't really hear um Democrats um within

1258
01:11:56,160 --> 01:12:00,237
the party trying to self police
and distance themselves from that. Uh you

1259
01:12:00,320 --> 01:12:04,317
saw I mean all the major figures in the
Democrat party have endorsed Mammi. So

1260
01:12:04,400 --> 01:12:05,517
yeah, I mean that's where that party

1261
01:12:05,600 --> 01:12:10,397
seems to be headed. Um I I I think that

1262
01:12:10,480 --> 01:12:13,277
um partly it's where their base is at. I

1263
01:12:13,360 --> 01:12:16,717
think partly it it might be um a a

1264
01:12:16,800 --> 01:12:20,237
misread of of uh it sort of it could be

1265
01:12:20,320 --> 01:12:25,837
kind of like a partial reaction to to
Trump where they feel like um you know

1266
01:12:25,920 --> 01:12:28,957
establishment politics has kind
of failed and so they need a populism of

1267
01:12:29,040 --> 01:12:33,117
the left to compete with a populism
of the right and so I think that's maybe

1268
01:12:33,200 --> 01:12:37,757
part of the calculation for why they're
going this direction but I don't you

1269
01:12:37,840 --> 01:12:40,477
know fundamentally I don't think it works.
I don't think socialism works. I

1270
01:12:40,560 --> 01:12:42,717
don't think the um you know, defund the

1271
01:12:42,800 --> 01:12:44,397
police, empty all the jails policies

1272
01:12:44,480 --> 01:12:47,837
work. Um so, you know, I think we're

1273
01:12:47,920 --> 01:12:51,597
about to get another um you know, uh

1274
01:12:51,680 --> 01:12:54,397
case uh a teaching moment in New York.

1275
01:12:54,480 --> 01:12:55,677
Unfortunately, it's not going to be good

1276
01:12:55,760 --> 01:12:58,957
for the city. Uh but, uh you know, we

1277
01:12:59,040 --> 01:13:03,117
we've seen this movie before, but uh
but yeah, that's that's where I mean, it

1278
01:13:03,200 --> 01:13:05,997
does appear that's where the Democrat
party is. Um I don't completely get it.

1279
01:13:06,080 --> 01:13:07,677
I mean, other people have made this

1280
01:13:07,760 --> 01:13:09,277
observation, but they do seem to be on

1281
01:13:09,360 --> 01:13:14,357
the 20% side of every 8020 issue. Um,

1282
01:13:14,440 --> 01:13:15,197
[snorts]

1283
01:13:15,280 --> 01:13:19,357
uh, you know, opening the border, um,

1284
01:13:19,440 --> 01:13:21,517
you know, on the, um, soft on crime

1285
01:13:21,600 --> 01:13:23,677
stuff, you know, releasing all the the

1286
01:13:23,760 --> 01:13:27,597
repeat offenders and, um,

1287
01:13:27,680 --> 01:13:32,557
and just sort of this, um, you know,
anti- capitalist approach um, you know,

1288
01:13:32,640 --> 01:13:35,917
which I think will be disastrous for the
economy. I mean, this but this is this

1289
01:13:36,000 --> 01:13:39,437
is kind of where the party's at right now.
It is it's a little scary because

1290
01:13:39,520 --> 01:13:43,837
it does mean that if we lose elections
in places where we do lose elections,

1291
01:13:43,920 --> 01:13:47,757
it's like, you know, there you could end
up with something really horrible, not

1292
01:13:47,840 --> 01:13:51,357
just like, you know, we're not just
playing in the 40 yard lines anymore in

1293
01:13:51,440 --> 01:13:55,197
American politics.
Uh and that that is a little bit scary.

1294
01:13:55,280 --> 01:13:58,237
>> Yeah. >> And I I do
think that, you know, if it

1295
01:13:58,320 --> 01:14:01,437
weren't for Donald Trump, I think
in a way we we might already be there. Um,

1296
01:14:01,520 --> 01:14:03,037
you know, I think, you know, but we have

1297
01:14:03,120 --> 01:14:06,717
to make sure that this um

1298
01:14:06,800 --> 01:14:12,397
this um the Trump revolution continues.

1299
01:14:12,480 --> 01:14:17,197
La lastly, um we just talked about New
York. Um recently in an episode all in

1300
01:14:17,280 --> 01:14:21,037
in San Francisco, you you you know
endorsed bringing the the National

1301
01:14:21,120 --> 01:14:24,077
Guard. You know, Ben off had his comments.
He he sort of, you know, went

1302
01:14:24,160 --> 01:14:25,917
back and forth forth in those comments.

1303
01:14:26,000 --> 01:14:28,637
I'm curious if speaking of teaching
moments, I'm curious if you see San

1304
01:14:28,720 --> 01:14:33,677
Francisco as savable in in some sense
and and what what needs to be true to to

1305
01:14:33,760 --> 01:14:36,797
get there if so. >> Well,
Daniel Luri is the best mayor

1306
01:14:36,880 --> 01:14:39,197
we've had in decades. So, I think he's

1307
01:14:39,280 --> 01:14:44,317
doing a very good job um within the
constraints that San Francisco presents.

1308
01:14:44,400 --> 01:14:48,317
Uh so, the the the mayor job,
unfortunately, we have a weak mayor in

1309
01:14:48,400 --> 01:14:51,517
San Francisco. I I don't mean him.
I just mean like the way it's all set up.

1310
01:14:51,600 --> 01:14:53,997
the board of supervisors has, you know,

1311
01:14:54,080 --> 01:14:57,197
a ton of power and um over time they've

1312
01:14:57,280 --> 01:15:02,157
been able to kind of transfer power from
the mayor to to themselves. Um and then

1313
01:15:02,240 --> 01:15:05,597
of course you got all these leftwing
judges. I mean, it's just amazing to me

1314
01:15:05,680 --> 01:15:10,317
that um there's a case right now,
this is a case that galvanized me several

1315
01:15:10,400 --> 01:15:12,317
years ago, the case of Troy Mallister,

1316
01:15:12,400 --> 01:15:15,997
who was a repeat offender who uh killed

1317
01:15:16,080 --> 01:15:17,917
pe two people on New Year's Eve. I think

1318
01:15:18,000 --> 01:15:21,197
it was like uh 2020.

1319
01:15:21,280 --> 01:15:24,397
Uh and um he was arrested four times,

1320
01:15:24,480 --> 01:15:27,677
you know, uh in the year before that. Uh

1321
01:15:27,760 --> 01:15:31,597
he he he ended up killing these two
people and he had a very very long

1322
01:15:31,680 --> 01:15:33,597
criminal history. He had committed armed

1323
01:15:33,680 --> 01:15:37,037
robbery before, stolen many cars.

1324
01:15:37,120 --> 01:15:39,757
Uh and he should have been in jail.
He should not have been released. But he

1325
01:15:39,840 --> 01:15:43,517
was basically released thanks to the
zero bail policies of Chase Bodin, who

1326
01:15:43,600 --> 01:15:47,277
was then the district attorney who we
got recalled. There was a huge outcry. I

1327
01:15:47,360 --> 01:15:49,197
mean, even in San Francisco for there to

1328
01:15:49,280 --> 01:15:51,677
be a recall of a polit I mean, you got

1329
01:15:51,760 --> 01:15:54,557
to be like seriously leftwing to um

1330
01:15:54,640 --> 01:15:56,237
>> to basically alienate San Francisco and

1331
01:15:56,320 --> 01:15:59,357
Chase managed to be so far out there

1332
01:15:59,440 --> 01:16:01,997
that he alienated even San Francisco.

1333
01:16:02,080 --> 01:16:04,877
And yet, I don't know why Tory Mallister

1334
01:16:04,960 --> 01:16:09,277
isn't sentenced already and in jail
for 20 years plus, but his case is still

1335
01:16:09,360 --> 01:16:13,357
pending through the courts, never ending.
And there's a leftwing judge

1336
01:16:13,440 --> 01:16:16,557
who's considering just giving him
diversion. basically means you just get

1337
01:16:16,640 --> 01:16:21,677
released maybe with an ankle bracelet
or something. Um that's insane. So I mean

1338
01:16:21,760 --> 01:16:24,317
that's what we're dealing with in San
Francisco. I mean like crazy leftwing

1339
01:16:24,400 --> 01:16:25,357
judges who want to release all the

1340
01:16:25,440 --> 01:16:29,757
criminals. Uh and um you know and so I

1341
01:16:29,840 --> 01:16:33,757
just wonder like is Daniel up against
too many constraints and therefore I

1342
01:16:33,840 --> 01:16:36,877
know he doesn't want the president
to send in the National Guard but maybe

1343
01:16:36,960 --> 01:16:41,757
ultimately it would be helpful. Um but
in any event, I think the president has

1344
01:16:41,840 --> 01:16:46,637
has agreed to kind of hold
off on that um out of you,

1345
01:16:46,720 --> 01:16:49,597
you know, Daniel had a good conversation
with the president and asked him to hold

1346
01:16:49,680 --> 01:16:52,237
back and um and you know, the president

1347
01:16:52,320 --> 01:16:54,877
uh agreed and is giving him time to

1348
01:16:54,960 --> 01:16:57,997
implement his solutions. And look, if if

1349
01:16:58,080 --> 01:17:03,197
um if Daniel and his team can keep
making progress and fix the problems

1350
01:17:03,280 --> 01:17:06,477
without the National Guard having
to come in, then so much the better. Um,

1351
01:17:06,560 --> 01:17:09,517
we'll just see if and and I know he
wants to and like I said, he's the best

1352
01:17:09,600 --> 01:17:13,277
mayor we've had in decades. It's just
a question of whether he'll be too

1353
01:17:13,360 --> 01:17:17,117
constrained by the other
powers that be in the city.

1354
01:17:17,200 --> 01:17:18,797
>> David, thank you so much
for coming on the podcast.

1355
01:17:18,880 --> 01:17:21,917
>> Yeah. >> Fantastic. Thank you,

1356
01:17:22,000 --> 01:17:25,357
>> and thank you for the work. It's we we

1357
01:17:25,440 --> 01:17:29,437
as much as anybody appreciate the work
that you've done to uh fix the things in

1358
01:17:29,520 --> 01:17:32,877
the past and and put us on a
great road to the future. Yep.

1359
01:17:32,960 --> 01:17:36,477
>> Well, thanks. And I appreciate what you
guys have done as well. So, thank you for

1360
01:17:36,560 --> 01:17:39,517
for your support, everything you're doing.
So, >> yeah, >> I appreciate it.

1361
01:17:39,600 --> 01:17:43,710
>> Definitely. [music]

1362
01:17:57,410 --> 01:17:59,430
>> [music]