1
00:00:01,360 --> 00:00:04,957
Very well. Um, welcome everybody.

2
00:00:05,040 --> 00:00:08,957
Welcome to Journal House and a big
welcome to to our audiences that are

3
00:00:09,040 --> 00:00:13,757
joining us online. But above all, a big
welcome to Dario Amade, the chief exec

4
00:00:13,840 --> 00:00:16,077
of Anthropic.
>> And thank you for having me.

5
00:00:16,160 --> 00:00:19,197
>> Not at all. So Dario, um, we're at

6
00:00:19,280 --> 00:00:23,917
Davos. There's a lot going on, but I
wanted to start with a a big picture

7
00:00:24,000 --> 00:00:25,917
question, which I'll characterize like

8
00:00:26,000 --> 00:00:29,357
this. It feels to me that this time last

9
00:00:29,440 --> 00:00:33,117
year, everybody was very excited about
AI and everyone was talking about what

10
00:00:33,200 --> 00:00:37,677
AI can do, its potential,
its capabilities. It feels to me as though

11
00:00:37,760 --> 00:00:39,437
the debate has shifted somewhat this

12
00:00:39,520 --> 00:00:42,397
year to be more less what can AI do to

13
00:00:42,480 --> 00:00:45,837
what is AI doing to the world. Um, and I

14
00:00:45,920 --> 00:00:49,517
I know that you think a lot about these
things. So my question is do you think

15
00:00:49,600 --> 00:00:53,437
businesses, policy makers, governments,
whatever are doing enough to prepare for

16
00:00:53,520 --> 00:00:54,397
the impact?

17
00:00:54,480 --> 00:00:57,917
>> No. Um I'll explain the longer version

18
00:00:58,000 --> 00:01:01,517
now. um uh you know I've been watching

19
00:01:01,600 --> 00:01:04,797
this field for 15 years and I've been in

20
00:01:04,880 --> 00:01:07,597
this field for 10 years and and one of

21
00:01:07,680 --> 00:01:09,997
the things I've most noticed is that

22
00:01:10,080 --> 00:01:14,760
there's been a a surprisingly

23
00:01:16,400 --> 00:01:19,437
uh h on on a a uh you know the same

24
00:01:19,520 --> 00:01:23,357
trajectory whereas the kind
of public opinion and

25
00:01:23,440 --> 00:01:27,037
the reaction of the public has
oscillated wildly. I would say that in

26
00:01:27,120 --> 00:01:31,277
in in two different ways. One is the
capabilities of the technology. Every

27
00:01:31,360 --> 00:01:35,837
three to six months we have this reversal
of of polarity where the media

28
00:01:35,920 --> 00:01:39,357
is incredibly excited about what the
technology can do. It's going to change

29
00:01:39,440 --> 00:01:42,877
everything and then it's you know it's
all a bubble. It's all it's all going to

30
00:01:42,960 --> 00:01:45,277
fall apart. And and what I see is this

31
00:01:45,360 --> 00:01:48,637
smooth exponential line where similar to

32
00:01:48,720 --> 00:01:53,357
Moors law for compute, we basically have
a Moore's law for intelligence where the

33
00:01:53,440 --> 00:01:56,717
model is getting more and more
cognitively capable every every few

34
00:01:56,800 --> 00:02:01,197
months. And that that march has just
been constant. The up and down the we

35
00:02:01,280 --> 00:02:05,357
invented a new thing. It's all going
to crash. It's hitting a wall. It's it's

36
00:02:05,440 --> 00:02:10,077
going to go crazy. That that is a public
perception phenomenon. That's on the

37
00:02:10,160 --> 00:02:14,477
capability of the technology. I think
there's a similar thing on the polarity

38
00:02:14,560 --> 00:02:16,317
of whether the technology is good or

39
00:02:16,400 --> 00:02:22,717
bad. Um uh you know in 2023 and 2024

40
00:02:22,800 --> 00:02:27,117
there was a lot of concern about AI
right there was you know concern AI you

41
00:02:27,200 --> 00:02:29,277
know AI is going to AI are going to take

42
00:02:29,360 --> 00:02:31,677
over. There was a lot of talk about AI

43
00:02:31,760 --> 00:02:35,277
risk AI misuse. Then in 2025 the

44
00:02:35,360 --> 00:02:38,797
political wind shifted as you say to AI

45
00:02:38,880 --> 00:02:41,197
opportunity and now they're sort of

46
00:02:41,280 --> 00:02:43,757
shifting back and and I think throughout

47
00:02:43,840 --> 00:02:48,397
throughout all of this the the approach
that I have tried to take and the

48
00:02:48,480 --> 00:02:49,917
approach that anthropic has tried to

49
00:02:50,000 --> 00:02:52,797
take is is is one of constancy of saying

50
00:02:52,880 --> 00:02:56,157
that there is balance here uh and and uh

51
00:02:56,240 --> 00:02:58,317
balance of a very strange form because I

52
00:02:58,400 --> 00:03:01,677
think the technology is very extreme in

53
00:03:01,760 --> 00:03:05,917
in what it's capable apable of doing
but I think it's positive impacts and its

54
00:03:06,000 --> 00:03:09,997
negative impacts that you know they
they both exist right I wrote this essay

55
00:03:10,080 --> 00:03:13,357
machines of loving grace about a year
and a half ago it had a very radical

56
00:03:13,440 --> 00:03:15,437
view of the upside of AI that you know

57
00:03:15,520 --> 00:03:17,677
it would it would help us to you know

58
00:03:17,760 --> 00:03:21,037
cure cancer eradicate tropical diseases

59
00:03:21,120 --> 00:03:25,437
you know kind of bring bring economic
development to you know parts of the

60
00:03:25,520 --> 00:03:28,957
world that haven't seen it and I my view
hasn't changed I believe all of those

61
00:03:29,040 --> 00:03:32,317
things um but the the other side of it

62
00:03:32,400 --> 00:03:35,677
which you know I'm I'm I'm now uh uh uh

63
00:03:35,760 --> 00:03:37,277
you know kind of writing writing more

64
00:03:37,360 --> 00:03:39,597
about and and you know may may release

65
00:03:39,680 --> 00:03:42,877
something about soon is yeah there you

66
00:03:42,960 --> 00:03:45,437
know bad things will happen as well if

67
00:03:45,520 --> 00:03:47,677
we just take just for an example as as

68
00:03:47,760 --> 00:03:49,917
one of the risks we take the economic

69
00:03:50,000 --> 00:03:53,757
side of it um my my view is the

70
00:03:53,840 --> 00:03:56,237
signature of this technology is it's

71
00:03:56,320 --> 00:03:57,837
going to take us to a world where we

72
00:03:57,920 --> 00:04:01,277
have very high GDP growth and

73
00:04:01,360 --> 00:04:04,637
potentially also very high unemployment

74
00:04:04,720 --> 00:04:08,717
and inequality. Now, that's not
a combination we've we've almost ever seen

75
00:04:08,800 --> 00:04:12,877
before, right? You you think of it as high
GDP growth, that's lots of stuff to

76
00:04:12,960 --> 00:04:16,717
do, lots of jobs for everyone. It's
always been like that in the past. We've

77
00:04:16,800 --> 00:04:18,877
never had a technology that's this

78
00:04:18,960 --> 00:04:20,717
disruptive. So, the idea that we could

79
00:04:20,800 --> 00:04:23,997
have five or 10% GDP growth, but also,

80
00:04:24,080 --> 00:04:27,037
you know, 10 10% unemployment, it's it's

81
00:04:27,120 --> 00:04:29,357
not logically inconsistent at all. It's

82
00:04:29,440 --> 00:04:31,277
just never happened that that that that

83
00:04:31,360 --> 00:04:34,157
way before and I'm I'm really quite uh

84
00:04:34,240 --> 00:04:36,797
you know for those both reasons excited

85
00:04:36,880 --> 00:04:39,197
and worried. If I take an example

86
00:04:39,280 --> 00:04:43,277
something like AI coding um you know the

87
00:04:43,360 --> 00:04:47,437
latest model release claude opus 4.5 um

88
00:04:47,520 --> 00:04:50,157
I have some uh uh engineers some

89
00:04:50,240 --> 00:04:53,837
engineering leads within enthropic who

90
00:04:53,920 --> 00:04:58,557
have basically said to me I don't write
any code anymore. I I you know I just

91
00:04:58,640 --> 00:05:02,317
let Opus do the work and I edit it.
We just released a new thing called Claude

92
00:05:02,400 --> 00:05:04,557
Co-work. We can go into go you know we

93
00:05:04,640 --> 00:05:07,597
can go into that later but this was a a

94
00:05:07,680 --> 00:05:09,677
version of our tool cla code for

95
00:05:09,760 --> 00:05:14,797
non-coding. This was built in a week and
a half almost entirely with Claude Opus.

96
00:05:14,880 --> 00:05:17,677
There are still things for the software
engineers to do right. It's like even if

97
00:05:17,760 --> 00:05:21,437
the software engineers are only doing
10% of it they can you know they still

98
00:05:21,520 --> 00:05:25,037
have a job to do or they can take
a level up that's not going to last

99
00:05:25,120 --> 00:05:29,597
forever. the models are going to do more
and more and so there's an incredible

100
00:05:29,680 --> 00:05:32,797
you this is a microcosm you can
see there's an incredible amount of

101
00:05:32,880 --> 00:05:34,637
productivity here software is going to

102
00:05:34,720 --> 00:05:37,997
become cheap maybe essentially free the

103
00:05:38,080 --> 00:05:41,437
premise that you need to amortize a

104
00:05:41,520 --> 00:05:45,197
piece of software you build across
millions of users that may start to be

105
00:05:45,280 --> 00:05:48,317
false like for this meeting it might

106
00:05:48,400 --> 00:05:52,397
cost a few cents to just say I don't
know let's let's let's make some app so

107
00:05:52,480 --> 00:05:53,917
people can talk to each other or you

108
00:05:54,000 --> 00:05:57,037
know I it just it it just may be very

109
00:05:57,120 --> 00:05:59,597
flexible and recyclable but but at the

110
00:05:59,680 --> 00:06:01,997
same time there are whole jobs whole

111
00:06:02,080 --> 00:06:06,637
careers that we built built for decades
that that may not be be present and you

112
00:06:06,720 --> 00:06:10,717
know I think we can deal with it.
I think we can we can adjust to it but I

113
00:06:10,800 --> 00:06:16,077
don't I don't think there's an awareness
at all of what of what is coming here

114
00:06:16,160 --> 00:06:19,677
and the magnitude of it.
>> So how that's it's so interesting when

115
00:06:19,760 --> 00:06:21,917
you say that. So how do you think in a

116
00:06:22,000 --> 00:06:25,357
world of of high GDP growth but also

117
00:06:25,440 --> 00:06:29,197
high unemployment you know what does
that do to society and you say there are

118
00:06:29,280 --> 00:06:32,397
you know people aren't thinking about
it now can you give concrete examples of

119
00:06:32,480 --> 00:06:36,717
how society might organize
itself to adapt to such a world

120
00:06:36,800 --> 00:06:39,037
>> yeah so you know I
think I think there's

121
00:06:39,120 --> 00:06:43,837
a few things the the first thing that
that we've done that we've focused on

122
00:06:43,920 --> 00:06:48,557
and this is not a solution so much as it
is a first step is we have this thing

123
00:06:48,640 --> 00:06:52,237
called the anthropic economic index index.
We've had it for about a year.

124
00:06:52,320 --> 00:06:56,477
We've updated it, I think, four or five
times now. Uh, and what that does is

125
00:06:56,560 --> 00:06:59,197
it's a real-time index that lets you

126
00:06:59,280 --> 00:07:03,837
track um, you know, what our model
Claude is being used for. It goes across

127
00:07:03,920 --> 00:07:08,877
all the conversations and kind of uses
Claude in a privacy preserving way to

128
00:07:08,960 --> 00:07:12,797
you know kind of kind of statistically
query how Claude is being used. What are

129
00:07:12,880 --> 00:07:14,717
the tasks it's being used for? To what

130
00:07:14,800 --> 00:07:17,757
extent it is is it automating versus

131
00:07:17,840 --> 00:07:22,237
augmenting tasks? um what industries
is it being used in? How is it diffusing

132
00:07:22,320 --> 00:07:26,717
across states in the United States
and countries countries in the world? We

133
00:07:26,800 --> 00:07:31,357
we've just added kind of more and more
detail here and and and my view is until

134
00:07:31,440 --> 00:07:35,837
we can measure the shape of this economic
transition. Any policy is going

135
00:07:35,920 --> 00:07:37,757
to be blind and and misinformed, right?

136
00:07:37,840 --> 00:07:40,637
Many many policies have gone wrong
because they're they're based on

137
00:07:40,720 --> 00:07:42,397
premises that are that are fundamentally

138
00:07:42,480 --> 00:07:45,837
incorrect. So So that's that's step one.

139
00:07:45,920 --> 00:07:47,757
Step two is I think we need to think

140
00:07:47,840 --> 00:07:52,877
very carefully about how do we allow
people to adapt right people can adapt

141
00:07:52,960 --> 00:07:57,277
more quickly or they can adapt more
slowly. This can mean adapting to use

142
00:07:57,360 --> 00:07:59,757
the technology within existing jobs.

143
00:07:59,840 --> 00:08:03,197
This can mean adapting you know from one

144
00:08:03,280 --> 00:08:05,837
job to another job. For example, I think

145
00:08:05,920 --> 00:08:10,397
there are probably going to be more jobs
in the physical world and less jobs in

146
00:08:10,480 --> 00:08:14,397
the knowledge work economy. Now maybe
eventually robotics you know makes

147
00:08:14,480 --> 00:08:18,317
progress but I think that's on a that's
on a slower trajectory. So so that's

148
00:08:18,400 --> 00:08:21,277
one. Are there jobs that have you know

149
00:08:21,360 --> 00:08:25,917
still kind of you know really value
a human touch? Um some of them do some of

150
00:08:26,000 --> 00:08:29,437
them don't. We may find out how
important how important that is in the

151
00:08:29,520 --> 00:08:33,437
market and where it's most important
at the level of companies. What are the

152
00:08:33,520 --> 00:08:37,357
moes when software becomes cheap and then
subsequently the rest of knowledge

153
00:08:37,440 --> 00:08:40,797
work becomes cheap? We don't know.
We've never quite asked that question. and

154
00:08:40,880 --> 00:08:43,997
we've thought about modes in a certain
way. So there's there's going to be a

155
00:08:44,080 --> 00:08:48,397
huge scramble at the level of at the level
of companies. So you know teaching

156
00:08:48,480 --> 00:08:52,637
people to adapt, teaching them what
to expect, I think is the second step. And

157
00:08:52,720 --> 00:08:54,957
and the third step is I think there's

158
00:08:55,040 --> 00:08:56,877
there's going to need to be some role

159
00:08:56,960 --> 00:08:59,677
for government in in a displacement

160
00:08:59,760 --> 00:09:03,997
that's that's this macroeconomically
large. I just I just don't see how I

161
00:09:04,080 --> 00:09:07,997
don't see how it doesn't happen. The pie
is going to grow much larger, right?

162
00:09:08,080 --> 00:09:11,997
like the money is going to be there like
you know we may you know the budget may

163
00:09:12,080 --> 00:09:16,637
balance without us doing anything
because there's so much growth um uh the

164
00:09:16,720 --> 00:09:20,557
issue is distributing it to the right
people and so and so I think I think

165
00:09:20,640 --> 00:09:24,397
this is probably a time to worry less
about disincentivizing growth and and

166
00:09:24,480 --> 00:09:27,357
worry more about making making sure that

167
00:09:27,440 --> 00:09:31,517
everyone gets gets gets a part of that
growth um which which you know which I

168
00:09:31,600 --> 00:09:35,037
know is the opposite of the prevailing
sentiment now but I think technological

169
00:09:35,120 --> 00:09:39,597
reality is about to change in a way
that forces our ideas to change.

170
00:09:39,680 --> 00:09:43,437
>> So obviously in your in your desire
to create this greater sense of urgency,

171
00:09:43,520 --> 00:09:46,877
are you speaking to people in the
administration? I mean Anthropic hasn't

172
00:09:46,960 --> 00:09:50,317
always been the sort of first on the
guest list for this administration, but

173
00:09:50,400 --> 00:09:53,117
do do you have people there
that you're talking to?

174
00:09:53,200 --> 00:09:55,917
>> I've I I I have said it to them myself.

175
00:09:56,000 --> 00:09:59,277
Um uh and you know to to be clear, there

176
00:09:59,360 --> 00:10:03,117
are plenty of things there are plenty
of things we agree on, right? You know, I

177
00:10:03,200 --> 00:10:05,117
think the AI action plan that the

178
00:10:05,200 --> 00:10:07,997
administration put out in the middle of

179
00:10:08,080 --> 00:10:12,477
this year actually had some some,
you know, some very good ideas here. You

180
00:10:12,560 --> 00:10:16,637
know, I think I think, you know,
we probably probably agreed with the vast

181
00:10:16,720 --> 00:10:21,197
vast majority of it. But I think most of
all, we just want to say these things in

182
00:10:21,280 --> 00:10:24,637
public and kind of ha and kind of have
a public debate about them, right? We

183
00:10:24,720 --> 00:10:27,037
don't we don't control policy. I think

184
00:10:27,120 --> 00:10:29,597
the most useful thing we can do is

185
00:10:29,680 --> 00:10:32,317
describe to the world what we're seeing

186
00:10:32,400 --> 00:10:34,957
and and provide data to the world and

187
00:10:35,040 --> 00:10:36,957
then and then you know it's it's left to

188
00:10:37,040 --> 00:10:40,077
the the public in in a democracy to to

189
00:10:40,160 --> 00:10:44,637
you know to take that data and and to use
it and to use it to drive policy. We

190
00:10:44,720 --> 00:10:46,157
can't drive policy on our own.

191
00:10:46,240 --> 00:10:47,597
>> Are you going to be
talking to officials

192
00:10:47,680 --> 00:10:50,957
while you're here? Have you have
you been along to USA House yet?

193
00:10:51,040 --> 00:10:53,437
>> Uh I've not I've not been to USA House.

194
00:10:53,520 --> 00:10:57,197
you know I I will I will I I you know
I will be talking to officials during my

195
00:10:57,280 --> 00:11:00,397
during my trip to Davos.

196
00:11:00,480 --> 00:11:02,957
>> Good. So um just to go back to uh

197
00:11:03,040 --> 00:11:07,437
Anthropic then. So you founded Anthropic
um specifically because you were worried

198
00:11:07,520 --> 00:11:11,997
about that you didn't think that OpenAI
was taking safety seriously enough. Now

199
00:11:12,080 --> 00:11:15,037
some people say that you know
the competitive pressures mean that you've

200
00:11:15,120 --> 00:11:19,437
gone more hawkish now. I mean, do are
those competitive pressures, you know,

201
00:11:19,520 --> 00:11:23,357
to keep up with China and keep ahead
of China, all the rest of it, have they do

202
00:11:23,440 --> 00:11:25,917
you think they have compromised
your safety principles?

203
00:11:26,000 --> 00:11:27,997
>> So, we've taken a very different route

204
00:11:28,080 --> 00:11:30,797
than some of the other players have. Uh,

205
00:11:30,880 --> 00:11:32,957
I I think one of the good choices we

206
00:11:33,040 --> 00:11:37,197
made early was to be a company that is

207
00:11:37,280 --> 00:11:39,757
focused on enterprise rather than

208
00:11:39,840 --> 00:11:44,317
consumer. Um, and I think, you know,
it's it's very hard to fight your own

209
00:11:44,400 --> 00:11:48,397
business incentives. it's easier
to choose a business model where there's

210
00:11:48,480 --> 00:11:52,877
less need to fight your own business
incentives. So, you know, I have a lot

211
00:11:52,960 --> 00:11:55,197
of worries about consumer AI that it

212
00:11:55,280 --> 00:11:57,597
kind of leads to needing to maximize

213
00:11:57,680 --> 00:12:00,557
engagement. It leads to slop. You know,

214
00:12:00,640 --> 00:12:03,117
we've seen a lot of stuff around ads

215
00:12:03,200 --> 00:12:07,677
from from some of the other players.
You know, Enthropic is not a player that

216
00:12:07,760 --> 00:12:12,557
works like that or needs to work like
that. We just sell things to businesses

217
00:12:12,640 --> 00:12:16,797
and those things directly have value,
right? We we don't need to monetize a

218
00:12:16,880 --> 00:12:20,957
billion free users. We don't need
to maximize engagement for a billion free

219
00:12:21,040 --> 00:12:25,597
users because we're in some, you know,
death race with with some other, you

220
00:12:25,680 --> 00:12:29,517
know, some other some other large player.
And and so I think that has let

221
00:12:29,600 --> 00:12:33,677
us think more carefully. But but even
with that, we have made sacrifices. You

222
00:12:33,760 --> 00:12:35,677
know, we do all these tests on our

223
00:12:35,760 --> 00:12:40,717
models that others have not done. um
uh you know some other players have done

224
00:12:40,800 --> 00:12:42,317
them but but you know I think we've been

225
00:12:42,400 --> 00:12:45,517
the most uh you know aggressive in you

226
00:12:45,600 --> 00:12:50,237
know when we when we run tests that you
know show up concerning behaviors in our

227
00:12:50,320 --> 00:12:54,877
model you know these things around
deception, blackmail, syphancy that we

228
00:12:54,960 --> 00:12:56,317
show in tests and then are present in

229
00:12:56,400 --> 00:12:59,757
all of the models but you know we make

230
00:12:59,840 --> 00:13:01,917
sure to always talk to the public about

231
00:13:02,000 --> 00:13:06,797
about these things. Um and you know we
we've we you know we've pioneered the

232
00:13:06,880 --> 00:13:10,957
science of mechanistic interpretability
for looking inside models. So you know

233
00:13:11,040 --> 00:13:14,237
have we been have we been perfect?
Of course not. I think we've done a

234
00:13:14,320 --> 00:13:18,317
generally good job. I mean you mentioned
China. I you know I think that's not

235
00:13:18,400 --> 00:13:20,877
about competition. That is that is about

236
00:13:20,960 --> 00:13:23,597
actually the public benefit mission is

237
00:13:23,680 --> 00:13:28,077
I'm I'm worried that if autocracies
lead in this technology it will be a bad

238
00:13:28,160 --> 00:13:29,917
outcome for every single
person in this room.

239
00:13:30,000 --> 00:13:33,757
>> What are your specific concern there?
Is it about the chips about you know

240
00:13:33,840 --> 00:13:36,397
sharing data around chips or

241
00:13:36,480 --> 00:13:38,637
>> Yeah. Well, I think I mean you know the

242
00:13:38,720 --> 00:13:43,517
the the the kind of means is selling
the chips, right? That's the that's the um

243
00:13:43,600 --> 00:13:47,357
thing that I think will have the most
impact on on who is ahead and and who's

244
00:13:47,440 --> 00:13:49,677
not. But you know the the concern and

245
00:13:49,760 --> 00:13:54,397
you know it's not about any particular
country or certainly not the people in

246
00:13:54,480 --> 00:13:56,477
any country. It's about it's about a

247
00:13:56,560 --> 00:13:59,837
form of government. Um I am concerned

248
00:13:59,920 --> 00:14:03,597
that AI may be uniquely well suited to

249
00:14:03,680 --> 00:14:08,317
autocracy and to deepening
the repression that we see in autocracies.

250
00:14:08,400 --> 00:14:11,597
We already see it in the kind of
surveillance state that is possible with

251
00:14:11,680 --> 00:14:15,917
today's technology. But if you think
of the extent to which AI can make

252
00:14:16,000 --> 00:14:18,397
individualized propaganda, can break

253
00:14:18,480 --> 00:14:21,597
into any computer system in the world,

254
00:14:21,680 --> 00:14:25,437
um, uh, can, uh, you know, surveil

255
00:14:25,520 --> 00:14:29,357
everyone, everyone in a population,
detect descent everywhere and suppress

256
00:14:29,440 --> 00:14:32,397
it, you know, make, you know, a a huge

257
00:14:32,480 --> 00:14:34,637
army of drones that could go after each

258
00:14:34,720 --> 00:14:39,437
each each individual person. It's it's
it's really scary. It's it's it's really

259
00:14:39,520 --> 00:14:44,157
scary and we have to stop it.
But again, is that something that you feel

260
00:14:44,240 --> 00:14:47,677
governments aren't paying enough
attention to? I mean I you know I think

261
00:14:47,760 --> 00:14:52,077
I think it's I think it's fair to say
that you know obviously you know

262
00:14:52,160 --> 00:14:53,357
different countries think of themselves

263
00:14:53,440 --> 00:14:55,197
as having having geopolitical

264
00:14:55,280 --> 00:14:59,277
adversaries but the specific focus on we

265
00:14:59,360 --> 00:15:01,597
we don't want autocracies to get this

266
00:15:01,680 --> 00:15:03,757
powerful technology and we should have

267
00:15:03,840 --> 00:15:06,717
targeted policies like we don't need to

268
00:15:06,800 --> 00:15:09,357
fight them like we just need to not sell

269
00:15:09,440 --> 00:15:13,597
these chips um uh that you know I I

270
00:15:13,680 --> 00:15:16,637
think there's not enough focus on that.

271
00:15:16,720 --> 00:15:20,637
I want to talk a bit more about Claude
because I think it's fair to say it's

272
00:15:20,720 --> 00:15:23,037
having a real moment.
I mean we >> it's having a moment.

273
00:15:23,120 --> 00:15:24,877
>> It is having a moment
and we we recently

274
00:15:24,960 --> 00:15:26,957
reported on how engineers and regular

275
00:15:27,040 --> 00:15:30,077
users are like are getting clawpilled.

276
00:15:30,160 --> 00:15:33,677
Um and I just wondered how you feel
about the state of the business today

277
00:15:33,760 --> 00:15:37,437
versus a year ago. Yeah,
I mean this is one of these things

278
00:15:37,520 --> 00:15:39,277
that's been the growth of the business

279
00:15:39,360 --> 00:15:42,557
has been fast but but but kind of on on

280
00:15:42,640 --> 00:15:47,357
the same smooth exponential curve as the
technology. So you know we you know we

281
00:15:47,440 --> 00:15:49,437
have this this this revenue curve that

282
00:15:49,520 --> 00:15:52,237
in 2023 went from zero to roughly 100

283
00:15:52,320 --> 00:15:54,797
100 million went in 2024 went from

284
00:15:54,880 --> 00:15:59,357
roughly 100 million to roughly billion
2025 went from roughly a billion to

285
00:15:59,440 --> 00:16:03,357
roughly 10 billion. Not exactly.
These are rounded numbers, but but that is

286
00:16:03,440 --> 00:16:06,957
that is roughly it. You know, through
that, if you go on Twitter, every couple

287
00:16:07,040 --> 00:16:10,077
months, it's like, "Oh my god,
Anthropic's changing the world. Oh my

288
00:16:10,160 --> 00:16:13,037
god, you know, Anthropics totally
destroy, you know, just just the

289
00:16:13,120 --> 00:16:15,517
excitability of the of of the moment."

290
00:16:15,600 --> 00:16:19,997
But but we just watch it and we watch
this curve. It's fast. It's constantly

291
00:16:20,080 --> 00:16:21,757
progressing. It's given us confidence.

292
00:16:21,840 --> 00:16:25,517
We never know for sure if it's going to
continue. It might not. But but that has

293
00:16:25,600 --> 00:16:29,277
been empirically what we have observed
the whole time. And then there are these

294
00:16:29,360 --> 00:16:32,717
moments where even though the curve
is smooth, there's a breakout moment. And

295
00:16:32,800 --> 00:16:34,957
so right now I think there's a breakout

296
00:16:35,040 --> 00:16:37,837
moment around claude code among

297
00:16:37,920 --> 00:16:42,557
developers. Um you know this this thing
about being able to make whole apps and

298
00:16:42,640 --> 00:16:44,557
doing things end to end. Again that

299
00:16:44,640 --> 00:16:47,037
advanced gradually but with our most

300
00:16:47,120 --> 00:16:52,077
recent model opus 4.5 it just kind
of reached an inflection point right where

301
00:16:52,160 --> 00:16:55,757
the you know the the improvement was
gradual but you know it's it's just like

302
00:16:55,840 --> 00:16:59,357
you know boiling the fraud. you know,
you see you see the gradual improvement

303
00:16:59,440 --> 00:17:03,437
and then there's a specific point
at which suddenly that's the point people

304
00:17:03,520 --> 00:17:07,357
notice. I think the second thing
that has that has maybe accelerated that

305
00:17:07,440 --> 00:17:10,077
further is we looked at claude code and

306
00:17:10,160 --> 00:17:12,877
one of the things we we noticed is there

307
00:17:12,960 --> 00:17:14,797
were a lot of people inside enthropic

308
00:17:14,880 --> 00:17:17,197
and outside enthropic who were not

309
00:17:17,280 --> 00:17:20,557
technical but who realized that claude

310
00:17:20,640 --> 00:17:24,397
code could do these incredible agentic
tasks for you. It couldn't just write

311
00:17:24,480 --> 00:17:26,557
code. It could also organize your to-do

312
00:17:26,640 --> 00:17:31,917
list or plan your projects or organize
your folders or, you know, process a

313
00:17:32,000 --> 00:17:35,757
bunch of information and kind
of summarize. So, the the idea that not

314
00:17:35,840 --> 00:17:38,477
just a chatbot, but a gentic tasks were

315
00:17:38,560 --> 00:17:43,837
were needed. Um, non-technical people
were realizing it and they wanted it so

316
00:17:43,920 --> 00:17:46,957
much that they were wrestling with the
command line, right? Nontechnical, you

317
00:17:47,040 --> 00:17:50,237
know, they have no reason if you're
not a programmer. It's such a terrible

318
00:17:50,320 --> 00:17:53,037
interface to use if you're not
a programmer. But people were going

319
00:17:53,120 --> 00:17:55,517
through and using it anyway. And so
I looked at that and I said that looks

320
00:17:55,600 --> 00:17:59,037
like unmet demand. And so we we used we

321
00:17:59,120 --> 00:18:01,037
used claude code again in like two

322
00:18:01,120 --> 00:18:03,997
weeks, you know, to make basically kind

323
00:18:04,080 --> 00:18:07,837
of a version with a better UI that's

324
00:18:07,920 --> 00:18:10,957
customized for tasks that tasks other

325
00:18:11,040 --> 00:18:13,997
than other than code. And you know you

326
00:18:14,080 --> 00:18:15,917
know we released it and you know within

327
00:18:16,000 --> 00:18:18,397
like a day it you know it had like you

328
00:18:18,480 --> 00:18:21,837
know you know you know most of the the

329
00:18:21,920 --> 00:18:25,837
metrics on it were like four times
as much as anything anything we'd ever

330
00:18:25,920 --> 00:18:29,837
released. Uh so you know those are the two
moments. I don't know that these are

331
00:18:29,920 --> 00:18:33,597
new capabilities but you know there
was just one of these kind of consensus

332
00:18:33,680 --> 00:18:37,757
moments where people got really excited
and it's it's it's driving adoption

333
00:18:37,840 --> 00:18:41,677
really fast. I think people are catching
up to what the technology is is capable

334
00:18:41,760 --> 00:18:44,637
of because it's reached a certain point
and because we built interfaces that

335
00:18:44,720 --> 00:18:47,277
have made it accessible.
>> Can you tell us a bit about how you

336
00:18:47,360 --> 00:18:51,357
personally in your life,
your family life, use a Gentic AI?

337
00:18:51,440 --> 00:18:55,997
>> Yeah. Um, so you know when I'm writing
like you know an essay or something or

338
00:18:56,080 --> 00:18:59,677
like you know things I say in front of
the company um I feel like a fair amount

339
00:18:59,760 --> 00:19:03,837
of my job is writing and so I kind of help
I have Claude you know come up with

340
00:19:03,920 --> 00:19:08,717
sources, help me help me with my writing
that that kind of thing. And then

341
00:19:08,800 --> 00:19:11,517
obviously you're having this great
moment and I think there you're it's

342
00:19:11,600 --> 00:19:14,637
widely expected that you're going to IPO
this year. Can you tell us a bit about

343
00:19:14,720 --> 00:19:17,597
your plans for that? >> Yeah,
I mean you know we we don't know

344
00:19:17,680 --> 00:19:21,277
for sure we don't know for sure what
we're going to do and you know I would

345
00:19:21,360 --> 00:19:23,437
say we're we're more focused on just

346
00:19:23,520 --> 00:19:26,477
keeping the revenue curve going better

347
00:19:26,560 --> 00:19:30,477
selling the models to people you know
warning about the societal impacts and

348
00:19:30,560 --> 00:19:33,357
bringing bringing the good societal
impacts. So, you know, that's the that's

349
00:19:33,440 --> 00:19:36,797
that's kind that's kind of the highest
that's kind of the highest priority

350
00:19:36,880 --> 00:19:40,557
right now. But, you know, I'm not saying
anything novel if I say that this is an

351
00:19:40,640 --> 00:19:42,797
industry with very high capital demands.

352
00:19:42,880 --> 00:19:47,037
Um uh and you know that the there's only
so much at some point that the private

353
00:19:47,120 --> 00:19:49,757
markets the private markets can provide.

354
00:19:49,840 --> 00:19:54,077
>> So, another model that's absolutely
having a moment is Gemini and it sort of

355
00:19:54,160 --> 00:19:58,237
surged to the top of the app store
recently and OpenAI declared code red

356
00:19:58,320 --> 00:19:59,917
and so everyone was got very excited

357
00:20:00,000 --> 00:20:02,877
about that. Do you worry about your

358
00:20:02,960 --> 00:20:05,357
ability to compete against uh Gemini

359
00:20:05,440 --> 00:20:08,077
given the sheer size of Google?

360
00:20:08,160 --> 00:20:10,237
>> So I think I think
this is another place

361
00:20:10,320 --> 00:20:12,637
where just just being different helps.

362
00:20:12,720 --> 00:20:15,517
So you know the enterprise strategy

363
00:20:15,600 --> 00:20:20,477
Google and open AI are fighting it out
in consumer right it they it is both is

364
00:20:20,560 --> 00:20:22,317
existential to both of them. Existential

365
00:20:22,400 --> 00:20:27,517
to open AI because that's their whole
business existential to Google because

366
00:20:27,600 --> 00:20:30,317
they have the search business and that's
that's what's being disrupted by this.

367
00:20:30,400 --> 00:20:34,557
So they they need to you know replace
replace themselves and and fight the

368
00:20:34,640 --> 00:20:36,877
disruption. So that's always their first

369
00:20:36,960 --> 00:20:41,117
priority and you know they they seem

370
00:20:41,200 --> 00:20:43,437
more f they seem much more focused on

371
00:20:43,520 --> 00:20:45,997
that than kind of operating in the

372
00:20:46,080 --> 00:20:50,557
enterprise. It's been great to see what
what Gemini is is capable of you know

373
00:20:50,640 --> 00:20:53,517
capable of in consumer. You know I think
I think they're you know I think they're

374
00:20:53,600 --> 00:20:57,277
going about it a different way. I was you
know I was just on a panel with with

375
00:20:57,360 --> 00:21:00,797
Demis who leads research at Google. you
know, I think I think he's a great guy.

376
00:21:00,880 --> 00:21:03,837
I've known him for 15 years,
so I'm rooting for him.

377
00:21:03,920 --> 00:21:07,197
>> Um, what you talk about differences.
One difference is I believe is that

378
00:21:07,280 --> 00:21:11,437
Anthropic doesn't have the ability
to generate videos and photos. Do you see

379
00:21:11,520 --> 00:21:13,597
that as a a potential weakness?

380
00:21:13,680 --> 00:21:17,357
>> Um, I you know, I think for enterprise
business, you know, there's there's not

381
00:21:17,440 --> 00:21:19,837
really a demand for like, you know, you

382
00:21:19,920 --> 00:21:22,797
know, photos of, you know, the cats

383
00:21:22,880 --> 00:21:26,397
riding donkeys or, you know, whatever

384
00:21:26,480 --> 00:21:28,477
whatever consumer video people want.

385
00:21:28,560 --> 00:21:32,237
There's maybe an edge case around like
slides and presentations, but if we ever

386
00:21:32,320 --> 00:21:35,917
need it, we can just buy a, you know, we
can just contract a model from from from

387
00:21:36,000 --> 00:21:38,557
from contract a model from someone else.

388
00:21:38,640 --> 00:21:41,837
So, you know, I don't I don't know
what will happen. I don't know what will

389
00:21:41,920 --> 00:21:45,917
happen in the future, but I at least
don't anticipate needing this. Um, and I

390
00:21:46,000 --> 00:21:49,277
think there are problems associated with
this. Like, you know, I think I think,

391
00:21:49,360 --> 00:21:52,557
you know, we look at the amount of short
form video out there, like a lot of it's

392
00:21:52,640 --> 00:21:54,477
fake, a lot of it's pretty pretty

393
00:21:54,560 --> 00:21:57,357
addictive, a lot of it's slop. Um, not

394
00:21:57,440 --> 00:22:01,757
to say that all of it is bad or that
necessarily doing it means you're bad,

395
00:22:01,840 --> 00:22:04,077
but but it's not it's not a part of the

396
00:22:04,160 --> 00:22:08,957
market that I'm I'm like, you know,
tripping over myself to to get involved

397
00:22:09,040 --> 00:22:11,677
in. >> [snorts] >> You mentioned
that you were on a panel

398
00:22:11,760 --> 00:22:15,997
with Disa Savis and when we were
chatting earlier yesterday, you talked

399
00:22:16,080 --> 00:22:19,757
you said something that I thought
was very interesting that scientists are

400
00:22:19,840 --> 00:22:23,277
approaching the AI AI era or scientists
who are leading these big AI companies

401
00:22:23,360 --> 00:22:28,157
are approaching the era differently from
sort of uh tech entrepreneurs. Can you

402
00:22:28,240 --> 00:22:30,077
say a bit more about
what you mean by that?

403
00:22:30,160 --> 00:22:32,317
>> Yeah. Well, so you know when I when you

404
00:22:32,400 --> 00:22:34,797
think about this technology, it's really

405
00:22:34,880 --> 00:22:37,757
it's really the intersection of research

406
00:22:37,840 --> 00:22:42,237
that has going been going on for many
decades, much of which was academic in

407
00:22:42,320 --> 00:22:44,237
nature until you know a decade decade

408
00:22:44,320 --> 00:22:48,717
and a half ago. um and the kind of scale

409
00:22:48,800 --> 00:22:52,237
needed to um you know needed to develop

410
00:22:52,320 --> 00:22:56,237
and deploy these technologies over
the last decade and a half which has only

411
00:22:56,320 --> 00:23:00,477
come from the large scale kind
of internet and social media companies

412
00:23:00,560 --> 00:23:04,797
right they have the infrastructure they
have the cash so we've seen a world in

413
00:23:04,880 --> 00:23:09,117
which some of the companies are
essentially led by people who who have a

414
00:23:09,200 --> 00:23:12,317
scientific background that's
my background that's Dennis's background

415
00:23:12,400 --> 00:23:16,717
some of them are led by the generation
of entrepreneurs that social media. Um,

416
00:23:16,800 --> 00:23:18,877
and I I I think that's very different

417
00:23:18,960 --> 00:23:21,677
that um, you know, scientists, there's a

418
00:23:21,760 --> 00:23:23,837
long tradition of scientists thinking

419
00:23:23,920 --> 00:23:27,277
about the effects of the technology they

420
00:23:27,360 --> 00:23:29,197
build, of thinking of themselves as

421
00:23:29,280 --> 00:23:34,637
having responsibility for the technology
they build, not ducking responsibility,

422
00:23:34,720 --> 00:23:38,157
right? They're they're they're mo
they're motivated in the first place by

423
00:23:38,240 --> 00:23:40,557
creating something for the world. And so

424
00:23:40,640 --> 00:23:42,957
then they they worry in the cases where

425
00:23:43,040 --> 00:23:45,757
that something can go wrong. Um I think

426
00:23:45,840 --> 00:23:48,397
the motivation of entrepreneurs

427
00:23:48,480 --> 00:23:53,357
particularly the generation of uh
uh social media entrepreneurs is very

428
00:23:53,440 --> 00:23:57,757
different. The selection effects that
operated on them the way in which they

429
00:23:57,840 --> 00:24:01,437
interacted with you might say
manipulated consumers is very different.

430
00:24:01,520 --> 00:24:04,557
Um and and so I think that that leads to

431
00:24:04,640 --> 00:24:07,197
that leads to different attitudes.

432
00:24:07,280 --> 00:24:11,517
Um now we've been taking some questions
from readers who submitted their

433
00:24:11,600 --> 00:24:14,557
questions online but before we do that I
just wanted to ask you one more thing

434
00:24:14,640 --> 00:24:18,717
which is that you know again big picture
tensions are running very high at the

435
00:24:18,800 --> 00:24:23,357
moment between the US and the EU. Do you
wonder about how that might impact how

436
00:24:23,440 --> 00:24:26,237
you operate your business
should things escalate.

437
00:24:26,320 --> 00:24:29,997
>> Look I mean you know we we have we
have always you know we only speak for

438
00:24:30,080 --> 00:24:33,597
ourselves. We've always thought
of ourselves as as as, you know, kind of

439
00:24:33,680 --> 00:24:36,157
kind of our own thing and and and and

440
00:24:36,240 --> 00:24:38,637
and independent, right? We don't go out

441
00:24:38,720 --> 00:24:42,557
of our way to be for or against anyone.

442
00:24:42,640 --> 00:24:45,437
But when we, you know, when we disagree

443
00:24:45,520 --> 00:24:47,757
on policy, we say so. When we agree on

444
00:24:47,840 --> 00:24:50,557
policy, we say so. And we we really keep

445
00:24:50,640 --> 00:24:53,037
it focused on AI. And and so, you know,

446
00:24:53,120 --> 00:24:57,677
I, you know, I haven't seen any
reluctance in folks in other parts of

447
00:24:57,760 --> 00:25:01,757
the world to work with us, right?
We're our own thing. We're providing AI

448
00:25:01,840 --> 00:25:03,997
models. We try to do that responsibly.

449
00:25:04,080 --> 00:25:07,357
>> I mean, there's been a lot of talk
this week about AI sovereignty. I'm not

450
00:25:07,440 --> 00:25:09,677
entirely sure what
everybody seems to have.

451
00:25:09,760 --> 00:25:11,197
>> I don't know what it means either.

452
00:25:11,280 --> 00:25:12,987
>> You don't have your own definition.

453
00:25:13,070 --> 00:25:14,557
[laughter]

454
00:25:14,640 --> 00:25:17,597
>> Good. Okay. Well, look, so um we have

455
00:25:17,680 --> 00:25:20,477
solicited questions from readers online.

456
00:25:20,560 --> 00:25:22,557
Um so I'm going to start now with one

457
00:25:22,640 --> 00:25:25,677
from Trevor Lumis. Um and his question

458
00:25:25,760 --> 00:25:27,997
is what is the single most important

459
00:25:28,080 --> 00:25:30,317
technical breakthrough still missing to

460
00:25:30,400 --> 00:25:32,957
make frontier AI reliably safe and

461
00:25:33,040 --> 00:25:36,557
controllable in real world deployment?

462
00:25:36,640 --> 00:25:39,517
>> So I think we need
to make more progress

463
00:25:39,600 --> 00:25:43,197
on mechanistic interpretability which
is the science of looking inside the

464
00:25:43,280 --> 00:25:45,677
models. One of the problems when we

465
00:25:45,760 --> 00:25:51,117
train these models is that we don't know
you can't be sure they're going to do

466
00:25:51,200 --> 00:25:54,557
what you think they're going to do. You
can talk to the model in one context. It

467
00:25:54,640 --> 00:25:59,117
can say all kinds of things. Just as
with a human, that may not be a faithful

468
00:25:59,200 --> 00:26:03,437
representation of what they're actually
thinking. If they tell you, "I'm doing X

469
00:26:03,520 --> 00:26:06,957
because why?" They might be doing X
for a completely different reason. They

470
00:26:07,040 --> 00:26:10,557
might be lying about doing X. Like,
we're very used to these problems with

471
00:26:10,640 --> 00:26:15,117
humans, but but they exist
with AI as well. And so any kind of

472
00:26:15,200 --> 00:26:17,517
phenomenological testing or training we

473
00:26:17,600 --> 00:26:21,037
can't be certain of. But similar to how

474
00:26:21,120 --> 00:26:25,597
you know you can learn things about
human brains by doing an MRI or an X-ray

475
00:26:25,680 --> 00:26:29,437
that you can't learn just by talking
to a human, the science of looking inside

476
00:26:29,520 --> 00:26:31,997
the AI models. I am convinced that this

477
00:26:32,080 --> 00:26:35,357
ultimately holds the key to making the

478
00:26:35,440 --> 00:26:37,837
model safe and controllable because it's

479
00:26:37,920 --> 00:26:39,837
the only ground truth we have.

480
00:26:39,920 --> 00:26:42,957
>> Right. Okay. Um,
I have another question

481
00:26:43,040 --> 00:26:45,597
here from Jim O' Connell. How will AI

482
00:26:45,680 --> 00:26:50,477
affect current K12 educational
achievement gaps? Very practical

483
00:26:50,560 --> 00:26:52,477
question there from no doubt a parent.

484
00:26:52,560 --> 00:26:56,397
>> Yeah. So, you know, I I my you know,
there there's kind of the short-term

485
00:26:56,480 --> 00:26:59,997
stuff about, you know, people using
AI for cheating, which I think is, you

486
00:27:00,080 --> 00:27:04,717
know, I think is problematic. But,
you know, in in in relative terms, okay,

487
00:27:04,800 --> 00:27:09,997
fine, you can have a kind of a different
way of um uh you know, of teaching using

488
00:27:10,080 --> 00:27:12,877
AI. And you know, we've we've thought
about that. We've released versions of

489
00:27:12,960 --> 00:27:15,197
of Claude for for education that are

490
00:27:15,280 --> 00:27:20,637
kind of designed around that. But I
think the harder problem behind that is

491
00:27:20,720 --> 00:27:24,077
okay, what skills are we actually
teaching in in the world of AI? What

492
00:27:24,160 --> 00:27:26,077
does education look like look like in

493
00:27:26,160 --> 00:27:28,397
the world of AI? And it's it's not so

494
00:27:28,480 --> 00:27:33,277
easy because the disruption is broad
and we don't you know, if someone asked me

495
00:27:33,360 --> 00:27:37,517
what exactly what career should I go
into, the the uncomfortable truth is I

496
00:27:37,600 --> 00:27:41,517
I'm I'm not sure. I can't tell
the direction that it's going to I can't

497
00:27:41,600 --> 00:27:43,117
tell the direction that it's going to go

498
00:27:43,200 --> 00:27:46,237
yet. I I will say that I think we should

499
00:27:46,320 --> 00:27:48,957
go back to some concepts that we had

500
00:27:49,040 --> 00:27:50,877
earlier about education like we've we've

501
00:27:50,960 --> 00:27:53,597
had a very kind of like um like

502
00:27:53,680 --> 00:27:56,477
economically inflected almost mercenary

503
00:27:56,560 --> 00:28:00,877
notion notion of education. Um and and
and I think one of the things that we

504
00:28:00,960 --> 00:28:04,877
should do is we should maybe move away
move away from that notion back to the

505
00:28:04,960 --> 00:28:07,357
idea that like you know education is is

506
00:28:07,440 --> 00:28:09,837
is designed to shape you as a person is

507
00:28:09,920 --> 00:28:12,317
designed to build character is is you

508
00:28:12,400 --> 00:28:15,357
know is is is is kind of designed to

509
00:28:15,440 --> 00:28:19,357
enrich you and and like make you
a better person. I think that's actually a

510
00:28:19,440 --> 00:28:21,837
safer foundation for for
education in the future.

511
00:28:21,920 --> 00:28:25,437
>> That sounds I'm rather envious of
the kids kids who are yet to be educated.

512
00:28:25,520 --> 00:28:28,637
It's the kind of education I think we'd
all have liked to have. Um, so to be

513
00:28:28,720 --> 00:28:31,517
fair to everybody in the room then
I just I think we've got time for one

514
00:28:31,600 --> 00:28:36,237
question if anybody would
like to ask a question. Yeah.

515
00:28:36,320 --> 00:28:38,077
This lady here.

516
00:28:38,160 --> 00:28:41,277
>> Yeah. Yeah. No, no, he's the mic. Um, I

517
00:28:41,360 --> 00:28:45,677
wanted to ask from a point of view
of the AI labs, uh, what kind of

518
00:28:45,760 --> 00:28:49,677
responsibility do you hold when there
are economies, countries, and people

519
00:28:49,760 --> 00:28:51,997
that are being left behind? um would

520
00:28:52,080 --> 00:28:54,877
that expand into structurally involving

521
00:28:54,960 --> 00:28:58,797
them um slowing down or actually making

522
00:28:58,880 --> 00:29:00,637
sure that they're not being left out?

523
00:29:00,720 --> 00:29:05,117
>> Yeah, I worry about that on a whole
bunch of scales and it's you know it's

524
00:29:05,200 --> 00:29:07,117
not just country versus country.

525
00:29:07,200 --> 00:29:11,117
Certainly I worry about the developing
world versus the developed world. Um

526
00:29:11,200 --> 00:29:14,077
where you know sometimes the the
developed world developing world will

527
00:29:14,160 --> 00:29:15,677
get passed by by by technological

528
00:29:15,760 --> 00:29:18,077
revolutions. But I but you know I also

529
00:29:18,160 --> 00:29:21,117
worry about divisions within a country.

530
00:29:21,200 --> 00:29:23,037
It it has occurred to me as I've looked

531
00:29:23,120 --> 00:29:26,077
across our customers as the startups are

532
00:29:26,160 --> 00:29:30,637
so fast to adopt AI and the traditional
enterprises because they're bigger

533
00:29:30,720 --> 00:29:35,117
because they do a specific thing they
move much slower. Um and we can see it

534
00:29:35,200 --> 00:29:39,677
in our economic data. We can see the
diffusion of the technology from states

535
00:29:39,760 --> 00:29:43,517
within the US that adopt it quickly
and states that move slowly. It is

536
00:29:43,600 --> 00:29:47,197
diffusing. It's getting out there. But
there's there's no question that there's

537
00:29:47,280 --> 00:29:50,477
a differential here. If I were
to describe the nightmare and then then

538
00:29:50,560 --> 00:29:54,957
I'll try to describe some you know what
I think of as a solutions. The nightmare

539
00:29:55,040 --> 00:29:57,037
would be that there's there there's like

540
00:29:57,120 --> 00:30:00,317
this emerging zeroth world country of

541
00:30:00,400 --> 00:30:03,277
like 10 million people that's like seven

542
00:30:03,360 --> 00:30:07,437
million people in the b you know in like
Silicon Valley and you know 3 million

543
00:30:07,520 --> 00:30:11,277
people kind of scattered scattered
scattered throughout that that you know

544
00:30:11,360 --> 00:30:15,677
is kind of forming its own economy
and is becoming decoupled or disconnected

545
00:30:15,760 --> 00:30:20,477
right maybe the 10% GDP growth looks
like 50% GDP growth in that part of like

546
00:30:20,560 --> 00:30:25,357
this technology is so crazy it can pull
things apart that way. Um I think that

547
00:30:25,440 --> 00:30:28,637
would be a really bad world.
I would almost say that it was a dystopian

548
00:30:28,720 --> 00:30:33,517
world, right? And and we should think
about how to stop that. Um there are,

549
00:30:33,600 --> 00:30:35,437
you know, a number of things anthropic

550
00:30:35,520 --> 00:30:38,557
is thinking about or doing. Um one is as

551
00:30:38,640 --> 00:30:42,797
regards the developing world, we're we
are starting to do a lot of work around

552
00:30:42,880 --> 00:30:46,317
public health. Um we've announced
stuff with, you know, Ronda Ministry of

553
00:30:46,400 --> 00:30:50,797
Education. We're doing a a lot of work
with the Gates Foundation. Um you know,

554
00:30:50,880 --> 00:30:53,917
it it would I wrote about this in machines
of loving grace. it it would be

555
00:30:54,000 --> 00:30:57,357
really great to get these fast economic
growth rates which in theory should be

556
00:30:57,440 --> 00:31:01,517
even faster because it's catch-up
growth um in the developing world that I

557
00:31:01,600 --> 00:31:03,037
predict we're going to get in in the

558
00:31:03,120 --> 00:31:05,437
developed world um you know within

559
00:31:05,520 --> 00:31:08,557
within countries we need to you know we

560
00:31:08,640 --> 00:31:13,197
need to think about you know how how
not to have a part of the world that just

561
00:31:13,280 --> 00:31:16,637
that just kind of decouples right
how do we get the economic growth to

562
00:31:16,720 --> 00:31:20,957
Mississippi that that you know that is
coming to this contained contained area

563
00:31:21,040 --> 00:31:24,797
of of of of Silicon Valley And so, you
know, there we've done work around kind

564
00:31:24,880 --> 00:31:29,517
of e economic mobility and e economic
opportunity. But I think both of these

565
00:31:29,600 --> 00:31:32,557
again are going to need to have some
some involvement of the government.

566
00:31:32,640 --> 00:31:34,877
We're going to find that that you know

567
00:31:34,960 --> 00:31:37,917
ideology will will not survive the

568
00:31:38,000 --> 00:31:39,597
nature of this technology. It won't

569
00:31:39,680 --> 00:31:45,037
survive reality. The things I'm talking
about, you know, while you could today

570
00:31:45,120 --> 00:31:48,157
say, "Oh, yeah, they're they're like
politically coded in some way," they're

571
00:31:48,240 --> 00:31:51,437
going to become bipartisan and universal
because everyone will recognize the

572
00:31:51,520 --> 00:31:55,277
necessity of it. Just mark my words,
we come back, if not next year, the year

573
00:31:55,360 --> 00:31:57,917
after, everyone's going to think this.

574
00:31:58,000 --> 00:32:01,437
>> Well, you've managed to end on a more
or less positive note. So, I'm going to

575
00:32:01,520 --> 00:32:04,957
draw a line there and say thank you
very much, Dario. That was really, really

576
00:32:05,040 --> 00:32:09,050
fascinating. >> Thank you
for having me. [applause]