WEBVTT

1
00:00:00.160 --> 00:00:04.799
Some great guests here. I think
you guys know me and opposite Prime Minister,

2
00:00:05.120 --> 00:00:08.039
but if you guys could introduce yourselves
and say a bit about you and

3
00:00:08.480 --> 00:00:11.839
you know you've done both done amazing
things, so please let me shy.

4
00:00:12.119 --> 00:00:15.720
Sure thing. So I'm Greg Brockman. I'm the president and co founder of

5
00:00:15.720 --> 00:00:20.039
Open Aie. But we started Opening
Eye out of my apartment in twenty fifteen,

6
00:00:20.120 --> 00:00:23.679
twenty sixteen. So yeah, it's
been a really exciting time to just,

7
00:00:24.320 --> 00:00:26.960
you know, be able to help
not just move this field forward,

8
00:00:27.000 --> 00:00:29.839
but try to help steer it in
a positive direction. I think that's really

9
00:00:29.839 --> 00:00:32.960
foundational to how I think about AI. What got me excited about this field

10
00:00:33.280 --> 00:00:36.920
is the fact it's going to be
so transformative and I think it can benefit

11
00:00:36.960 --> 00:00:40.039
everyone in huge ways, but we
also really need to mitigate the downsides.

12
00:00:40.119 --> 00:00:43.759
And so I think that's something that
really resonated with me about the earlier conversation.

13
00:00:43.840 --> 00:00:47.399
I hope we'll talk about a lot
upcoming minutes. And since Greg was

14
00:00:47.439 --> 00:00:50.560
too modest to say so, I'm
going to add something to your introduction of

15
00:00:50.679 --> 00:00:55.600
yourself there, which is just a
few years ago when Greg and his colleagues

16
00:00:55.759 --> 00:00:59.000
were saying, yo, in a
few years we can act. We believe

17
00:00:59.039 --> 00:01:04.560
we can actually build AI that can
master human language and common knowledge. A

18
00:01:04.680 --> 00:01:08.920
lot of my colleagues and AI were
like, that's crazy science fiction. It's

19
00:01:08.920 --> 00:01:14.400
going to take decades, but you
did it. We literally had an intern

20
00:01:14.480 --> 00:01:19.680
in twenty six teams our very first
summer who we had this conversation about how

21
00:01:19.719 --> 00:01:22.599
to solve the touring test, and
we had all these ideas and whatever,

22
00:01:22.640 --> 00:01:23.400
and we were like, all right, three years, we'll solve the touring

23
00:01:23.439 --> 00:01:26.640
test. And that intern told me
I'm out, like I'm not joining full

24
00:01:26.640 --> 00:01:30.280
time, like you guys were crazy. And three years later we had GPT

25
00:01:30.359 --> 00:01:32.519
two, so we didn't quite solve
the Turing desk. So he was right,

26
00:01:32.799 --> 00:01:34.719
But I think that we had far
more progress than I think most people

27
00:01:34.760 --> 00:01:41.000
thought possible. So I'm Max tech
Mark, professor doing AI research at MIT,

28
00:01:41.599 --> 00:01:46.120
and I think it was fought on
what we heard earlier here from you,

29
00:01:46.280 --> 00:01:51.840
that artificial intelligence will be both a
blessing and a curse. I've very

30
00:01:51.879 --> 00:01:55.120
for a long time and very excited
about the blessing part, how we can

31
00:01:55.280 --> 00:01:57.719
use it to solve all the problems
that we haven't been smart enough to solve

32
00:01:57.760 --> 00:02:02.079
before. And go for beyond and
help life flourished for billions of years on

33
00:02:02.120 --> 00:02:07.319
Earth and beyond. And I also
I'm very concerned about the curse part.

34
00:02:07.680 --> 00:02:12.400
So nine years ago I founded this
nonprofit called the Future Life Institute. Together

35
00:02:12.439 --> 00:02:17.000
with Elon we launched the first ever
attempt to the mainstream AI safety by launching

36
00:02:17.039 --> 00:02:22.680
a worldwide grants program and educate and
do conferences and so on that you Greg

37
00:02:22.879 --> 00:02:28.520
participated also to make sure we can
steer this every more powerful technology away from

38
00:02:28.560 --> 00:02:32.479
the curse towards the blessing direction.
Well, can I say that I want

39
00:02:32.479 --> 00:02:37.120
to plug this book? He wrote
this book and I read it. It's

40
00:02:37.120 --> 00:02:39.319
a good book. It's a very
good book. Was it chat guptwo or

41
00:02:39.599 --> 00:02:44.080
I have an alibi because not even
Greg was fast enough to have it written

42
00:02:44.280 --> 00:02:49.199
by the time chat gupt ain't good
enough yet to do this. But from

43
00:02:49.199 --> 00:02:52.400
what I hear you guys saying,
it's good. Yeah, give it a

44
00:02:52.400 --> 00:02:54.800
minute. It was thirty seconds.
Well you see, Well let's talk about

45
00:02:54.800 --> 00:02:58.159
that, because you know, I
think that the kind of things that you

46
00:02:58.360 --> 00:03:01.280
ring here are I think a monumental. You know, when I see the

47
00:03:01.319 --> 00:03:06.599
talent on this stage, excluding myself, I'm reminded of what JFK. John

48
00:03:06.680 --> 00:03:09.280
Kennedy said, you know, when
he brought his Gifted team into the White

49
00:03:09.280 --> 00:03:14.719
House. He said, this is
the most gifted team ever since that's sat

50
00:03:14.800 --> 00:03:19.120
here, ever since Thomas Jefferson dined
alone. So I think you're about to

51
00:03:19.120 --> 00:03:22.840
surpass that, and you are changing
history. And I'm sure you know when

52
00:03:22.840 --> 00:03:25.919
such power comes enormous responsibility. And
I think that's the crux of what we're

53
00:03:25.919 --> 00:03:31.560
talking about here is how do we
inject a measure of responsibility and ethics into

54
00:03:31.639 --> 00:03:37.719
this, into this exponentially changing development. How do you do that? That's

55
00:03:37.759 --> 00:03:40.400
really what I'd like to know from
you. I think that this should be

56
00:03:40.400 --> 00:03:45.840
a very interesting discussion. And you
know, I think I know almost everyone

57
00:03:45.919 --> 00:03:50.280
in AI, and I think I
really can't think of anyone who's got bad

58
00:03:50.280 --> 00:03:53.039
motivations and AI. I think everyone
I know an AI has good motivations.

59
00:03:53.120 --> 00:03:57.960
They want to create technology that helps
the world. But one you can also

60
00:03:58.000 --> 00:04:00.639
say, if you think of someone
like Einstein, who was obviously a very

61
00:04:00.879 --> 00:04:05.159
sort of peace loving person, and
you know, he didn't think that there

62
00:04:05.159 --> 00:04:09.680
would be a nuclear bomb created as
a result of his discoveries in physics.

63
00:04:10.039 --> 00:04:13.120
He wasn't thinking, let me,
let's make super weapons. He was thinking,

64
00:04:13.199 --> 00:04:15.680
let's try to understand that the truth
of the universe. So it's just

65
00:04:15.759 --> 00:04:19.879
important too. I always want to
just preface, you know, any criticism

66
00:04:20.040 --> 00:04:25.360
of AI development with it not being
a criticism of the people doing it,

67
00:04:25.480 --> 00:04:30.079
but just fair in mind that just
as Einstein didn't expect his work in physics

68
00:04:30.079 --> 00:04:33.120
to lead to nuclear weapons, we
need to be cautious that even with good

69
00:04:33.199 --> 00:04:38.720
intentions, the best of intentions,
that we could create something bad, that

70
00:04:38.319 --> 00:04:41.879
that that's one of the possible outcomes, and we want to try to take

71
00:04:41.920 --> 00:04:46.959
whatever actions we can to ensure that
the future of humanity is good and that

72
00:04:46.360 --> 00:04:51.240
is much more blessing than That's definitely
why I work on it. I think

73
00:04:51.240 --> 00:04:55.680
that not for evil it turns out, But I think that balance is really

74
00:04:55.720 --> 00:04:59.439
key, and I think that one
of the things that is important is to

75
00:04:59.519 --> 00:05:03.199
really calibrate the choices with respect to
the technology, right because I think this

76
00:05:03.279 --> 00:05:08.399
technology it has this way of being
very surprising, the way that people pictured

77
00:05:08.600 --> 00:05:12.360
like you read asthmav you read kind
of all the science fiction you read Ellie

78
00:05:12.360 --> 00:05:14.879
as you're a Kowski, and kind
of all this thinking people are doing about

79
00:05:15.000 --> 00:05:17.439
very smart machines that are smarter than
humans, and somehow none of it really

80
00:05:17.480 --> 00:05:21.560
predicted GPT three, right. Somehow
the technology that we're building, it's just

81
00:05:21.639 --> 00:05:26.079
surprising. It has very different properties. And I think that even from you

82
00:05:26.120 --> 00:05:29.040
think about like chess, people used
to think chess would be we solve that,

83
00:05:29.079 --> 00:05:30.319
we solve human intelligence, and it
turns out that's the first thing to

84
00:05:30.360 --> 00:05:33.000
go. And so I think that
what we're really seeing is that we need

85
00:05:33.040 --> 00:05:36.399
to be calibrated and see not just
what the current systems are doing. And

86
00:05:36.399 --> 00:05:40.240
this is one of the reasons we
deployed chat GPT is it's just really important

87
00:05:40.279 --> 00:05:42.319
to not get in our heads and
think, oh, this is how it's

88
00:05:42.319 --> 00:05:44.439
all going to play out, this
is how people are going to misuse it.

89
00:05:44.439 --> 00:05:46.720
But you need to actually see that. You need to actually have people

90
00:05:46.000 --> 00:05:49.399
spending a lot of time thinking about
where the slots in, where it works,

91
00:05:49.399 --> 00:05:53.199
where it doesn't, but also to
really try to peer ahead. And

92
00:05:53.240 --> 00:05:55.839
so we've been putting a lot of
effort into capability prediction. One of the

93
00:05:55.879 --> 00:05:59.240
things that I think was kind of
undersung in our GPT four release is that

94
00:05:59.240 --> 00:06:02.519
we actually had very good predictions of
exactly how good GPD four would be before

95
00:06:02.560 --> 00:06:05.639
we even trained it, and that's
something we can never do with previous models.

96
00:06:05.920 --> 00:06:10.000
And I think that this and doing
that it's a sidal scale, Like

97
00:06:10.040 --> 00:06:12.759
I think that's one answer for how
do we actually get this right? Because

98
00:06:12.800 --> 00:06:15.240
if you can see around the corner, then you have an answer and an

99
00:06:15.279 --> 00:06:17.120
AI. It's always been the opposite. We've always been taken by surprise.

100
00:06:17.480 --> 00:06:21.199
Well, I want to ask you
because there are sort of layers of questions

101
00:06:21.199 --> 00:06:27.759
here. The Max's book takes you
to the existential question of whether you know

102
00:06:28.040 --> 00:06:33.959
you project basically machine intelligence or human
intelligence into the cosmos, human intelligence turned

103
00:06:33.959 --> 00:06:39.600
into machine intelligence into the cosmos and
so on. That's a big philosophical question.

104
00:06:39.680 --> 00:06:42.319
I'd like to think we have about
six years for that, okay,

105
00:06:42.439 --> 00:06:45.639
But the other questions that I'm dealing
with as a leader of a country that

106
00:06:45.800 --> 00:06:50.519
is blessed with a lot of AI
talent rated about number two and talent relative

107
00:06:50.560 --> 00:06:55.319
to population size and AI. But
I wanted to be among the top three

108
00:06:55.439 --> 00:06:59.560
because it's important to advance the country. But I think also for it raises

109
00:06:59.680 --> 00:07:03.199
questions that we raised before. But
here's another one. Okay, as chat

110
00:07:03.199 --> 00:07:08.680
GUP four or chat. GPT eight
comes into being, and you have Hollywood

111
00:07:08.720 --> 00:07:13.720
strikes already about screenwriters and directors and
actors. Okay, what do you think

112
00:07:13.800 --> 00:07:16.240
will happen to the job market?
I mean, we had this conversation,

113
00:07:16.360 --> 00:07:19.199
what do you think will happen?
I mean, you know, you can

114
00:07:19.199 --> 00:07:24.000
tell me, you can comfort me
with all previous revolutions that there were more

115
00:07:24.079 --> 00:07:27.759
jobs, many more jobs created than
lost. But you really believe that now?

116
00:07:27.959 --> 00:07:29.920
Yeah, I think it's I think
it's far from obvious. I think

117
00:07:29.920 --> 00:07:32.120
it is a correct question. And
I also think that it's important to kind

118
00:07:32.160 --> 00:07:35.879
of look at the whole picture of
for sure, their jobs are going to

119
00:07:35.920 --> 00:07:39.040
go away, what's going to be
the balance of like, you know,

120
00:07:39.120 --> 00:07:41.759
marginal job creation? And you can
tell the story of oh, it's going

121
00:07:41.759 --> 00:07:44.879
to be very comforting you look at
every previous technology. I do think AI

122
00:07:45.000 --> 00:07:47.720
is not like every previous technology,
right that this is usually we kind of

123
00:07:47.800 --> 00:07:54.600
are seeing automation of mechanical parts of
creation, and here it's almost this creative

124
00:07:54.639 --> 00:07:57.680
aspect of creation. But I think
that we want to even go one step

125
00:07:57.720 --> 00:08:01.000
further from where we've been, right
and not just okay, like you know,

126
00:08:01.040 --> 00:08:03.800
I could have been an artist and
now an AI is a competing artist,

127
00:08:03.040 --> 00:08:05.720
but you want AI that can help
us come up with ideas and solve

128
00:08:05.759 --> 00:08:09.920
problems that we just couldn't before.
And so what does that world look like?

129
00:08:09.959 --> 00:08:11.399
And you know, if your post
scarcity world but people don't have to

130
00:08:11.439 --> 00:08:16.120
work to survive, you know what
set of people? Because people I identify

131
00:08:16.160 --> 00:08:18.160
themselves, you know, your whole
identity is around your work. You identify

132
00:08:18.240 --> 00:08:22.959
yourself by your job, yes,
exactly, And so what will happen then?

133
00:08:22.000 --> 00:08:24.680
And what does that look like?
And so I don't come with an

134
00:08:24.680 --> 00:08:28.160
answer to this question of exactly what
the balance will be like. I think

135
00:08:28.160 --> 00:08:33.799
there will be fundamental changes and how
we even relate to work and to meaning

136
00:08:35.240 --> 00:08:37.879
that will happen and may result in
a landscape that just looks very different.

137
00:08:37.879 --> 00:08:41.480
That the answer to that question is, you know, if you're a hunter

138
00:08:41.519 --> 00:08:43.000
gatherer and someone's like, you know, what's going to happen to our hunter

139
00:08:43.039 --> 00:08:46.559
gatherer job post agriculture of revolution,
it's just a weird question, right,

140
00:08:46.559 --> 00:08:50.480
It's like the framework changes on you. And so that's that's not a cop

141
00:08:50.480 --> 00:08:52.080
out, right. I think that
it might be the case that actually many

142
00:08:52.120 --> 00:08:54.639
more jobs go away than get created, and there's a lot of chaos and

143
00:08:54.679 --> 00:08:58.039
turmoil, but I think we got
to be ahead of it. And the

144
00:08:58.080 --> 00:09:00.679
more that we can predict it,
the more that we can greet answers.

145
00:09:00.879 --> 00:09:03.240
And we have teams that try,
right, But I think that we're just

146
00:09:03.279 --> 00:09:05.919
getting started and we need help.
Well, it'll change our economic models.

147
00:09:05.960 --> 00:09:11.759
Okay. I'm a clear free market
guy. I liberated and I helped liberate

148
00:09:11.759 --> 00:09:15.279
the Israeli economy to make it a
sort of a technological powerhouse. But that's

149
00:09:15.279 --> 00:09:18.720
because of free markets, and free
markets means open competition. And it's fine,

150
00:09:18.960 --> 00:09:22.799
but you stop at the monopolies and
monopolies edge. Okay, well you

151
00:09:22.879 --> 00:09:24.559
do that. I had a conversation
with one of your colleagues, Peter Tail,

152
00:09:24.639 --> 00:09:28.720
and he said to me, oh, it's all scale advantages, it's

153
00:09:28.720 --> 00:09:31.399
all monopolies. I said, well, yeah, I believe that, but

154
00:09:31.639 --> 00:09:33.519
I think we have to stop it
at a certain point because we don't want

155
00:09:33.519 --> 00:09:39.279
it to press competition. Okay.
Well, Ai is producing you know these

156
00:09:39.679 --> 00:09:43.919
this wall and you have these trillion
dollar companies that are produced to one overnight,

157
00:09:45.120 --> 00:09:48.960
okay, and they concentrate enormous wealth
and power with smaller and smaller number

158
00:09:48.960 --> 00:09:54.080
of people. Okay, and the
question is what do you do about that

159
00:09:54.159 --> 00:09:58.480
monopoly power? What do you do
about competition? And if you are going

160
00:09:58.559 --> 00:10:01.120
to if you are going to cannibalize
a lot more jobs than you create,

161
00:10:01.440 --> 00:10:07.360
then we have to change the structure
of our economic policies and our political policies

162
00:10:07.399 --> 00:10:11.000
to take care of the people who
are not going to find jobs. We're

163
00:10:11.000 --> 00:10:13.919
not going to contribute added value to
the economy. We have to make sure

164
00:10:13.960 --> 00:10:16.720
that they have a living, a
decent one, and they have all the

165
00:10:16.759 --> 00:10:20.399
services. Will probably have the money
to do that, but it requires a

166
00:10:20.519 --> 00:10:24.159
challenging model, certainly for a free
market guy, a free market disciple like

167
00:10:24.320 --> 00:10:28.320
me. And that's coming very fast. It's not going to come slowly,

168
00:10:28.360 --> 00:10:31.080
and we have time to adjust.
I like how you challenge the question what's

169
00:10:31.120 --> 00:10:35.080
going to happen? I like to
continue on that thing because I think if

170
00:10:35.320 --> 00:10:39.000
we asked this very passive kind of
question, what's going to happen to us,

171
00:10:39.320 --> 00:10:43.039
let's just sit here and eat popcorn
and wait for the future to happen,

172
00:10:43.360 --> 00:10:46.879
then we're steading straight for the curse. I think this is a sort

173
00:10:46.879 --> 00:10:50.720
of very passive approach which I think
is just gonna you know, We've done

174
00:10:50.759 --> 00:10:54.039
this experiment before when a new species
showed up on the planet that was smarter

175
00:10:54.120 --> 00:10:58.799
than all the other ones, and
what happened was then the underthal automentics think,

176
00:10:58.840 --> 00:11:01.000
if we just very passively go into
this, I think it's very likely

177
00:11:01.080 --> 00:11:05.360
humans will go extinct. Also,
I think that's and I think that's the

178
00:11:05.399 --> 00:11:09.799
wrong attitude. Even in economics in
a country, you don't just sit back

179
00:11:09.840 --> 00:11:15.480
and wait to see what happens and
get child labor and the super monopoly of

180
00:11:15.519 --> 00:11:18.559
doom. No, you ask yourself
what do you want to happen, and

181
00:11:18.600 --> 00:11:22.120
then you put in place institutions,
you ban child labor, you put an

182
00:11:22.120 --> 00:11:24.919
anti trust. And in this case, I think the analogous question is how

183
00:11:24.919 --> 00:11:30.080
do we make sure that with this
ever more powerful tech, it's we who

184
00:11:30.120 --> 00:11:33.440
control the AI rather than the other
way around. Right, So I was

185
00:11:33.559 --> 00:11:37.200
honored that you had this book.
If this book still hasn't put you to

186
00:11:37.240 --> 00:11:39.919
sleep, because this is a technical
nerd paper I wrote with this one's written

187
00:11:39.919 --> 00:11:45.159
with technological approach to how we can
control machines that are even smarter than us

188
00:11:45.159 --> 00:11:48.799
by actually having other AIS formally prove
that the I is safe. And the

189
00:11:48.879 --> 00:11:52.360
reason why I'm so excited about the
proving part is, you know, the

190
00:11:52.399 --> 00:11:56.799
only way you can trust something much
smarter than you to still do what you

191
00:11:56.840 --> 00:12:00.879
want is if you can prove it, Because no one, how clever it

192
00:12:00.960 --> 00:12:03.679
is, it can never do what's
provably impossible, and it turns out it's

193
00:12:03.679 --> 00:12:07.919
super It's much harder to come up
with proofs about math or about what a

194
00:12:07.960 --> 00:12:11.320
program is going to do than it
is to check that the proof is correct.

195
00:12:11.600 --> 00:12:16.480
You can write a three hundred line
computer program which will check any proof

196
00:12:16.559 --> 00:12:20.519
claiming anything about anything else. We
can understand that if we force AI to

197
00:12:20.639 --> 00:12:26.200
prove things to ourselves, to us
humans that's going to meet our specifications,

198
00:12:26.519 --> 00:12:30.799
then we actually have a reason a
way we can control things that are smarter

199
00:12:30.879 --> 00:12:33.519
than us. How do you know? How do you know that the proof

200
00:12:33.720 --> 00:12:37.519
is not disputable? That's the beauty
of mathematics. You can actually check even

201
00:12:37.519 --> 00:12:41.840
if it's a brilliant proof, rather
than I couldn't come up with myself.

202
00:12:41.879 --> 00:12:43.759
I can just see I just plug
this into this and this, Yeah,

203
00:12:45.000 --> 00:12:48.679
you know it checks out. And
there's a big science in doing this with

204
00:12:48.840 --> 00:12:54.840
proving things about programs for cybersecurity because
companies are so fed up with losing millions

205
00:12:54.840 --> 00:12:58.320
of dollars by begetting hacked all the
time because yet another bug was found and

206
00:13:00.120 --> 00:13:03.000
or a shell or whatever. That
there's been years of investment. Greg and

207
00:13:03.039 --> 00:13:07.840
I had a fun conversation actually just
before we went on stage here about how

208
00:13:07.120 --> 00:13:11.480
this field of trying to prove that
things are safe is still in the stone

209
00:13:11.519 --> 00:13:16.320
age and there's pre large language model
phase where it's most the humans trying to

210
00:13:16.320 --> 00:13:18.919
make the proofs. I think it's
ripe for a revolution where the kind of

211
00:13:18.919 --> 00:13:24.840
technology that Open Eyes building can totally
turbocharge our ability to prove things about the

212
00:13:24.879 --> 00:13:28.240
code. And I think there's a
bigger picture too, because in my mind,

213
00:13:28.559 --> 00:13:31.600
if you think about cybersecurity, it's
always a cat and mouse game.

214
00:13:31.679 --> 00:13:35.000
You have attackers, you have defenders. Each one's leveling up defeating the other.

215
00:13:35.440 --> 00:13:37.720
Is there ever going to be a
time when the defenders totally win?

216
00:13:37.879 --> 00:13:41.200
Is there going to be a time
when the attackers totally win? And I

217
00:13:41.240 --> 00:13:43.399
think with AI that balance can change. And I think that formal verifications an

218
00:13:43.440 --> 00:13:48.200
example where maybe you can have defenders
just win once and for all. And

219
00:13:48.240 --> 00:13:50.960
I think there's a lot of asterisks
there. I think it's definitely a maybe,

220
00:13:50.039 --> 00:13:54.200
But I think that at least at
a technical level, formal verification has

221
00:13:54.240 --> 00:13:58.759
been this idea of if you could
mathematically prove these systems are safe and good

222
00:13:58.799 --> 00:14:01.440
and that you don't have bugs,
and you're linings, colonel, then all

223
00:14:01.519 --> 00:14:03.480
these bugs we've seen we were talking
about heart bleed and shell shock and all

224
00:14:03.519 --> 00:14:07.120
these these bugs from the past ten
years, those would be solved, those

225
00:14:07.120 --> 00:14:09.519
would go away. You would know
there'd be no future one of those.

226
00:14:09.759 --> 00:14:11.799
Now you need to be very careful
because you need to make sure that your

227
00:14:11.879 --> 00:14:16.240
undergoing machine model is like actually correct
in a bunch of details like that.

228
00:14:16.519 --> 00:14:20.759
But it's just interesting to think that
these fundamental things we believe about how security

229
00:14:20.799 --> 00:14:22.639
works might just be up it.
And we've never had an opportunity for this.

230
00:14:22.799 --> 00:14:26.320
That will give us a lot of
free time exactly. And it's worth

231
00:14:26.360 --> 00:14:28.039
adding for the nerds listening to this
to like the code. You know,

232
00:14:28.399 --> 00:14:33.799
normally, debugging, all that ever
does is is all it can never accomplish

233
00:14:35.200 --> 00:14:37.399
is proved that there is something wrong. If you don't find any bugs,

234
00:14:37.600 --> 00:14:43.240
is there one more? What when
you prove it? When it's what's called

235
00:14:43.559 --> 00:14:48.759
formal verification, As Greg said,
that means guaranteed you'll seal the deal.

236
00:14:48.960 --> 00:14:52.039
Seal the deal, all right.
That assumes, by the way, that

237
00:14:52.080 --> 00:14:54.919
mathematics is, you know, that's
it. It's the law of the universe.

238
00:14:56.200 --> 00:15:00.000
It's not going to change. You're
not going to get alien mathematics.

239
00:15:00.240 --> 00:15:01.759
You don't need to assume that.
Actually, well you're sort of getting the

240
00:15:01.840 --> 00:15:05.159
girls incompleteness theorem and things like that. That has comes in a big way

241
00:15:05.200 --> 00:15:09.960
when you're dealing with infinities of various
kinds. All the computers, even Gregg's

242
00:15:09.039 --> 00:15:11.919
very biggest servers, it's all finite. I would say, you still got

243
00:15:13.000 --> 00:15:16.039
to watch out for the alien mathematics. But you can totally formalize what it

244
00:15:16.080 --> 00:15:20.360
means that it's impossible to hack into
his laptop, and if you have a

245
00:15:20.440 --> 00:15:24.120
rigorous proof, I think that's heals
the deal. So I think that beyond

246
00:15:24.200 --> 00:15:28.600
these, you know, these enormous
questions that you raise, I think from

247
00:15:28.639 --> 00:15:33.600
me, the immediate problem is to
fashion a national policy, which I tend

248
00:15:33.639 --> 00:15:37.039
to announce in a few months,
and you guys are simulated me. I

249
00:15:37.080 --> 00:15:39.840
have to say this visit, but
you sort of think, Okay, what

250
00:15:41.000 --> 00:15:43.919
is it that you what is it
that you were trying to achieve. You're

251
00:15:43.960 --> 00:15:50.200
trying to better all the systems transportation, health, education, agriculture, production,

252
00:15:50.320 --> 00:15:56.159
everything for government, for the private
sector, for philanthropy, for academia

253
00:15:56.159 --> 00:15:58.639
and so on. And what is
the role of the government in this?

254
00:15:58.120 --> 00:16:02.080
And you know, various kind trees
are seeking solutions to this. What do

255
00:16:02.080 --> 00:16:04.320
you do? What kind of instruments
do you build? I think the larger

256
00:16:04.440 --> 00:16:11.679
question is you still have the monopolization
of AI power in small concentrations because of

257
00:16:11.679 --> 00:16:15.440
what you said before, Elon,
you said, basically, it's still big.

258
00:16:15.679 --> 00:16:18.600
You know, you need the big
data, you need the big you

259
00:16:18.600 --> 00:16:22.039
know, the big computational power,
you need the talent for creating the algorithms,

260
00:16:22.240 --> 00:16:26.399
and that's concentrated in a few countries. We have another test, not

261
00:16:26.440 --> 00:16:30.159
only to make sure that we take
care of ourselves, but how we have

262
00:16:30.279 --> 00:16:34.840
these benefits percolate to you know,
the rest the scientific and technological have nots.

263
00:16:36.080 --> 00:16:38.679
And I think that's that's another real
issue. But so far, you

264
00:16:38.720 --> 00:16:42.080
know the advances that we've made,
at least in Israel, I don't know

265
00:16:42.159 --> 00:16:45.799
anything from disconquy. Remember that you
used to have something like that, And

266
00:16:47.120 --> 00:16:51.440
I don't know that a drug,
the camera pill that goes into your intestine

267
00:16:51.559 --> 00:16:55.960
or any other thing that you have, from ICQ to cherry tomatoes that benefits

268
00:16:55.960 --> 00:16:59.480
everybody. Drip irogation, okay,
in Ai, it seems to me that

269
00:16:59.519 --> 00:17:04.599
you're gonna have concentration of power that
will create a bigger and bigger distance between

270
00:17:04.880 --> 00:17:10.440
the haves and the have nots.
And that's another thing that causes tremendous instability

271
00:17:10.440 --> 00:17:12.519
in our world. And I don't
know if you have an idea how you

272
00:17:12.599 --> 00:17:15.359
overcome that. Well, first of
all, I want to say just on

273
00:17:15.400 --> 00:17:21.039
the talent density front, we have
a number of Israeli engineers, scientist managers

274
00:17:21.079 --> 00:17:23.519
and they are top notch, no
question. We have a manager who runs

275
00:17:23.559 --> 00:17:27.680
one of our core teams who is
ex Special Forces and there's nothing that phases

276
00:17:27.759 --> 00:17:32.119
him. So I think there's something
about the Israeli way that I think is

277
00:17:32.119 --> 00:17:34.839
actually very conducive to this field.
And I think that it is really important

278
00:17:36.039 --> 00:17:40.319
to counteract this rich, get richer
phenomenon, at if it comes at the

279
00:17:40.319 --> 00:17:42.759
expense of everyone else. Right,
I think that there's a the world we

280
00:17:42.759 --> 00:17:45.680
should shoot for is one where all
the boats are rising, and I think

281
00:17:45.680 --> 00:17:48.079
that there will be some sort of
inequities in the landscape, and I think

282
00:17:48.119 --> 00:17:52.000
that's something to address, and that's
something that I think that leaders such as

283
00:17:52.039 --> 00:17:53.680
yourself, I think need to be
very cognizant about. You know, I

284
00:17:53.680 --> 00:17:56.960
think that we've talked about UBI like
opening actually has run. I think the

285
00:17:56.960 --> 00:18:03.119
world's most comprehensive UBI. Something people
talk about universal basic income, the idea

286
00:18:03.119 --> 00:18:04.680
of just the government just distributes money, right, that's just what it does.

287
00:18:04.720 --> 00:18:08.880
If most people are not working,
producing, you know, dollars themselves,

288
00:18:10.039 --> 00:18:11.920
and there's a world of plenty.
Is that a mechanism that can work?

289
00:18:12.119 --> 00:18:15.720
Maybe it can, maybe it can't. There's a lot of assumptions that

290
00:18:15.759 --> 00:18:18.640
would go into a world where it
works. It's clearly I think also has

291
00:18:18.720 --> 00:18:21.839
a lot of downsides to it.
But I think that we need to have

292
00:18:21.920 --> 00:18:26.480
creative solutions because again, I think
that just like defense versus offense might fundamentally

293
00:18:26.599 --> 00:18:30.599
change economic production and what it takes
to actually build even a stable society I

294
00:18:30.640 --> 00:18:34.240
think is going to fundamentally change.
Well, Ellen, you said when I

295
00:18:34.240 --> 00:18:37.680
asked you this question one of our
nighttime conversations. I asked you, well,

296
00:18:37.920 --> 00:18:41.160
what are you gonna do when people
don't have jobs and we shell out

297
00:18:41.200 --> 00:18:45.559
money for them and you said.
I think you said, oh, what's

298
00:18:45.599 --> 00:18:48.680
bad and living in paradise? Right, I think you said that, Yeah,

299
00:18:48.799 --> 00:18:52.119
I mean it's I mean this is
a question of like, you know,

300
00:18:52.119 --> 00:18:55.119
the concept of heaven. Now,
in heaven, I don't think you

301
00:18:55.200 --> 00:18:59.519
need to work in heaven. Haven't
read that anywhere, So they're not laughing,

302
00:19:00.480 --> 00:19:04.119
you know if you want so,
you know, the very positive scenario

303
00:19:04.279 --> 00:19:08.799
of AI is actually in a lot
of ways the description of heaven in that

304
00:19:10.200 --> 00:19:12.640
really nobody would need to work.
I wouldn't even quote universal basic income.

305
00:19:12.680 --> 00:19:17.839
I'd say it's probably universal high income. I'm describing the best case scenario here,

306
00:19:17.920 --> 00:19:19.400
so I'm not saying this is definitely
what occurred. There's a range of

307
00:19:19.400 --> 00:19:25.799
scenarios from very negative to very positive. Very positive scenario is basically sounds like

308
00:19:25.839 --> 00:19:27.200
heaven. You can have whatever you
want, you don't need to work,

309
00:19:27.240 --> 00:19:30.960
you have no obligations. Any illness
you have can be cured. And when

310
00:19:32.000 --> 00:19:33.200
do we die? Well, that's
a good question. I think it'll be

311
00:19:33.240 --> 00:19:36.519
a choice. I think it probably
ends up being somewhat of a choice.

312
00:19:36.880 --> 00:19:40.200
So you want this world, well, I'm not saying I don't know.

313
00:19:40.640 --> 00:19:44.559
There's some there's a sort of a
philosophical debate is the concept of heaven as

314
00:19:44.559 --> 00:19:47.440
it is normally described. Actually,
what do you want? I mean,

315
00:19:47.480 --> 00:19:51.839
do you want to be in a
future or world where you know the in

316
00:19:51.880 --> 00:19:55.240
the AI sense that a I could
do everything better than you by far?

317
00:19:55.599 --> 00:19:59.720
That's that's because you know you have, like man, search for meaning questions

318
00:20:00.200 --> 00:20:03.200
thing good work and what meaning if
you if, like I said, if

319
00:20:03.400 --> 00:20:07.200
I can do everything better than you
do, we live a life of hedonism.

320
00:20:07.599 --> 00:20:11.279
I don't know, it's hard to
say, but that is where things

321
00:20:11.319 --> 00:20:15.200
ahead. In a positive scenario is
that there would be no scarcity of goods

322
00:20:15.240 --> 00:20:18.519
and services. The robots and computers
can make as much as much of whatever

323
00:20:18.559 --> 00:20:22.680
you want, for as many people
as you want, with really no limitation.

324
00:20:22.920 --> 00:20:26.799
Any scarcity we have would be artificial, meaning like if you define that

325
00:20:26.039 --> 00:20:32.640
in particular artwork done by a human
is unique and special, then it's artificial

326
00:20:32.680 --> 00:20:37.319
scarcity. But there will be no
actual scarcity unless we define it arbitrarily,

327
00:20:37.559 --> 00:20:41.359
define something to be scarce. So
I think for a lot of people that

328
00:20:41.519 --> 00:20:45.119
is a good, good future.
I might be personally a bit frustrated because

329
00:20:45.119 --> 00:20:47.119
I'm like, well, what am
I supposed to do? Now? You'll

330
00:20:47.119 --> 00:20:51.079
think of think of something. I
think that this is a very interesting set

331
00:20:51.119 --> 00:20:53.720
of challenges. What would we do
if people have no jobs? But I

332
00:20:53.759 --> 00:20:57.960
think it's important to remember that we're
not automatically going to end up in this

333
00:20:59.079 --> 00:21:02.240
situation in the way we're thinking about
it now. We could also, I

334
00:21:02.279 --> 00:21:06.119
think, very likely end up in
a situation where people have no jobs because

335
00:21:06.119 --> 00:21:08.279
there are no people anymore, where
we literally go extinct. You know,

336
00:21:08.319 --> 00:21:12.039
if we had a statement cut that
came out in May, right that I

337
00:21:12.039 --> 00:21:17.400
think Sam Olmen and you both signed
right saying and ours left under Lyon came

338
00:21:17.400 --> 00:21:19.279
out with it a couple of days
ago saying we have to take seriously human

339
00:21:19.319 --> 00:21:23.440
extinction. And if this sounds like
very crazy science fiction, you know we

340
00:21:23.440 --> 00:21:27.079
were joking. But just a few
years ago all your success sounds like science

341
00:21:27.079 --> 00:21:30.519
fiction. Two and you said that
we don't have a lot of time.

342
00:21:30.720 --> 00:21:34.920
These are things that could happen relatively
soon. And there's how it's very important

343
00:21:36.200 --> 00:21:38.920
to me, because I want the
good future, to think about how we

344
00:21:40.000 --> 00:21:42.400
steer away from those sort of scenarios. One way in which we could end

345
00:21:42.480 --> 00:21:47.960
up and this kind of bad future
is pretty obvious. The idea that we

346
00:21:48.000 --> 00:21:52.799
can lose control to something more intelligent
is not new at all. That Alan

347
00:21:52.839 --> 00:21:56.720
Turing himself said this in the fifties. That's kind of the natural thing to

348
00:21:56.720 --> 00:21:59.920
expect. That's what always happened on
our planet before. When a news more

349
00:22:00.119 --> 00:22:03.119
speechies came in, it took control
over the older ones, and then you

350
00:22:03.160 --> 00:22:07.480
know, bye bye and Neanderthals by
by mammoths whatever. More concretely, I

351
00:22:07.519 --> 00:22:12.200
think, even though Hollywood is full
of scenarios where someone somehow loses control of

352
00:22:14.000 --> 00:22:18.440
AI and it takes over, there
are more sneaky ways in which we could

353
00:22:18.519 --> 00:22:21.880
end up in this sort of bad
state, even more willingly. So for

354
00:22:21.920 --> 00:22:25.000
example, right now, I'm a
big fan of the free market too,

355
00:22:25.240 --> 00:22:29.799
you know, But it's the fact
that if people are competing against other people,

356
00:22:30.079 --> 00:22:33.240
they will often seed control voluntarily at
the machines to get an edge on

357
00:22:33.279 --> 00:22:38.240
the other guys. So companies who
replace workers with machines out compete companies who

358
00:22:38.279 --> 00:22:44.440
don't. Armies who seed control to
machines out compete armies who don't. Once

359
00:22:44.480 --> 00:22:48.519
we have really powerful any of that, that's better at running companies and countries

360
00:22:48.519 --> 00:22:52.319
and humans, you know, than
countries that replaced their prime ministers by AI

361
00:22:52.440 --> 00:22:55.920
might outcompete others, and then we
end up in this future where all these

362
00:22:56.319 --> 00:22:59.319
things are happening, but it's not
our future anymore, and we're like,

363
00:22:59.359 --> 00:23:02.079
wait a minute, where did we
go along wrong? Here? It was

364
00:23:02.119 --> 00:23:04.279
supposed to be for us. But
that's the point, you know, forgiving

365
00:23:04.319 --> 00:23:08.039
for borrowing a page from your book, literally a page from your book,

366
00:23:08.240 --> 00:23:14.960
I'm confessing I'm a speciest Okay,
Okay, So I'm for the human species.

367
00:23:15.279 --> 00:23:18.559
Okay. We can talk about other
evolutions and you know, machine launched

368
00:23:18.839 --> 00:23:22.119
intelligence and so on, but I'm
you know, I'm for you know,

369
00:23:22.319 --> 00:23:26.079
in the Jewish tradition, there's a
saying, how do we translate this?

370
00:23:26.559 --> 00:23:30.359
The impoverished of your own city,
of your own town, proceed the impoverished

371
00:23:30.400 --> 00:23:33.519
of other towns. So the human
species, as far as I'm concerned,

372
00:23:33.640 --> 00:23:37.720
is something that I'm vitally interested in
maintaining. And I'll tell you that the

373
00:23:37.759 --> 00:23:41.640
problem we have is not merely extinction, which I hope you can, maybe

374
00:23:41.680 --> 00:23:47.720
you can extemporaries just elaborate on.
But here's something else. Even if we

375
00:23:47.799 --> 00:23:51.319
don't go extinct, how do we
define ourselves? You talked about that,

376
00:23:51.480 --> 00:23:53.799
Greg, Okay, so we defined
ourselves. We had this myth in the

377
00:23:53.799 --> 00:23:59.039
Bible, we had heaven, you
know, paradise you can just pick off

378
00:23:59.079 --> 00:24:00.880
the trees, you know, pick
off the fruit from the trees. You'd

379
00:24:00.880 --> 00:24:04.880
have to do nothing, and then
the serpent comes and messes things up,

380
00:24:06.119 --> 00:24:10.240
and then we're condemned to toil for
our labors. Life is a struggle.

381
00:24:10.400 --> 00:24:14.200
It's defined as a struggle to define
all the time, as a struggle where

382
00:24:14.200 --> 00:24:17.839
you're competing with the forces of nature, or with other human beings or with

383
00:24:17.880 --> 00:24:21.920
animals, and you constantly better your
position. This is how the human race

384
00:24:22.240 --> 00:24:25.880
has defined itself, and our self
definition is based on that, both as

385
00:24:25.920 --> 00:24:30.759
individuals, as nations, as humanity
as a whole. Now that could be

386
00:24:30.880 --> 00:24:34.359
challenged, and it is challenged.
So it's both our self definition throughout human

387
00:24:34.640 --> 00:24:40.279
history and evolution that is being challenged, and also the question of our continued

388
00:24:40.319 --> 00:24:44.519
existence, which you know you can
please explain why you think we're in danger

389
00:24:44.559 --> 00:24:45.480
of extension. Well, well,
I was gonna say. I mean,

390
00:24:45.519 --> 00:24:49.440
I think this whole arc in my
mind is all about paradigm shift, right,

391
00:24:49.480 --> 00:24:55.160
And I think that even the question
of what would that heaven push agi

392
00:24:55.240 --> 00:24:57.599
positive future look like? I think
even that is hard for us to imagine

393
00:24:57.599 --> 00:25:02.359
what the true upside could be because
it's not just material abundance. It's not

394
00:25:02.359 --> 00:25:06.519
just we all get nice cars or
something hopefully I'll tell us, but also

395
00:25:06.680 --> 00:25:08.240
that everyone gets a great education,
right, you know, everyone has that

396
00:25:08.279 --> 00:25:11.480
story of that one teacher that they
had who paid attention to them and inspire

397
00:25:11.519 --> 00:25:15.079
them on a specific subject. Imagine
if you had that teacher for anything you

398
00:25:15.119 --> 00:25:18.799
wanted constantly, and just how much
better if people would we be? How

399
00:25:18.880 --> 00:25:22.839
much better would we relate to others? Like that is just one example of

400
00:25:22.839 --> 00:25:26.079
the kinds of upsides that are possible. And I think the question of well

401
00:25:26.079 --> 00:25:27.400
what does it mean to be a
human? You know, if you back

402
00:25:27.440 --> 00:25:32.160
to how do you predict what's going
to come next? Actually the thinker who

403
00:25:32.200 --> 00:25:36.440
I think had the best foresight about
how the AI revolution was going to play

404
00:25:36.440 --> 00:25:41.160
out is actually Ray Kurtzwile you know
he yeah, and his book Singularity is

405
00:25:41.200 --> 00:25:44.039
near gets like a lot of shit. I think that people that you know,

406
00:25:44.119 --> 00:25:45.440
kind of assume it is going to
be this almost like just sort of

407
00:25:45.440 --> 00:25:48.920
religious text, but instead it's a
very dry analytical text. And he just

408
00:25:48.960 --> 00:25:52.240
looks at the compute curves and he
says this is the fundamental unlock or of

409
00:25:52.279 --> 00:25:56.680
intelligence. Everyone thought that was crazy, and now it's basically true. It's

410
00:25:56.680 --> 00:25:59.119
basically common wisdom. And part of
what he says is, look, what's

411
00:25:59.119 --> 00:26:00.119
going to happen is in twenty thirties. First of all, he says,

412
00:26:00.160 --> 00:26:03.640
AGI twenty twenty nine. Yeah,
I keep telling people it seems to be

413
00:26:03.599 --> 00:26:07.440
almost exactly right. It's spooky.
It's spooky. Twenty thirties is when the

414
00:26:07.480 --> 00:26:11.680
merge happens. So we've got neurallink
coming, and you know, maybe other

415
00:26:11.720 --> 00:26:15.400
systems like that, And what does
it mean once you actually are kind of

416
00:26:15.440 --> 00:26:19.960
merging with an intelligence. Well,
neuralalink is necessarily moves slower than AI because

417
00:26:21.079 --> 00:26:22.519
when when have you put a device
in a human? You have to be

418
00:26:22.599 --> 00:26:26.480
incredibly careful. So I think it's
not clear to me that the neuralink will

419
00:26:26.519 --> 00:26:30.319
be ready before a GI. I
think AGI is probably going to happen first,

420
00:26:30.640 --> 00:26:34.559
but you know, we'll have the
sort of AGI singularity, you know,

421
00:26:34.640 --> 00:26:38.400
so as digital superintelligence is called like
a singularity, like a black hole,

422
00:26:38.440 --> 00:26:41.640
because just like with a black hole, is difficult to predict what happens

423
00:26:41.680 --> 00:26:45.279
after you pass the event. Arison
of black hole, and we are currently

424
00:26:45.680 --> 00:26:49.200
so you know, circling the event
horizon of the black hole, that is

425
00:26:49.440 --> 00:26:55.359
at digital superintelligence, the event horizon. So I bet part of the reason

426
00:26:55.559 --> 00:26:59.440
for neurallink, although initially we're very
helpful to people who have brain or spine

427
00:26:59.480 --> 00:27:03.680
injuries, just doing basic stuff like
enabling people who are tetraplegics or like a

428
00:27:03.720 --> 00:27:08.920
Stephen Hawking to communicate actually even faster
than I'm communicating with a fully functional body.

429
00:27:10.200 --> 00:27:18.039
So and ultimately to improve the bandwidth
between the cortex and your AI version

430
00:27:18.039 --> 00:27:21.640
of yourself. In fact, I
think a lot of people perhaps don't quite

431
00:27:21.640 --> 00:27:25.119
realize that they're already a syebalk.
So you've got your lymic system, your

432
00:27:25.160 --> 00:27:29.000
sort of basic drives, your cortex, which is the thinking and planning,

433
00:27:29.240 --> 00:27:30.920
and then you have tertiary layer,
which is your computers, your devices,

434
00:27:30.960 --> 00:27:36.480
your phones, laptops, all the
servers that exist, the applications. And

435
00:27:36.519 --> 00:27:40.160
in fact, I think probably a
lot of people have found that if you

436
00:27:40.240 --> 00:27:42.960
leave your cell phone, take it
away, if you get if you forget

437
00:27:42.960 --> 00:27:47.240
your cell phone, it's like missing
Limb syndrome. You know, you've like

438
00:27:47.480 --> 00:27:49.480
read that thing, go Losing your
cell phone is like missing Lin syndrome.

439
00:27:49.759 --> 00:27:53.559
So, because it is your cell
phone is extension extension of yourself, the

440
00:27:53.599 --> 00:27:59.279
limitation is badwidth. So you the
rate which you can input or i should

441
00:27:59.319 --> 00:28:03.240
say output in nation into your phone
or computer is very slow. So the

442
00:28:03.279 --> 00:28:07.160
phone, it's really just the speed
of your thumb movements, and with you

443
00:28:07.200 --> 00:28:11.160
know, best case scenario, you're
a speed typist on a keyboard. But

444
00:28:11.240 --> 00:28:14.880
even that data rate is very slow, which we're talking about tens, maybe

445
00:28:15.119 --> 00:28:19.079
hundreds of bits per second, whereas
a computer can communicate in trillions of bits

446
00:28:19.079 --> 00:28:23.960
per second. So so the and
this is admittedly somewhat of a you know,

447
00:28:25.000 --> 00:28:27.240
hail Mary shot or whatever lung shot, is that if you can improve

448
00:28:27.279 --> 00:28:33.640
the bad with between your cortex and
the your digital tertiary self, then you

449
00:28:33.640 --> 00:28:38.240
can achieve better cohesion between what humans
want and what AI does. At least

450
00:28:38.279 --> 00:28:41.319
that's one theory. I'm not saying
this is a short thing. It's just

451
00:28:41.559 --> 00:28:45.359
one potential iron in the fire.
If ultimately, you know, one hundreds

452
00:28:45.400 --> 00:28:51.200
of millions or billions people get a
high bound with interface to their digital tertiary

453
00:28:51.200 --> 00:28:56.000
self, their AI self effectively,
then that that seems like that probably leads

454
00:28:56.039 --> 00:29:00.240
to a better future for humanity.
So that's this is I'm getting somewhat esoteric.

455
00:29:00.480 --> 00:29:03.079
Hopefully some of this is making sense, and like I said, it's

456
00:29:03.079 --> 00:29:07.039
a long shot. But you know, what I find interesting is that I've

457
00:29:07.160 --> 00:29:11.359
I've not found any human who wishes
to delete either their cortex or their limits

458
00:29:11.400 --> 00:29:14.720
system. I probably I suspect nobody
in this room wants to do that.

459
00:29:14.920 --> 00:29:18.599
It's a role or their cell phone. Well, now I've met people who

460
00:29:18.599 --> 00:29:21.680
want to who don't use a cell
phone. It's rare. Okay, Well,

461
00:29:21.759 --> 00:29:25.640
I confess for years I avoided this, and I avoided it. I

462
00:29:25.680 --> 00:29:30.200
didn't even have Not only did I
not have a cell phone within my immediate

463
00:29:30.319 --> 00:29:34.039
radius, I didn't have a television
set. My phones were thirty five year

464
00:29:34.039 --> 00:29:38.599
old phones because I was Prime Minister
Visual and I knew what this means.

465
00:29:38.920 --> 00:29:42.880
Okay, I mean maybe your audience
doesn't know, but we knew. Now,

466
00:29:44.319 --> 00:29:48.079
having spent some time in the opposition, I lapsed into the use of

467
00:29:48.079 --> 00:29:51.519
the cell phone, and I find
it very hard to dissociate from it.

468
00:29:51.559 --> 00:29:53.720
For the reasons that you said.
It's become an extension of, you know,

469
00:29:53.960 --> 00:29:59.680
a repository of memory, the ability
to get information very fast, which

470
00:29:59.680 --> 00:30:03.640
you say correctly, is very slow, but compared to what it was before.

471
00:30:03.440 --> 00:30:07.039
Remember you had research assistance, have
to go to libraries, things like

472
00:30:07.079 --> 00:30:11.400
that. Where is that that's gone? But this will be seen as very

473
00:30:11.519 --> 00:30:15.119
vastly primitive compared to what you're talking
about. Yeah, I'll just add something

474
00:30:15.240 --> 00:30:18.079
is that to max? But the
I mean, it's it's sort of a

475
00:30:18.200 --> 00:30:21.839
funny thing like if you assume,
like a best case scenario, imagine if

476
00:30:21.680 --> 00:30:25.799
if you're the AI and you're trying
to you just want the human to tell

477
00:30:25.839 --> 00:30:30.000
you what it wants, just please
spit it out. But it's speaking so

478
00:30:30.119 --> 00:30:34.000
slowly like a tree. Okay,
Like trees communicate, Okay, they if

479
00:30:34.000 --> 00:30:37.319
you watch a tree, like a
you know, sped up a version of

480
00:30:37.319 --> 00:30:41.599
a tree growing, it's actually communicating. It's communicating with the soil. It's

481
00:30:41.680 --> 00:30:45.480
trying to find the sunlight. You
know, it's reacting to other trees and

482
00:30:45.559 --> 00:30:49.319
that kind of thing very slowly.
But from a tree standpoint, it's you

483
00:30:49.319 --> 00:30:52.799
know, not that slow. So
so what I'm saying is we don't want

484
00:30:52.799 --> 00:30:56.279
to be a tree. That's that's
the idea behind a high band with the

485
00:30:56.440 --> 00:31:00.519
neural interface is just even when the
AI it desperately wants to do good,

486
00:31:00.559 --> 00:31:04.200
good things for us that we can
actually communicate several orders of attitude fast,

487
00:31:04.200 --> 00:31:07.160
and we currently can. I love
that you mentioned the trees there, because

488
00:31:07.200 --> 00:31:11.720
if anyone is still struggling to understand
why it's so likely that we could get

489
00:31:11.720 --> 00:31:17.680
wiped out by AI, you know, as superintelligent AI, we are like

490
00:31:17.839 --> 00:31:22.119
trees. They are as much faster
than us, and they're thinking than we

491
00:31:22.200 --> 00:31:23.680
are. The trees. So if
some trees in the rainforest are a little

492
00:31:23.720 --> 00:31:26.960
bit worried that some humans are going
to come chop them down, you know,

493
00:31:27.200 --> 00:31:30.640
and they're like, Oh, don't
worry, We're so smart, we'll

494
00:31:30.680 --> 00:31:33.640
stop those humans. Good luck with
that, right. I love how you

495
00:31:33.799 --> 00:31:37.160
went philosophical here and talked about what
it means to be human, and I

496
00:31:37.200 --> 00:31:40.400
wanted to follow up on that,
and but before that, maybe just continue

497
00:31:40.480 --> 00:31:44.759
that. Okay, So, how
you're assuming that the danger of human extinction

498
00:31:45.240 --> 00:31:51.880
is that you have machines, supermachines, superintelligent machines that outpass humans by orders

499
00:31:51.880 --> 00:31:57.680
and orders of magnitude who will not
necessarily value the continued human existence as we

500
00:31:57.759 --> 00:32:01.440
know it, and basically program is
out. That's basically what you're saying,

501
00:32:01.599 --> 00:32:07.279
so, if we were to be
so dumb as to create the entities that

502
00:32:07.279 --> 00:32:09.200
are much more powerful than us that
don't share our goals, yeah, that's

503
00:32:09.240 --> 00:32:13.359
a really bad idea. I mean, in Israel, I think you understand

504
00:32:13.680 --> 00:32:17.400
really well that if you have a
bunch of other beings who are intelligent and

505
00:32:17.880 --> 00:32:22.119
don't share your values, you know, it can go really badly, right,

506
00:32:22.359 --> 00:32:25.400
and let's not make that mistake by
giving too much power to machines don't

507
00:32:25.400 --> 00:32:30.359
share our values. But of course
we can tackle that in various ways.

508
00:32:30.480 --> 00:32:34.160
So coming back to what can we
concretely do, I just want to say

509
00:32:34.200 --> 00:32:37.680
one thing about the philosophical point about
being human, just for myself. One

510
00:32:37.720 --> 00:32:42.240
of the things that I really cherish
the most about being human is the agency

511
00:32:42.440 --> 00:32:45.799
that I can actually make choices and
make a difference. And one of the

512
00:32:45.839 --> 00:32:51.519
most beautiful things about what you were
all doing with technology is giving us more

513
00:32:51.559 --> 00:32:54.359
agency. Back in the Middle Ages, we had a life expectancy of freaking

514
00:32:54.440 --> 00:32:59.680
thirty years and knowing that we might
starve to death next year because of the

515
00:32:59.720 --> 00:33:02.000
weather. There, you know what
kind of it's so much more inspiring now

516
00:33:02.000 --> 00:33:06.559
that we're becoming more of the captain
of our own ship with technology able to

517
00:33:06.599 --> 00:33:08.880
actually have more agency as a species. So I say, let's use that

518
00:33:08.920 --> 00:33:13.599
agency to create a wonderful future,
not to throw it away. And three

519
00:33:13.680 --> 00:33:16.359
things I wanted to just suggest for
concrete action items we can do here.

520
00:33:16.640 --> 00:33:20.039
I'll say a little more about each
of them. One has to do with

521
00:33:20.119 --> 00:33:23.839
pausing risky stuff, one has to
do with controlling the machines, and one

522
00:33:23.880 --> 00:33:29.160
has to do with inspiration for the
future. So for the controlling, we

523
00:33:29.240 --> 00:33:32.680
already talked quite a lot about the
importance of staying in control over these machines,

524
00:33:32.960 --> 00:33:37.119
so I don't need to doell on
it anymore. But we are I

525
00:33:37.160 --> 00:33:39.920
think it's saying things like, oh, what is going to happen to us

526
00:33:40.000 --> 00:33:44.880
bring out the popcorn. That's too
passive for me. Let's celebrate our agency

527
00:33:44.920 --> 00:33:47.400
and ask ourselves not what we think
will happen, but what we want to

528
00:33:47.400 --> 00:33:52.559
happen. And clearly we want to
control these machines we're building them. Second

529
00:33:52.559 --> 00:33:57.240
thing, so, I, you
know, agree with Gregg on many things,

530
00:33:57.279 --> 00:34:00.960
and I greatly respect your work.
One thing where we have a friendly

531
00:34:00.000 --> 00:34:04.400
disagreement is you know, I signed
this pause letter saying I think we should

532
00:34:04.440 --> 00:34:07.079
pause the riskiest things, and you
don't. I think that it would be

533
00:34:07.119 --> 00:34:10.960
a good strategy for us to emulate
what biotech does and just have clear safety

534
00:34:12.039 --> 00:34:14.880
standards. You know, if you
want to release a new drug, you

535
00:34:14.920 --> 00:34:17.400
can't just sell it in the supermarket. You have to do a clinical trial,

536
00:34:17.480 --> 00:34:21.559
persuade some experts somehow that this is
safe. And I would like to

537
00:34:21.599 --> 00:34:24.760
see really high stakes, very powerful
systems, some sort of system like this

538
00:34:25.039 --> 00:34:29.960
with standards, and then the companies
have and it's on them to, you

539
00:34:30.000 --> 00:34:31.840
know, prove that they're safe before
they get released it. Who does the

540
00:34:31.920 --> 00:34:37.519
standards? Well, and we've solved
this, It's been solved in Israel.

541
00:34:37.559 --> 00:34:40.199
It's been solved in the US for
biotech, not for biotech. But here,

542
00:34:40.480 --> 00:34:44.599
you know, it's a's a do
you want it? Regulators? Who

543
00:34:44.639 --> 00:34:46.960
do you want it? Legislatures?
I mean, it's not the only choice,

544
00:34:46.960 --> 00:34:51.199
but you know, it's a pretty
pretty clear dichotomy. I don't think

545
00:34:51.239 --> 00:34:53.920
we need to reinvent this wheel.
The people who decide whether a drug gets

546
00:34:53.920 --> 00:34:58.800
approved in Israel is not people in
parliament in the Kinnessets, right, it's

547
00:34:58.880 --> 00:35:02.199
Israeli scientists, but the government involved
in approving them. Exactly we can emulate

548
00:35:02.239 --> 00:35:06.440
that, and Elon can tell us
all about how we deal with the cars.

549
00:35:06.559 --> 00:35:08.199
I don't want to dwell on,
and I just think this is not

550
00:35:08.320 --> 00:35:14.480
rocket science. We can put these
standards in place and then we can continue

551
00:35:14.880 --> 00:35:16.960
doing all sorts of wonderful stuff with
a tech. Some stuff will wait a

552
00:35:16.960 --> 00:35:21.440
little longer for but it'll be worth
the wait because it'll be safe. The

553
00:35:21.480 --> 00:35:25.000
last thing I just wanted to say
is I think for us to really seize

554
00:35:25.000 --> 00:35:30.239
this agency and work together, not
just between companies but also across borders,

555
00:35:30.519 --> 00:35:36.360
we really need a shared positive vision. And this is something I really appreciate

556
00:35:36.400 --> 00:35:38.800
so much about You, Elon have
the guts more than most to talk about

557
00:35:38.880 --> 00:35:45.199
like really audacious positive visions for the
future. What is it that makes people

558
00:35:45.320 --> 00:35:51.000
collaborate and put little squabbles aside.
It's always a shared positive vision where people

559
00:35:51.119 --> 00:35:53.599
realize that if we collaborate, we'll
all be better off. You know,

560
00:35:53.880 --> 00:35:59.840
technology is not a zero some game, right. You can take countries like

561
00:36:00.320 --> 00:36:05.840
Sweden and Denmark that had been killing
each other for centuries and flat GDP,

562
00:36:06.039 --> 00:36:08.920
you know, and then technology came
along and they both got dramatically better off.

563
00:36:09.039 --> 00:36:13.920
It's so obvious that the technology that
you're building. You're building you know,

564
00:36:14.280 --> 00:36:19.039
can make all humans just spectacularly better
off with AI, and not just

565
00:36:19.119 --> 00:36:22.159
on this planet. I like the
list before we started you had there about

566
00:36:22.199 --> 00:36:25.360
how it can revolutionize healthcare and so
many other things. And I would add

567
00:36:25.400 --> 00:36:29.440
to that, you know, why
limit ourselves to this planet when there's this

568
00:36:29.639 --> 00:36:32.360
beautiful universe where life can flourish for
billions of years. If the more we

569
00:36:32.440 --> 00:36:37.559
can can just appreciate how huge the
upside is if we do this right,

570
00:36:37.800 --> 00:36:43.280
you know, the more incentive we
will have to actually be a little patient

571
00:36:43.280 --> 00:36:45.800
where we need to be a little
bit patient, make everything safe and therefore

572
00:36:46.000 --> 00:36:52.159
get this wonderful future. Well,
maybe we're sort of lagging behind in our

573
00:36:52.519 --> 00:36:57.119
social and political development. This is
like having nuclear technology and the Stone Age,

574
00:36:57.360 --> 00:36:59.760
what do you do with it?
Which I think is you know,

575
00:37:00.000 --> 00:37:02.039
because more That's the thing that I
talked about in terms of the pace of

576
00:37:02.079 --> 00:37:07.719
development, pacing basically what solutions we
need to put in place to maximize the

577
00:37:07.760 --> 00:37:12.679
benefits and limits the risks. I
think we're there. I think we should

578
00:37:12.719 --> 00:37:15.760
have a meeting of minds. You
were in you were in the Senate the

579
00:37:15.760 --> 00:37:17.679
other day. I presume you were
talking about this. I mean, yeah,

580
00:37:17.800 --> 00:37:22.000
you don't have to divulge what you
can divulge, but divulge would you

581
00:37:22.039 --> 00:37:23.000
can? Well, a lot of
people in the meeting, so I don't

582
00:37:23.039 --> 00:37:28.239
think it was secret. It was
just a discussion on AI regulation. And

583
00:37:28.360 --> 00:37:31.599
actually Sam Often from open I was
there and a number of eye leaders were

584
00:37:31.719 --> 00:37:36.239
were there. And I think generally
it would be a good idea to have

585
00:37:36.320 --> 00:37:42.800
some kind of AI regulatory agency and
start off with a team that gathers insight

586
00:37:43.000 --> 00:37:46.760
to get a maximum understanding. Then
you have some proposed rulemaking, and then

587
00:37:47.159 --> 00:37:51.440
eventually you have regulations that are put
in place. And this is something we

588
00:37:51.480 --> 00:37:53.639
have for everything that is potential danger
to the public. So if it is

589
00:37:53.960 --> 00:37:57.920
you know, if we're in drugs
in drug administration, we got aircraft with

590
00:37:58.079 --> 00:38:01.760
fa and rockets. You know,
there's ever anything that is a danger to

591
00:38:01.800 --> 00:38:06.159
the public. Over time, we
have learned that often the hard way,

592
00:38:06.559 --> 00:38:10.639
after many people have died, to
have a regulatory agency to protect public safety.

593
00:38:12.079 --> 00:38:15.320
I'm not someone who thinks that regulations
some panacea where it's only good.

594
00:38:15.440 --> 00:38:19.559
Of course, there are some downsides
to regulation and that things move a bit

595
00:38:19.599 --> 00:38:22.840
slower, and sometimes you get regulatory
capture and that kind of thing. But

596
00:38:22.840 --> 00:38:28.440
but on balance, I think the
public would not want to get rid of

597
00:38:28.679 --> 00:38:32.079
most regulatory agencies. And you can
think of it also as like the regulatory

598
00:38:32.119 --> 00:38:36.760
agency being like a referee. You
know, what sports game doesn't have a

599
00:38:36.800 --> 00:38:39.599
referee? You need someone to make
sure that that people are playing fairly,

600
00:38:40.000 --> 00:38:45.840
not breaking the rules. And that's
why basically every sport has a referee of

601
00:38:45.880 --> 00:38:50.719
one kind or another. So that's
the rationale for AI safety. And I've

602
00:38:50.719 --> 00:38:52.280
been pushing this all around the world, and when I was in China a

603
00:38:52.320 --> 00:38:55.920
few months ago meeting with some of
the senior leadership, that my primary topic

604
00:38:57.119 --> 00:39:01.719
was AI safety and regulation and they, after we had a long discussion,

605
00:39:01.840 --> 00:39:07.360
agreed that this merit to AI regulation
and immediately took action in this regard.

606
00:39:07.679 --> 00:39:12.760
So so sometimes it will get this
comment of like, well, if the

607
00:39:12.800 --> 00:39:16.280
West does AI regulation surely, then
what about what if China doesn't and then

608
00:39:16.360 --> 00:39:21.519
leaves ahead. And I think they
are also taking it very seriously, because

609
00:39:21.559 --> 00:39:24.360
you know the opposite of what of
a moral constraints you program into the system?

610
00:39:24.559 --> 00:39:27.360
I mean, Greg, I don't
know what you what are your thoughts

611
00:39:27.400 --> 00:39:30.719
on this? Yeah, I mean
I think again it's just like this technology

612
00:39:30.800 --> 00:39:32.639
is almost backwards from how you'd expect
it to work. Like I think that

613
00:39:32.719 --> 00:39:36.679
a lot of people talk about the
paper clips problem of if you think of

614
00:39:36.719 --> 00:39:39.119
that as if we literally program into
a machine make as many paper clips as

615
00:39:39.159 --> 00:39:42.559
possible, and it gets really smart, and it's like, well, there's

616
00:39:42.599 --> 00:39:45.239
all these houses and factories and other
things, they're not producing paper clips.

617
00:39:45.239 --> 00:39:47.840
They're all going to be just paper
clip production. Now, all these humans

618
00:39:47.840 --> 00:39:51.639
that are in the way paper clip
would be a better arrangement of their atoms.

619
00:39:51.800 --> 00:39:54.199
And that's just not the technology path
we actually seem to be on.

620
00:39:54.480 --> 00:39:58.440
You look at GPT three, and
it is a system that gets the nuance,

621
00:39:58.519 --> 00:40:00.519
It gets the common sense. It
has other problems, right, it

622
00:40:00.519 --> 00:40:05.039
makes mistakes that no human ever would
like kind of loses this chain of thought

623
00:40:05.079 --> 00:40:07.280
and says all these weird things and
sometimes make things up like we have real

624
00:40:07.280 --> 00:40:10.360
problems we have to encounter. But
I think that for me, that the

625
00:40:10.519 --> 00:40:14.960
hard thing that I think gets lost
in some of the conversations is the nuance

626
00:40:15.039 --> 00:40:19.159
right of really being coupled to what's
the technology we actually see and particularly how

627
00:40:19.199 --> 00:40:21.800
does it depart from our intuitions about
where we thought it would go. I

628
00:40:21.800 --> 00:40:23.559
actually think it's even interesting. You
know, Max mentioned the pause letter.

629
00:40:23.679 --> 00:40:27.800
The pause letter, one of the
key parts of it was to not train

630
00:40:27.880 --> 00:40:30.679
GPD five for the next six months, and we looked at that and we're

631
00:40:30.719 --> 00:40:32.840
like, well, we're not training
GPT five for the next six months anyway.

632
00:40:34.079 --> 00:40:36.519
And so you know, it's just
like again if you kind of miss

633
00:40:36.639 --> 00:40:39.159
the sort of nuances of what's going
on at a mechanical level. And some

634
00:40:39.199 --> 00:40:43.679
of this does require getting like really
in the weeds and coming and talking to

635
00:40:44.039 --> 00:40:46.880
us and also opening up and showing
kind of in progress work. And I

636
00:40:46.880 --> 00:40:51.679
think that's something that will be very
important as you think about moving to these

637
00:40:51.760 --> 00:40:54.400
very powerful systems. And I also
think there's another blessing and curse of how

638
00:40:54.400 --> 00:40:59.360
the technology is playing out, where
we talk about digital superintelligence and we also

639
00:40:59.400 --> 00:41:02.400
talk about chat. These are very
different things, but somehow there's going to

640
00:41:02.400 --> 00:41:07.079
be a smooth continuing between them.
And so I think that the blessing is

641
00:41:07.119 --> 00:41:08.599
that we see a little bit into
the future. We have a little taste

642
00:41:08.639 --> 00:41:13.480
of a system that you know,
I talked about a personalized tutor for everyone.

643
00:41:13.719 --> 00:41:15.079
You kind of have it now.
It's not a perfect tutor. It's

644
00:41:15.079 --> 00:41:17.280
got a lot of problems, but
like you know, you can go use

645
00:41:17.280 --> 00:41:21.440
shat GPT for free to ask about
anything and it'll give you something. Sometimes

646
00:41:21.440 --> 00:41:25.199
it's actually be pretty insightful. Well
fair enough, I'll withdraw the seven and

647
00:41:25.239 --> 00:41:28.599
a half, but but you'll get
there. You'll get to a ten.

648
00:41:28.760 --> 00:41:30.000
We will, we will, we
will. Definitely, that's right, And

649
00:41:30.039 --> 00:41:34.400
I think it's great for people to
see it while we can still laugh about

650
00:41:34.440 --> 00:41:36.599
it. Right. I think that
when it starts coming up with the new

651
00:41:36.639 --> 00:41:38.320
ideas, you type in some problem
you're having at home, trying to figure

652
00:41:38.320 --> 00:41:40.320
out how to do something, and
it just comes back with, oh,

653
00:41:40.360 --> 00:41:43.360
you should say this to this person, and you should do that, you

654
00:41:43.400 --> 00:41:45.800
should you know, here's say you
should revenge your schedule. And it's just

655
00:41:45.840 --> 00:41:49.719
like great ideas, Like that's gonna
feel so different. And then one day,

656
00:41:49.719 --> 00:41:52.199
when we're actually having a system that
can actually just autonomously go off and

657
00:41:52.239 --> 00:41:55.920
do a bunch of research and say, here's this crazy chemical formula, like

658
00:41:55.920 --> 00:41:59.679
please mix them together in these ways. You want to make sure that thing

659
00:42:00.119 --> 00:42:02.400
trustworthy right before we start mixing those
chemicals. And so I think that this

660
00:42:02.519 --> 00:42:07.800
question of how do we separate out
the sort of diffusion of today's AI technology

661
00:42:07.840 --> 00:42:10.760
and application and really, you know, all the debates that we're having,

662
00:42:10.840 --> 00:42:15.920
let those happen, open source technologies
like everyone's building. We should not squelch

663
00:42:15.000 --> 00:42:19.719
those, we should actually promote those. But I think that also looking ahead,

664
00:42:19.800 --> 00:42:22.360
being ahead of the game here and
saying there is going to be a

665
00:42:22.400 --> 00:42:24.960
point and it's where we're hitting that
wall, where we're going to be building

666
00:42:24.960 --> 00:42:29.559
the most massive machines that humans have
ever built. Like Youan talks about these

667
00:42:29.760 --> 00:42:32.760
massive rows of data centers with all
these computers, like like it's one of

668
00:42:32.760 --> 00:42:36.440
those things where you kind of have
to see it to really feel it.

669
00:42:36.519 --> 00:42:38.440
Like I've walked through some of these
data center halls and you just get a

670
00:42:38.480 --> 00:42:42.400
sense of the scale of what's being
built. And it's very different when it's

671
00:42:42.440 --> 00:42:45.360
abstractly just like you're looking at,
oh is this many GPUs and you know,

672
00:42:45.400 --> 00:42:47.840
looking at numbers, it's like you
just realize the mechanical force that is

673
00:42:47.880 --> 00:42:52.519
behind what you're building for cognitive intelligence. And I think that kind of thing

674
00:42:52.519 --> 00:42:57.119
we need international answers for. I
fully agree with you. I think,

675
00:42:57.159 --> 00:43:00.639
Chris, of all you know,
with all due respect of the pause button,

676
00:43:00.840 --> 00:43:02.000
we ain't going to get it.
It's not going to happen. I

677
00:43:02.039 --> 00:43:06.400
think you can't put the genie back
into the bottle, and I'm not sure

678
00:43:06.440 --> 00:43:07.880
you can slow it down. The
only thing you can do is just to

679
00:43:08.119 --> 00:43:12.039
accelerate our own efforts. And I
think you're right. It has to be

680
00:43:12.079 --> 00:43:15.159
international. I think there. I
think you start with a like minded and

681
00:43:15.239 --> 00:43:17.800
like valued states. Start with that. And frankly, I think you know,

682
00:43:19.039 --> 00:43:21.679
obviously, you know what AI stands
for. It stands for America and

683
00:43:21.760 --> 00:43:25.039
Israel. Obviously, I'm just plugging
my own country. But it's a it's

684
00:43:25.039 --> 00:43:29.920
not an empty plug, because I
think that there's a concentration of talent an

685
00:43:30.000 --> 00:43:32.320
Israel that makes it more than a
mouse compared to the American elephant. It's

686
00:43:32.320 --> 00:43:37.639
more than that. It's a very
nimble mammal, very fast compact, and

687
00:43:37.679 --> 00:43:44.960
it's it's also got this feed of
military AI which moves very fast and takes

688
00:43:45.320 --> 00:43:49.280
a very large part of the population. The talents there screens it and it

689
00:43:49.360 --> 00:43:52.039
trains it very quickly. So I
think that you start with a like minded

690
00:43:52.079 --> 00:43:58.239
states, you definitely should have a
very robust conversation. First of all,

691
00:43:58.440 --> 00:44:02.320
decide among us what is something there
are things that we wouldn't want to share,

692
00:44:02.519 --> 00:44:07.440
and it's obvious related to our security. There are standards that we have,

693
00:44:07.559 --> 00:44:12.639
ethical standards and other regulations that we
have to achieve quickly among between industry

694
00:44:12.639 --> 00:44:16.440
and governments quickly. And then I
think we have to conduct a robust discussion

695
00:44:16.760 --> 00:44:22.159
with the other powers of the world
based on their self interest, as you

696
00:44:22.239 --> 00:44:24.639
began to do. And I think
that's a pioneering work, and I think

697
00:44:24.639 --> 00:44:30.159
we have a shot maybe at getting
to some degree of control over our future,

698
00:44:30.199 --> 00:44:34.480
which could be amazing. You know, I'm not sure I want to

699
00:44:34.519 --> 00:44:37.800
live in heaven. It could be
boring. I don't mean the boring company.

700
00:44:37.840 --> 00:44:39.840
I mean it could be boring,
But I think we're headed. I'd

701
00:44:39.880 --> 00:44:45.920
much rather have a boring heaven than
an interesting hell. So here at Tesla,

702
00:44:45.360 --> 00:44:47.960
they have a different name for pause
button. They call it the break.

703
00:44:49.360 --> 00:44:52.960
And when we launched this pause letter, right, I personally didn't think

704
00:44:53.000 --> 00:44:57.559
that there was immediately going to cause
a pause for me. The real purpose

705
00:44:57.599 --> 00:45:00.880
of it was the mainstream the conversation
about these issues, which really did happen.

706
00:45:01.199 --> 00:45:06.039
And you know, I really appreciate
you and Sam and your colleagues for

707
00:45:06.159 --> 00:45:09.760
saying very openly that there may come
a time when you will want to pause

708
00:45:09.840 --> 00:45:15.199
certain things if you meet certain criteria. Right now, if Elon says that

709
00:45:15.480 --> 00:45:19.559
may come a time when he's going
to slow down his Tesla, I would

710
00:45:19.599 --> 00:45:22.840
feel I would trust that more if
I knew that Tesla actually had breaks in

711
00:45:22.880 --> 00:45:25.079
it, so that they had like
a mechanism by which you could actually slow

712
00:45:25.119 --> 00:45:28.960
it down. So this is something
I'm very curious to actually to hear it

713
00:45:29.000 --> 00:45:34.760
from you about if what criteria and
would you say that you all have can

714
00:45:35.079 --> 00:45:37.639
can you say it in the future, say it already now, Like what

715
00:45:37.679 --> 00:45:39.800
are some things which would make you
want to pause? And and then won't

716
00:45:39.840 --> 00:45:45.159
it be difficult with the actual mechanism, like to resist commercial pressures to continue

717
00:45:45.199 --> 00:45:47.599
even if you and Sam want to
pause. Well, I actually kind of

718
00:45:47.639 --> 00:45:52.519
agree with both of you right that, I think that my picture of how

719
00:45:52.559 --> 00:45:55.159
the future is going to play out
in terms of AI development is not as

720
00:45:55.199 --> 00:45:59.159
simple as oh, you're scaling models
and then you see something a little bit

721
00:45:59.199 --> 00:46:01.440
weird and then you like stop all
scaling and you all go home or something,

722
00:46:01.519 --> 00:46:04.760
right'. It's it's definitely not going
to be that. I think that

723
00:46:04.800 --> 00:46:09.199
we need to be making maximum velocity
efforts towards a good future, right,

724
00:46:09.239 --> 00:46:12.719
And I think that is something that
you don't want to pause. You actually

725
00:46:12.719 --> 00:46:15.000
want to put the foot on the
pedal. Now, of course, the

726
00:46:15.079 --> 00:46:17.960
question is what actions actually get you
towards there, which ones are against it,

727
00:46:19.000 --> 00:46:21.880
you know, Like I think that
nuance is something that I think is

728
00:46:21.920 --> 00:46:23.480
what we should be debating. But
that's why I actually bounce a little bit

729
00:46:23.519 --> 00:46:27.800
off of the framing of a pause, because I think even to the point

730
00:46:27.800 --> 00:46:30.920
of GPD four, doing all this
prediction work ahead of time, like if

731
00:46:30.920 --> 00:46:32.280
that prediction work was a little bit
wobbly, weren't quite there, and we

732
00:46:32.280 --> 00:46:35.960
actually do this all the time,
that there's like some experiment that came back

733
00:46:36.000 --> 00:46:37.639
a little bit strange, we're not
really sure we understand it. We actually

734
00:46:37.719 --> 00:46:40.519
spend a lot of time to fix
that. And it's not even you know,

735
00:46:40.519 --> 00:46:43.960
even if you forget about the safety
reasons, and actually one of the

736
00:46:44.000 --> 00:46:45.519
reasons we do it is safety motivated. But if you even forget about that,

737
00:46:45.519 --> 00:46:47.480
you're just not going to build a
good system. And so I think

738
00:46:47.519 --> 00:46:53.239
that there's a lot of places in
AI where there's actually great alignment between capability

739
00:46:53.280 --> 00:46:57.679
and safety, between building something useful
and something that is safe. It's not

740
00:46:57.719 --> 00:47:00.400
perfect, but I think it is
actually surprising where there are divergences. Actually,

741
00:47:00.440 --> 00:47:05.199
the single biggest divergence we've seen has
not been owe too much pressure to

742
00:47:05.199 --> 00:47:07.119
deploy quickly because I actually think that
getting things out, Like you can see,

743
00:47:07.159 --> 00:47:10.440
the market it hates hallucinations, It
hates it when the model makes something

744
00:47:10.480 --> 00:47:13.800
up. All the jokes we made
about oh it doesn't give you good insights,

745
00:47:13.840 --> 00:47:15.039
the market hates that stuff, right, And actually I think it's very

746
00:47:15.079 --> 00:47:19.920
aligned with safety to fix them.
But what the market actually pushes us against

747
00:47:20.000 --> 00:47:22.400
is actually when it comes into conflict
with other values, for example privacy.

748
00:47:22.639 --> 00:47:27.440
Right, So that the way we
launch the initial gpt APIs was that we

749
00:47:27.440 --> 00:47:29.880
said, hey, we're going to
monitor everything. We're going to record everything,

750
00:47:29.960 --> 00:47:31.599
like if something sketchy comes up,
we'll go and look and we'll be

751
00:47:31.599 --> 00:47:34.519
able to look back in the logs. We'll be able to see what you

752
00:47:34.559 --> 00:47:37.719
did. People hate that, right, like you know you kind of want

753
00:47:37.760 --> 00:47:39.880
that for everyone else, but for
yourself you want privacy. And so how

754
00:47:39.880 --> 00:47:44.039
do you balance those two factors?
And we actually see this very much in

755
00:47:44.159 --> 00:47:47.119
terms of to actually win deals to
decide to be able to give companies the

756
00:47:47.119 --> 00:47:50.119
assurance that hey, we're not going
to retain any of your data, will

757
00:47:50.159 --> 00:47:52.920
delete it immediately. That removes our
ability to do any investigation post talk.

758
00:47:53.000 --> 00:47:55.519
And so this is actually a place
where I think government can be very helpful,

759
00:47:55.519 --> 00:47:59.880
where norms can be very helpful to
say, hey, even though you

760
00:48:00.079 --> 00:48:04.719
know that this is what would be
you know commercially most viable, that this

761
00:48:04.760 --> 00:48:06.559
is what the rule is going to
be, and everyone has to play with

762
00:48:06.559 --> 00:48:08.360
it. You don't ever race to
the bottom. So I think that again,

763
00:48:08.400 --> 00:48:12.079
it's all going to be nuanced,
it's all going to be I think,

764
00:48:12.239 --> 00:48:15.559
like I've just accepted that this field
it's totally counterintuitive, Like everything we

765
00:48:15.639 --> 00:48:20.199
learn this is like it just would
make you just like if you told yourself

766
00:48:20.199 --> 00:48:22.000
ten years ago you have said,
there's no way. That sounds totally crazy,

767
00:48:22.239 --> 00:48:25.639
and somehow we have to navigate through
that. Well, the only thing

768
00:48:25.679 --> 00:48:30.039
that I can add to this conversation
is to invite you all to as well.

769
00:48:30.280 --> 00:48:34.000
I'd like to, in fact continue
this conversation, in fact, bring

770
00:48:34.119 --> 00:48:38.280
a few more of your colleagues and
a few leaders to discuss this international strategy

771
00:48:38.599 --> 00:48:44.719
for like minded states to begin with
that, you know, first like minded

772
00:48:44.719 --> 00:48:46.880
states, then the world. Okay, how about that as a plan of

773
00:48:46.880 --> 00:48:52.320
action. But I think that I
think that the work that you're doing is

774
00:48:52.360 --> 00:48:54.920
amazing. Nothink short of it.
You know, Elon was saying, what

775
00:48:55.039 --> 00:49:00.280
is a human being? It's sort
of this two legged thing ape like true,

776
00:49:00.320 --> 00:49:05.320
with a cortex and a computer and
its wants, desires and a brain.

777
00:49:05.400 --> 00:49:07.679
But the brain is developing in ways
that we couldn't possibly imagine, and

778
00:49:07.800 --> 00:49:12.639
it just doesn't stop. Well,
I mean, from my other comments,

779
00:49:12.679 --> 00:49:16.519
like the thing that paps should provide
some hope for digital not being controllable in

780
00:49:16.519 --> 00:49:21.679
a very very specific way but acting
in a way that is beneficial to humans

781
00:49:21.760 --> 00:49:25.440
is that our cortex and our limc
system, the cortex is actually in service

782
00:49:25.440 --> 00:49:29.840
to the limits system. So you
know, the cortex, the thinking part,

783
00:49:29.920 --> 00:49:32.039
is constantly trying to make the instinctual
part happy, even though it is

784
00:49:32.119 --> 00:49:37.199
much smarter than the instinctual part.
And so I think there's hope there that

785
00:49:37.519 --> 00:49:40.400
digital superintelligence, even though it is
much smarter than our cortex, will none

786
00:49:40.480 --> 00:49:45.199
less want to make the cortex happy
and ultimately LIMICS system happy. I mean

787
00:49:45.239 --> 00:49:49.159
not to get too off color here, but the sheer amount of thinking and

788
00:49:49.559 --> 00:49:53.599
about two that people have done to
have sex is a lot, but not

789
00:49:53.800 --> 00:49:59.079
sex will procreation, just for the
to make the limics system happy. So

790
00:49:59.440 --> 00:50:02.159
you know, given how much you
know thought has gone into getting laid essentially,

791
00:50:02.320 --> 00:50:06.280
but for the sole purpose of getting
laid, not procreation, and all

792
00:50:06.320 --> 00:50:08.599
that does is make the limbic system, you know, give us block of

793
00:50:08.679 --> 00:50:14.480
joy that gives. I think that's
actually suggests that there is a good future

794
00:50:14.480 --> 00:50:19.039
where even if we don't understand the
digital superintelligence, just as the Olympic system

795
00:50:19.079 --> 00:50:22.599
does not under understated the cortex,
it might still actually at the end of

796
00:50:22.599 --> 00:50:23.920
the day, if you're trying to
make our limbic system and cortex happy,

797
00:50:24.199 --> 00:50:27.519
is that a happy note to end? I think it's a great note to

798
00:50:27.599 --> 00:50:30.840
end on. Thank you, Thank
you everyone, Thank you athletes. So

799
00:50:30.880 --> 00:50:30.599
that's really amid