1
00:00:09,760 --> 00:00:13,237
Daniela, welcome to Stanford
and View from the Top.

2
00:00:13,320 --> 00:00:15,517
>> Thank you so much for having me.

3
00:00:15,600 --> 00:00:17,957
>> We're so excited to hear to be with you

4
00:00:18,040 --> 00:00:19,757
here today. As you can see, we have a

5
00:00:19,840 --> 00:00:23,197
full house and I want to start by asking

6
00:00:23,280 --> 00:00:28,157
everyone in the audience, please raise
your hand if you're a user of Claude.

7
00:00:28,240 --> 00:00:32,677
OH, WOW! >> [screaming] >> OKAY,
BUT I FEEL LIKE IF YOU had Sam

8
00:00:32,760 --> 00:00:36,397
Altman visit and you asked about ChatGPT,

9
00:00:36,480 --> 00:00:38,077
would you all have raised your hands?

10
00:00:38,160 --> 00:00:42,317
Like, let's be real. Okay. >> Still,
you have a very supportive crowd

11
00:00:42,400 --> 00:00:45,437
down here.
And you and your brother helped build

12
00:00:45,520 --> 00:00:47,877
one of the most important AI companies

13
00:00:47,960 --> 00:00:52,917
in the world. Yet, neither of you
grew up planning for any of this. Your

14
00:00:53,000 --> 00:00:57,437
background is in the arts. You studied
English literature and you your early

15
00:00:57,520 --> 00:01:01,997
career was in politics.
Can you share more about what was your

16
00:01:02,080 --> 00:01:05,877
original career plan? >> Oh my god.
First of all, so nice of you

17
00:01:05,960 --> 00:01:09,597
to use the word plan in there.
Um I don't know that I necessarily would

18
00:01:09,680 --> 00:01:14,157
have described it that way at any point
in time, but I I I really think, you

19
00:01:14,240 --> 00:01:17,997
know, this is sort of a story that I
hear from a lot of people who either

20
00:01:18,080 --> 00:01:21,517
wind up, you know, founding a company
or just doing something a little bit

21
00:01:21,600 --> 00:01:25,757
unexpected or unconventional with their
life. It was following what was most

22
00:01:25,840 --> 00:01:30,437
interesting to me at the time and the
intersection of what was I good at, what

23
00:01:30,520 --> 00:01:34,637
was I interested in, and what was going
to have a big impact in the world.

24
00:01:34,720 --> 00:01:37,717
So, for me, I think coming out of college,
that looked like, by the way, I

25
00:01:37,800 --> 00:01:41,837
graduated in 2009, which was like not
the most fun year to be a graduate. You

26
00:01:41,920 --> 00:01:46,197
were like, I have a literature degree
and no skills. Who will hire me?

27
00:01:46,280 --> 00:01:49,637
Um but at the time,
I felt this very strong

28
00:01:49,720 --> 00:01:54,197
pull towards wanting to make the world
better. I think that was always that

29
00:01:54,280 --> 00:01:57,917
sort of defining feature of both both
me and Dario from a young age, and where

30
00:01:58,000 --> 00:02:01,517
that started out for me was in in
international development, working in

31
00:02:01,600 --> 00:02:05,917
global health. And I think my desire
there was really to figure out like, how

32
00:02:06,000 --> 00:02:10,197
do we build a world that is, you know,
fair, where everybody has access to

33
00:02:10,280 --> 00:02:14,677
basic things like food and water
and medicine. And even though that's, you

34
00:02:14,760 --> 00:02:19,357
know, not what I directly work on now,
I think that early grounding sort of gave

35
00:02:19,440 --> 00:02:22,237
me this foundation for thinking about
how do you do good in the world, right?

36
00:02:22,320 --> 00:02:26,597
How do you build something that is
of consequence and has a real purpose

37
00:02:26,680 --> 00:02:30,637
behind where you're spending, you know,
50 or 60 of your hours per week, right?

38
00:02:30,720 --> 00:02:34,077
What you're working on. Um and so it was
kind of a winding journey from there. I

39
00:02:34,160 --> 00:02:37,557
worked on Capitol Hill after that.
I worked on a campaign. Uh and eventually

40
00:02:37,640 --> 00:02:39,957
I ended up coming back to Silicon
Valley. I'm originally from San

41
00:02:40,040 --> 00:02:43,877
Francisco and started uh working at
Stripe, which was this tiny company that

42
00:02:43,960 --> 00:02:47,237
at the time like nobody had heard of.
My friends on Capitol Hill were like,

43
00:02:47,320 --> 00:02:48,397
"You're leaving to go do what?

44
00:02:48,480 --> 00:02:51,637
Payments?" At like now it looks like
a great decision, but at the time it was

45
00:02:51,720 --> 00:02:55,237
about 40 people. Um and then from there
things really just snowballed into

46
00:02:55,320 --> 00:02:58,077
working at OpenAI and then
co-founding Anthropic.

47
00:02:58,160 --> 00:02:59,957
>> You've moved across
fields without being

48
00:03:00,040 --> 00:03:03,717
constrained by what you
studied or had done before.

49
00:03:03,800 --> 00:03:07,477
Where does that mindset come from? That
your background doesn't have to define

50
00:03:07,560 --> 00:03:10,517
your next move. >> So, you know,
it's really interesting. I

51
00:03:10,600 --> 00:03:14,437
think in some ways I really think
of myself as a generalist, right? That if

52
00:03:14,520 --> 00:03:16,837
you were to sort of look through
my background, you would be like, what is

53
00:03:16,920 --> 00:03:19,557
this lady actually good at, right?
She doesn't She doesn't have a She doesn't

54
00:03:19,640 --> 00:03:21,517
have a law degree. She's not a computer

55
00:03:21,600 --> 00:03:24,717
scientist. But I think this concept that

56
00:03:24,800 --> 00:03:27,637
um the ability to be curious and learn

57
00:03:27,720 --> 00:03:31,757
across a lot of disciplines, to have
a strong foundation of of wanting to have

58
00:03:31,840 --> 00:03:35,797
impact regardless of of the area
that you're working on, I think that is I

59
00:03:35,880 --> 00:03:39,597
think that's an underrated quality.
And I see it a lot in the people that we

60
00:03:39,680 --> 00:03:42,717
hire at Anthropic, and the really
talented people that I've worked with in

61
00:03:42,800 --> 00:03:44,797
the technology industry more broadly.

62
00:03:44,880 --> 00:03:48,437
People who are curious and smart
and they want to learn and they want to be

63
00:03:48,520 --> 00:03:50,437
helpful. That is like the first of all,

64
00:03:50,520 --> 00:03:55,517
the description of like every role
except for like engineer at a startup.

65
00:03:55,600 --> 00:03:58,997
Like you're like it's like what you know,
whatever. People there like I have

66
00:03:59,080 --> 00:04:01,477
this degree, I have that degree
and those are sort of the qualities that

67
00:04:01,560 --> 00:04:04,997
you're looking for.
But I think for me it was always very It

68
00:04:05,080 --> 00:04:09,197
was very interest and impact driven. So
I was like, okay, there's something that

69
00:04:09,280 --> 00:04:13,077
feels really wrong to me about the fact
that you know, I was born in America. I

70
00:04:13,160 --> 00:04:15,957
had access to all of the sort
of basic things that we

71
00:04:16,040 --> 00:04:19,037
take for granted in life. Some people
around the world just weren't They just

72
00:04:19,120 --> 00:04:20,237
That's just not where they were born.

73
00:04:20,320 --> 00:04:23,717
That's not what they were born into. How
do I make that more fair? Like how do we

74
00:04:23,800 --> 00:04:27,717
make that better as a broader
community of the world?

75
00:04:27,800 --> 00:04:30,837
And and from there I was like, I'm just
not having the level of impact that I

76
00:04:30,920 --> 00:04:34,397
want to. I need some skills. And so I
went and I worked on a campaign and I

77
00:04:34,480 --> 00:04:38,077
was like, wow, a small number of people
that work really hard who are young and

78
00:04:38,160 --> 00:04:41,717
driven can really change the world. Not
that surprising that eventually that led

79
00:04:41,800 --> 00:04:44,437
me to Silicon Valley where
you're like, you can do that but

80
00:04:44,520 --> 00:04:48,397
it turns out that at a startup you have
a lot more money and it's a lot easier

81
00:04:48,480 --> 00:04:52,157
going than the kind of 80 hours a week
of working on a campaign. But I think

82
00:04:52,240 --> 00:04:55,397
those kind of core qualities around

83
00:04:55,480 --> 00:04:59,677
you know, really following your passion
because you just want to do more when

84
00:04:59,760 --> 00:05:02,317
you care about the thing that you're
working on either intellectually or from

85
00:05:02,400 --> 00:05:04,997
an implicit implicit sense of meaning.

86
00:05:05,080 --> 00:05:06,877
>> And your career in AI started when you

87
00:05:06,960 --> 00:05:09,757
joined OpenAI in 2018 when it was still

88
00:05:09,840 --> 00:05:12,597
a relatively small research lab.

89
00:05:12,680 --> 00:05:16,357
Suddenly you were in rooms where people
were talking about neural networks and

90
00:05:16,440 --> 00:05:18,757
transformers and scaling laws.

91
00:05:18,840 --> 00:05:22,077
How did you learn to speak that language?

92
00:05:22,160 --> 00:05:26,477
>> So I think I was I was well trained
on two dimensions. I think the first is I

93
00:05:26,560 --> 00:05:29,997
had already spent almost six years at
Stripe and so I had worked with a lot of

94
00:05:30,080 --> 00:05:33,157
engineers. Obviously research
and engineering are different but I think

95
00:05:33,240 --> 00:05:35,997
there was some overlap and some grounding.

96
00:05:36,080 --> 00:05:38,237
The second is that you know, I I grew up

97
00:05:38,320 --> 00:05:41,077
with a very talented technology oriented

98
00:05:41,160 --> 00:05:45,517
physicist who's my sibling and obviously
my co-founder at Anthropic among five

99
00:05:45,600 --> 00:05:48,517
others who are also all
engineers or researchers.

100
00:05:48,600 --> 00:05:51,637
Um but I think the number one thing I
would say is both of those experiences

101
00:05:51,720 --> 00:05:53,477
just instilled in me the sense to just

102
00:05:53,560 --> 00:05:55,877
not be afraid of technology, right? It

103
00:05:55,960 --> 00:05:59,397
was ultimately it's it's a set of skills

104
00:05:59,480 --> 00:06:03,117
that are really highly prized, I think,
but it's something that anybody can

105
00:06:03,200 --> 00:06:05,637
learn. And the sort of basics behind it,

106
00:06:05,720 --> 00:06:07,917
I think the the the terminology and the

107
00:06:08,000 --> 00:06:13,357
jargon can feel overwhelming at first,
but if you just ask enough questions and

108
00:06:13,440 --> 00:06:16,837
if you have people who are kind enough
to be patient with you, which I I had in

109
00:06:16,920 --> 00:06:20,517
my life. I was very lucky for that. Um
you I just kept asking questions until I

110
00:06:20,600 --> 00:06:24,077
felt like I could understand it. Um and I
think you know, the second part of it

111
00:06:24,160 --> 00:06:27,917
was also just like knowing my lane
and their lane. So, there were a lot of

112
00:06:28,000 --> 00:06:31,237
things that uh that the researchers
I was like I I probably couldn't have

113
00:06:31,320 --> 00:06:35,517
trained GP I certainly couldn't have
trained GPT-2 or GPT-3, but like I I

114
00:06:35,600 --> 00:06:38,517
brought things to the table that they
didn't know how to do as well. And so, I

115
00:06:38,600 --> 00:06:42,037
really think understanding what your
comparative advantage is and knowing how

116
00:06:42,120 --> 00:06:45,717
you fit into the broader ecosystem, um
that takes a lot of skills that are kind

117
00:06:45,800 --> 00:06:49,797
of interpersonal. Um curiosity, I think
is is a very inherent skill, but one

118
00:06:49,880 --> 00:06:53,557
that you can learn and train. And I
think all of those sort of put together

119
00:06:53,640 --> 00:06:57,877
really put me in this position to be able
to be successful in that type of role.

120
00:06:57,960 --> 00:07:01,757
>> And in December of 2020, you,
your brother, along with a group of

121
00:07:01,840 --> 00:07:04,557
colleagues left OpenAI.

122
00:07:04,640 --> 00:07:09,760
Why did you and Dario
decide to start Anthropic?

123
00:07:11,920 --> 00:07:13,637
>> I'm going to take a big sip of water.

124
00:07:13,720 --> 00:07:18,160
That one. Take a I'll
take one more, actually.

125
00:07:18,960 --> 00:07:22,117
So, um there were seven of us that are

126
00:07:22,200 --> 00:07:26,557
originally left. Me and Dario and our
five co-founders. And then a number of

127
00:07:26,640 --> 00:07:28,157
people that kind of came over shortly

128
00:07:28,240 --> 00:07:30,277
shortly thereafter. And I I think for

129
00:07:30,360 --> 00:07:34,077
us, it really it really came down to

130
00:07:34,160 --> 00:07:38,277
just the focus on what it was we
really wanted the ultimate impact of the

131
00:07:38,360 --> 00:07:43,317
technology to be.
And I think in our own, you know, very

132
00:07:43,400 --> 00:07:46,997
different ways, all seven
of us are people that have

133
00:07:47,080 --> 00:07:48,717
a lot of integrity. We're people who

134
00:07:48,800 --> 00:07:51,397
care a lot about the impact of what it

135
00:07:51,480 --> 00:07:54,197
is that we build. And I think eventually

136
00:07:54,280 --> 00:07:56,157
it felt like it was just easier for us

137
00:07:56,240 --> 00:07:59,797
to sort of create the type of vision

138
00:07:59,880 --> 00:08:04,877
that we saw outside of the company we
were in and in a new company. And I I

139
00:08:04,960 --> 00:08:07,437
frequently say cuz I'm I know it'll
shock you, I'm actually asked this

140
00:08:07,520 --> 00:08:11,797
question not infrequently. Um and

141
00:08:11,880 --> 00:08:15,637
you know, we really, I think, were
running towards something versus running

142
00:08:15,720 --> 00:08:17,117
away from something. Like we had this

143
00:08:17,200 --> 00:08:20,037
vision in our heads of wanting to create

144
00:08:20,120 --> 00:08:23,077
an organization where the values that

145
00:08:23,160 --> 00:08:27,877
matter to us around safety,
around responsibility, um

146
00:08:27,960 --> 00:08:30,677
were the kind of forefront of what
we were doing. That's why we chose to

147
00:08:30,760 --> 00:08:34,077
incorporate it as a public benefit
corporation. That took a while to figure

148
00:08:34,160 --> 00:08:37,997
out what was the right kind of form
factor to express, look, we are going to

149
00:08:38,080 --> 00:08:41,477
be a commercial entity. We think there's
going to be a lot of economic value

150
00:08:41,560 --> 00:08:45,437
that's going to be created by artificial
intelligence, but it's really important

151
00:08:45,520 --> 00:08:47,437
to us that we do this the right way.

152
00:08:47,520 --> 00:08:50,997
And I think that was something that kind
of united the seven of us. We had We had

153
00:08:51,080 --> 00:08:54,917
all worked on a combination of both
capabilities and safety and policy work

154
00:08:55,000 --> 00:08:58,637
when we were at OpenAI. And it felt
like it was just easier to kind of create

155
00:08:58,720 --> 00:09:01,157
this structure in in a new form.

156
00:09:01,240 --> 00:09:04,357
>> And you're building Anthropic not only
with your brother, but with five other

157
00:09:04,440 --> 00:09:08,997
co-founders. Many of us here are about
to choose co-founders for the first

158
00:09:09,080 --> 00:09:12,397
time, and we all know
how often that ends badly.

159
00:09:12,480 --> 00:09:15,237
What does it take to make it work?

160
00:09:15,320 --> 00:09:18,957
>> So, you know, I think of myself as just

161
00:09:19,040 --> 00:09:21,677
extremely lucky. Like the seven of us, I

162
00:09:21,760 --> 00:09:25,637
think, are a very special group on on a

163
00:09:25,720 --> 00:09:29,797
couple of dimensions. I think
the first thing I would say is like the

164
00:09:29,880 --> 00:09:33,997
interpersonal relationships, I know this
is going to sound not that surprising,

165
00:09:34,080 --> 00:09:36,157
matter a lot more than you think. Like

166
00:09:36,240 --> 00:09:38,757
how do you guys have conflict together?

167
00:09:38,840 --> 00:09:44,437
So, for example, Dario and I have been
fighting and getting over it for almost

168
00:09:44,520 --> 00:09:48,157
40 years. >> [laughter] >> Because
he's my brother. And like, I

169
00:09:48,240 --> 00:09:51,797
used to steal his toys, right? So, like,
we know how to work through conflict

170
00:09:51,880 --> 00:09:54,677
together and like, there's no question
like, we will love each other at the end

171
00:09:54,760 --> 00:09:56,957
of that, no matter what. Um with our

172
00:09:57,040 --> 00:10:00,317
co-founders, it's like, I've known Jared

173
00:10:00,400 --> 00:10:02,317
for like 15 years. I've known Chris for

174
00:10:02,400 --> 00:10:07,637
15 years. Tom and Sam were roommates.
Um Jared and Sam worked together at

175
00:10:07,720 --> 00:10:11,117
Stanford when they were in their PhD
program. So, there was this kind of long

176
00:10:11,200 --> 00:10:16,197
history. Um both Dario and I actually
managed all of the other co-founders and

177
00:10:16,280 --> 00:10:19,957
like, I think they had either reported
to one of us or both of us. I think the

178
00:10:20,040 --> 00:10:23,437
majority reported to both of us at OpenAI.
So, we already had this kind of

179
00:10:23,520 --> 00:10:26,557
existing structure and framework of like,
we knew how to give each other

180
00:10:26,640 --> 00:10:28,397
feedback. We knew how to work together.

181
00:10:28,480 --> 00:10:33,117
We'd like, understood like, who we
were as people. And I I think the other

182
00:10:33,200 --> 00:10:38,157
really important thing is just making
sure you have a very strong sense of

183
00:10:38,240 --> 00:10:41,317
what it is you're trying to do and that
that picture is the same. Like, if you

184
00:10:41,400 --> 00:10:45,837
locked yourself and and your co-founder
in another room and you wrote down or

185
00:10:45,920 --> 00:10:48,437
drew a picture of like, what it is you're
trying to build, you're not going

186
00:10:48,520 --> 00:10:50,797
to walk out and like, one has drawn
a unicorn and the other has drawn a

187
00:10:50,880 --> 00:10:54,157
platypus, right? Like, that's the type
of situation where you think you're

188
00:10:54,240 --> 00:10:57,637
doing the same thing, but I think it
just doesn't end well. And I think for

189
00:10:57,720 --> 00:11:01,037
us, this kind of vision of what it is we
wanted to build, in some ways we were

190
00:11:01,120 --> 00:11:03,837
lucky because we had we had been in an
environment where we were like, "Ooh,

191
00:11:03,920 --> 00:11:07,637
it's not quite this. We want to do this
other thing." But, we were like, in the

192
00:11:07,720 --> 00:11:11,437
same zone already. So, we're sort of
pre-selected for that level of interest.

193
00:11:11,520 --> 00:11:15,077
But, I think really being able to like,
pressure test to the degree that you

194
00:11:15,160 --> 00:11:19,597
can, like, instead of like, starting
a company together, like, go on vacation

195
00:11:19,680 --> 00:11:21,517
together. Just like, see how that goes.

196
00:11:21,600 --> 00:11:24,917
Share a room with them. Be like,
"How did that go?" Um and if you're like,

197
00:11:25,000 --> 00:11:28,637
"Man, all I want to do is spend more
time with you." Great. If you're like,

198
00:11:28,720 --> 00:11:31,437
"Really I'm going to need a vacation
to recover from my vacation." You just

199
00:11:31,520 --> 00:11:33,517
might it might be the wrong choice.

200
00:11:33,600 --> 00:11:37,397
>> I want to circle back to what you
just said a few moments ago. Anthropic is

201
00:11:37,480 --> 00:11:41,437
deeply associated with AI safety,
but I want to make sure that everyone

202
00:11:41,520 --> 00:11:43,757
understands what that actually means.

203
00:11:43,840 --> 00:11:46,797
So, when you say AI safety,
what do you mean?

204
00:11:46,880 --> 00:11:49,477
>> Yeah, this is a great question. And I
think this is a term that's gotten like

205
00:11:49,560 --> 00:11:53,157
almost a little bit overloaded
in the past few years because it's sort of

206
00:11:53,240 --> 00:11:55,637
it's kind of a catchall
that I think has been

207
00:11:55,720 --> 00:12:00,157
reached for in a lot of contexts.
But I think to us the the kind of highest

208
00:12:00,240 --> 00:12:04,037
level of framing is just taking a form
of like radical responsibility for the

209
00:12:04,120 --> 00:12:08,757
technology that we're developing. And
the analog that we often use or that we

210
00:12:08,840 --> 00:12:11,757
often point to is social media
companies, which like it's

211
00:12:11,840 --> 00:12:16,237
so in vogue to like crap all over them
in public now, so I'm going to do that.

212
00:12:16,320 --> 00:12:18,477
But basically, you know, like if you

213
00:12:18,560 --> 00:12:21,557
imagine going back in time, I actually

214
00:12:21,640 --> 00:12:24,917
think developers who created these these

215
00:12:25,000 --> 00:12:28,877
technology companies, they weren't like
I'm setting out to cause a pandemic of

216
00:12:28,960 --> 00:12:30,877
eating disorders for teen girls. Right?

217
00:12:30,960 --> 00:12:34,237
That was not their intention. But they
were like, what are the metrics that I'm

218
00:12:34,320 --> 00:12:37,277
trying to optimize for? Right? I'm
trying to build a company. I would like

219
00:12:37,360 --> 00:12:40,997
to see rapid rapid scale growth. Like
let's just build towards that. And there

220
00:12:41,080 --> 00:12:44,277
wasn't this sort of at the time it just
wasn't necessary cuz we'd never seen

221
00:12:44,360 --> 00:12:47,637
something on the level of scale that we
have seen today, how quickly these

222
00:12:47,720 --> 00:12:50,997
companies grew, how many people adopt

223
00:12:51,080 --> 00:12:55,517
the technologies so quickly. But if you
could imagine, you could go back in time

224
00:12:55,600 --> 00:12:57,757
and say like, wow, you're starting

225
00:12:57,840 --> 00:13:02,557
Facebook or Instagram or Snapchat
or Twitter, and you're like, what if I

226
00:13:02,640 --> 00:13:06,837
really try to think through all of the
ways that this could go wrong? Like what

227
00:13:06,920 --> 00:13:10,957
if what if I could think about what
all the unintended externalities are and

228
00:13:11,040 --> 00:13:14,437
like really just kind of tried
in advance to like prevent some of those

229
00:13:14,520 --> 00:13:19,317
from happening. And it's a little bit
unfair because in AI, we've had this

230
00:13:19,400 --> 00:13:23,517
whole generation of technologies before
us where they've like gotten to up

231
00:13:23,600 --> 00:13:26,597
and we've gotten to be like, haha, we're
not going to do that thing again. But

232
00:13:26,680 --> 00:13:30,517
that is a huge privilege, right? We're
able to say, "Okay, you guys made this

233
00:13:30,600 --> 00:13:34,317
mistake. We are not going to make that
mistake this time, right? We're going to

234
00:13:34,400 --> 00:13:38,877
be careful and we're going to say, how
do we make sure to think about the other

235
00:13:38,960 --> 00:13:43,557
things that might not go wrong because
we understand the technology better?" Um

236
00:13:43,640 --> 00:13:48,437
how can we how can we
imagine a world where like

237
00:13:48,520 --> 00:13:51,997
everything goes right, but also a world
where everything goes wrong? And I think

238
00:13:52,080 --> 00:13:56,517
for us safety means like all of the big
stuff, so preventing chemical and

239
00:13:56,600 --> 00:14:00,157
biological weapons from being developed
using our technology, which by the way,

240
00:14:00,240 --> 00:14:05,037
they they could have the potential to do.
Um but also like cyber, right? Cyber

241
00:14:05,120 --> 00:14:08,517
warfare, there's we've been in the news
a lot lately about our decision to not

242
00:14:08,600 --> 00:14:11,557
um you know, release our Mythos class
model because of the potential for cyber

243
00:14:11,640 --> 00:14:14,277
warfare. There's also
a lot of work that happens

244
00:14:14,360 --> 00:14:16,157
around things like user wellness, child

245
00:14:16,240 --> 00:14:21,517
safety, um a lot of misinformation,
election integrity work. This is not

246
00:14:21,600 --> 00:14:25,117
something new that we invented. We've
been able to stand on the shoulders of

247
00:14:25,200 --> 00:14:29,437
previous safety and security teams who
worked on this at some of the other most

248
00:14:29,520 --> 00:14:32,997
consequential technology companies
in history and say, "How do we learn from

249
00:14:33,080 --> 00:14:35,837
you, right? How do we do this better?"

250
00:14:35,920 --> 00:14:38,317
>> Anthropic is an AI safety company that

251
00:14:38,400 --> 00:14:43,157
also has to generate revenue.
How do you manage the tension between

252
00:14:43,240 --> 00:14:46,917
the two? >> We're asked
this question a lot, too.

253
00:14:47,000 --> 00:14:49,637
And I think

254
00:14:49,720 --> 00:14:54,237
in general, the two don't come into
conflict as much as you would expect. So

255
00:14:54,320 --> 00:14:58,877
what I would say is most
businesses in particular, which is

256
00:14:58,960 --> 00:15:00,197
by the way, the majority of our revenue

257
00:15:00,280 --> 00:15:02,757
comes from businesses, are not looking

258
00:15:02,840 --> 00:15:06,477
to have models that are unsafe, right?

259
00:15:06,560 --> 00:15:10,717
They're not like, "Wow, we would love
for Claude to hallucinate more." Or it

260
00:15:10,800 --> 00:15:14,917
would be great if um Claude just
produced like harmful outputs when

261
00:15:15,000 --> 00:15:20,117
you're asking it a question. And so I
think until like pretty recently, this

262
00:15:20,200 --> 00:15:23,477
these things were just like 100%
in alignment, You were like, it's actually

263
00:15:23,560 --> 00:15:28,557
really good for business to be safe
because businesses are correctly the

264
00:15:28,640 --> 00:15:31,397
risk of us. Right? They're like,
we don't want AI technologies that are

265
00:15:31,480 --> 00:15:34,237
going to be super
unpredictable or unreliable.

266
00:15:34,320 --> 00:15:38,237
That said, I think we're now entering an
era where the capabilities of the models

267
00:15:38,320 --> 00:15:40,237
are developing so rapidly that the

268
00:15:40,320 --> 00:15:42,437
tension is about time.

269
00:15:42,520 --> 00:15:45,877
So, it's not necessarily the case that

270
00:15:45,960 --> 00:15:49,957
the models can't do amazing things. It's
just we don't fully understand at this

271
00:15:50,040 --> 00:15:51,797
stage and I think it will be more the

272
00:15:51,880 --> 00:15:55,197
case going forward. Like, how

273
00:15:55,280 --> 00:15:58,397
how serious are the risks? What are all

274
00:15:58,480 --> 00:16:02,797
of the risks and how do we help mitigate
them? That sometimes means that we take

275
00:16:02,880 --> 00:16:06,597
slightly unusual actions like we did
with Project Last Wing and say this new

276
00:16:06,680 --> 00:16:10,437
class of model, um, it would be great if
we could just release this to all of our

277
00:16:10,520 --> 00:16:14,397
customers. They all would love to use it,
but we're just not confident enough

278
00:16:14,480 --> 00:16:17,037
yet. Like, we need a little more time to

279
00:16:17,120 --> 00:16:20,397
do some of the work to make

280
00:16:20,480 --> 00:16:25,197
the models safer to use, um,
but that's uncomfortable. Right? It's

281
00:16:25,280 --> 00:16:27,797
uncomfortable to say that to your
customers. Right? They're like, look, we

282
00:16:27,880 --> 00:16:31,117
all believe in cyber defense, but like
I really want access to that model. And I

283
00:16:31,200 --> 00:16:33,917
think this is the place where we just
come back to the mission. Right? We say,

284
00:16:34,000 --> 00:16:37,077
okay, we understand that desire. We want
to get this technology to you as quickly

285
00:16:37,160 --> 00:16:42,477
as possible, but it is irresponsible
of us to release it until we are confident

286
00:16:42,560 --> 00:16:46,117
that all of the patching that
needs to be done has been done.

287
00:16:46,200 --> 00:16:47,837
>> We count online that when it comes to

288
00:16:47,920 --> 00:16:51,557
AI, there's a lot of fear. Fear that AI

289
00:16:51,640 --> 00:16:53,957
means fewer jobs because of less need

290
00:16:54,040 --> 00:16:58,157
for human judgment. Is that fear valid?

291
00:16:58,240 --> 00:17:01,717
>> I think this is actually
a very complicated question. Um,

292
00:17:01,800 --> 00:17:03,797
my sense is that this probably will not

293
00:17:03,880 --> 00:17:06,917
surprise people here. Um, AI is going to

294
00:17:07,000 --> 00:17:10,437
change what types of jobs are available

295
00:17:10,520 --> 00:17:12,837
and what types of jobs

296
00:17:12,920 --> 00:17:15,197
people do. So,

297
00:17:15,280 --> 00:17:20,157
today, I think there are jobs that exist
that did not exist 5 years ago because

298
00:17:20,240 --> 00:17:25,197
of AI. I also think there will be some
jobs that will not exist in the future

299
00:17:25,280 --> 00:17:29,957
because of AI. But what we
have seen so far today is that

300
00:17:30,040 --> 00:17:33,237
according to our economic index, where
we study like how are people actually

301
00:17:33,320 --> 00:17:36,077
using artificial intelligence technology

302
00:17:36,160 --> 00:17:38,797
right now, it mostly looks like

303
00:17:38,880 --> 00:17:41,517
complementary skills, right? So, you

304
00:17:41,600 --> 00:17:45,277
have artificial intelligence as

305
00:17:45,360 --> 00:17:47,957
an enabler of work. You don't have it as

306
00:17:48,040 --> 00:17:50,757
a replacer of work. Um except in a very

307
00:17:50,840 --> 00:17:55,037
vanishingly small number of cases,
which is mostly customer service.

308
00:17:55,120 --> 00:17:58,997
Sorry if you have to email Comcast.
It will never be a human again, probably.

309
00:17:59,080 --> 00:18:00,717
Um but I don't know that that was

310
00:18:00,800 --> 00:18:02,797
actually different 5 years ago. So, um I

311
00:18:02,880 --> 00:18:06,077
think I think in reality, what I expect

312
00:18:06,160 --> 00:18:10,357
is like there will be a number of

313
00:18:10,440 --> 00:18:13,557
types of work that will feel a lot like

314
00:18:13,640 --> 00:18:17,597
they rhyme with a job that exists today,
but they're not necessarily the same as

315
00:18:17,680 --> 00:18:20,717
a job that exists today. And I think we
just don't know the shape of what all of

316
00:18:20,800 --> 00:18:24,237
that is, right? Today, I think the thing
that's most talked about is coding,

317
00:18:24,320 --> 00:18:27,717
right? Software developers. Um I always,
you know, in business meetings, right?

318
00:18:27,800 --> 00:18:31,917
People will say to me like we're
talking shop about Claude, and

319
00:18:32,000 --> 00:18:35,437
then like two thirds of the way through
the conversation, like a CEO will kind

320
00:18:35,520 --> 00:18:38,677
of conspiratorially lean
across the table and say,

321
00:18:38,760 --> 00:18:44,077
you know, my daughter is a
sophomore at Stanford. Like, what

322
00:18:44,160 --> 00:18:47,637
should she study? Right?
Like, she was going to be a CS

323
00:18:47,720 --> 00:18:50,637
major. Should she not major in computer
science? And I think, you know, the

324
00:18:50,720 --> 00:18:54,757
truth is we don't know. But my guess
is software developers will still exist,

325
00:18:54,840 --> 00:18:56,877
but like they won't write as much code.

326
00:18:56,960 --> 00:19:00,277
Right? Like a lot of what software
developers do is much bigger than just

327
00:19:00,360 --> 00:19:04,317
hands-on keyboard. They're talking
to product managers. They're working

328
00:19:04,400 --> 00:19:07,877
closely with customers. And I think
the percentage of that work is going to

329
00:19:07,960 --> 00:19:10,877
expand. Um and I think the sort of

330
00:19:10,960 --> 00:19:13,197
things that can be more easily done by

331
00:19:13,280 --> 00:19:17,997
AI, that will contract, but my sense is
that is going to create a very different

332
00:19:18,080 --> 00:19:21,917
scope of what's possible.
>> What needs to happen in education,

333
00:19:22,000 --> 00:19:26,957
leadership, society, that people
feel prepared and excited and not just

334
00:19:27,040 --> 00:19:28,517
anxious?

335
00:19:28,600 --> 00:19:30,997
>> So I think there's a few things here. I

336
00:19:31,080 --> 00:19:34,037
think the first one is

337
00:19:34,120 --> 00:19:38,357
we need to start and lead from a place
of like humility and not knowing the

338
00:19:38,440 --> 00:19:42,517
answers, but doing the research. And I
think at Anthropic, something that we've

339
00:19:42,600 --> 00:19:45,717
always aimed to do is be as kind
of radically transparent about all of this

340
00:19:45,800 --> 00:19:47,957
as as we possibly can be, right? We've

341
00:19:48,040 --> 00:19:53,037
always said, "Look, we we don't have all
the answers. We do need to study this so

342
00:19:53,120 --> 00:19:56,637
that we're able to tell people what
we see coming." And I think, you know,

343
00:19:56,720 --> 00:19:59,597
sometimes fairly people can say like,
"Wow, you guys are just this is like a

344
00:19:59,680 --> 00:20:02,837
lot of negativity, right? You're like,
here's here's what we think might happen

345
00:20:02,920 --> 00:20:06,437
in the future." But I think it's more
important that we start the conversation

346
00:20:06,520 --> 00:20:10,237
sooner because we don't want people
to be caught off guard, right? We're we're

347
00:20:10,320 --> 00:20:14,197
publishing the academic index to say,
"Here's how people are using artificial

348
00:20:14,280 --> 00:20:18,877
intelligence today because we want
people to have an understanding of like

349
00:20:18,960 --> 00:20:22,197
where do we think this is going?"
So I think that's step one, is we have to

350
00:20:22,280 --> 00:20:25,837
actually all agree on what reality
is to the degree that's possible.

351
00:20:25,920 --> 00:20:31,077
I think step two is, you know, we have
to be creative and experimental at many

352
00:20:31,160 --> 00:20:34,757
different layers. So how do we have
artificial intelligence really be

353
00:20:34,840 --> 00:20:39,277
something that is a grounding and a
unifying force for people outside of

354
00:20:39,360 --> 00:20:42,317
just like, "Oh, I'm using it at my
kind of job," right? Which is really

355
00:20:42,400 --> 00:20:45,237
important, but I think in some ways
we need to sort of be rethinking the

356
00:20:45,320 --> 00:20:48,077
paradigm of this connection between

357
00:20:48,160 --> 00:20:53,917
work and meaning and like social life,
like all of these things I think are

358
00:20:54,000 --> 00:20:57,277
going to look very different in the future
and we need to practice. And then

359
00:20:57,360 --> 00:21:00,557
I think the third one, which is really
outside of the realm of what a

360
00:21:00,640 --> 00:21:05,317
technology company can do alone,
is this is going to become a social and

361
00:21:05,400 --> 00:21:09,797
political issue. People are going
to care if it feels like their jobs are

362
00:21:09,880 --> 00:21:12,557
being displaced by AI. People are
already care about this, right? It comes

363
00:21:12,640 --> 00:21:15,637
up in polling. People are like,
I have anxiety about what artificial

364
00:21:15,720 --> 00:21:18,997
intelligence is going to mean for my
future, for my kids' future, right? That

365
00:21:19,080 --> 00:21:23,277
CEO that's leaning across the table from
me, it's not just them. And so, I think

366
00:21:23,360 --> 00:21:26,677
there is a broader discussion that needs
to happen at many different levels of

367
00:21:26,760 --> 00:21:30,477
government, with civil society, with
universities, by the way, to say like,

368
00:21:30,560 --> 00:21:33,557
what does this mean? Where What is the
type of world that we want to be able to

369
00:21:33,640 --> 00:21:36,877
build where artificial intelligence is
capable of doing many of the things that

370
00:21:36,960 --> 00:21:41,157
humans do today. >> At the core,
it is all about adoption.

371
00:21:41,240 --> 00:21:45,997
Here at Stanford, we live in the AI, but
Stanford and Silicon Valley are not the

372
00:21:46,080 --> 00:21:49,877
whole world.
What currently hinders AI adoption

373
00:21:49,960 --> 00:21:53,597
outside of this bubble? >> You know,
I think this is a I think this

374
00:21:53,680 --> 00:21:57,117
is a really, really great call out.
Um it's so interesting because it feels

375
00:21:57,200 --> 00:22:00,237
like, you know, certainly to us
at Anthropic, and I'm sure at Stanford,

376
00:22:00,320 --> 00:22:03,917
too, it's like the only thing people
want to talk about is AI. Certainly the

377
00:22:04,000 --> 00:22:08,277
only thing people want to talk to me about
is AI, but that's probably a a me problem.

378
00:22:08,360 --> 00:22:11,077
>> [laughter] >> But I I
think I think that feeling

379
00:22:11,160 --> 00:22:13,597
you're totally right, that even in other

380
00:22:13,680 --> 00:22:17,037
parts of America, it's it's not it's not

381
00:22:17,120 --> 00:22:21,277
something that people are comfortable
with yet on the whole, and it's not even

382
00:22:21,360 --> 00:22:25,277
something that people like necessarily
know how to use with very high fluency.

383
00:22:25,360 --> 00:22:29,037
Right? So, you read these like really
astounding, super impressive numbers

384
00:22:29,120 --> 00:22:31,157
about how many people are using AI

385
00:22:31,240 --> 00:22:34,317
tools. Um but there's there's

386
00:22:34,400 --> 00:22:36,397
there's a demographic component to that.

387
00:22:36,480 --> 00:22:38,477
Right? So, it's generally people who are

388
00:22:38,560 --> 00:22:41,117
college educated, not uniquely. Uh it's

389
00:22:41,200 --> 00:22:45,757
more men than women. Um there's racial
demographics involved. There's like, um

390
00:22:45,840 --> 00:22:49,517
wealth demographics involved. And if you
look around the world, it's really not

391
00:22:49,600 --> 00:22:55,317
equally distributed. And so, I think
what's interesting is that you pair that

392
00:22:55,400 --> 00:22:58,437
with some other data that we've
collected, which is that people in

393
00:22:58,520 --> 00:23:02,357
developing countries
are much more optimistic

394
00:23:02,440 --> 00:23:04,637
about AI than people in

395
00:23:04,720 --> 00:23:06,357
higher income countries. So, the global

396
00:23:06,440 --> 00:23:11,757
south is almost universally like,
"Wow, this is a huge opportunity for us,

397
00:23:11,840 --> 00:23:15,197
right? This is This is the moment where
perhaps we could have an equalizing

398
00:23:15,280 --> 00:23:18,157
force that will make things more fair."

399
00:23:18,240 --> 00:23:22,877
But, I think in the US and in Europe and
in parts of Asia, people have a lot more

400
00:23:22,960 --> 00:23:26,197
anxiety. They're like, "I like things
the way they are. I don't want AI to

401
00:23:26,280 --> 00:23:29,117
come in and disrupt that.
That doesn't sound as good to me."

402
00:23:29,200 --> 00:23:31,837
What do we do with this information?
I have no idea. But, I think it's really

403
00:23:31,920 --> 00:23:33,517
interesting to say like there are

404
00:23:33,600 --> 00:23:36,237
different questions of access and

405
00:23:36,320 --> 00:23:39,357
adoption around the technology.

406
00:23:39,440 --> 00:23:44,077
And I think we we are we are actually
still very early in the game. And that's

407
00:23:44,160 --> 00:23:47,717
the thing that I think can be missed
in Silicon Valley, in our bubble, that

408
00:23:47,800 --> 00:23:51,397
we're like, "It's already Everybody is
a software engineer. It's like, I'm using

409
00:23:51,480 --> 00:23:55,557
cloud code. I'm using Codex." That is
not the vast majority of developers in

410
00:23:55,640 --> 00:23:57,677
the world at large. And so, I think the

411
00:23:57,760 --> 00:24:00,637
the the race is is still like we're at

412
00:24:00,720 --> 00:24:04,797
the like the gun just went off to start
the race. And I think there's a lot of

413
00:24:04,880 --> 00:24:09,837
opportunity to still positively shape
how this technology is going to be used

414
00:24:09,920 --> 00:24:13,317
and developed, what access looks like,
and just what the values that are baked

415
00:24:13,400 --> 00:24:15,917
into it are going to ultimately be.

416
00:24:16,000 --> 00:24:18,637
>> So, let's forward
to the future where AI

417
00:24:18,720 --> 00:24:21,477
is far more widely adopted.

418
00:24:21,560 --> 00:24:25,117
What are the things we risk losing

419
00:24:25,200 --> 00:24:29,117
if we start delegating too much to AI?

420
00:24:29,200 --> 00:24:32,597
>> Yeah. Um we did
this very large qualitative

421
00:24:32,680 --> 00:24:36,597
survey at Anthropic. I think it's the
largest qualitative study ever done that

422
00:24:36,680 --> 00:24:41,237
we know about. We talked to 81,000
people about their use of artificial

423
00:24:41,320 --> 00:24:44,957
intelligence. And so, some of them were
Claude users, some of them were users of

424
00:24:45,040 --> 00:24:48,037
other of other AI tools. Um

425
00:24:48,120 --> 00:24:50,397
And what was interesting is, you know,

426
00:24:50,480 --> 00:24:53,477
people people have a lot of different

427
00:24:53,560 --> 00:24:56,837
feelings about AI, right? And again, you

428
00:24:56,920 --> 00:25:00,557
know, cuts differently depending
on where you are and and what you do, but

429
00:25:00,640 --> 00:25:05,357
some people are like, I've never been
able, you know, it it enabled me to do

430
00:25:05,440 --> 00:25:09,157
things that I never thought I could do,
right? I think for myself as an example,

431
00:25:09,240 --> 00:25:12,117
like, I didn't think I could build
a website, right? And now using Claude,

432
00:25:12,200 --> 00:25:14,917
I'm like, oh man, that's so easy. Like,
I just click a couple buttons and Claude

433
00:25:15,000 --> 00:25:18,117
like built a website for me. What?
That would have That would have taken me

434
00:25:18,200 --> 00:25:21,197
probably a year if I tried to do it
by myself, and it would not have been a

435
00:25:21,280 --> 00:25:23,437
very good website. Um

436
00:25:23,520 --> 00:25:26,917
there's some people though who express a

437
00:25:27,000 --> 00:25:32,117
feeling of there's not like a a specific
term for it, but I think there might be,

438
00:25:32,200 --> 00:25:36,837
or there might be in another language
one day. Um like, I don't I don't engage

439
00:25:36,920 --> 00:25:38,477
my brain because I don't have to.

440
00:25:38,560 --> 00:25:40,757
>> Yeah. >> So, it's not
the same feeling as like

441
00:25:40,840 --> 00:25:45,037
scrolling on your phone, but it's like,
I could have reached for this idea, I

442
00:25:45,120 --> 00:25:48,757
could have thought through it, but it
was so much easier to not do it and to

443
00:25:48,840 --> 00:25:53,357
just trust what the AI tool was giving me.
And I think this is the source, I I

444
00:25:53,440 --> 00:25:55,037
actually believe, of a lot of the

445
00:25:55,120 --> 00:25:58,957
anxiety around AI. As humans like to, I

446
00:25:59,040 --> 00:26:02,037
think, have an inherent desire to learn,

447
00:26:02,120 --> 00:26:05,517
to be curious, to want to expand the

448
00:26:05,600 --> 00:26:08,117
aperture of things that they know about.

449
00:26:08,200 --> 00:26:13,197
And AI in some ways enables that,
but if you if used incorrectly can sort of

450
00:26:13,280 --> 00:26:16,157
disable that, right? It's like, I've
done this sometimes, right? I was like,

451
00:26:16,240 --> 00:26:18,837
oh, I could like look this up and figure
it out myself, but I'll just ask an AI

452
00:26:18,920 --> 00:26:21,557
tool and then I'll blindly trust that what
it says is correct. It's not always

453
00:26:21,640 --> 00:26:22,877
correct, by the way. Sometimes Claude is

454
00:26:22,960 --> 00:26:26,197
wrong. Heretical to say, but a fact. Um

455
00:26:26,280 --> 00:26:29,277
and I I think I think the anxiety there

456
00:26:29,360 --> 00:26:31,557
is around

457
00:26:31,640 --> 00:26:34,637
how do we actually set some guardrails

458
00:26:34,720 --> 00:26:38,997
in place so that it's just it's not
impossible to do that, but you actually

459
00:26:39,080 --> 00:26:43,117
have to really be trying to do it,
right? Like, I think some of the work

460
00:26:43,200 --> 00:26:45,917
that we do with universities is maybe an

461
00:26:46,000 --> 00:26:50,477
interesting microcosm for this. We have
this concept of learning mode. Maybe

462
00:26:50,560 --> 00:26:51,797
some of you even use it. I know if we're

463
00:26:51,880 --> 00:26:54,677
at the GSP yet, but um

464
00:26:54,760 --> 00:26:58,397
faculty and professors and students,

465
00:26:58,480 --> 00:27:03,637
like one version of this is
like you put your homework

466
00:27:03,720 --> 00:27:07,317
in chat GPT, I'm going to use that one
instead, and you're like, "Haha, it just

467
00:27:07,400 --> 00:27:10,317
gave me the answer." There's a word
for that, it's called cheating, right? And

468
00:27:10,400 --> 00:27:14,837
you're like, "That was great." There's
another version where you use Claude in

469
00:27:14,920 --> 00:27:20,197
learning mode and you're like, "I don't
I'm I I'm stuck, right? I'm trying to

470
00:27:20,280 --> 00:27:22,837
write this essay and I like there's
something about the format that doesn't

471
00:27:22,920 --> 00:27:26,357
feel right to me."
And Claude is this sort of

472
00:27:26,440 --> 00:27:28,197
patient tutor. It's almost like you have

473
00:27:28,280 --> 00:27:30,597
an individualized professor who like

474
00:27:30,680 --> 00:27:35,237
knows you and understands like what you
most want to learn and why this class is

475
00:27:35,320 --> 00:27:38,477
important to you. It's like, "Let me
help get you unstuck, right? Do you want

476
00:27:38,560 --> 00:27:41,037
to go back and read this section together?
Could we talk through this?" I

477
00:27:41,120 --> 00:27:44,677
think that's the version where like
these tools can make you smarter, they

478
00:27:44,760 --> 00:27:48,917
can make you um expand the set
of things that you think

479
00:27:49,000 --> 00:27:52,317
you can learn. And then I
think there's the version

480
00:27:52,400 --> 00:27:54,837
that's just turn your brain off.

481
00:27:54,920 --> 00:27:57,997
And I'm hoping that as
an industry we're going to

482
00:27:58,080 --> 00:28:00,997
choose to do the second versus the first.

483
00:28:01,080 --> 00:28:03,477
>> So, if you had to prioritize,

484
00:28:03,560 --> 00:28:06,677
what human skills most likely going to

485
00:28:06,760 --> 00:28:10,717
be more important in an AI-driven world?

486
00:28:10,800 --> 00:28:11,757
>> So,

487
00:28:11,840 --> 00:28:16,117
I I have my own views on this, which is

488
00:28:16,200 --> 00:28:17,757
I think

489
00:28:17,840 --> 00:28:21,757
like we talked about, a lot of specific

490
00:28:21,840 --> 00:28:25,117
like task-oriented things, right?
Like I'm a financial analyst or I'm a

491
00:28:25,200 --> 00:28:30,157
developer or I'm a copy editor,
um those jobs are going to change a lot.

492
00:28:30,240 --> 00:28:31,597
And a lot of that work I think will be

493
00:28:31,680 --> 00:28:34,397
able to be done by by AI tools.

494
00:28:34,480 --> 00:28:38,477
But I think ultimately there's this very

495
00:28:38,560 --> 00:28:42,757
real phenomenon, which is that humans
like to be with other humans.

496
00:28:42,840 --> 00:28:46,877
We like to spend time together
with each other. We like to

497
00:28:46,960 --> 00:28:50,797
um learn from each other. We like
to be creative. We like to spend time

498
00:28:50,880 --> 00:28:55,517
understanding the other person and we're
social creatures, right? And I think I'm

499
00:28:55,600 --> 00:28:58,957
imagining that in a world where AI
is able to do a lot more of the productive

500
00:28:59,040 --> 00:29:03,037
day-to-day work that we do,
those skills are going to become a lot

501
00:29:03,120 --> 00:29:05,517
more important and much more prized.

502
00:29:05,600 --> 00:29:09,517
Because ultimately, like if you're in a
work environment and you're like, "Well,

503
00:29:09,600 --> 00:29:12,997
I could just ask Claude like write
a bunch of code for me." You are You are

504
00:29:13,080 --> 00:29:16,477
going to choose to talk to the developer
that's going to explain to you like why

505
00:29:16,560 --> 00:29:21,237
something broke or why we chose to build,
you know, a tool the way that we

506
00:29:21,320 --> 00:29:25,517
did. And I think sort of expanding
this outside of the realm of just the

507
00:29:25,600 --> 00:29:27,557
technology industry, the example I often

508
00:29:27,640 --> 00:29:31,157
use is in medicine. Right today,

509
00:29:31,240 --> 00:29:35,317
we hire doctors who are really good
diagnosticians. We're like, "Hey, can

510
00:29:35,400 --> 00:29:37,957
you tell me what's wrong with me?
I don't feel well." And you're like

511
00:29:38,040 --> 00:29:40,597
basically paying this doctor to say like,
"Here's a set of things that could

512
00:29:40,680 --> 00:29:43,997
be wrong with you. This is the one that's
the most likely. Let me run some tests."

513
00:29:44,080 --> 00:29:46,917
Guess what? AI is going to get
really good at doing that.

514
00:29:47,000 --> 00:29:51,277
But the thing that an AI tool can't do
is actually like look at you and examine

515
00:29:51,360 --> 00:29:55,237
you and also like help
understand how you're

516
00:29:55,320 --> 00:29:57,197
feeling and help you feel better.

517
00:29:57,280 --> 00:30:00,517
Right? There's a There's a reasonable
body of medical literature that

518
00:30:00,600 --> 00:30:03,197
indicates that people who have a good
relationship with their doctor, they

519
00:30:03,280 --> 00:30:07,517
just like their doctor, um have better
clinical outcomes than people who do not

520
00:30:07,600 --> 00:30:10,477
like their doctor. Like that's really
hard to explain. But like what's

521
00:30:10,560 --> 00:30:13,877
probably going on?
Probably the doctor like

522
00:30:13,960 --> 00:30:16,677
tries a little bit harder to understand
what's what's wrong with you. Maybe they

523
00:30:16,760 --> 00:30:18,997
run a set of tests that were unexpected.

524
00:30:19,080 --> 00:30:23,157
Um and I think those skills, right? That
bedside manner is going to be like five

525
00:30:23,240 --> 00:30:26,877
times more important in a world where
you're not trying to cram that into one

526
00:30:26,960 --> 00:30:31,477
of seven things that you're looking for
to make a doctor qualified to treat you.

527
00:30:31,560 --> 00:30:34,077
>> And when you are
thinking about the future,

528
00:30:34,160 --> 00:30:38,757
what AI use cases you're
most personally excited about?

529
00:30:38,840 --> 00:30:43,997
>> Oh man. Um I think I think
for me personally, um I'm I mean

530
00:30:44,080 --> 00:30:47,797
I'm a career manager. So, I spend most
of my time with people. And we have this

531
00:30:47,880 --> 00:30:50,277
sort of I don't know,
again, it's maybe one day

532
00:30:50,360 --> 00:30:53,317
there's going to be a word for it, but
there's this phenomenon where everybody

533
00:30:53,400 --> 00:30:56,997
thinks that AI is not going to come
for their job because they're so special.

534
00:30:57,080 --> 00:30:59,357
And I am I am extremely guilty of this.

535
00:30:59,440 --> 00:31:01,837
I was like, "People love people.

536
00:31:01,920 --> 00:31:04,517
People are going to want to report to me.
They're not going to want to report

537
00:31:04,600 --> 00:31:07,757
to Claude, obviously." Um but I actually

538
00:31:07,840 --> 00:31:10,037
think Claude is incredibly powerful as a

539
00:31:10,120 --> 00:31:15,157
management coach and as a person
to help you be a better leader. So, I use

540
00:31:15,240 --> 00:31:18,797
Claude. We write performance reviews
at Anthropic. And I've uploaded, you know,

541
00:31:18,880 --> 00:31:21,757
a lot of the people are that I work with
have reported to me for like, you know,

542
00:31:21,840 --> 00:31:25,117
three or four years. So, in general,
it's like you're the same person,

543
00:31:25,200 --> 00:31:28,157
they're the same person. You give them
feedback, but like, how much has really

544
00:31:28,240 --> 00:31:30,077
changed just like in the past 6 months?

545
00:31:30,160 --> 00:31:33,877
But I think Claude has been really
powerful in helping me to spot patterns

546
00:31:33,960 --> 00:31:38,197
about somebody. Um so, if you you know,
more data is better, but if you look

547
00:31:38,280 --> 00:31:41,997
back over the course of 3 to 4 years
of time of working with somebody, and

548
00:31:42,080 --> 00:31:45,197
you're like, "Wow, you know, you guys
have been circling around like this

549
00:31:45,280 --> 00:31:48,917
topical issue for the past 3 to 4
years." Like, maybe they need some

550
00:31:49,000 --> 00:31:52,357
additional coaching. Um or maybe they
need somebody sort of outside of you.

551
00:31:52,440 --> 00:31:55,317
It's just the type of thing that I think
tends to get missed because you're just

552
00:31:55,400 --> 00:31:56,957
in it day-to-day.

553
00:31:57,040 --> 00:31:59,357
And in the in the opposite direction,

554
00:31:59,440 --> 00:32:02,397
Claude is great at giving you feedback.

555
00:32:02,480 --> 00:32:06,117
So, like, I have I upload all of my
reports, upward feedback for me. And

556
00:32:06,200 --> 00:32:09,597
sometimes Claude will kind of very
kindly be like, "It sounds like you

557
00:32:09,680 --> 00:32:12,517
haven't improved on this in the past year.
Like, maybe you should get some

558
00:32:12,600 --> 00:32:17,237
extra coaching, Danielle." Um but but I
think those I think Claude's ability to

559
00:32:17,320 --> 00:32:20,797
kind of coach and help people be
the best versions of themselves, I think

560
00:32:20,880 --> 00:32:24,077
there's a version of that that makes
sense in the workplace, uh in people's

561
00:32:24,160 --> 00:32:27,797
personal lives, that I think could
be done kind of, you know, quite quite

562
00:32:27,880 --> 00:32:31,757
carefully, but that I think could be
really powerful. Um and then the second

563
00:32:31,840 --> 00:32:35,677
is I have two little kids. So, I have
a almost 5-year-old and almost 1-year-old.

564
00:32:35,760 --> 00:32:39,477
And I have to tell you, number one,
best thing Claude has ever done is help me

565
00:32:39,560 --> 00:32:42,357
through potty training. Um that was

566
00:32:42,440 --> 00:32:46,517
not a fun experience. And Claude
made it just like a little bit it was

567
00:32:46,600 --> 00:32:50,077
empathetic, like very actionable. There
were some diagrams, I don't need to tell

568
00:32:50,160 --> 00:32:53,157
you guys, but um it was really really

569
00:32:53,240 --> 00:32:57,117
useful. And I think Claude's ability
to help in particular um overwhelmed

570
00:32:57,200 --> 00:33:01,557
parents is going to be really powerful
because there's so much bad information.

571
00:33:01,640 --> 00:33:03,797
It's like every time you Google
is something wrong with your kid, the

572
00:33:03,880 --> 00:33:08,077
answer is yes. And I think Claude
is a lot more measured and like can be

573
00:33:08,160 --> 00:33:11,477
interactive in a way that I
think is really, really helpful.

574
00:33:11,560 --> 00:33:13,077
>> Daniela, before we turn to student

575
00:33:13,160 --> 00:33:15,997
questions, when you think about the next

576
00:33:16,080 --> 00:33:18,837
generation of AI leaders and builders

577
00:33:18,920 --> 00:33:21,717
who are with us here today in this room,

578
00:33:21,800 --> 00:33:26,880
what's the one thing you hope
they take from your journey?

579
00:33:27,520 --> 00:33:30,197
>> I would say, can I do two? >> Go ahead.

580
00:33:30,280 --> 00:33:33,877
>> Okay. Um I would say
the first is it sounds so

581
00:33:33,960 --> 00:33:36,637
trite and so I almost like I almost

582
00:33:36,720 --> 00:33:41,277
won't say it, but I truly think like
following something that you really care

583
00:33:41,360 --> 00:33:45,317
about, something that you are passionate
about is the most important thing you

584
00:33:45,400 --> 00:33:48,157
can do. Um there are so many great ideas

585
00:33:48,240 --> 00:33:52,557
and if you don't feel the sort of like
burning feeling of like this is a thing

586
00:33:52,640 --> 00:33:54,877
that needs to exist in the world and I I

587
00:33:54,960 --> 00:33:57,157
I will just run through walls to be able

588
00:33:57,240 --> 00:34:00,517
to do it, um it's just it matters for

589
00:34:00,600 --> 00:34:04,877
the times when like it's not fun and when
it sucks. Like you just have to be

590
00:34:04,960 --> 00:34:08,397
able to say like I remember why
this matters to me, I remember why it's

591
00:34:08,480 --> 00:34:12,157
important, whether it's to me personally
or it's because of a type of change that

592
00:34:12,240 --> 00:34:16,317
I want to see happen in the world. There
will be time, there certainly have been

593
00:34:16,400 --> 00:34:20,197
at Anthropic, times where we're like
this is not the most fun part, right?

594
00:34:20,280 --> 00:34:24,237
There are parts that are tough. And so
being able to relate it back to why you

595
00:34:24,320 --> 00:34:26,917
decided to do this in the first place
and what matters to you I think is so

596
00:34:27,000 --> 00:34:31,637
important. And then the second I would
say is I I for you know, this generation

597
00:34:31,720 --> 00:34:33,357
in particular and I think really in the

598
00:34:33,440 --> 00:34:35,957
past, you know, 5 to 10 years, this

599
00:34:36,040 --> 00:34:38,917
concept that like

600
00:34:39,000 --> 00:34:41,557
being in business doesn't have to be in

601
00:34:41,640 --> 00:34:44,157
tension with with doing good. I think

602
00:34:44,240 --> 00:34:46,997
that is very that is a very new idea and

603
00:34:47,080 --> 00:34:49,517
I think it is really special and I have

604
00:34:49,600 --> 00:34:51,877
been so impressed at the sort of

605
00:34:51,960 --> 00:34:54,677
generation of

606
00:34:54,760 --> 00:34:57,837
founders and you know, just like

607
00:34:57,920 --> 00:35:00,557
creators who are thinking in that in

608
00:35:00,640 --> 00:35:02,797
that way, right? Like there's this kind

609
00:35:02,880 --> 00:35:07,837
of marriage of innovation
and social impact. I think

610
00:35:07,920 --> 00:35:11,877
Stanford has always been exceptional
at this, but I think that is a very new

611
00:35:11,960 --> 00:35:16,317
concept and I think there's like more
appetite for it today, right? I think

612
00:35:16,400 --> 00:35:20,597
there can sort of be this feeling
of like only the kind of like mean sucky

613
00:35:20,680 --> 00:35:25,437
people can like build a business.
I just don't think that's true and I think

614
00:35:25,520 --> 00:35:27,437
increasingly I feel that the desire to

615
00:35:27,520 --> 00:35:32,117
do good is a strong

616
00:35:32,200 --> 00:35:35,917
it's a strong correlate
with actually doing well.

617
00:35:36,000 --> 00:35:39,000
>> Turn to the students.

618
00:35:39,480 --> 00:35:43,317
>> Hi Daniela, my name is Brandon. I'm
a second-year MBA student here. Thank you

619
00:35:43,400 --> 00:35:47,117
for joining us.
There's a debate about whether we're in

620
00:35:47,200 --> 00:35:51,037
an AI bubble and people usually mean
three different things when they say

621
00:35:51,120 --> 00:35:54,757
that. Company valuations,
how much companies

622
00:35:54,840 --> 00:35:58,717
are spending on infrastructure,
or whether the pace of AI progress is

623
00:35:58,800 --> 00:36:02,237
actually sustainable.
Which of those three are you most

624
00:36:02,320 --> 00:36:05,837
worried about and which do you
think people are most wrong about?

625
00:36:05,920 --> 00:36:08,117
>> Mhm, that's a great question.

626
00:36:08,200 --> 00:36:13,197
Um, yeah, we need like different like we
need like air bubbles and glass bubble.

627
00:36:13,280 --> 00:36:17,237
They're very different.
I see I see what you're saying. Um,

628
00:36:17,320 --> 00:36:22,237
I think I think the one
that probably I don't

629
00:36:22,320 --> 00:36:26,237
know if I would say I'm the most worried
about, but I think is is is a is valid

630
00:36:26,320 --> 00:36:28,037
as a concern about this industry in

631
00:36:28,120 --> 00:36:32,637
general is it is a high capital

632
00:36:32,720 --> 00:36:37,837
expenditure business. And that inherently
brings some risks along with it.

633
00:36:37,920 --> 00:36:41,397
This is probably stuff you all already
know, but it is really expensive to

634
00:36:41,480 --> 00:36:45,797
train these models. It takes a lot
of compute and that compute is in scarce

635
00:36:45,880 --> 00:36:49,317
supply and when you put those things
together like there's a lot of demand,

636
00:36:49,400 --> 00:36:51,877
there's not a lot of supply.
I'm not [laughter] an economics

637
00:36:51,960 --> 00:36:56,467
professor, but I think that means
the price goes up. Um and so

638
00:36:56,550 --> 00:36:59,837
>> [laughter] >> So the compute

639
00:36:59,920 --> 00:37:03,797
that is sort of the lifeblood
of these companies.

640
00:37:03,880 --> 00:37:07,197
Um you have to buy really far in advance.

641
00:37:07,280 --> 00:37:09,997
And so you're essentially making a bet
on the future. You're like we think

642
00:37:10,080 --> 00:37:13,797
we're going to need this much
compute at this period of time.

643
00:37:13,880 --> 00:37:17,677
That's a really that's a really big
expenditure to make and like it is it is

644
00:37:17,760 --> 00:37:20,997
a little bit harrowing to work at any
of these companies. I think if I think if

645
00:37:21,080 --> 00:37:23,397
someone doesn't tell you that maybe
with the exception of Google because they

646
00:37:23,480 --> 00:37:27,357
have their public company have so
much money. But I think certainly for

647
00:37:27,440 --> 00:37:31,317
Anthropic, for Open AI like you're kind
of making a calculated bet that you're

648
00:37:31,400 --> 00:37:35,277
going to be able to pay
that money back over time.

649
00:37:35,360 --> 00:37:38,917
We obviously are very bullish on this.
I think the revenue from both of those

650
00:37:39,000 --> 00:37:41,797
companies is like unbelievable.
It's something that I

651
00:37:41,880 --> 00:37:45,637
think you know we hear all the time like
venture capital is nothing like this has

652
00:37:45,720 --> 00:37:48,837
ever happened before, right?
It's impossible to imagine a business you

653
00:37:48,920 --> 00:37:52,237
know getting to the kind of revenue
numbers that are being talked about on

654
00:37:52,320 --> 00:37:56,637
such a short time scale. And if that
ever were to change, there would be a

655
00:37:56,720 --> 00:37:59,437
problem, right? Both of these companies
have bought a lot of compute for the

656
00:37:59,520 --> 00:38:03,557
future. It's very expensive. And so
I think that is probably the risk that

657
00:38:03,640 --> 00:38:06,717
feels it's not crazy
to be worried about that.

658
00:38:06,800 --> 00:38:10,317
We obviously think we're in a very
good position. I think the industry as a

659
00:38:10,400 --> 00:38:13,157
whole is, but that could
change anytime. And I

660
00:38:13,240 --> 00:38:16,477
think it's important to remember that
like this is ultimately a bet, right?

661
00:38:16,560 --> 00:38:19,317
It's ultimately the industry thinking
hey this is going to have a lot of

662
00:38:19,400 --> 00:38:23,960
returns to it, but like we
could absolutely be wrong.

663
00:38:25,000 --> 00:38:29,997
>> Hi, I'm Daniel Yash, MBA student. Uh my
question is, what does Anthropic believe

664
00:38:30,080 --> 00:38:33,997
is the right balance between government
regulation and AI innovation? And what

665
00:38:34,080 --> 00:38:37,437
do you wish governments around
the world were doing differently?

666
00:38:37,520 --> 00:38:42,637
>> Great question.
Um So I I think this is an area where

667
00:38:42,720 --> 00:38:46,077
um I think the conversation about it has

668
00:38:46,160 --> 00:38:48,317
been you know, sadly as sort of is the

669
00:38:48,400 --> 00:38:53,117
case I think today just just in the sort
of political climate, it's like hard to

670
00:38:53,200 --> 00:38:57,077
have a nuanced discussion.
And I I think that's a shame

671
00:38:57,160 --> 00:38:59,197
because I think it is a really nuanced

672
00:38:59,280 --> 00:39:01,677
question.

673
00:39:01,760 --> 00:39:06,397
I think sensible regulation will
need to be part of the story for

674
00:39:06,480 --> 00:39:09,757
artificial intelligence, right? It's
a very different technology than any other

675
00:39:09,840 --> 00:39:13,197
technology that's ever been built before.
And even the last generation of

676
00:39:13,280 --> 00:39:16,197
technologies probably could have
benefited from slightly more regulation

677
00:39:16,280 --> 00:39:19,117
in my personal opinion. Um that being

678
00:39:19,200 --> 00:39:23,837
said, we're not blind to the fact
that like you need to have some room to

679
00:39:23,920 --> 00:39:27,677
maneuver as a company to be able to try
things if you want to come up with the

680
00:39:27,760 --> 00:39:31,037
next generation of incredible products
that people want to use and that are

681
00:39:31,120 --> 00:39:33,197
adopted. And I think this is one of

682
00:39:33,280 --> 00:39:38,237
these things that my actual most
kind of critical hope for the

683
00:39:38,320 --> 00:39:41,437
conversation is that it doesn't become
politicized, which I fear it already

684
00:39:41,520 --> 00:39:45,717
has, right? It's sort of like, oh,
like regulation bad, innovation good, or

685
00:39:45,800 --> 00:39:50,077
innovation bad, regulation good. I think
it's just really complicated. Um I think

686
00:39:50,160 --> 00:39:53,557
there are areas of regulation that just
don't make a lot of sense. And then I

687
00:39:53,640 --> 00:39:57,077
think there are areas of regulation
that are absolutely critical to make the

688
00:39:57,160 --> 00:40:00,117
technology be developed in a way that
is good for people and that will prevent

689
00:40:00,200 --> 00:40:03,757
bad things from happening to the people
that rely on it every day and to the

690
00:40:03,840 --> 00:40:07,237
broader world.
My hope is that in an ideal world what

691
00:40:07,320 --> 00:40:09,397
that would look like is

692
00:40:09,480 --> 00:40:12,517
technology companies and regulators like

693
00:40:12,600 --> 00:40:15,157
working hand-in-hand because we have the

694
00:40:15,240 --> 00:40:20,397
information about how the technology can
be abused because we see it every day,

695
00:40:20,480 --> 00:40:23,637
right? We have safeguards teams and
security teams that look at like how are

696
00:40:23,720 --> 00:40:27,277
people poking on this and what are
the what are the actual risks, but

697
00:40:27,360 --> 00:40:31,397
regulators know how to provide
a framework and a system that can actually

698
00:40:31,480 --> 00:40:36,037
be followed and and enforced. And so
maybe it's overly optimistic, but I I

699
00:40:36,120 --> 00:40:39,197
still hold out hope that that future's
possible where I think the two sides can

700
00:40:39,280 --> 00:40:43,477
find common ground and say like how do
we ensure that we're able to develop the

701
00:40:43,560 --> 00:40:46,717
next, you know, amazing technology
that doesn't even exist yet or the next

702
00:40:46,800 --> 00:40:49,837
company that's going to be like
whatever, the next Google or the next

703
00:40:49,920 --> 00:40:52,117
meta, um but that we just put some

704
00:40:52,200 --> 00:40:57,760
common sense regulations in place
that that help protect people.

705
00:40:58,200 --> 00:41:01,317
>> Hi Daniela, thanks for being here.
My name's Jackie Kimmel, I'm an MSX

706
00:41:01,400 --> 00:41:04,077
student. My question is AI's gaining

707
00:41:04,160 --> 00:41:09,037
access to increasingly sensitive
personal data such as our health data.

708
00:41:09,120 --> 00:41:12,677
Uh what do you think individuals should
actually be doing to protect their

709
00:41:12,760 --> 00:41:18,197
privacy? >> I think this is also
an excellent question. Um

710
00:41:18,280 --> 00:41:22,637
you know, the first thing that I'll say
is you would be surprised at how common

711
00:41:22,720 --> 00:41:25,877
of a use case it is to ask Claude
medical questions. I think it's like one

712
00:41:25,960 --> 00:41:29,477
of our most common just casual use cases.
People I use it all the time,

713
00:41:29,560 --> 00:41:33,077
right? I'm like this is wrong with my kid,
this is wrong with me, help me.

714
00:41:33,160 --> 00:41:36,357
I think there's I think there's two I
think there's two sides to this. I think

715
00:41:36,440 --> 00:41:38,757
the first is it's actually

716
00:41:38,840 --> 00:41:42,597
it's actually the companies' jobs to use

717
00:41:42,680 --> 00:41:45,517
and protect your data with care.

718
00:41:45,600 --> 00:41:49,997
I think that's extremely important.
Um I bluntly think like people should hold

719
00:41:50,080 --> 00:41:54,237
companies accountable for using their
data carefully, um because it is very

720
00:41:54,320 --> 00:41:56,837
personal. And I think for example, our

721
00:41:56,920 --> 00:41:59,997
decision to not put ads in Claude, um is

722
00:42:00,080 --> 00:42:03,917
partly predicated on this belief that AI
technology is just different, right?

723
00:42:04,000 --> 00:42:06,797
People have conversations with AI tools

724
00:42:06,880 --> 00:42:11,437
that are are much more personal
than even what you would put on a an

725
00:42:11,520 --> 00:42:14,157
Instagram account or um

726
00:42:14,240 --> 00:42:18,557
on social media of any form.
And so I think the first is like with that

727
00:42:18,640 --> 00:42:21,197
knowledge comes more responsibility
for the technology companies to actually

728
00:42:21,280 --> 00:42:24,837
protect your data. Um the second is I

729
00:42:24,920 --> 00:42:28,077
think from a personal perspective like

730
00:42:28,160 --> 00:42:31,997
I don't know that I have the perfect
answer. Um I can certainly tell you like

731
00:42:32,080 --> 00:42:35,437
a lot of people use it
for medical questions.

732
00:42:35,520 --> 00:42:37,957
I would really think about safety from

733
00:42:38,040 --> 00:42:43,157
the perspective of like don't take
the models on faith about medical things.

734
00:42:43,240 --> 00:42:46,037
Um in my own experience Claude has been

735
00:42:46,120 --> 00:42:48,637
right more often than my doctors about

736
00:42:48,720 --> 00:42:51,637
complex medical cases and I would never

737
00:42:51,720 --> 00:42:56,597
do something without checking with a
licensed medical professional. Right? We

738
00:42:56,680 --> 00:43:00,157
we um we are very open about the fact that

739
00:43:00,240 --> 00:43:04,277
like the models make things up sometimes.
They get confused. They don't

740
00:43:04,360 --> 00:43:05,957
know you they can't examine you, right?

741
00:43:06,040 --> 00:43:07,917
Just like I was saying on stage. So I

742
00:43:08,000 --> 00:43:09,477
think like having some healthy

743
00:43:09,560 --> 00:43:13,677
skepticism is extremely correct. Um but

744
00:43:13,760 --> 00:43:18,397
think of it as like if you had a friend
who was a really good doctor who maybe

745
00:43:18,480 --> 00:43:21,517
wasn't a specialist. So you're like
I'm seeing a specialist and I would like

746
00:43:21,600 --> 00:43:25,477
help kind of being guided in this
conversation with my doctor. I think

747
00:43:25,560 --> 00:43:30,037
Claude is a great tool for that. Um
it's great for helping you think of like

748
00:43:30,120 --> 00:43:33,197
things that might be going on that you
might not know. But I think my number

749
00:43:33,280 --> 00:43:36,357
one recommendation is please do not just
do medical things that any I I mean I

750
00:43:36,440 --> 00:43:37,637
know you're all too smart to do this.

751
00:43:37,720 --> 00:43:41,477
But like any AI tool just
says like go do X like

752
00:43:41,560 --> 00:43:45,277
look at it with some skepticism
and actually talk to a professional.

753
00:43:45,360 --> 00:43:47,797
>> Thank you to the students. And Daniela,

754
00:43:47,880 --> 00:43:52,677
before we let you go we have to do our
view from the top rapid fire tradition.

755
00:43:52,760 --> 00:43:54,877
>> Love it. >> Are you ready?
>> I'm so ready.

756
00:43:54,960 --> 00:44:01,557
>> Okay, let's go. What would you
major in if you were back in college?

757
00:44:01,640 --> 00:44:05,237
>> [laughter] >> Can I if I
say business does that just

758
00:44:05,320 --> 00:44:08,477
get me out of this [laughter] entirely?

759
00:44:08,560 --> 00:44:13,197
Um What would I major in? I I would
probably major in literature again. I

760
00:44:13,280 --> 00:44:16,957
know that sounds crazy.
I loved I like to read.

761
00:44:17,040 --> 00:44:20,077
>> What's your favorite thing
about working with your brother?

762
00:44:20,160 --> 00:44:23,827
>> Oh, um Thanksgiving
dinner. No, I'm just

763
00:44:23,910 --> 00:44:24,517
>> [laughter]

764
00:44:24,600 --> 00:44:28,037
>> Um I would say
that we like know so much

765
00:44:28,120 --> 00:44:32,117
about each other and we're able to have
like sometimes we're able to say things

766
00:44:32,200 --> 00:44:35,237
to each other that nobody else in the
company can. Right? Like sometimes we

767
00:44:35,320 --> 00:44:38,397
can get away with something that no
one else feels like they could.

768
00:44:38,480 --> 00:44:41,197
>> What about Lee's favorite thing
about working with your brother?

769
00:44:41,280 --> 00:44:44,557
>> Um Thanksgiving dinner. Uh no, um

770
00:44:44,640 --> 00:44:47,957
I think just like the um needing to also

771
00:44:48,040 --> 00:44:52,077
have separation between like
our personal relationship and our work

772
00:44:52,160 --> 00:44:56,077
relationship. So, we build in time every
week when we hang out outside of the

773
00:44:56,160 --> 00:44:59,557
office, but I think it that's hard.
Just that like we were siblings for a long

774
00:44:59,640 --> 00:45:02,997
time before we were co-founders. We will
be siblings for a long time after we're

775
00:45:03,080 --> 00:45:06,237
co-founders and making sure to just like
con- continue to water that relationship

776
00:45:06,320 --> 00:45:10,077
outside of work. >> Favorite
book you found at the library

777
00:45:10,160 --> 00:45:14,037
in your office. >> Ooh, oh man.
Um I don't know if I've

778
00:45:14,120 --> 00:45:17,357
discovered a new book, which maybe means
I should get I was like, I like reading.

779
00:45:17,440 --> 00:45:19,757
Just kidding. Um I should probably

780
00:45:19,840 --> 00:45:24,397
probably go look more, but um I did
I was reminded of a favorite book

781
00:45:24,480 --> 00:45:27,917
there. Maybe that counts. Um there's
a book called The Guns of August. I don't

782
00:45:28,000 --> 00:45:31,037
know if there's any World War I
like I'm not saying I'm seeing some

783
00:45:31,120 --> 00:45:34,757
blank faces, so maybe not. It's a great
book. If you're curious about World War I,

784
00:45:34,840 --> 00:45:39,357
um I I picked it up from the library and
actually reread it um because I read it

785
00:45:39,440 --> 00:45:42,517
I think maybe right after college
or something. But I think it's a really

786
00:45:42,600 --> 00:45:44,237
important study of just like the

787
00:45:44,320 --> 00:45:47,157
individual people and personalities that

788
00:45:47,240 --> 00:45:51,677
resulted in and sort of led to um you
know, the beginnings of what became

789
00:45:51,760 --> 00:45:55,517
World War I and just how
much like so many individual

790
00:45:55,600 --> 00:45:58,797
events and and sort of people and
personalities hinged on this like really

791
00:45:58,880 --> 00:46:03,357
ultimately sort of tragic and terrible
thing that happened. So, anyway,

792
00:46:03,440 --> 00:46:05,037
Blinkist shoutout of read that book.

793
00:46:05,120 --> 00:46:08,757
It's great. >> Amazing.
>> And if Anthropic had ended up with a

794
00:46:08,840 --> 00:46:11,117
different name, what would it be?

795
00:46:11,200 --> 00:46:16,497
>> Oh, man. Um, wow, we went through
some like truly tragic ideas before

796
00:46:16,580 --> 00:46:19,277
[laughter] we came up with Anthropic.
I think for some reason we were like very

797
00:46:19,360 --> 00:46:23,077
into birds at the time.
We were like Sparrow Systems. I think

798
00:46:23,160 --> 00:46:26,797
no idea where that one came from,
but um, I think actually now that I'm

799
00:46:26,880 --> 00:46:28,757
remembering, I think some of the like

800
00:46:28,840 --> 00:46:30,917
model names from the early days were all

801
00:46:31,000 --> 00:46:34,357
birds. So, there was like Bert, which

802
00:46:34,440 --> 00:46:36,797
then became like we had a Snuffleupagus.

803
00:46:36,880 --> 00:46:40,717
That's not a bird, but like then anyway,
we had this kind of like bird thing. I

804
00:46:40,800 --> 00:46:43,877
don't know. There were some
terrible bird names. Thankfully,

805
00:46:43,960 --> 00:46:47,797
better decisions prevailed and we named
it Anthropic, which is now impossible to

806
00:46:47,880 --> 00:46:50,037
imagine as not her name. >> Imagine.

807
00:46:50,120 --> 00:46:53,117
And finally, not a rapid-fire question.

808
00:46:53,200 --> 00:46:55,117
Best advice you ever received.

809
00:46:55,200 --> 00:46:59,557
>> Ooh. Drop the

810
00:46:59,640 --> 00:47:01,957
ant. No, um, I'm trying to think of

811
00:47:02,040 --> 00:47:03,517
um,

812
00:47:03,600 --> 00:47:07,237
best advice I've ever received.

813
00:47:07,320 --> 00:47:11,917
I think probably, you know, I will
say when we were thinking about leaving

814
00:47:12,000 --> 00:47:15,197
and it was it was

815
00:47:15,280 --> 00:47:19,677
it's now in retrospect. People are like,
"Of course you guys you all left and you

816
00:47:19,760 --> 00:47:24,397
founded Anthropic." It didn't feel that
way at the time. We were like, "This is

817
00:47:24,480 --> 00:47:28,117
a really, you know, this is kind of a
crazy thing to do. Maybe we should maybe

818
00:47:28,200 --> 00:47:32,197
we should stay. Maybe we could make
this work." Um, but we talked I talked to

819
00:47:32,280 --> 00:47:34,197
one, you know, sort of friend and mentor

820
00:47:34,280 --> 00:47:38,197
outside of work and she was like,

821
00:47:38,280 --> 00:47:41,517
"Honestly, I don't think you really
need to be on the phone with me. Like you

822
00:47:41,600 --> 00:47:43,277
already know what the right answer is."

823
00:47:43,360 --> 00:47:47,797
And I think that is I think in general
like when you're in a moment of

824
00:47:47,880 --> 00:47:49,077
"Is this the right thing for my life?"

825
00:47:49,160 --> 00:47:52,117
often you actually know what the right
answer is. Um, and I think that was

826
00:47:52,200 --> 00:47:55,517
really good advice.
>> Daniela, it's been a pleasure.

827
00:47:55,600 --> 00:47:59,680
>> Thank you. >> Thank you so much.
Thanks for having me.

828
00:48:04,650 --> 00:48:06,670
>> [music]

829
00:48:13,820 --> 00:48:15,840
[music]

830
00:48:19,060 --> 00:48:21,080
[music]

831
00:48:26,920 --> 00:48:28,940
[music]