1
00:00:00,000 --> 00:00:03,917
Our revenue has been going up 10x per
year. It's currently, I think it was

2
00:00:04,000 --> 00:00:05,197
reported this morning, which I'm happy

3
00:00:05,280 --> 00:00:07,677
to confirm is correct, is approaching

4
00:00:07,760 --> 00:00:10,957
approaching 7 billion run rate. Um,

5
00:00:11,040 --> 00:00:12,877
that's gone up 10x a year for the for

6
00:00:12,960 --> 00:00:15,917
the last 3 years. We are using Claude to

7
00:00:16,000 --> 00:00:17,997
help build [music] products on top of

8
00:00:18,080 --> 00:00:20,317
Claude and to help train the next

9
00:00:20,400 --> 00:00:23,837
Claude. I was uh originally interested

10
00:00:23,920 --> 00:00:26,557
in science, not in uh not in tech per

11
00:00:26,640 --> 00:00:29,437
se. So you know I did my did my uh uh

12
00:00:29,520 --> 00:00:31,917
graduate work in uh neuroscience. I was

13
00:00:32,000 --> 00:00:34,637
a posttock in computational biology at

14
00:00:34,720 --> 00:00:39,037
at Stanford and you know as I was working
hard on solving the problems of

15
00:00:39,120 --> 00:00:41,037
biology you know trying to understand

16
00:00:41,120 --> 00:00:46,157
the diseases we we we face like uh like
cancer and to understand biology one of

17
00:00:46,240 --> 00:00:50,797
the things I noticed is that the
complexity of biology felt like it was

18
00:00:50,880 --> 00:00:54,237
beyond human scale you know and I
was starting to despair a little bit that

19
00:00:54,320 --> 00:00:58,397
that we would be able to make progress
at any speed just because of you know if

20
00:00:58,480 --> 00:01:01,997
you've ever seen one of these diagrams
of human metabolis capitalism. It's just

21
00:01:02,080 --> 00:01:06,737
it's just beyond human scale. But almost
exactly at the time that I was wondering

22
00:01:06,820 --> 00:01:11,197
[clears throat] about this out of Google,
out of University of Toronto

23
00:01:11,280 --> 00:01:15,997
came the first work on deep neural
networks. Uh and I looked at it, you

24
00:01:16,080 --> 00:01:19,037
know, I I replicated some of the results
myself. I said, "My god, this stuff

25
00:01:19,120 --> 00:01:22,477
really works. This stuff's really going
to go there." So, I went into AI. I

26
00:01:22,560 --> 00:01:27,437
ended up working at um uh Google.
I joined Open AI in the early days. I led

27
00:01:27,520 --> 00:01:30,477
research there for for for several years.

28
00:01:30,560 --> 00:01:35,117
um uh and and and you know was one
of the people who kind of documented this

29
00:01:35,200 --> 00:01:39,917
trend of exponential scaling of the
technology um you know myself and the

30
00:01:40,000 --> 00:01:42,797
folks who eventually became
the anthropic co-founders were among the

31
00:01:42,880 --> 00:01:46,077
first to uh study the scaling of the

32
00:01:46,160 --> 00:01:48,477
technology and so you know around 2019

33
00:01:48,560 --> 00:01:51,757
or 2020 had really two realizations one

34
00:01:51,840 --> 00:01:55,197
was wow AI is is really going to you

35
00:01:55,280 --> 00:01:58,797
know it's going to it's going to scale
to the moon and it's really going to be

36
00:01:58,880 --> 00:02:01,517
you know this whole thing is going
to reshape the economy. It'll take a few

37
00:02:01,600 --> 00:02:03,037
years, but it's going to reshape the

38
00:02:03,120 --> 00:02:04,877
economy. And you know, they are almost

39
00:02:04,960 --> 00:02:08,877
there now. Um uh and and the second was

40
00:02:08,960 --> 00:02:12,717
look, there's, you know, these systems
are different from ordinary software.

41
00:02:12,800 --> 00:02:15,997
You don't write every line of code,
right? Building ordinary software, it's

42
00:02:16,080 --> 00:02:17,677
like building a skyscraper or something.

43
00:02:17,760 --> 00:02:21,677
You make the blueprint, you make
everything to design. This is ironically

44
00:02:21,760 --> 00:02:25,597
a bit more like biological. It's organic
or something. You're growing these

45
00:02:25,680 --> 00:02:30,637
models. You set a highle recipe. It's
like growing a plant or baking a cake or

46
00:02:30,720 --> 00:02:33,837
something like that. And that
unpredictability combined with the

47
00:02:33,920 --> 00:02:38,077
incredible economic impact means that
safety and responsibility are going to

48
00:02:38,160 --> 00:02:42,157
be very important to make sure that the
models behave predictably as well as as

49
00:02:42,240 --> 00:02:45,677
we deploy them and they have enormous
impact on the world. We you know we we

50
00:02:45,760 --> 00:02:48,957
need to make sure that we deploy them
responsibly. And so those were kind of

51
00:02:49,040 --> 00:02:51,677
the two pillars on which myself and the

52
00:02:51,760 --> 00:02:54,397
seven co-founders um who left OpenAI to

53
00:02:54,480 --> 00:02:57,917
to found Anthropic um you know generally

54
00:02:58,000 --> 00:02:59,917
generally based our generally based our

55
00:03:00,000 --> 00:03:02,557
view uh and you know the the the rest is

56
00:03:02,640 --> 00:03:05,197
history. Enthropic has been um you know

57
00:03:05,280 --> 00:03:08,157
around for 4 and 1/2 years um we're

58
00:03:08,240 --> 00:03:12,877
growing very quickly. um you know I
think we're we're at this point um by

59
00:03:12,960 --> 00:03:17,677
revenue have at least a plurality
perhaps the majority of uh um you know

60
00:03:17,760 --> 00:03:21,757
enterprise enterprise customers
and enterprise revenue while most of the

61
00:03:21,840 --> 00:03:25,517
other players are are focused on the
consumer side and we find that you know

62
00:03:25,600 --> 00:03:28,557
as as you know I think you would know
from running an enterprise focused

63
00:03:28,640 --> 00:03:32,797
company these these values
of trust and responsibility

64
00:03:32,880 --> 00:03:36,717
you know are really are synergistic
with serving the enterprises right these

65
00:03:36,800 --> 00:03:40,397
values help us to serve the enterprise
and Having a business model that serves

66
00:03:40,480 --> 00:03:43,837
the enterprise helps
to reinforce our values. [snorts]

67
00:03:43,920 --> 00:03:46,477
>> One of the one of the things um and you

68
00:03:46,560 --> 00:03:48,077
went pretty fast into kind of starting

69
00:03:48,160 --> 00:03:51,117
the company and kind of shot out of uh

70
00:03:51,200 --> 00:03:54,557
uh uh shot through uh Google and API to

71
00:03:54,640 --> 00:03:58,397
anthropic, but you didn't mention
your sister and that was like one of the

72
00:03:58,480 --> 00:04:01,757
things that first really shocked me when
we started to get to know you was you

73
00:04:01,840 --> 00:04:05,917
and your sister together are working on
this company and can you kind of tell us

74
00:04:06,000 --> 00:04:09,117
the origin story? >> Yeah.
Yeah. So, you know, there are

75
00:04:09,200 --> 00:04:12,477
seven co-founders and and me
and Dianiela are are two of them. As I

76
00:04:12,560 --> 00:04:16,317
mentioned, we both, you know, we both
grew up in San Francisco. We went to to

77
00:04:16,400 --> 00:04:19,997
LOL High School and you know, even even
very early on, we really figured that we

78
00:04:20,080 --> 00:04:21,997
wanted to do >> My mom went
to LOL and she's sitting

79
00:04:22,080 --> 00:04:25,757
right there. So, she's going
to be very happy about that.

80
00:04:25,840 --> 00:04:30,157
>> Amazing. Um, uh, no, no, it was it
was a great place. I actually didn't over

81
00:04:30,240 --> 00:04:33,677
Daniela was four years younger than me,
so we were never there at the same year,

82
00:04:33,760 --> 00:04:36,957
but I was there for four and then,
you know, the next four years she was she

83
00:04:37,040 --> 00:04:39,677
was there. So um uh yeah you know very

84
00:04:39,760 --> 00:04:44,317
early on you know we we we had these you
know these ambitions to you know to to

85
00:04:44,400 --> 00:04:47,437
do something to do something good
for the world to have some kind of grand

86
00:04:47,520 --> 00:04:51,597
project and and to work together.
I think I think many many people you know

87
00:04:51,680 --> 00:04:55,757
kind of or many siblings I guess have
these have these ambitions but you know

88
00:04:55,840 --> 00:04:58,077
some somehow things found their way to

89
00:04:58,160 --> 00:05:00,237
us actually actually being able to do

90
00:05:00,320 --> 00:05:03,757
it. So um uh you know she joined me at

91
00:05:03,840 --> 00:05:07,997
at she joined me at OpenAI in uh uh 2018

92
00:05:08,080 --> 00:05:12,237
um and then the two of us along with
some folks we worked with many of whom

93
00:05:12,320 --> 00:05:14,797
had been longtime uh uh friends decided

94
00:05:14,880 --> 00:05:19,277
to start Antropics. So me and Daniela
essentially essentially run the company

95
00:05:19,360 --> 00:05:24,077
together. You know she's responsible
for the dayto-day running of the company.

96
00:05:24,160 --> 00:05:28,157
you know, kind of operate, you know, how
how the company is put put together, how

97
00:05:28,240 --> 00:05:32,717
the org runs, what its processes are,
does, you know, kind of kind of a fair

98
00:05:32,800 --> 00:05:36,957
amount of the the the, you know, kind
of the business on the on the the business

99
00:05:37,040 --> 00:05:41,677
on the structural side. Um, uh, you know,
I I I I I just, you know, couldn't

100
00:05:41,760 --> 00:05:45,437
couldn't run the company without her and
without her without her doing all those

101
00:05:45,520 --> 00:05:47,837
things. Um you know I tend to think

102
00:05:47,920 --> 00:05:50,557
about the strategic side where the where

103
00:05:50,640 --> 00:05:53,517
the research is going what our kind of

104
00:05:53,600 --> 00:05:57,757
you know plans are as a company
strategically how we think about safety

105
00:05:57,840 --> 00:06:01,757
how we think about the company's uh uh
role in the world and and the vision and

106
00:06:01,840 --> 00:06:04,557
and I find that's a very good um that's

107
00:06:04,640 --> 00:06:05,997
a very good split that's a very good

108
00:06:06,080 --> 00:06:09,037
separation um because both of us can do

109
00:06:09,120 --> 00:06:13,277
both pieces but it's it's really hard
to switch between a kind of operational

110
00:06:13,360 --> 00:06:17,277
mode and a vision mode and I think both
people can be more effective if you if

111
00:06:17,360 --> 00:06:19,277
you split. >> Well,
thanks for allowing me to come

112
00:06:19,360 --> 00:06:22,877
back to that because I think it's such
a great part of the company and the

113
00:06:22,960 --> 00:06:26,157
culture and how you're working. And last
year she was here and now this year

114
00:06:26,240 --> 00:06:28,877
you're here. Um, and we've known you all

115
00:06:28,960 --> 00:06:33,277
from the beginning. I mean, I think
Salesforce owns like a some little point

116
00:06:33,360 --> 00:06:35,997
or something of anthropic. So,
we're very happy to be an investor.

117
00:06:36,080 --> 00:06:39,197
>> Very early investor. Very >>
very early investor. Believe in you.

118
00:06:39,280 --> 00:06:42,077
>> I think you I think you invested
when the valuation was 4 billion. Is that

119
00:06:42,160 --> 00:06:43,997
correct? >> I I don't know,
but it was too high.

120
00:06:44,080 --> 00:06:48,477
whatever it was. I haven't >> Before
you say it's too high, you should

121
00:06:48,560 --> 00:06:50,557
calculate how many times
it's gone up since then.

122
00:06:50,640 --> 00:06:53,357
>> Okay. We made money [laughter] so far.

123
00:06:53,440 --> 00:06:58,397
>> Well, um, no, I I'm thrilled obviously
that we're part of your journey and, uh,

124
00:06:58,480 --> 00:07:02,637
you know, we love working with you.
We love working with Daniela um, so much.

125
00:07:02,720 --> 00:07:06,877
And, you know, I think kind of getting
back to this story of you and your

126
00:07:06,960 --> 00:07:09,437
sister and your friends, you know,
starting this company here in San

127
00:07:09,520 --> 00:07:13,037
Francisco and then you came to us. We
were starting to consolidate some of our

128
00:07:13,120 --> 00:07:16,237
real estate here. And then you,
you know, we had a conversation about you

129
00:07:16,320 --> 00:07:19,837
taking over the Slack building. We had
just bought Slack, >> which we have done.

130
00:07:19,920 --> 00:07:21,277
>> And now you're in the Slack building.

131
00:07:21,360 --> 00:07:25,197
So, you're right on the park with us
looking at the tower from your window.

132
00:07:25,280 --> 00:07:27,597
>> I I believe my office is what I I was

133
00:07:27,680 --> 00:07:29,357
told it was formerly Steuart's office.

134
00:07:29,440 --> 00:07:32,557
>> Formerly Stuart's office.
Very good. And

135
00:07:32,640 --> 00:07:36,557
I would just, you know, it just just
occurred to me like here you are. You're

136
00:07:36,640 --> 00:07:40,797
the San Franciscan. Um,
you're right on you're right on the

137
00:07:40,880 --> 00:07:43,597
park with us. You're right in the tower
with us. We're an investor in the

138
00:07:43,680 --> 00:07:48,477
company. Um, it's, you know, obviously
a critical part of Slack today, right? In

139
00:07:48,560 --> 00:07:52,077
terms of how you couldn't use Slack
without it, but you're at the very

140
00:07:52,160 --> 00:07:55,517
bleeding edge of these models. It's
really incredible and you've really been

141
00:07:55,600 --> 00:07:58,557
visionary uh, right from
the very beginning. Now,

142
00:07:58,640 --> 00:08:02,317
when I didn't know you when you were at
OpenAI, I only met you after you started

143
00:08:02,400 --> 00:08:04,477
the company. So, just now kind of take

144
00:08:04,560 --> 00:08:09,597
us through your expectation.
What has been your biggest surprise,

145
00:08:09,680 --> 00:08:12,957
you know, since it's only been what,
four and a half, five years since you

146
00:08:13,040 --> 00:08:15,437
kind of walked out the door, said

147
00:08:15,520 --> 00:08:20,317
goodbye to Sam, right? And then you
just said, "Okay, Daniela, we're going to

148
00:08:20,400 --> 00:08:23,837
we're going to get this going." What
What was your biggest shock right there?

149
00:08:23,920 --> 00:08:25,277
>> Yeah, I mean,
you know, there were there

150
00:08:25,360 --> 00:08:26,797
were so many moments cuz I think that,

151
00:08:26,880 --> 00:08:28,637
you know, the last four and a half years

152
00:08:28,720 --> 00:08:32,317
were were just so surprising overall. Um

153
00:08:32,400 --> 00:08:35,197
you know I think I think at a high level

154
00:08:35,280 --> 00:08:39,517
there's a dichotomy where on one hand
everything we've seen over the last four

155
00:08:39,600 --> 00:08:43,677
and a half years was in some sense
predicted by those straight lines on

156
00:08:43,760 --> 00:08:47,117
graphs that I made with the scaling
laws right these enormous data center

157
00:08:47,200 --> 00:08:49,197
buildouts the way that our revenue is

158
00:08:49,280 --> 00:08:54,717
going up you know 10x per year the the use
across the economy you could you you

159
00:08:54,800 --> 00:08:58,077
know you can see it all in these
scribbles on the page but actually

160
00:08:58,160 --> 00:09:01,677
seeing it play out is is wild and and a

161
00:09:01,760 --> 00:09:05,757
lot of the individual things that that I'm
seeing happening I just I just never

162
00:09:05,840 --> 00:09:07,917
would have I never would have expected.

163
00:09:08,000 --> 00:09:10,397
I guess if I were to pick one most

164
00:09:10,480 --> 00:09:13,837
surprising thing, it is, you know, we

165
00:09:13,920 --> 00:09:18,237
left one company and started another
company. And uh something you see over

166
00:09:18,320 --> 00:09:19,997
time is that is that as you build a

167
00:09:20,080 --> 00:09:23,037
company, you really do see through the

168
00:09:23,120 --> 00:09:25,997
people you hire, through the choices

169
00:09:26,080 --> 00:09:30,317
that the company makes, you kind
of see the paths diverge and you see

170
00:09:30,400 --> 00:09:34,477
differentiation, right? OpenAI has gone
very much in the consumer direction.

171
00:09:34,560 --> 00:09:38,717
We've gone very much in the enterprise
direction. we've taken, you know, just

172
00:09:38,800 --> 00:09:42,557
just the way we approach things
is is very different. And so you have this

173
00:09:42,640 --> 00:09:45,837
market structure where there are these
two companies that are like it's sort of

174
00:09:45,920 --> 00:09:48,957
a natural experiment. They're they're
just taking a very different approach.

175
00:09:49,040 --> 00:09:53,277
Both exist in the ecosystem, but but
the approach they take is very different.

176
00:09:53,360 --> 00:09:56,637
>> And then you have to reflect back
at your time at Google as well. They're

177
00:09:56,720 --> 00:10:01,517
also very much with Gemini and you know
very much a top model as well. How do

178
00:10:01,600 --> 00:10:05,597
you look at them? Do you look at them
as a blend then? Yeah, I mean, you know,

179
00:10:05,680 --> 00:10:09,437
Google's a partner to us. They're
an investor in us. Obviously, their models

180
00:10:09,520 --> 00:10:13,757
compete with us, compete with us at the
same time. But I think this co-optition

181
00:10:13,840 --> 00:10:16,957
dynamic is becoming more and more
common, right? As we have all these

182
00:10:17,040 --> 00:10:18,877
layers, we have like the chip layer, the

183
00:10:18,960 --> 00:10:22,317
cloud layer, the the kind of model

184
00:10:22,400 --> 00:10:26,317
layer, the the like the like go
to market and distribution layer, and

185
00:10:26,400 --> 00:10:30,157
various companies are involved in more
than one of these layers. and and so I

186
00:10:30,240 --> 00:10:34,717
think it's not uncommon for companies to
cooperate in one area and and compete in

187
00:10:34,800 --> 00:10:38,957
another area. It's fairly normal.
I think it's also been helped by you know

188
00:10:39,040 --> 00:10:42,557
I I I know quite well some of the folks
behind Google. I've been you know

189
00:10:42,640 --> 00:10:47,117
friends with with Deisabis for uh who
you know who leads uh who leads research

190
00:10:47,200 --> 00:10:50,797
at Google for something like 15 years,
right? I think he was maybe one of the

191
00:10:50,880 --> 00:10:54,477
only people who was earlier than
me in in seeing the potential of the

192
00:10:54,560 --> 00:10:58,637
technology. He also has a neuroscience
background like me. So I you know I

193
00:10:58,720 --> 00:11:02,077
think that helps us to get along even
as there are some competitive aspects

194
00:11:02,160 --> 00:11:05,197
between the companies.
>> What was your reaction when he won the

195
00:11:05,280 --> 00:11:09,597
Nobel Prize this year?
>> Uh I I was I was very happy for uh

196
00:11:09,680 --> 00:11:13,517
Demis. Uh uh actually you know I
was writing this essay about the potential

197
00:11:13,600 --> 00:11:18,077
of AI in biology called machines
of loving grace. And two days before I was

198
00:11:18,160 --> 00:11:20,077
about to publish it of course Demis and

199
00:11:20,160 --> 00:11:22,557
his colleagues Alphafold won the won the

200
00:11:22,640 --> 00:11:27,117
Nobel Prize. So um uh you know I I I
ended up dedicating the essay to the

201
00:11:27,200 --> 00:11:29,197
2024 you know w w w w w w w w w w w w w
w w w w w w w w w w w w w w w w w w w w

202
00:11:29,280 --> 00:11:29,677
w w w w w w w w w w w w w w w w w w w w
w w w w w w w w w w w w w w w w w w w w

203
00:11:29,760 --> 00:11:30,077
w w w w w w w w w w w w w w w w w w w w
w w w w w w w w w w w w w w w w w w w w

204
00:11:30,160 --> 00:11:33,277
w w w w w w w w of the Nobel Prize in
chemistry for for Alphold because it was

205
00:11:33,360 --> 00:11:36,077
an example of the future
that I was talking about.

206
00:11:36,160 --> 00:11:39,997
>> Very [snorts] exciting. So just
getting back to you and Daniela and you're

207
00:11:40,080 --> 00:11:43,997
working on and growing the company
and up it goes and you're doing it right

208
00:11:44,080 --> 00:11:47,597
here in San Francisco, you know, what
are some of the characteristics and the

209
00:11:47,680 --> 00:11:50,877
models that you're looking at now that
are most surprising? But what's really

210
00:11:50,960 --> 00:11:55,037
exciting about, you know, as you as you
see the next next generations of these

211
00:11:55,120 --> 00:11:58,557
models unfold? >> Yeah.
You know, I think broadly it's the

212
00:11:58,640 --> 00:12:03,037
jump to being able to do agentic
abilities. Being able to do things.

213
00:12:03,120 --> 00:12:04,237
>> What was it? I didn't hear the word.

214
00:12:04,320 --> 00:12:07,357
>> Oh, sorry. Agentic abilities. Um, uh,

215
00:12:07,440 --> 00:12:09,997
uh, uh, turning the models into agents

216
00:12:10,080 --> 00:12:13,117
that can Oh. Oh, I see. You've been I

217
00:12:13,200 --> 00:12:17,677
>> don't worry about it, Dario. Don't
worry. I I wasn't. I just wanted I'm a

218
00:12:17,760 --> 00:12:20,397
little weak in my right ear.
genuinely thought you didn't hear.

219
00:12:20,480 --> 00:12:26,120
>> I see. I understand. [laughter] >> Um,

220
00:12:26,320 --> 00:12:29,037
>> uh, yeah, >> it
was very funny, wasn't it?

221
00:12:29,120 --> 00:12:32,797
>> Definitely a buzz word and you've,
you know, perhaps contributed more more to

222
00:12:32,880 --> 00:12:36,157
mo more than most to to making it
to making it a buzz word. But I, you know,

223
00:12:36,240 --> 00:12:37,677
I think I think it's I think it's a real

224
00:12:37,760 --> 00:12:41,277
thing, which is being able to do tasks

225
00:12:41,360 --> 00:12:43,277
end to end in a way that was not

226
00:12:43,360 --> 00:12:45,757
possible before. And I think we see it

227
00:12:45,840 --> 00:12:49,837
more uh you know in code than uh you

228
00:12:49,920 --> 00:12:52,237
know we see it in in any other area. But

229
00:12:52,320 --> 00:12:54,477
I think code is just a leading indicator

230
00:12:54,560 --> 00:12:57,277
honestly in other tasks in tasks like

231
00:12:57,360 --> 00:12:59,837
financial services in tasks like you

232
00:12:59,920 --> 00:13:04,877
know health and medical in tasks like uh
you know insurance or manufacturing many

233
00:13:04,960 --> 00:13:08,797
of the regulated industries that we're
working together on our on our on our

234
00:13:08,880 --> 00:13:13,677
partnership agents are agents are just
as you know valuable. you just as much

235
00:13:13,760 --> 00:13:17,277
want to do things end to end, right?
You think of the you know the sales and go

236
00:13:17,360 --> 00:13:21,517
to market and you know customer
management workflows. You want to string

237
00:13:21,600 --> 00:13:24,397
a bunch of things together and you want
the models to pull in a bunch of things

238
00:13:24,480 --> 00:13:27,437
in the context and learn this, do that,

239
00:13:27,520 --> 00:13:31,677
access this information. Um so it's it's
just as valuable. It's just the coders

240
00:13:31,760 --> 00:13:35,037
are early adopters. So you know they've
they've kind of given us a glimpse of

241
00:13:35,120 --> 00:13:39,037
the future because they they adopt
things so quickly. But I think we're

242
00:13:39,120 --> 00:13:42,477
going to see a similar
spike in in other areas.

243
00:13:42,560 --> 00:13:46,637
So I want to ask you I have a few a few
questions in that area. So the f you

244
00:13:46,720 --> 00:13:49,997
know when we first started getting
in probably I think one of our very first

245
00:13:50,080 --> 00:13:54,637
conversations we were starting
to talk about this theory and also AGI

246
00:13:54,720 --> 00:13:56,877
especially and what the potential in the

247
00:13:56,960 --> 00:13:59,597
future and AGI at that time and this is

248
00:13:59,680 --> 00:14:04,157
kind of now we go back several years was
kind of being defined as yeah the model

249
00:14:04,240 --> 00:14:07,837
is almost kind of recursive.
It's building itself. It's growing. It's

250
00:14:07,920 --> 00:14:10,557
evolving. It knows it is coding itself.

251
00:14:10,640 --> 00:14:14,877
it can code and that it's it's enhancing
and it's off and going off and you know

252
00:14:14,960 --> 00:14:19,117
we had Brett Adcock on the stage earlier
today and he was talking about how kind

253
00:14:19,200 --> 00:14:23,277
of like had a scene out of you know we
are Bob we are Legion that the robot had

254
00:14:23,360 --> 00:14:24,957
taken off and was now you know

255
00:14:25,040 --> 00:14:27,437
self-replicating into the future and you

256
00:14:27,520 --> 00:14:30,877
know an infinite number of uh you know

257
00:14:30,960 --> 00:14:35,037
uh robots it was it was incredibly

258
00:14:35,120 --> 00:14:40,477
compelling would go off and that was
that you know to this to that point a

259
00:14:40,560 --> 00:14:44,077
few years ago when we were talking it
was like well now the model is kind of

260
00:14:44,160 --> 00:14:47,837
building itself it's growing itself it's
develop it didn't turn out exactly like

261
00:14:47,920 --> 00:14:51,837
that though right or it's not
we're not at that point yet are we

262
00:14:51,920 --> 00:14:55,277
>> yeah so you know I would separate two
things there there's I think inevitably

263
00:14:55,360 --> 00:14:59,437
as we you know talk about AI in the
future this kind of sci-fi sheen on on

264
00:14:59,520 --> 00:15:03,357
on on you know the on the whole thing
and that's you know in a way it's

265
00:15:03,440 --> 00:15:07,597
helpful as a metaphor but also it can be
distracting because you know it just it

266
00:15:07,680 --> 00:15:09,277
just kind of conjures everyone's mind

267
00:15:09,360 --> 00:15:12,077
into into this kind of uh you know into

268
00:15:12,160 --> 00:15:14,797
this kind of uh you know future dreamy

269
00:15:14,880 --> 00:15:19,197
mode. Um uh but you know I think there
are some things that are real. One thing

270
00:15:19,280 --> 00:15:22,557
that's real is that you know every 3
months we make a model that's smarter

271
00:15:22,640 --> 00:15:26,317
than the month than the model we made
before. And indeed the models are slowly

272
00:15:26,400 --> 00:15:30,237
getting to where they're you know better
than most humans at most things. Maybe

273
00:15:30,320 --> 00:15:34,477
at some point better all humans than all
things. The other thing I think is real

274
00:15:34,560 --> 00:15:37,037
is you know the model is not building

275
00:15:37,120 --> 00:15:41,597
itself in the sense that like I have
a specific copy of the model that's like

276
00:15:41,680 --> 00:15:45,437
making a specific copy of the the you
know the the next model you know we

277
00:15:45,520 --> 00:15:47,597
don't have our self-replicating you know

278
00:15:47,680 --> 00:15:50,237
nanobots or robots or or whatever but

279
00:15:50,320 --> 00:15:52,637
but what absolutely is happening is that

280
00:15:52,720 --> 00:15:55,117
within anthropic and elsewhere we are

281
00:15:55,200 --> 00:15:58,637
using claude to help build products on

282
00:15:58,720 --> 00:16:01,357
top of claude and to help train the next

283
00:16:01,440 --> 00:16:06,237
claude to help serve serve claude faster.
Um, you know, 6 months ago I I

284
00:16:06,320 --> 00:16:10,557
made this prediction that, you know, in
in 6 months 90% of code would be written

285
00:16:10,640 --> 00:16:14,877
by by by AI models. Some people think
that prediction is wrong, but within

286
00:16:14,960 --> 00:16:18,557
Enthropic and within a number
of companies that we work with. That is

287
00:16:18,640 --> 00:16:23,117
absolutely true. Um now 90 you know >>
so you're saying that 90% of all code

288
00:16:23,200 --> 00:16:26,237
adenthropics being written
by the by the model today

289
00:16:26,320 --> 00:16:29,437
>> on on many teams you
know not uniformly every

290
00:16:29,520 --> 00:16:33,117
>> is that something that you've
kind of provided something kind of

291
00:16:33,200 --> 00:16:37,037
>> something auditable around that or like
shown how that's happened. Yeah, the

292
00:16:37,120 --> 00:16:42,317
teams play an editing and supervisory
role, but you know Claude can both you

293
00:16:42,400 --> 00:16:45,757
can say write this feature and Claude
will do it and then the human kind of

294
00:16:45,840 --> 00:16:48,157
looks over the code or asks for help

295
00:16:48,240 --> 00:16:50,397
from another another model, another copy

296
00:16:50,480 --> 00:16:55,757
of Claude. Um uh the other thing is
sometimes we have these longunning bugs.

297
00:16:55,840 --> 00:16:59,517
you know, when we were training our our
the most recently released Claude, it

298
00:16:59,600 --> 00:17:03,357
was like, you know, the cluster
was broken. There was this bug, engineers

299
00:17:03,440 --> 00:17:07,517
were spending a few days on it, and you
know, we just we just went to Claude and

300
00:17:07,600 --> 00:17:10,317
we said, I don't know, just kind of play
around with the cluster and see if you

301
00:17:10,400 --> 00:17:13,757
can figure out what's wrong. And,
you know, Claude found this just super

302
00:17:13,840 --> 00:17:17,277
obscure bug that the engineers had missed.
So, this this was one area where

303
00:17:17,360 --> 00:17:19,197
it was like Claude is like having a very

304
00:17:19,280 --> 00:17:21,357
talented teammate. So if claude is now

305
00:17:21,440 --> 00:17:24,477
running writing 90% of all of the code

306
00:17:24,560 --> 00:17:29,277
in enthropic do you need less engineers
if 90% of all the code when will it be

307
00:17:29,360 --> 00:17:33,597
95% or 99%. >> Yeah.
So you know I you know I think

308
00:17:33,680 --> 00:17:38,077
people can misinterpret it as you know
you know like Claude writing 90% of the

309
00:17:38,160 --> 00:17:41,677
code means we're firing 90% of the
software engineers. But you know if

310
00:17:41,760 --> 00:17:44,877
you're familiar with the principle of >>
well you have made some very aggressive

311
00:17:44,960 --> 00:17:51,650
comments around firing executives
>> that that so [laughter]

312
00:17:51,920 --> 00:17:53,677
um >> I just want to put
two and two together

313
00:17:53,760 --> 00:17:55,357
to get four. Okay. [laughter]

314
00:17:55,440 --> 00:18:00,717
>> Um uh so uh um you know if >>
one thing I've learned in knowing you

315
00:18:00,800 --> 00:18:03,757
for so long is I have to come back
and kind of like get that qualifying

316
00:18:03,840 --> 00:18:08,157
question again. Yes. Because you're go
you know you you're you're going fast.

317
00:18:08,240 --> 00:18:12,557
So trying to I'm trying to keep up
with you and it's amazing but I wanted to

318
00:18:12,640 --> 00:18:14,477
just make sure I understand
exactly what you're saying.

319
00:18:14,560 --> 00:18:16,077
>> Yes, absolutely. Let me let me let me

320
00:18:16,160 --> 00:18:19,437
take them one by one. So um uh you know

321
00:18:19,520 --> 00:18:22,717
under comparative advantage if Claude is

322
00:18:22,800 --> 00:18:26,157
writing 90% of the code what that means

323
00:18:26,240 --> 00:18:30,957
usually is you need just as many
software engineers you might need more

324
00:18:31,040 --> 00:18:35,277
because they can they can then be more
leveraged. they can focus on the 10%

325
00:18:35,360 --> 00:18:37,277
that's editing the code or writing the

326
00:18:37,360 --> 00:18:40,157
10% that's the hardest or supervising a

327
00:18:40,240 --> 00:18:44,557
group of a of AI models. And so what
happens is, you know, you can just end

328
00:18:44,640 --> 00:18:46,957
up being 10 times more productive

329
00:18:47,040 --> 00:18:50,237
because the remaining 10% is is now a

330
00:18:50,320 --> 00:18:53,997
complement to the to the AI system
that's that's getting much better at at

331
00:18:54,080 --> 00:18:56,477
getting much better at what
it's producing. So So

332
00:18:56,560 --> 00:18:59,597
>> it's a rebalancing of the work
is what you're really saying.

333
00:18:59,680 --> 00:19:00,877
>> It's it's not a replacement.

334
00:19:00,960 --> 00:19:02,077
>> It's not a replacement. It's a

335
00:19:02,160 --> 00:19:05,277
rebalancing. Okay. Now, as that 90% goes

336
00:19:05,360 --> 00:19:10,317
to 95 99 totally automated end to end,
there may be some point at which all

337
00:19:10,400 --> 00:19:12,637
bets are off. >> What's
your time frame for that? You

338
00:19:12,720 --> 00:19:16,637
It's a pretty big jump there. >> It's
very hard to say. Um, but you know,

339
00:19:16,720 --> 00:19:20,397
we're looking at it. We're looking at it
closely. We're always looking for ways

340
00:19:20,480 --> 00:19:24,397
to encourage complimentarity. >> You see
that in the short term or the long term?

341
00:19:24,480 --> 00:19:27,117
>> Um, I you know, >> is that a
fiveyear goal or a one-year goal?

342
00:19:27,200 --> 00:19:29,277
>> I'm optimistic about 10ear goal. I'm

343
00:19:29,360 --> 00:19:31,357
optimistic about complimentarity in the

344
00:19:31,440 --> 00:19:34,317
short run. I you know I I I I do worry

345
00:19:34,400 --> 00:19:37,437
as we get out to two years 5 years that

346
00:19:37,520 --> 00:19:39,997
across the economy not that there won't

347
00:19:40,080 --> 00:19:42,317
be anything for humans to do but that

348
00:19:42,400 --> 00:19:44,637
there will be enough disruption that

349
00:19:44,720 --> 00:19:47,677
that will have a you know a kind of

350
00:19:47,760 --> 00:19:49,837
labor disruption that is you know not

351
00:19:49,920 --> 00:19:52,397
fundamentally different from what we've

352
00:19:52,480 --> 00:19:54,717
seen with previous technologies but that

353
00:19:54,800 --> 00:20:00,157
operates faster that is broader across
the economy because it relates to all. I

354
00:20:00,240 --> 00:20:01,917
think that's in the next
two to five years.

355
00:20:02,000 --> 00:20:03,357
>> Two to five years I would say.

356
00:20:03,440 --> 00:20:05,117
>> Okay. And at what
point just coming back

357
00:20:05,200 --> 00:20:10,557
to my previous question, at what point
is Enthropic the model writing 100% of

358
00:20:10,640 --> 00:20:14,397
the code? >> Yeah. Again,
it's it's a tricky question

359
00:20:14,480 --> 00:20:18,237
because maybe even if the model's
writing 100% of the code, the humans are

360
00:20:18,320 --> 00:20:23,800
still supervising the model
or prompting it or clarif

361
00:20:26,480 --> 00:20:28,557
it's it's it's it's a surprisingly

362
00:20:28,640 --> 00:20:33,437
clever thing. I don't, you know, I I'm
I I I am concerned particularly about

363
00:20:33,520 --> 00:20:37,517
people's ability to adapt as the jobs
are changing fast. The jobs are changing

364
00:20:37,600 --> 00:20:41,037
so fast, but comparative comparative
advantage is powerful. That's going to

365
00:20:41,120 --> 00:20:43,277
be one of our tools in this.
>> Well, you've mentioned healthcare

366
00:20:43,360 --> 00:20:47,357
several times, but we've seen these kind
of very conflicting studies recently in

367
00:20:47,440 --> 00:20:51,357
healthcare around accuracy, especially
because these are word models. There

368
00:20:51,440 --> 00:20:53,357
seems to be some level of accuracy

369
00:20:53,440 --> 00:20:58,877
that's not 100%. and were, you know,
you know, and then when applied to some of

370
00:20:58,960 --> 00:21:02,317
these kind of other things like
radiology is one where we saw these very

371
00:21:02,400 --> 00:21:04,637
interesting, you know, studies where we

372
00:21:04,720 --> 00:21:09,437
just didn't see the accuracy of a
radiologist reflected in what the model

373
00:21:09,520 --> 00:21:13,517
was doing. Is that how you see it going
forward or how did you reflect back on

374
00:21:13,600 --> 00:21:17,357
that? >> Yeah, I I would
say where we are now is

375
00:21:17,440 --> 00:21:20,957
I think of the models and I think a lot
of people who use the models in practice,

376
00:21:21,040 --> 00:21:22,717
>> you know where I'm going, right?

377
00:21:22,800 --> 00:21:26,797
I'm trying to like understand where are
we really in the evolution of the world.

378
00:21:26,880 --> 00:21:30,877
>> Where's the arc? So, so I think a lot
of I I think and a lot of people who use

379
00:21:30,960 --> 00:21:33,357
the models in practice, I think think of

380
00:21:33,440 --> 00:21:36,077
the models as maybe a bit like a second

381
00:21:36,160 --> 00:21:40,717
opinion, right? Where you don't
necessarily trust them alone. Just like

382
00:21:40,800 --> 00:21:44,557
if a doctor says something to me, I you
know, I don't I don't necessarily take

383
00:21:44,640 --> 00:21:47,917
that as fact. They're an expert, but
and they're often right, but they're not

384
00:21:48,000 --> 00:21:50,957
always right. Um uh so
you know I'll tell you

385
00:21:51,040 --> 00:21:54,957
>> especially in this one category
radiology it seems strange like it just

386
00:21:55,040 --> 00:21:57,917
can't get its >> I was going
to say can't get its head

387
00:21:58,000 --> 00:22:01,277
around it but that's not really
the right metaphor but doesn't have

388
00:22:01,360 --> 00:22:05,197
>> it does there's something about either
it doesn't have enough data or it's not

389
00:22:05,280 --> 00:22:07,917
getting it's too nuanced. What is it?

390
00:22:08,000 --> 00:22:11,437
>> Yeah. So let me let me actually tell
you a story about about one case where it

391
00:22:11,520 --> 00:22:15,277
where it where it did work. Um so
you know we're talking about you know

392
00:22:15,360 --> 00:22:18,557
founding the company with with my sister.
She recently had her second

393
00:22:18,640 --> 00:22:22,397
child while she was pregnant. Um, you
know, pregnancy can suppress your immune

394
00:22:22,480 --> 00:22:26,557
system. Uh, was was viral. For some
reason, it didn't occur to to any of

395
00:22:26,640 --> 00:22:29,917
them that it could be a bacterial
infection. She took all her data. She

396
00:22:30,000 --> 00:22:34,637
uploaded it to Claude and, you know,
Claude, you know, systematically listed

397
00:22:34,720 --> 00:22:37,837
the possibilities in the way
that the doctors had somehow missed.

398
00:22:37,920 --> 00:22:41,517
>> That is amazing, right? How many of us
have already used the models in some way

399
00:22:41,600 --> 00:22:43,677
to look at our healthcare?
Raise your hand.

400
00:22:43,760 --> 00:22:45,837
>> Yeah, I think we're
all having exactly that.

401
00:22:45,920 --> 00:22:48,717
>> That's that's a lot of people.
It's amazing, right? Where you can kind of

402
00:22:48,800 --> 00:22:52,397
say, especially how many of you are the
hypochondrics that are doing that? Raise

403
00:22:52,480 --> 00:22:55,677
your hand. Yeah, [laughter] I got a few.

404
00:22:55,760 --> 00:22:59,117
Okay, I know I'm >> But but
but here's the point, which is

405
00:22:59,200 --> 00:23:02,397
Claude listed possibilities. If she had

406
00:23:02,480 --> 00:23:05,757
just blindly taken the advice of Claude,

407
00:23:05,840 --> 00:23:09,757
that would not have gone well. Um Claude
was good for saying, "Hey, have you

408
00:23:09,840 --> 00:23:14,317
considered this? Have you considered
that?" So, so it has this assistive

409
00:23:14,400 --> 00:23:17,197
augmentative ability now >>
especially in regards to words, right?

410
00:23:17,280 --> 00:23:20,237
It's very good on words. >> Words.
Yeah. Yeah. I mean, you know,

411
00:23:20,320 --> 00:23:23,997
this specialized stuff of reading
images like I think that's more about

412
00:23:24,080 --> 00:23:27,757
specialized models like if you don't
have the right data maybe it's not doing

413
00:23:27,840 --> 00:23:31,597
well. It's less reflective of the overall
>> is that the crux of the issue like in

414
00:23:31,680 --> 00:23:33,517
radiology cuz like when we're saying

415
00:23:33,600 --> 00:23:35,917
it's going to improve healthare but we

416
00:23:36,000 --> 00:23:39,197
also saw this weird study where doctors

417
00:23:39,280 --> 00:23:44,157
who were using the models some in some
cases were relying too much of the

418
00:23:44,240 --> 00:23:48,477
models and then ending up with bad
outcomes. On the other side then we saw

419
00:23:48,560 --> 00:23:52,797
models that were like trying to read
radiology but then were not able to

420
00:23:52,880 --> 00:23:56,957
achieve accuracy. there was like this
kind of strange bifurcation between the

421
00:23:57,040 --> 00:24:00,317
two things. >> Yeah.
You know, I think there's a lot

422
00:24:00,400 --> 00:24:04,877
still left to work out in what the role
of the human is and what the role of the

423
00:24:04,960 --> 00:24:09,357
model is. And of course, that's changing
as the models improve rapidly. But, you

424
00:24:09,440 --> 00:24:12,477
know, I think it's something you often
see in new technologies where there's

425
00:24:12,560 --> 00:24:16,797
this great technology and then these
practical issues of exactly how do you

426
00:24:16,880 --> 00:24:21,197
use it, exactly how do you interact with
it, they take surprisingly long to work

427
00:24:21,280 --> 00:24:23,437
out. And you know if you look back at

428
00:24:23,520 --> 00:24:25,997
you know the early days when people you

429
00:24:26,080 --> 00:24:30,557
know kind of you know deployed deployed
new technologies I don't know it's it's

430
00:24:30,640 --> 00:24:33,037
just it's often >> you think
this is going to be about

431
00:24:33,120 --> 00:24:34,877
humans and agents working together.

432
00:24:34,960 --> 00:24:36,957
>> Yeah. No no no exactly just just the

433
00:24:37,040 --> 00:24:40,077
collaboration model needs to be worked

434
00:24:40,160 --> 00:24:43,997
out a bit more than it has in these
practical areas. code is good to point

435
00:24:44,080 --> 00:24:46,557
to like code happens so fast, right?

436
00:24:46,640 --> 00:24:49,757
Developers are early adopters and so

437
00:24:49,840 --> 00:24:53,677
that cycle we've gone through it
in like months, right? You can see people

438
00:24:53,760 --> 00:24:58,077
talking in blog posts about, oh,
you know, what's the best way to use cloud

439
00:24:58,160 --> 00:25:02,717
code? Do I want it to suggest and I make
edits or I write code and it makes edits

440
00:25:02,800 --> 00:25:06,397
and these people riff off each other
and you know, over the course of three or

441
00:25:06,480 --> 00:25:10,557
four months they work out what the best
way to use quad code is. I think that

442
00:25:10,640 --> 00:25:14,077
same thing is happening in healthcare
and radiology, but because there are a

443
00:25:14,160 --> 00:25:18,237
lot of barriers in these industries
because they're high stakes, that that

444
00:25:18,320 --> 00:25:23,197
process is working itself out much slower.
So that's really kind of getting

445
00:25:23,280 --> 00:25:26,717
down to it, right? That it's
a collaborative thing like here's the best

446
00:25:26,800 --> 00:25:29,197
way to use cloud code. Cloud code is

447
00:25:29,280 --> 00:25:31,517
writing the code, but here's really the

448
00:25:31,600 --> 00:25:34,477
best way as a developer now to work on

449
00:25:34,560 --> 00:25:38,797
cloud code. It's not writing a 100%
of the code. It's writing code now. I'm

450
00:25:38,880 --> 00:25:42,877
like checking here's here's and now now
I'm going to have to maintain the code.

451
00:25:42,960 --> 00:25:46,237
I'm going to have to provide all
the capabilities the whole infrastructure

452
00:25:46,320 --> 00:25:48,877
around the code.
Is this the right way to think about it?

453
00:25:48,960 --> 00:25:50,397
>> I I think that's the right way to think

454
00:25:50,480 --> 00:25:53,277
about it right now. Um uh you know as as

455
00:25:53,360 --> 00:25:56,157
the models get stronger then the

456
00:25:56,240 --> 00:26:01,037
collaboration may change to more distant
supervision by the human. Um so it's

457
00:26:01,120 --> 00:26:05,117
it's very hard to say what will happen
even six months from now. But but I

458
00:26:05,200 --> 00:26:08,317
think what you've said is a very
accurate description of where we are now

459
00:26:08,400 --> 00:26:13,437
as of you know our our you know
as of set 4.5 which was the latest model

460
00:26:13,520 --> 00:26:16,797
released at >> which is probably one
of the best coders that we've seen right

461
00:26:16,880 --> 00:26:21,117
>> it it you know every every we you know
we have these benchmarks that are like

462
00:26:21,200 --> 00:26:25,917
you know what fraction of you know
a random sample of GitHub PRs can this

463
00:26:26,000 --> 00:26:29,597
thing write from a description
and that's over the last year and a half

464
00:26:29,680 --> 00:26:32,317
that's gone from like 5% which was

465
00:26:32,400 --> 00:26:37,357
basically the thing can't code
at all to something like you know 77%.

466
00:26:37,440 --> 00:26:41,517
Uh so it's it's it's really >>
and it's also yielding also this whole

467
00:26:41,600 --> 00:26:46,077
other ecosystem of things on top of it
as well like even we showed agent force

468
00:26:46,160 --> 00:26:50,317
fives yesterday that I think our customers
will benefit from or other you

469
00:26:50,400 --> 00:26:53,597
know other tools lovable these things
have been built on these things right

470
00:26:53,680 --> 00:26:57,197
has that been a surprise to you to see
a whole ecosystem on top of the model

471
00:26:57,280 --> 00:27:00,477
>> I mean that's been that's been great
to see there are there are many different

472
00:27:00,560 --> 00:27:04,237
models right if you look at the GitHub
co-pilots of the world you know that's

473
00:27:04,320 --> 00:27:07,677
kind of complimentary to GitHub if you
look at the cursors of the world. That's

474
00:27:07,760 --> 00:27:12,637
kind of an IDE model. If you look at the
lovables and replets, that's more of a

475
00:27:12,720 --> 00:27:16,877
consumer model. It allows people who are
not traditionally software developers to

476
00:27:16,960 --> 00:27:20,797
write code. Um, you know, there's
like a, you know, Verscell is like a

477
00:27:20,880 --> 00:27:25,597
different a different way of looking
at it. Augment code, wind surf, you know,

478
00:27:25,680 --> 00:27:28,957
there's a whole ecosystem. And then
there's the command line tools like like

479
00:27:29,040 --> 00:27:31,837
claude code. Uh, so I think it's too

480
00:27:31,920 --> 00:27:36,077
early to say, you know, what's going
to be the winner there or or even if there

481
00:27:36,160 --> 00:27:39,037
is one winner, it's just possible
there's just different things that work

482
00:27:39,120 --> 00:27:41,997
for different people or or different
>> there's a whole civilization.

483
00:27:42,080 --> 00:27:43,917
>> There's a whole eos capability, right?

484
00:27:44,000 --> 00:27:48,157
It's a whole civilization like it's
an EO it's even it's broader like it's

485
00:27:48,240 --> 00:27:50,797
really coming across the whole span. It

486
00:27:50,880 --> 00:27:53,277
it's really impressive, you know? It's

487
00:27:53,360 --> 00:27:55,917
it's it's so fundamental that it's able

488
00:27:56,000 --> 00:28:00,477
to yield all this incredible innovation
from others. Yeah. No, this I mean

489
00:28:00,560 --> 00:28:03,917
>> which one of those is the most
like the one you did not expect?

490
00:28:04,000 --> 00:28:06,157
>> Which one is the one
that I most did not

491
00:28:06,240 --> 00:28:11,037
expect? Um I think the model of kind of

492
00:28:11,120 --> 00:28:13,597
allowing people to write code end to end

493
00:28:13,680 --> 00:28:16,237
like allowing non-coders to write code.

494
00:28:16,320 --> 00:28:19,917
It totally makes sense once I see it
and it's doing really well as a business

495
00:28:20,000 --> 00:28:23,437
model. But but I I I just when
I was conceptualizing all the

496
00:28:23,520 --> 00:28:27,037
>> which product do you love the best?
Uh yeah, this is this is one of those

497
00:28:27,120 --> 00:28:31,117
questions that you know you know it's
c your customers are like your children.

498
00:28:31,200 --> 00:28:34,157
You you you you really shouldn't give
an answer. So I'm not going to give an

499
00:28:34,240 --> 00:28:36,830
answer. [laughter]

500
00:28:41,360 --> 00:28:43,380
[snorts]

501
00:28:46,600 --> 00:28:50,157
[applause] All right.
So are you getting a sense

502
00:28:50,240 --> 00:28:54,237
for Daario, by the way? [laughter]
Are you getting a sense for Daario?

503
00:28:54,320 --> 00:28:58,877
Okay. So that's I'm trying to like you
know help them to understand you and how

504
00:28:58,960 --> 00:29:00,237
you're thinking about this and where

505
00:29:00,320 --> 00:29:03,357
we're going. So now here you are you

506
00:29:03,440 --> 00:29:05,997
know you're entrepreneur you've left

507
00:29:06,080 --> 00:29:08,877
open AI this thing is taking off. It's

508
00:29:08,960 --> 00:29:12,877
you and Daniela you know um you're in

509
00:29:12,960 --> 00:29:17,837
the Slack building you're get where
is it going to be in five years?

510
00:29:17,920 --> 00:29:22,157
What does it look like? We've seen zero
to five years now. You don't have to

511
00:29:22,240 --> 00:29:24,957
tell me as an investor and there's
nobody hold going to hold you

512
00:29:25,040 --> 00:29:27,197
accountable for it. But just if you just

513
00:29:27,280 --> 00:29:30,637
fantasized out and now it's not 2025 but

514
00:29:30,720 --> 00:29:33,997
it's 2030 and you know where's

515
00:29:34,080 --> 00:29:36,237
Enthropic? What does it look like?

516
00:29:36,320 --> 00:29:40,157
Forget about all the other companies
and all the other competitors whatever.

517
00:29:40,240 --> 00:29:42,717
Nobody cares. Where's Enthropic 5 years

518
00:29:42,800 --> 00:29:47,677
from now? What's your dream big what's
your big dream? So, you know, I I think

519
00:29:47,760 --> 00:29:50,957
I think maybe maybe one of the best
ways to think about this is, you know, I

520
00:29:51,040 --> 00:29:54,637
wrote this essay called Machines of Loving
Grace, where where it's really it

521
00:29:54,720 --> 00:29:57,757
was written from a societal perspective,
but, you know, of course, the company is

522
00:29:57,840 --> 00:30:01,357
trying to enact that and and and
contribute to it, and so you can back

523
00:30:01,440 --> 00:30:05,677
out from that where we where we hope
the the company will be. So, you know, I I

524
00:30:05,760 --> 00:30:08,637
really thought of it as like the upside
of AI, right? There's lots of concerns

525
00:30:08,720 --> 00:30:10,397
with AI. I I you know I'm one of the

526
00:30:10,480 --> 00:30:13,037
loudest in voicing them but but I am you

527
00:30:13,120 --> 00:30:16,477
know I am I am I am incredible believer

528
00:30:16,560 --> 00:30:20,717
in the upside of AI you know I think
I think more so than people who code or

529
00:30:20,800 --> 00:30:23,117
describe themselves as more optimistic I

530
00:30:23,200 --> 00:30:25,917
have just a really expansive view I am

531
00:30:26,000 --> 00:30:28,077
perhaps most excited about what it can

532
00:30:28,160 --> 00:30:30,877
do for curing diseases um you know going

533
00:30:30,960 --> 00:30:33,117
back to my background in biology like

534
00:30:33,200 --> 00:30:35,357
you know we look at these tough diseases

535
00:30:35,440 --> 00:30:38,717
like cancer like Alzheimer's um you know

536
00:30:38,800 --> 00:30:43,197
like like you know various kinds
of metabolic disease like I I I just think

537
00:30:43,280 --> 00:30:47,117
AI is the right thing to cure them
and so I think of all the work we can do

538
00:30:47,200 --> 00:30:50,237
working with pharmaceutical companies
working with scientific research

539
00:30:50,320 --> 00:30:54,157
institutions and working with the health
care system to kind of enact it out and

540
00:30:54,240 --> 00:30:57,597
you know you again our partnership
is about regulated industries and one of

541
00:30:57,680 --> 00:31:02,077
the regulated industries I think
especially is is healthcare but you know

542
00:31:02,160 --> 00:31:05,997
that is one of the most positive
applications but I think that we will

543
00:31:06,080 --> 00:31:08,237
see applications across the industry I'm

544
00:31:08,320 --> 00:31:10,717
very excited about financial services.

545
00:31:10,800 --> 00:31:13,437
I'm very excited about insurance. I'm

546
00:31:13,520 --> 00:31:15,917
very excited about taking basically

547
00:31:16,000 --> 00:31:19,437
every enterprise um and learning to

548
00:31:19,520 --> 00:31:24,077
serve every enterprise with smarter and
smarter models. And the skill ceiling is

549
00:31:24,160 --> 00:31:27,917
very high. One way to say it is what
is an enterprise? It's kind of a super

550
00:31:28,000 --> 00:31:29,997
intelligence, right? It it is an entity

551
00:31:30,080 --> 00:31:31,917
that acts in the world in in in in in in

552
00:31:32,000 --> 00:31:35,517
in a way with a level of strategy and

553
00:31:35,600 --> 00:31:39,277
and knowledge and power in the world
that goes far beyond what any individual

554
00:31:39,360 --> 00:31:43,357
human can do. So, you know, even
as these models get smarter than than than

555
00:31:43,440 --> 00:31:47,597
any than any individual human, there's
a ceiling substantially higher than that

556
00:31:47,680 --> 00:31:51,917
in terms of doing the things that a
company tries to do in the world. Um,

557
00:31:52,000 --> 00:31:55,357
you know, most of which are good for
the world, most of which create economic

558
00:31:55,440 --> 00:32:00,157
value. Um and and and so then my my
vision of Enthropic is a company that

559
00:32:00,240 --> 00:32:01,837
serves the enterprise of the world that

560
00:32:01,920 --> 00:32:04,797
transforms every every enterprise puts

561
00:32:04,880 --> 00:32:07,357
AI at the center at the center of every

562
00:32:07,440 --> 00:32:09,837
enterprise. Um uh you know and and and

563
00:32:09,920 --> 00:32:12,077
and and you know I think I won't give

564
00:32:12,160 --> 00:32:14,557
specific numbers but like if you if you

565
00:32:14,640 --> 00:32:16,877
think about the actual market potential

566
00:32:16,960 --> 00:32:22,077
of that is very very high right our
revenue has been going up 10x per year.

567
00:32:22,160 --> 00:32:24,797
It's currently, I think it was reported
this morning, which I'm happy to confirm

568
00:32:24,880 --> 00:32:28,077
is correct, is approaching approaching 7

569
00:32:28,160 --> 00:32:32,797
billion run rate. Um, that's gone up
10x a year for the for the last 3 years.

570
00:32:32,880 --> 00:32:36,957
It's can't continue like that for much
longer, of course, but uh you know,

571
00:32:37,040 --> 00:32:40,797
>> as an investor, I'm looking forward
to the $70 billion year. You should not

572
00:32:40,880 --> 00:32:43,597
feel bad investing at 4 bill. You should

573
00:32:43,680 --> 00:32:47,837
not feel bad. Um uh uh uh so you know

574
00:32:47,920 --> 00:32:52,397
but I I I think the potential is is
is very large but but I think even even

575
00:32:52,480 --> 00:32:56,717
more important than that or perhaps even
necessary to that success. We want to be

576
00:32:56,800 --> 00:32:59,197
known for, you know, the company that is

577
00:32:59,280 --> 00:33:01,677
trustworthy, the company that behaves

578
00:33:01,760 --> 00:33:04,397
responsibly, the company that has a that

579
00:33:04,480 --> 00:33:07,517
has a reputation for, you know, doing

580
00:33:07,600 --> 00:33:11,917
what it says, saying what it does,
and building products that are trustworthy

581
00:33:12,000 --> 00:33:15,677
that, you know, if you if you if
you build on us, you won't get any

582
00:33:15,760 --> 00:33:18,237
surprises. You won't get anything crazy.

583
00:33:18,320 --> 00:33:21,277
You won't get the model, you know,
saying it's Hitler. you won't you won't

584
00:33:21,360 --> 00:33:24,477
get, you know, all kinds of um you won't

585
00:33:24,560 --> 00:33:28,317
get all kinds of crazy stuff. Um and
and I think that's very important to us

586
00:33:28,400 --> 00:33:31,437
being good stewards of this.
>> Are there certain companies that you're

587
00:33:31,520 --> 00:33:33,117
trying to imply that that might be?

588
00:33:33,200 --> 00:33:36,317
>> None whatsoever. >> There's
an entrepreneur down in Palo Alto.

589
00:33:36,400 --> 00:33:38,717
>> None at all. I don't know
what you're talking about.

590
00:33:38,800 --> 00:33:42,637
>> Okay, I see. [laughter] That's
a certain letter in the alphabet.

591
00:33:42,720 --> 00:33:45,357
>> I don't I don't I I have no
idea what you're talking about. Um

592
00:33:45,440 --> 00:33:50,437
>> Okay, that's before the letter Y.
I sorry I still don't know what you mean.

593
00:33:50,520 --> 00:33:53,677
[laughter] >> Okay,
one more thing. I'm going to take

594
00:33:53,760 --> 00:33:56,397
a couple I'm going to use moderator
privilege here and take a couple extra

595
00:33:56,480 --> 00:34:00,157
minutes with you. So number one, we see

596
00:34:00,240 --> 00:34:03,117
right now a tremendous fever on building

597
00:34:03,200 --> 00:34:06,797
data centers and you know we've seen a

598
00:34:06,880 --> 00:34:09,917
tremendous fever around the chips, the

599
00:34:10,000 --> 00:34:14,077
financing, the size and scale of the
data centers, the energy of the data

600
00:34:14,160 --> 00:34:15,757
centers. you haven't really addressed it

601
00:34:15,840 --> 00:34:18,077
in your comments per se besides the

602
00:34:18,160 --> 00:34:23,677
scaling laws but let's talk about it
just in part and parcel to enthropic and

603
00:34:23,760 --> 00:34:27,037
when you think about obviously one
of your core partners has been Amazon it's

604
00:34:27,120 --> 00:34:29,517
been very important for your growth very

605
00:34:29,600 --> 00:34:34,077
important company um even though you
do have you obviously relationship with

606
00:34:34,160 --> 00:34:38,157
others as well but that one has
been especially important to you

607
00:34:38,240 --> 00:34:41,197
know just give us a vision of what's
happening with these data centers right

608
00:34:41,280 --> 00:34:45,437
now what are we going to see in the next
few years with these data centers and

609
00:34:45,520 --> 00:34:48,397
how do you think about it as an
entrepreneur who's going to need data

610
00:34:48,480 --> 00:34:52,797
centers to grow? Who will you go to? How
will you do it? Will you build your own?

611
00:34:52,880 --> 00:34:56,077
Will you just continue to partner?
What what what is your plan?

612
00:34:56,160 --> 00:34:59,597
>> Yeah. So, you know, my my my view is is

613
00:34:59,680 --> 00:35:03,837
that there is going to be an enormous
need for compute for for all the reasons

614
00:35:03,920 --> 00:35:07,597
that uh you know that that we've laid out,
right? The the scaling laws suggest

615
00:35:07,680 --> 00:35:10,397
that we're going to need more and more
compute to train smarter and smarter

616
00:35:10,480 --> 00:35:13,757
models. And then as the revenue goes up,
we're gonna we're, you know, we're going

617
00:35:13,840 --> 00:35:17,117
to need to serve more. We're going
to need to serve more of our models. So

618
00:35:17,200 --> 00:35:22,317
when that revenue is not seven billion,
but a hundred billion or more. Um uh and

619
00:35:22,400 --> 00:35:25,837
you know, when when model training
requires millions of chips, you know,

620
00:35:25,920 --> 00:35:29,837
we're we're going to need many many
gigawatts of power to to train the

621
00:35:29,920 --> 00:35:31,917
models. So we're trying to secure that

622
00:35:32,000 --> 00:35:36,877
power basically wherever we can. Um you
know, and I'm confident in the company's

623
00:35:36,960 --> 00:35:38,557
uh I'm confident in the company's

624
00:35:38,640 --> 00:35:41,597
ability to do so. I I I do get a bit

625
00:35:41,680 --> 00:35:44,397
worried when the press focuses too much

626
00:35:44,480 --> 00:35:47,357
on the building of the data centers and

627
00:35:47,440 --> 00:35:51,677
deals to build data centers. Some
of these deals seem a little bit fishy.

628
00:35:51,760 --> 00:35:55,517
They seem like they're maybe double
counting. And I I also worry again

629
00:35:55,600 --> 00:35:57,117
>> deals are you ask talking about?

630
00:35:57,200 --> 00:35:58,157
>> I don't know what you're talking about.

631
00:35:58,240 --> 00:36:03,437
I have no idea. Um uh uh uh >> who
is double counting? These are all

632
00:36:03,520 --> 00:36:07,037
public companies. I I >> We're going
to want to know who's double counting.

633
00:36:07,120 --> 00:36:09,517
>> I don't I don't I don't know.
I don't know anything more than you.

634
00:36:09,600 --> 00:36:11,037
>> What about the ones who
are doing triple counting?

635
00:36:11,120 --> 00:36:14,557
>> I I don't There might be some
of that, too. Wow. All right.

636
00:36:14,640 --> 00:36:17,917
>> I don't know. Um uh uh uh but but but

637
00:36:18,000 --> 00:36:20,557
look look look at the end of the day,

638
00:36:20,640 --> 00:36:22,797
all this data center stuff is spending

639
00:36:22,880 --> 00:36:25,197
money, right? Um I think the thing

640
00:36:25,280 --> 00:36:27,997
that's most important is making money.

641
00:36:28,080 --> 00:36:30,797
Um and and so this is why I focus on,

642
00:36:30,880 --> 00:36:33,197
you know, the revenue and the actual

643
00:36:33,280 --> 00:36:35,357
work that we're doing with businesses.

644
00:36:35,440 --> 00:36:38,317
Andropic is growing faster than any

645
00:36:38,400 --> 00:36:42,477
other company in the space. The revenue
is growing faster and you know I think I

646
00:36:42,560 --> 00:36:46,557
think basically if if you know if the if
the demand is there if customers want

647
00:36:46,640 --> 00:36:50,317
the money and they're willing to pay you
then you are willing to pay your compute

648
00:36:50,400 --> 00:36:54,477
providers and you'll get the compute
you need at the end of the day. Daria, we

649
00:36:54,560 --> 00:36:56,637
could not be more grateful for you to be

650
00:36:56,720 --> 00:37:01,677
here, to be in San Francisco, to be
a native son, and to Daniela as well, and

651
00:37:01,760 --> 00:37:03,117
for everything that you're doing with

652
00:37:03,200 --> 00:37:09,240
Enthropic. Congratulations.
Please give Dario a round of applause.

653
00:37:16,200 --> 00:37:18,220
>> [music]