1
00:00:00,000 --> 00:00:02,637
live stream um to this conversation. I

2
00:00:02,720 --> 00:00:03,997
have to say I have been looking forward

3
00:00:04,080 --> 00:00:07,197
to for months. Uh I had was lucky enough

4
00:00:07,280 --> 00:00:09,837
to ch to moderate a conversation between

5
00:00:09,920 --> 00:00:13,597
Darede and Demis last year in Paris. Um

6
00:00:13,680 --> 00:00:17,437
which I'm afraid got most attention for
the fact that you two were squashed on a

7
00:00:17,520 --> 00:00:22,157
very small love seat while I sat
on an enormous sofa which was probably my

8
00:00:22,240 --> 00:00:26,237
screw- up. But I said at that point that
this was for me like you know chairing a

9
00:00:26,320 --> 00:00:29,837
conversation between the Beatles and the
Rolling Stones. and you have not had a

10
00:00:29,920 --> 00:00:33,837
conversation on stage since. So, this is,
you know, the sequel, the the the,

11
00:00:33,920 --> 00:00:35,517
you know, the bands get together again.

12
00:00:35,600 --> 00:00:37,997
I'm delighted. You need no introduction.

13
00:00:38,080 --> 00:00:42,557
Uh, the title of our conversation is the
day after AGI, which I think is perhaps

14
00:00:42,640 --> 00:00:45,277
slightly getting ahead of ourselves
because we should probably talk about

15
00:00:45,360 --> 00:00:49,117
how quickly and easily we will get there.
And I want to do a bit of a sort

16
00:00:49,200 --> 00:00:52,237
of update on that and then talk about
the consequences. So firstly on the

17
00:00:52,320 --> 00:00:56,877
timeline Dario you last year in Paris
said we'll have a model that can do

18
00:00:56,960 --> 00:00:58,717
everything a human could do at the level

19
00:00:58,800 --> 00:01:00,797
of a Nobel laureate across many fields

20
00:01:00,880 --> 00:01:05,917
by 2627 we're in 26.
Uh do you still stand by that timeline?

21
00:01:06,000 --> 00:01:09,117
>> So you know it's always hard to know
exactly when something will happen but

22
00:01:09,200 --> 00:01:11,037
but I don't I don't think that's going

23
00:01:11,120 --> 00:01:13,677
to turn out to be that far off. So um

24
00:01:13,760 --> 00:01:18,797
you know the the the mechanism whereby
I imagined it would happen is that we

25
00:01:18,880 --> 00:01:22,877
would make models that were good
at coding and good at AI research and we

26
00:01:22,960 --> 00:01:27,277
would use that to produce the next
generation of model and speed it up to

27
00:01:27,360 --> 00:01:29,757
create a loop that would that would uh

28
00:01:29,840 --> 00:01:31,917
increase the speed of model development.

29
00:01:32,000 --> 00:01:34,477
We are now in terms of you know the

30
00:01:34,560 --> 00:01:39,037
models that write code I have engineers
within enthropic who say I don't write

31
00:01:39,120 --> 00:01:41,277
any code anymore. I just I just let the

32
00:01:41,360 --> 00:01:43,757
model write the code. I edit it. I do

33
00:01:43,840 --> 00:01:46,397
the things around it. I think I don't

34
00:01:46,480 --> 00:01:49,357
know. We might be six to 12 months away

35
00:01:49,440 --> 00:01:51,837
from when the model is doing most maybe

36
00:01:51,920 --> 00:01:55,357
all of what SWES do end to end. And then

37
00:01:55,440 --> 00:01:57,117
it's a question of how fast does that

38
00:01:57,200 --> 00:02:02,397
loop close. Not every part of that loop
is something that can be sped up by AI,

39
00:02:02,480 --> 00:02:06,077
right? There's like chips, there's
manufacturer of chips, there's training

40
00:02:06,160 --> 00:02:09,517
time for the model. So it's, you know,
I I think there's a lot of uncertainty.

41
00:02:09,600 --> 00:02:15,117
It's easy to see how this could take
a few years. I don't I I it's very hard

42
00:02:15,200 --> 00:02:19,517
for me to see how it could take longer
than that. Um but if if I had to guess,

43
00:02:19,600 --> 00:02:23,437
I would guess that this goes faster
than people imagine. And that that key

44
00:02:23,520 --> 00:02:26,397
element of code and increasingly

45
00:02:26,480 --> 00:02:29,837
research going faster than we imagine.

46
00:02:29,920 --> 00:02:33,757
That's going to be the key driver.
It's it's really hard to predict again how

47
00:02:33,840 --> 00:02:37,517
much that exponential is going to speed
us up, but but something fast is going

48
00:02:37,600 --> 00:02:41,757
to happen. So you demis were a little
more cautious last year. You said a 50%

49
00:02:41,840 --> 00:02:45,917
chance of a system that can exhibit all
the cognitive capabilities humans can by

50
00:02:46,000 --> 00:02:48,637
the end of the decade. Um clearly in

51
00:02:48,720 --> 00:02:53,197
coding as Dario says it's been remarkable.
What is your sense of do you

52
00:02:53,280 --> 00:02:56,157
stand by your prediction
and what's changed in the past year?

53
00:02:56,240 --> 00:02:59,997
>> Yeah, look I I I I think I'm still on
the same kind of timeline. I think there

54
00:03:00,080 --> 00:03:01,917
has been remarkable progress. But I

55
00:03:02,000 --> 00:03:05,357
think some areas of uh uh um kind of

56
00:03:05,440 --> 00:03:09,997
engineering work, coding or so you could
say mathematics are a little bit easier

57
00:03:10,080 --> 00:03:13,757
to see how they would be automated
partly because they're verifiable what

58
00:03:13,840 --> 00:03:17,837
the output is. Um some areas of natural
science are much harder to do than that.

59
00:03:17,920 --> 00:03:21,197
You won't necessarily know if
the chemical compound you've built or this

60
00:03:21,280 --> 00:03:24,557
prediction about physics is correct.
It may be you may have to test it

61
00:03:24,640 --> 00:03:28,797
experimentally and that will all take
longer. So uh I also think there are

62
00:03:28,880 --> 00:03:31,277
some missing capabilities at the moment

63
00:03:31,360 --> 00:03:35,997
uh in terms of like not just solving
existing conjectures uh or existing

64
00:03:36,080 --> 00:03:39,677
problems but actually coming up with the
question in the first place or coming up

65
00:03:39,760 --> 00:03:43,117
with the theory or the hypothesis.
I think that's much much harder and I

66
00:03:43,200 --> 00:03:46,477
think that's the highest level
of scientific creativity and it's not

67
00:03:46,560 --> 00:03:49,277
clear. I think we will have those
systems so I don't think it's impossible

68
00:03:49,360 --> 00:03:53,277
but I think there may be one or two
missing ingredients. Um, it remains to

69
00:03:53,360 --> 00:03:56,397
be seen how, you know, first of all,
can this self-improvement loop that we're

70
00:03:56,480 --> 00:03:59,917
all working on actually close without
a human in the loop. I think there are

71
00:04:00,000 --> 00:04:02,797
also risks to that with to that kind
of system, by the way, which we should

72
00:04:02,880 --> 00:04:07,197
discuss and I'm sure we will. But the
the but but that could speed things up

73
00:04:07,280 --> 00:04:08,797
if that kind of system does work.

74
00:04:08,880 --> 00:04:11,437
>> We'll get to the risks in a minute.
But one other change I think of the past

75
00:04:11,520 --> 00:04:15,437
year has been a kind of change in the
pecking order of the race, if you will.

76
00:04:15,520 --> 00:04:19,037
This time a year ago, we just had
the deepseek moment and everyone was

77
00:04:19,120 --> 00:04:23,117
incredibly excited about what happened
there and there was still a sense, you

78
00:04:23,200 --> 00:04:28,557
know, that Google Deep Mind was kind of
lagging open AI. I would say that now uh

79
00:04:28,640 --> 00:04:31,597
it's looking quite different. I mean,
they've declared code red, right? Um

80
00:04:31,680 --> 00:04:33,917
it's been quite a quite a year. So, talk

81
00:04:34,000 --> 00:04:38,877
me through what specifically you've been
surprised by and how well you've done

82
00:04:38,960 --> 00:04:41,917
this year and whether you think and then
I'm going to ask you about the lineup.

83
00:04:42,000 --> 00:04:46,397
Well, look, I I think we were I
was always very confident we uh would get

84
00:04:46,480 --> 00:04:49,997
back to sort of the top of the the
leaderboards and and the soda type of

85
00:04:50,080 --> 00:04:54,157
models across the board because I think
we've always had like the deepest and

86
00:04:54,240 --> 00:04:57,837
broadest research bench and it was about
kind of marshalling that all together

87
00:04:57,920 --> 00:05:01,997
and um getting the intensity and focus
and the kind of startup mentality back

88
00:05:02,080 --> 00:05:04,077
to the whole organization and it's been

89
00:05:04,160 --> 00:05:09,117
a a lot of work and um but I think we're
and we're still a lot of work to do um

90
00:05:09,200 --> 00:05:13,357
but I think you can start seeing the the
the the the you know the the kind of um

91
00:05:13,440 --> 00:05:17,837
the progress that's been made in both
the models with Gemini 3 but also uh on

92
00:05:17,920 --> 00:05:22,317
the product side with Gemini app getting
increasing uh market share. So I feel

93
00:05:22,400 --> 00:05:27,357
like uh we're making great progress um
but there's a ton more work to do um and

94
00:05:27,440 --> 00:05:30,557
you know we're bringing to bear Google
deep mind's kind of like the engine room

95
00:05:30,640 --> 00:05:35,037
of Google where we're getting used to um
shipping our models more quickly into

96
00:05:35,120 --> 00:05:38,477
the product surfaces. One question
for you Daria on on this aspect of it

97
00:05:38,560 --> 00:05:41,997
because you've just or you're in the
process of you know a new round at an

98
00:05:42,080 --> 00:05:44,077
extraordinary valuation too. Um but you

99
00:05:44,160 --> 00:05:49,037
are unlike Demis a let's call it
an independent model maker and there is I

100
00:05:49,120 --> 00:05:52,077
think an increasing concern that the
independent model makers will not be

101
00:05:52,160 --> 00:05:56,317
able to continue for long enough until
you get to where the revenues come in.

102
00:05:56,400 --> 00:06:00,237
Um it's made very openly about open AI
but talk me through how you think about

103
00:06:00,320 --> 00:06:03,277
that and then we'll get to the AGI itself.
Yeah, I mean, you know, I think

104
00:06:03,360 --> 00:06:08,157
I think I think how we think about that
is, you know, as we've built better and

105
00:06:08,240 --> 00:06:12,717
better models, there's been a kind
of exponential relationship, not only

106
00:06:12,800 --> 00:06:16,477
between how much compute you put into
the model and how cognitively capable it

107
00:06:16,560 --> 00:06:20,557
is, but between how cognitively capable
it is and how much revenue it's able to

108
00:06:20,640 --> 00:06:24,797
generate. So, our revenues grown 10x
in the last three years from 0 to 100

109
00:06:24,880 --> 00:06:27,117
million in 2023, 100 million to a

110
00:06:27,200 --> 00:06:32,077
billion in 2024, and 1 billion to 10
billion in 2025. And so th those revenue

111
00:06:32,160 --> 00:06:35,277
numbers, you know, I don't know if that
curve will literally continue. It would

112
00:06:35,360 --> 00:06:40,157
be crazy if it did. Um, but those
numbers are starting to get not too far

113
00:06:40,240 --> 00:06:43,757
from, you know, the sca the scale
of the largest companies in the world. So

114
00:06:43,840 --> 00:06:46,157
there's there's there's always
uncertainty. You know, we're trying to

115
00:06:46,240 --> 00:06:50,477
bootstrap this from nothing. It's it's
a crazy thing, but but I have confidence

116
00:06:50,560 --> 00:06:55,437
that if we're able to produce the best
models in the things that we focus on,

117
00:06:55,520 --> 00:06:59,117
um, uh, then I think then I think things
will go well. And you know I I will I

118
00:06:59,200 --> 00:07:02,477
will generally say you know I think
I think it's been a good year for both

119
00:07:02,560 --> 00:07:07,357
both Google and Anthropic and I think
the thing we actually have in common is

120
00:07:07,440 --> 00:07:10,717
that they're you know they're both kind
of kind of kind of companies that are

121
00:07:10,800 --> 00:07:13,997
you know or the research part
of the company that are kind of led by

122
00:07:14,080 --> 00:07:19,117
researchers who focus on the models who
focus on solving important problems in

123
00:07:19,200 --> 00:07:20,637
the world right who have these kind of

124
00:07:20,720 --> 00:07:23,677
hard scientific problems as a as a north

125
00:07:23,760 --> 00:07:27,917
star and and and I think those are
the kind of companies companies that are

126
00:07:28,000 --> 00:07:31,437
going to succeed going forward and you
know I think I think we share that

127
00:07:31,520 --> 00:07:33,837
between us >> very much.
Uh I'm I'm going to resist

128
00:07:33,920 --> 00:07:36,477
the temptation to ask you what will
happen to the companies that are not led

129
00:07:36,560 --> 00:07:39,917
by researchers uh because
I know you won't answer it.

130
00:07:40,000 --> 00:07:45,277
But let's then go on to uh
the predictions area now and this we are

131
00:07:45,360 --> 00:07:48,157
supposed to be talking about the day
after AI. But let's talk about closing

132
00:07:48,240 --> 00:07:52,557
the loop. This the odds that you will
get models that will close the loop and

133
00:07:52,640 --> 00:07:55,997
be able to you know power themselves if
you will because that's the really the

134
00:07:56,080 --> 00:08:00,397
crux for the the winner takes all
threshold approach. Do you still believe

135
00:08:00,480 --> 00:08:04,797
that we are likely to see that or is
this going to be much more of a normal

136
00:08:04,880 --> 00:08:09,117
technology where followers
and catchup can can compete?

137
00:08:09,200 --> 00:08:11,677
>> Well, look, I definitely don't think
it's going to be a normal technology.

138
00:08:11,760 --> 00:08:15,517
So, I mean, there are aspects already
that as Dario mentioned that it's

139
00:08:15,600 --> 00:08:19,997
already helping with our coding
and and some aspects of research. The full

140
00:08:20,080 --> 00:08:23,677
closing of the loop though, I think
is an unknown. I mean, I think it's

141
00:08:23,760 --> 00:08:28,877
possible to do. you may need AGI itself
to be able to do that in some domains

142
00:08:28,960 --> 00:08:32,077
again where there these domains you
know where there's there's more messiness

143
00:08:32,160 --> 00:08:34,237
around them it's not so easy to verify

144
00:08:34,320 --> 00:08:39,437
your answer very quickly um there's
kind of MP hard domains so as soon as you

145
00:08:39,520 --> 00:08:43,357
start getting more and you know I also
include by the way for AGI physical AI

146
00:08:43,440 --> 00:08:46,557
robotics working all of these kind
of things and then you've got you know

147
00:08:46,640 --> 00:08:51,037
hardware in the loop uh that may uh
limit how fast the self-improvement

148
00:08:51,120 --> 00:08:55,357
systems can work but I think in coding
and mathematics and these kind areas. I

149
00:08:55,440 --> 00:08:57,997
can definitely see that working.
And then the question is more theoretical

150
00:08:58,080 --> 00:09:00,077
one is what is the limit of engineering

151
00:09:00,160 --> 00:09:03,197
and maths uh to solve uh the natural

152
00:09:03,280 --> 00:09:07,837
sciences. Dario, you um last year,
I think it was last year that you

153
00:09:07,920 --> 00:09:09,837
published Machines of Love and Grace um

154
00:09:09,920 --> 00:09:12,797
which was a very I would say upbeat

155
00:09:12,880 --> 00:09:17,277
essay about the potential that that you
were going to see unfold and you were

156
00:09:17,360 --> 00:09:19,597
talking about you know a a what was it a

157
00:09:19,680 --> 00:09:22,477
genius of data at country

158
00:09:22,560 --> 00:09:27,117
data center I'm told that you are
working on an update to this a new essay

159
00:09:27,200 --> 00:09:29,357
so you know wait for it guys it's not

160
00:09:29,440 --> 00:09:31,437
out yet but it is coming out but perhaps

161
00:09:31,520 --> 00:09:36,877
you can give us a sort of a sneak
preview of what a year later your big

162
00:09:36,960 --> 00:09:39,997
take is going to be. >> Yes.
So, you know, my take my take has

163
00:09:40,080 --> 00:09:43,917
not changed. It has always been my view
that, you know, AI is going to be

164
00:09:44,000 --> 00:09:47,037
incredibly powerful. I think Demis and I,
you know, kind of agree on that. It's

165
00:09:47,120 --> 00:09:51,437
just a question of exactly when um uh
and because it's incredibly powerful, it

166
00:09:51,520 --> 00:09:54,397
will do all these wonderful things like
the ones I talked about in Machines of

167
00:09:54,480 --> 00:09:58,237
Love and Grace. It, you know, will
help us cure cancer. It may help us to

168
00:09:58,320 --> 00:10:02,397
eradicate tropical diseases.
It will help us understand understand the

169
00:10:02,480 --> 00:10:07,277
universe. but that there are these, you
know, immense and grave risks that, you

170
00:10:07,360 --> 00:10:08,797
know, not that we can't address them.

171
00:10:08,880 --> 00:10:12,557
I'm not a doomer, but but that, you know,
we we we we we need to think about

172
00:10:12,640 --> 00:10:16,477
them and we need to address them. And I
wrote Machines of Loving Grace first. I'

173
00:10:16,560 --> 00:10:20,397
I'd love to give some uh a sophisticated
reason why I wrote that first, but it

174
00:10:20,480 --> 00:10:24,557
was just that the the positive essay
was easier and more fun to write than than

175
00:10:24,640 --> 00:10:28,797
the negative essay. Um, so, you know,
I finally spent some time on vacation and

176
00:10:28,880 --> 00:10:30,797
I was able to write an essay about the

177
00:10:30,880 --> 00:10:32,397
risks. And even when I'm writing about

178
00:10:32,480 --> 00:10:36,557
the risks, um, I I I try, you know, I I

179
00:10:36,640 --> 00:10:40,717
I'm like an optimistic person, right? So
even as I'm writing about these risks, I

180
00:10:40,800 --> 00:10:44,717
I I wrote about it in a way that was like,
how do we overcome these risks?

181
00:10:44,800 --> 00:10:48,717
How do we have a battle plan to fight
them? And and and the way I the way I

182
00:10:48,800 --> 00:10:53,117
framed it was, you know, there's
this scene from Carl Sean's Contact, the

183
00:10:53,200 --> 00:10:56,797
movie version of it, where, you know,
they they kind of discover alien life

184
00:10:56,880 --> 00:11:01,517
and this international panel that's like
interviewing um uh you know, people to,

185
00:11:01,600 --> 00:11:05,677
you know, to be humanity's
representative to meet the alien. Um uh

186
00:11:05,760 --> 00:11:09,117
and uh one one of the questions they
asked one of the candidates is, you

187
00:11:09,200 --> 00:11:12,637
know, if you could ask the aliens anyone
question, what it would what what what

188
00:11:12,720 --> 00:11:17,037
would it be? And one of one of the
characters says, "I would ask, how did

189
00:11:17,120 --> 00:11:21,997
you do it? How did you manage to get
through this technological adolescence

190
00:11:22,080 --> 00:11:26,157
without destroying yourselves? How did
you make it through?" And and and ever

191
00:11:26,240 --> 00:11:28,957
since I saw it, it was like 20 years ago,
I think I saw that movie, it's kind

192
00:11:29,040 --> 00:11:32,877
of stuck with me. And that that's
the frame that I used, which is which is

193
00:11:32,960 --> 00:11:35,437
that, you know, we we're we're we are

194
00:11:35,520 --> 00:11:37,997
knocking on the door of these incredible

195
00:11:38,080 --> 00:11:40,237
capabilities, right? the the ability to

196
00:11:40,320 --> 00:11:45,117
build basically machines out of sand,
right? I think I think it was inevitable

197
00:11:45,200 --> 00:11:47,037
that the instant we started working with

198
00:11:47,120 --> 00:11:50,957
fire. Um uh but but how we handle it is

199
00:11:51,040 --> 00:11:53,037
is not inevitable. And so I think the

200
00:11:53,120 --> 00:11:58,157
next few years we're going to be dealing
with, you know, how do we keep these

201
00:11:58,240 --> 00:11:59,837
systems under control that are highly

202
00:11:59,920 --> 00:12:02,637
autonomous and smarter than any human?

203
00:12:02,720 --> 00:12:05,917
How do we make sure that individuals

204
00:12:06,000 --> 00:12:09,757
don't misuse them? Right? I have worries
about things like bioteterrorism. How do

205
00:12:09,840 --> 00:12:13,757
we make sure that nation states don't
misuse them? That's why I've been so

206
00:12:13,840 --> 00:12:18,157
concerned about, you know, the CCP,
other authoritarian authoritarian

207
00:12:18,240 --> 00:12:21,437
governments. What are the economic
impacts? Right? I've talked about labor

208
00:12:21,520 --> 00:12:24,797
displacement a lot. And and you know,
what what haven't we thought of which

209
00:12:24,880 --> 00:12:28,477
which in many cases, you know, maybe may
be the the hardest thing to deal with at

210
00:12:28,560 --> 00:12:31,357
all. Um so, you know, I I'm I'm thinking

211
00:12:31,440 --> 00:12:34,397
through how to address those risks. And

212
00:12:34,480 --> 00:12:38,557
you know, for for each of these, it's
a mixture of things that we individually

213
00:12:38,640 --> 00:12:43,197
need to do as as leaders of the of of of
the companies and that we can do working

214
00:12:43,280 --> 00:12:46,397
together. And then there there's
going to need to be some role for wider

215
00:12:46,480 --> 00:12:50,637
societal institutions like the like
the government in in in addressing all of

216
00:12:50,720 --> 00:12:54,717
these. But, you know, I I I just feel
this urgency that, you know, every day,

217
00:12:54,800 --> 00:12:57,517
you know, there's there's all kinds
of crazy stuff going on in the outside

218
00:12:57,600 --> 00:13:00,557
world, outside AI, right? Um but but you

219
00:13:00,640 --> 00:13:03,197
know my my my view is this is happening

220
00:13:03,280 --> 00:13:08,397
so fast and is such a crisis we should
be devoting almost all of our effort to

221
00:13:08,480 --> 00:13:10,077
thinking about how to get through this.

222
00:13:10,160 --> 00:13:13,677
>> So I can't decide whether I'm more
surprised that you a take a vacation b

223
00:13:13,760 --> 00:13:17,597
when you take a vacation you think about
the risks of AI and c that your essay is

224
00:13:17,680 --> 00:13:21,997
framed in terms of are we going to get
through the technological adolescence of

225
00:13:22,080 --> 00:13:25,357
this technology without destroying
ourselves. So, I'm my head is slightly

226
00:13:25,440 --> 00:13:28,557
spinning, but you then and I can't wait
to read it, but you you you mentioned

227
00:13:28,640 --> 00:13:32,477
several areas that can guide the rest
of our conversation. Let's start with jobs

228
00:13:32,560 --> 00:13:35,277
um because you actually have been very
outspoken about that and I think you

229
00:13:35,360 --> 00:13:38,397
said that half of entry- level white
collar jobs could be gone within the

230
00:13:38,480 --> 00:13:39,997
next one to five years. But I'm going to

231
00:13:40,080 --> 00:13:43,517
turn to you Demis because so far we

232
00:13:43,600 --> 00:13:48,317
haven't actually seen any discernable
impact on the labor market. Um, yes,

233
00:13:48,400 --> 00:13:51,917
unemployment has ticked up in the US,
but all of the kind of economic studies

234
00:13:52,000 --> 00:13:56,157
I've looked at and that we've written
about suggest that this is overhiring

235
00:13:56,240 --> 00:13:57,837
post pandemic that it's really not

236
00:13:57,920 --> 00:14:00,797
AIdriven. If anything, people are hiring

237
00:14:00,880 --> 00:14:03,917
to build out AI capability.

238
00:14:04,000 --> 00:14:08,957
Do you think that this will be as you
know economists have always argued that

239
00:14:09,040 --> 00:14:12,317
it's not a lump of labor fallacy that
actually there will be new jobs created

240
00:14:12,400 --> 00:14:16,637
because so far the evidence seems
to suggest that? Yeah, I mean I I think in

241
00:14:16,720 --> 00:14:19,837
um the near term that is what will happen.
The kind of normal evolution

242
00:14:19,920 --> 00:14:21,757
when a breakthrough technology arrives.

243
00:14:21,840 --> 00:14:26,157
So some jobs will get disrupted but I
think new even more valuable perhaps

244
00:14:26,240 --> 00:14:28,237
more meaningful jobs will get created.

245
00:14:28,320 --> 00:14:32,477
Um I think we're going to see this year
the beginnings of maybe impacting the

246
00:14:32,560 --> 00:14:36,637
junior level entry level kind of jobs
internships this type of thing. I think

247
00:14:36,720 --> 00:14:40,637
there is some evidence I can feel
that ourselves maybe like a slowdown in

248
00:14:40,720 --> 00:14:44,077
hiring in that but I think that can be
more than compensated by the fact there

249
00:14:44,160 --> 00:14:48,317
are these amazing creative tools out
there pretty much available for everyone

250
00:14:48,400 --> 00:14:50,957
uh almost for free that if you know I

251
00:14:51,040 --> 00:14:53,757
was to talk to a a class of undergrads

252
00:14:53,840 --> 00:14:58,077
right now I would be telling them to get
really unbelievably proficient with

253
00:14:58,160 --> 00:15:01,757
these tools I think to the extent
that even those of us building it we're so

254
00:15:01,840 --> 00:15:05,517
busy building it it's hard to have also
time to really explore the almost the

255
00:15:05,600 --> 00:15:10,157
capability overhang even today's models
and products have let alone tomorrow's

256
00:15:10,240 --> 00:15:14,237
and I think that uh can be maybe better
than a traditional internship would have

257
00:15:14,320 --> 00:15:19,117
been in terms of you sort of leaprogging
uh yourself to be a useful uh in a

258
00:15:19,200 --> 00:15:22,797
useful in a profession so I think
there's that's what I see happening

259
00:15:22,880 --> 00:15:27,277
probably in the next 5 years um maybe we
again slightly differ on time scales on

260
00:15:27,360 --> 00:15:31,037
that but I think what happens after
AGI arrives that's a different question

261
00:15:31,120 --> 00:15:34,397
because I think really we would be
in uncharted territory at that point

262
00:15:34,480 --> 00:15:37,037
>> do you think it's going to take longer
than you thought last year when you said

263
00:15:37,120 --> 00:15:40,557
half of all white color jobs. >> I
have about the same view. I I actually

264
00:15:40,640 --> 00:15:45,037
agree with you and with Demis that at
the time I made the comment there was no

265
00:15:45,120 --> 00:15:47,917
impact on the labor market. I wasn't
saying there was an impact on the labor

266
00:15:48,000 --> 00:15:53,037
market at that moment. Um you know now
I think maybe we're starting to see just

267
00:15:53,120 --> 00:15:56,317
just the little beginnings of it you
know in software and coding. I even see

268
00:15:56,400 --> 00:15:59,117
it within within anthropic where you

269
00:15:59,200 --> 00:16:02,317
know I you know I can look forward I can

270
00:16:02,400 --> 00:16:06,477
kind of look forward to a time where on
the more junior end and then on the more

271
00:16:06,560 --> 00:16:10,557
on the more on the more on the more
intermediate end we actually need less

272
00:16:10,640 --> 00:16:13,437
and not more people and you know we're
thinking about how to deal with that

273
00:16:13,520 --> 00:16:16,397
within anthropic in a in a in a you know

274
00:16:16,480 --> 00:16:20,637
sense in a sensible way. Um I you know

275
00:16:20,720 --> 00:16:24,957
one to five years as of six months ago
I would stick with that. you know, if you

276
00:16:25,040 --> 00:16:29,037
kind of, you know, connect this to what
I said before, which is, you know, we we

277
00:16:29,120 --> 00:16:31,037
might have AI that's better than humans

278
00:16:31,120 --> 00:16:33,837
at at everything in, you know, maybe one

279
00:16:33,920 --> 00:16:36,157
to two years, maybe a little longer than

280
00:16:36,240 --> 00:16:38,797
that. The those don't seem to line up.

281
00:16:38,880 --> 00:16:43,917
The reason is that there's this there's
this lag and there's this replacement

282
00:16:44,000 --> 00:16:47,197
thing, right? I I know that the labor
market is adaptable, right? Just like

283
00:16:47,280 --> 00:16:51,517
you know 80% of people used to do
farming you know farming got automated

284
00:16:51,600 --> 00:16:55,517
and then they became factory workers
and then knowledge workers. So you know

285
00:16:55,600 --> 00:16:59,757
there is some level of adaptability here
as well right we should be economically

286
00:16:59,840 --> 00:17:03,917
sophisticated about how the labor
market works but my worry is as this

287
00:17:04,000 --> 00:17:08,157
exponential keeps compounding and I
don't think it's going to take that long

288
00:17:08,240 --> 00:17:12,557
again somewhere between between a year
and five years it will overwhelm our

289
00:17:12,640 --> 00:17:17,117
ability to adapt. I think I may be
saying the same thing Demis is just

290
00:17:17,200 --> 00:17:20,477
factored out of that that difference
we have about timelines which I think

291
00:17:20,560 --> 00:17:23,677
ultimately comes down to how how
fast you close the loop on CO and

292
00:17:23,760 --> 00:17:25,517
>> how much confidence do you have that

293
00:17:25,600 --> 00:17:28,637
governments get the scale of this and

294
00:17:28,720 --> 00:17:33,277
have are beginning to think about
what policy responses they need to have

295
00:17:33,360 --> 00:17:37,677
>> I don't think that that that it's
anywhere near enough work going on about

296
00:17:37,760 --> 00:17:40,637
this I'm I'm constantly surprised even
when I meet economists at places like

297
00:17:40,720 --> 00:17:44,557
this that they're not more of uh
professional economist, professors

298
00:17:44,640 --> 00:17:47,437
thinking about what happens um and not

299
00:17:47,520 --> 00:17:52,077
just sort of on the way to AGI but um
even if we get all the technical things

300
00:17:52,160 --> 00:17:55,437
right that Dario is talking about
and the job displacement is one question

301
00:17:55,520 --> 00:17:57,917
we're worried about the economics
of that but maybe there are ways to

302
00:17:58,000 --> 00:18:02,077
distribute this new productivity this
new wealth more fairly I don't know if

303
00:18:02,160 --> 00:18:05,037
we have the right institutions to do
that but that's what should happen at

304
00:18:05,120 --> 00:18:08,477
that point there should be you know we
may be in a post scarcity world but then

305
00:18:08,560 --> 00:18:11,357
there are even the things that keep
me up right now there are even bigger

306
00:18:11,440 --> 00:18:13,517
questions than that at that to do with

307
00:18:13,600 --> 00:18:18,637
meaning and um purpose and a lot of
the things that we get from our jobs not

308
00:18:18,720 --> 00:18:22,477
just economically that's one question
but I think that may be easier to solve

309
00:18:22,560 --> 00:18:27,197
strangely than uh what happens to the
human condition and humanity as a whole

310
00:18:27,280 --> 00:18:30,397
and I think I'm also optimistic we'll
come up with new answers there we do a

311
00:18:30,480 --> 00:18:32,957
lot of things today um from extreme

312
00:18:33,040 --> 00:18:35,037
sports to art that aren't necessarily

313
00:18:35,120 --> 00:18:38,237
directly to do with economic gain so I

314
00:18:38,320 --> 00:18:40,397
think we will find uh meaning and maybe

315
00:18:40,480 --> 00:18:42,797
there'll be even more sort sophisticated

316
00:18:42,880 --> 00:18:47,597
versions of those activities. Um, plus
I think we'll be exploring the stars. So,

317
00:18:47,680 --> 00:18:51,997
there'll be all of that to to factor
in as well for in terms of purpose. But I

318
00:18:52,080 --> 00:18:56,077
think it's really worth thinking now
even on my timelines of like 5 to 10

319
00:18:56,160 --> 00:18:59,517
years away. That isn't a lot
of time uh before this comes.

320
00:18:59,600 --> 00:19:01,837
>> How big do you think is the risk of a

321
00:19:01,920 --> 00:19:04,957
popular backlash against AI that will

322
00:19:05,040 --> 00:19:09,917
somehow kind of cause governments to do
what from your perspective might be

323
00:19:10,000 --> 00:19:13,757
stupid things? Because I'm just
thinking back to the era of you know

324
00:19:13,840 --> 00:19:18,157
globalization in the 1990s when when
there was indeed some displacement of

325
00:19:18,240 --> 00:19:20,397
jobs governments didn't do enough the

326
00:19:20,480 --> 00:19:25,277
public backlash was such that we've
ended up sort of where we are now. Uh do

327
00:19:25,360 --> 00:19:29,437
you think that there is a risk
that there will be a growing antipathy

328
00:19:29,520 --> 00:19:34,157
towards what you are doing and your
companies in the kind of body politic?

329
00:19:34,240 --> 00:19:37,917
>> Um I think there's definitely a risk.
I think um I think that's kind of

330
00:19:38,000 --> 00:19:41,277
reasonable. there's fear and there's
worries about these things like jobs and

331
00:19:41,360 --> 00:19:45,357
livelihoods. Um I think there's a couple
of things that I mean it's going to be

332
00:19:45,440 --> 00:19:49,197
very complicated the next few years
I think geopolitically but also the

333
00:19:49,280 --> 00:19:52,637
various factors here like we want to and
we're trying to do this with AlphaFold

334
00:19:52,720 --> 00:19:57,277
and our science work and isomorphic
our spinout company solve all disease cure

335
00:19:57,360 --> 00:20:01,037
diseases come up with new energy sources
I think as a society it's clear we'd

336
00:20:01,120 --> 00:20:05,197
want that I think maybe the balance of
what the industry is doing is not enough

337
00:20:05,280 --> 00:20:08,317
balance towards those types
of activities I think we should have a lot

338
00:20:08,400 --> 00:20:12,237
more examples I know Dario agrees with
me of like alpha fold like things that

339
00:20:12,320 --> 00:20:15,837
help sort of unequivocal good
in the world. And I think actually it's

340
00:20:15,920 --> 00:20:19,517
incumbent on the industry and and all
of us leading players to show that more,

341
00:20:19,600 --> 00:20:23,517
demonstrate that, not just talk about it,
but demonstrate that. Um and but

342
00:20:23,600 --> 00:20:27,677
then it's going to come with these other
intendent disruptions and um but I don't

343
00:20:27,760 --> 00:20:30,477
I think the other issue is the
geopolitical competition. There's

344
00:20:30,560 --> 00:20:33,757
obviously competition between
the companies but also US and China

345
00:20:33,840 --> 00:20:37,597
primarily. So unless there's
an international cooperation or or

346
00:20:37,680 --> 00:20:41,117
understanding around this um uh which I
think would be good actually in terms of

347
00:20:41,200 --> 00:20:44,637
things like minimum safety standards for
deployment I think Dario would agree on

348
00:20:44,720 --> 00:20:47,357
that as well. I think it's vitally needed.
This technology is going to be

349
00:20:47,440 --> 00:20:49,997
crossber border. It's going to affect
everyone. It's going to affect all of

350
00:20:50,080 --> 00:20:54,797
humanity. Um actually contact is one
of my favorite films as well. So funny

351
00:20:54,880 --> 00:20:59,677
enough, I didn't realize it was yours too,
Dario. But I I think um um you know

352
00:20:59,760 --> 00:21:03,757
those kind of things need to be worked
through. Um and and if we can maybe it

353
00:21:03,840 --> 00:21:07,437
would be good to have a bit of slow
a slightly slower pace than we're

354
00:21:07,520 --> 00:21:11,837
currently predicting even my timelines
so that we can get this right society.

355
00:21:11,920 --> 00:21:13,677
But that would require some coordination

356
00:21:13,760 --> 00:21:16,157
that is I I prefer your timelines.

357
00:21:16,240 --> 00:21:19,117
>> Yes, I will concede.

358
00:21:19,200 --> 00:21:23,357
>> But but Dario, let's turn to this now
because one thing since we last spoke uh

359
00:21:23,440 --> 00:21:27,597
in Paris, the geopolitical environment
has, if anything, I don't know,

360
00:21:27,680 --> 00:21:31,677
complicated, mad, crazy, whatever,
whatever phrase you want to use.

361
00:21:31,760 --> 00:21:36,077
Secondly, the US has a very different
approach now towards China. It's a much

362
00:21:36,160 --> 00:21:38,157
more it's a kind of no holds barred, go

363
00:21:38,240 --> 00:21:43,597
as fast as we can, but then sell chips
to China. Um and that is it. So you've

364
00:21:43,680 --> 00:21:48,317
got a different attitude towards
the United States. You've got a a very um

365
00:21:48,400 --> 00:21:51,597
strange relationship between the United
States and and Europe right now

366
00:21:51,680 --> 00:21:55,517
geopolitically against that. I mean I
hear you talk about it would be nice to

367
00:21:55,600 --> 00:21:59,117
have a CERN like organization. I mean
it's a million years from where we are

368
00:21:59,200 --> 00:22:02,637
from the real world. So in the real
world have the geopolitical risks

369
00:22:02,720 --> 00:22:07,277
increased and what if anything do you
think should be done about that? And and

370
00:22:07,360 --> 00:22:09,917
the administration seems to be doing
the opposite of what you were suggesting?

371
00:22:10,000 --> 00:22:12,637
Yeah, I mean, look, you know, we're
we're we're just trying to do the best

372
00:22:12,720 --> 00:22:15,437
we can to, you know, we're just we're
just one company and we're we're trying

373
00:22:15,520 --> 00:22:18,797
to operate in, you know, the the
environment that exists, no matter how

374
00:22:18,880 --> 00:22:22,717
no matter how crazy it is. But, you know,
I think I think at least my policy

375
00:22:22,800 --> 00:22:25,597
recommendations haven't changed that,

376
00:22:25,680 --> 00:22:28,957
you know, not selling chips is one of

377
00:22:29,040 --> 00:22:33,437
the, you know, one of the one of the
biggest things we can do um to, you

378
00:22:33,520 --> 00:22:37,917
know, make sure that we have the time
to handle this. Um, you know, you know, I

379
00:22:38,000 --> 00:22:40,397
said I said before, you know, I I I

380
00:22:40,480 --> 00:22:45,117
prefer Demis' timeline. I wish we had
five to 10 years, you know, so it's it's

381
00:22:45,200 --> 00:22:48,317
possible he's just right and I'm just
wrong, but but assume I'm right and it

382
00:22:48,400 --> 00:22:51,917
can be done in one to two years.
Why can't we slow down to to Demis'

383
00:22:52,000 --> 00:22:56,317
timeline? Well, no.
The but but but the reason the

384
00:22:56,400 --> 00:22:58,637
reason we the reason we can't do that is

385
00:22:58,720 --> 00:23:01,917
is you know because we have geopolitical

386
00:23:02,000 --> 00:23:04,877
adversaries building the same technology

387
00:23:04,960 --> 00:23:09,357
at a similar pace. It's very hard
to have an enforcable agreement where they

388
00:23:09,440 --> 00:23:11,437
slow down and we slow down and and so if

389
00:23:11,520 --> 00:23:14,397
we can just if we can just not sell the

390
00:23:14,480 --> 00:23:19,517
chips then this isn't a question
of competition between the US and China.

391
00:23:19,600 --> 00:23:22,477
This is a question of competition
between me and Demis which I'm very

392
00:23:22,560 --> 00:23:24,397
confident that we can work out.

393
00:23:24,480 --> 00:23:26,237
>> And what do you
make of the logic of the

394
00:23:26,320 --> 00:23:27,917
administration which as I understand it

395
00:23:28,000 --> 00:23:29,437
is we need to sell them chips because we

396
00:23:29,520 --> 00:23:33,597
need to bind them into US supply chains.

397
00:23:33,680 --> 00:23:37,677
>> So you know it's it's I I think it's I

398
00:23:37,760 --> 00:23:40,157
think it's a question not just of time

399
00:23:40,240 --> 00:23:41,917
scale but of the significance of the

400
00:23:42,000 --> 00:23:45,197
technology. Right? If this was telecom

401
00:23:45,280 --> 00:23:47,597
or something, then all this stuff about

402
00:23:47,680 --> 00:23:52,317
proliferating the US stack and you
know wanting to build our you know chips

403
00:23:52,400 --> 00:23:53,917
around the world to make sure that you

404
00:23:54,000 --> 00:23:57,677
know you know this c you know the you

405
00:23:57,760 --> 00:24:01,197
know these random countries in different
parts of the world you know build data

406
00:24:01,280 --> 00:24:03,357
centers that have Nvidia chips instead

407
00:24:03,440 --> 00:24:06,077
of Huawei chips you know I think of this

408
00:24:06,160 --> 00:24:08,317
more as like you know it's a decision

409
00:24:08,400 --> 00:24:11,357
are we going to you know sell nuclear

410
00:24:11,440 --> 00:24:14,157
weapons to North Korea Uh and you know

411
00:24:14,240 --> 00:24:18,477
because that produces some profit
for Boeing. Um you know where where we can

412
00:24:18,560 --> 00:24:22,477
say okay yeah these cases were made
by Boeing like the US is winning like this

413
00:24:22,560 --> 00:24:24,557
is great like I I I just you know that

414
00:24:24,640 --> 00:24:27,677
that analogy should just make clear how

415
00:24:27,760 --> 00:24:29,677
I see this trade-off that I just don't

416
00:24:29,760 --> 00:24:32,157
think it makes sense. Um and and we've

417
00:24:32,240 --> 00:24:37,437
done a lot of more aggressive stuff
to you know toward towards towards China

418
00:24:37,520 --> 00:24:40,957
and other players that that I think
is much less effective than this this one

419
00:24:41,040 --> 00:24:44,557
this one measure. One more area from me
and then I hope we'll have time for a

420
00:24:44,640 --> 00:24:47,037
question or two. The other area of

421
00:24:47,120 --> 00:24:49,197
potential risk that doomers worry about

422
00:24:49,280 --> 00:24:54,397
is a kind of all powerful malign AI.
Um and I think you've both been somewhat

423
00:24:54,480 --> 00:24:58,797
skeptical of the doomer approach
but in the last year we have seen you know

424
00:24:58,880 --> 00:25:00,477
these models showing themselves to be

425
00:25:00,560 --> 00:25:04,317
capable of deception duplicity. Uh do

426
00:25:04,400 --> 00:25:08,237
you think that do you think differently
about that risk now than you did a year

427
00:25:08,320 --> 00:25:12,637
ago? And is there something about
the way the models are evolving that we

428
00:25:12,720 --> 00:25:14,557
should put a little bit
more concern on that?

429
00:25:14,640 --> 00:25:17,757
>> Yeah, I mean you know since since
the beginning of anthropic we've kind of

430
00:25:17,840 --> 00:25:21,837
thought about this risk. I mean you know
our our our research at the beginning of

431
00:25:21,920 --> 00:25:25,837
it was very theoretical right? You know
we pioneered this idea of mechanistic

432
00:25:25,920 --> 00:25:29,917
interpretability which is looking inside
the model and and trying to understand

433
00:25:30,000 --> 00:25:33,677
looking inside its brain trying
to understand why it does what it does as

434
00:25:33,760 --> 00:25:37,597
it you know as as human neuroscientists
which we actually both have background

435
00:25:37,680 --> 00:25:40,477
in. um try try to understand try to

436
00:25:40,560 --> 00:25:44,637
understand the brain and I think as time
has gone on we've we've increasingly

437
00:25:44,720 --> 00:25:48,797
documented the you know bad behaviors
of the models when they emerge and are now

438
00:25:48,880 --> 00:25:53,277
working on trying to address them with
mechanistic interpretability. So I you

439
00:25:53,360 --> 00:25:56,877
know I think uh you know I I've always
been concerned about these these risks.

440
00:25:56,960 --> 00:26:00,797
I've talked to Demis many times. I
think he has also been um concerned about

441
00:26:00,880 --> 00:26:05,677
these risks. I think I have definitely
been and I I I would guess Demis as well

442
00:26:05,760 --> 00:26:10,157
although I'll let him speak for himself
skeptical of of doomerism which is you

443
00:26:10,240 --> 00:26:12,237
know we're doomed there's nothing we can

444
00:26:12,320 --> 00:26:14,957
do or this is the most likely outcome. I

445
00:26:15,040 --> 00:26:17,517
think this is a risk this is a risk that

446
00:26:17,600 --> 00:26:19,997
if we work all work together we can

447
00:26:20,080 --> 00:26:22,797
address we can learn through science to

448
00:26:22,880 --> 00:26:25,837
properly you know control and and direct

449
00:26:25,920 --> 00:26:27,837
these creations that we're building. But

450
00:26:27,920 --> 00:26:31,677
if we build them poorly, if we go, you

451
00:26:31,760 --> 00:26:36,397
know, if if if we're all racing and we
go so fast that there's no guard rails,

452
00:26:36,480 --> 00:26:38,237
then I think there is risk
of something going wrong.

453
00:26:38,320 --> 00:26:40,877
>> So, I'm going to give you a chance
to answer that in the context of of a

454
00:26:40,960 --> 00:26:42,637
slightly broader question, which is over

455
00:26:42,720 --> 00:26:44,797
the past year, have you grown more

456
00:26:44,880 --> 00:26:48,477
confident of the upside potential of the

457
00:26:48,560 --> 00:26:52,077
technology, science, all of the areas
that you have talked about a lot, or are

458
00:26:52,160 --> 00:26:56,477
you more worried about the risks that
we've been discussing? Look as I've been

459
00:26:56,560 --> 00:26:58,557
working on this for 20 plus years. So we

460
00:26:58,640 --> 00:27:00,877
we already knew the reason I've spent my

461
00:27:00,960 --> 00:27:05,677
whole career on AI is is the upsides
of solving basically the ultimate tool for

462
00:27:05,760 --> 00:27:09,117
science and understanding the universe
around us. I've I've sort of been

463
00:27:09,200 --> 00:27:13,277
obsessed with that since a kid and and
and building AI is the you know should

464
00:27:13,360 --> 00:27:16,317
be the ultimate tool for that if we
do it in the right way. The risks also

465
00:27:16,400 --> 00:27:19,597
we've been thinking about since the start
at least the start of deep mine 15

466
00:27:19,680 --> 00:27:23,997
years ago and um we kind of sort of foraw
that if you got the upsides it's a

467
00:27:24,080 --> 00:27:28,237
dual purpose technology so it could be
repurposed by say bad actors for harmful

468
00:27:28,320 --> 00:27:31,437
ends so we've needed to think about
that all the way through but I'm a big

469
00:27:31,520 --> 00:27:34,157
believer in human ingenuity um but the

470
00:27:34,240 --> 00:27:37,117
question is having the time and the

471
00:27:37,200 --> 00:27:41,597
focus and all the best minds
collaborating on it to solve these

472
00:27:41,680 --> 00:27:45,517
problems. I'm sure if we had that we
would solve the technical risk problem.

473
00:27:45,600 --> 00:27:48,957
It may be we don't have that and then
that will introduce risk because we'll

474
00:27:49,040 --> 00:27:52,477
be sort of it'll be fragmented. There'll
be different projects and people be

475
00:27:52,560 --> 00:27:55,917
racing each other. Then it's much harder
to make sure you know these systems that

476
00:27:56,000 --> 00:28:00,957
we produce will be technically safe. But I
I feel like that's a very tractable uh

477
00:28:01,040 --> 00:28:04,797
problem. >> If you if you
have the time I want to

478
00:28:04,880 --> 00:28:07,997
make sure there's one question gentlemen
keep it very short because we've got

479
00:28:08,080 --> 00:28:11,757
literally two minutes. Thanks for Hello.

480
00:28:11,840 --> 00:28:14,477
>> Yeah. No speak. >> Thanks very much.
I'm Philip, co-founder

481
00:28:14,560 --> 00:28:18,237
of Star Cloud Building Data Centers
in Space. Um, I wanted to ask a very

482
00:28:18,320 --> 00:28:19,997
slightly philosophical core question.

483
00:28:20,080 --> 00:28:23,357
The sort of strongest argument
for doomerism to me is the Fermy paradox,

484
00:28:23,440 --> 00:28:26,477
the idea that we don't see intelligent
life in our galaxy. I was wondering if

485
00:28:26,560 --> 00:28:28,237
you guys have any thoughts.
>> Yeah, I've thought a lot about that.

486
00:28:28,320 --> 00:28:29,837
That can't be the reason because we we

487
00:28:29,920 --> 00:28:32,957
we should see all the AIs that have So,

488
00:28:33,040 --> 00:28:36,957
just for everyone know the idea is well,
it's sort of unclear why that would

489
00:28:37,040 --> 00:28:40,157
happen, right? So if if the reason
there's a Fmy paradox there are no

490
00:28:40,240 --> 00:28:44,077
aliens because they get taken out by
their own technology we should be seeing

491
00:28:44,160 --> 00:28:48,237
paper clips coming towards us from some
part of the galaxy and apparently we

492
00:28:48,320 --> 00:28:50,397
don't we don't see any structures Dyson

493
00:28:50,480 --> 00:28:55,277
sphere is nothing whether they're AI
or natur or sort of biological so to me um

494
00:28:55,360 --> 00:28:57,677
there has to be a different answer
to fmy patterns I have my own theories

495
00:28:57,760 --> 00:28:59,117
about that but it's out of scope for the

496
00:28:59,200 --> 00:29:01,677
next minute but um you know I I just

497
00:29:01,760 --> 00:29:04,877
feel like uh that that I my prediction

498
00:29:04,960 --> 00:29:09,037
my feeling is that we're past the great
filter it probably multisellular life if

499
00:29:09,120 --> 00:29:13,837
I would have to guess was incredibly
hard for for biology to evolve that. Um

500
00:29:13,920 --> 00:29:16,957
so we're on you know there isn't
a comfort of like what's going to happen

501
00:29:17,040 --> 00:29:20,397
next. I think it's for us to write
as humanity what's going to happen next.

502
00:29:20,480 --> 00:29:23,197
>> This this could be a great discussion
but is out of scope for the next 36

503
00:29:23,280 --> 00:29:27,997
seconds. But what isn't 15 seconds each
what when we meet again I hope next year

504
00:29:28,080 --> 00:29:29,917
uh the three of us which I would love uh

505
00:29:30,000 --> 00:29:33,917
what will have changed by then? I well I

506
00:29:34,000 --> 00:29:36,317
think the biggest thing to watch is this

507
00:29:36,400 --> 00:29:40,157
issue of AI systems building AI systems

508
00:29:40,240 --> 00:29:42,157
how that goes whe that whether that goes

509
00:29:42,240 --> 00:29:45,357
one way or another that that will

510
00:29:45,440 --> 00:29:47,677
determine you know whether it's a few

511
00:29:47,760 --> 00:29:50,157
more years until we get there or or if

512
00:29:50,240 --> 00:29:53,357
we have you know you know if if we have

513
00:29:53,440 --> 00:29:57,757
wonders and and a great emergency
in front of us that we have to face

514
00:29:57,840 --> 00:30:00,717
>> AI systems building >> I agree
on that so we're we're keeping

515
00:30:00,800 --> 00:30:04,877
close touch about that um but also I
think um outside of that I think there

516
00:30:04,960 --> 00:30:08,637
are other interesting uh uh uh ideas
being researched like world models

517
00:30:08,720 --> 00:30:11,757
continual learning these are the things
I think that will need to be cracked if

518
00:30:11,840 --> 00:30:15,597
self-improvement doesn't sort of deliver
the goods on its own then we'll need

519
00:30:15,680 --> 00:30:19,197
these other things to work and then I
think things like robotics may have its

520
00:30:19,280 --> 00:30:22,077
sort of breakout moment >>
but maybe on the basis of what you've

521
00:30:22,160 --> 00:30:24,877
just said we should all be hoping that it
does take you a little bit longer and

522
00:30:24,960 --> 00:30:27,437
indeed everybody else to give us >>
I would prefer that I think that would

523
00:30:27,520 --> 00:30:28,637
be better for the world

524
00:30:28,720 --> 00:30:34,760
>> but you guys could do something
about that thank you So it's very MUCH