Sam Altman Unfiltered: ChatGPT, AI Risks and What’s Coming Next, 40 Questions in 50 Minutes

WEBVTT
1
00:00:01.800 --> 00:00:06.360
Welcome to another episode of Express Conversations, a podcast where
2
00:00:06.360 --> 00:00:10.759
eminent voices from the world of politics, governance, sports, business
3
00:00:10.880 --> 00:00:14.240
and arts talk to us about their life, work and
4
00:00:14.359 --> 00:00:17.839
issues that really matter. In this very special episode, we
5
00:00:17.879 --> 00:00:21.600
will hear from the CEO of Open Ai, Sam Altman.
6
00:00:22.039 --> 00:00:25.640
Our guest is in conversation with Ananthguenka of The Indian Express.
7
00:00:26.079 --> 00:00:30.440
In this exclusive interview, sam Altman addresses growing concerns about
8
00:00:30.519 --> 00:00:34.920
AI safety and job disruption, and whether artificial general intelligence
9
00:00:35.039 --> 00:00:37.880
is closer than we think. He also discusses the evolution
10
00:00:37.960 --> 00:00:41.640
of chat jubid, the competitive AI race, the responsibility of
11
00:00:41.719 --> 00:00:44.960
tech leaders in shaping powerful AI systems, and more.
12
00:00:50.719 --> 00:00:54.000
Sam, just looking at you, I feel like the chaos
13
00:00:54.039 --> 00:00:56.679
of India is getting to you. It's day two.
14
00:00:56.799 --> 00:00:57.840
That's been a busy two days.
15
00:00:59.399 --> 00:01:04.079
Very warm. Welcome to Express. Up there, guys, a very
16
00:01:04.120 --> 00:01:07.359
special way for us. Forbes magazine calls Opening I one
17
00:01:07.359 --> 00:01:09.959
of the most important companies in the world. If we
18
00:01:10.040 --> 00:01:14.319
look back at today twenty years from now, I don't
19
00:01:14.319 --> 00:01:15.920
think it will be an overstatement to say that the
20
00:01:15.920 --> 00:01:18.879
gentleman in front of me today is probably that one
21
00:01:18.920 --> 00:01:22.079
person who has made a tectonic shifts if anything close
22
00:01:22.120 --> 00:01:26.519
to the Industrial revolution, it probably is the gentleman in
23
00:01:26.519 --> 00:01:28.680
front of me right now. It all probably started with
24
00:01:28.799 --> 00:01:30.239
is that Is that a fair statement to make sense?
25
00:01:30.560 --> 00:01:35.599
No? You know, I think this is generally true for
26
00:01:35.640 --> 00:01:38.239
many companies. The person running the company gets way too
27
00:01:38.280 --> 00:01:40.799
much credit relative to the work everybody else does. But
28
00:01:41.040 --> 00:01:44.280
in our case, in particular, this has really been a
29
00:01:44.319 --> 00:01:47.120
story of a scientific discovery. There have been a handful
30
00:01:47.159 --> 00:01:52.519
of researchers that did a miracle and figured out a
31
00:01:52.599 --> 00:01:56.159
very deep thing about the way the world works. And
32
00:01:57.319 --> 00:02:00.879
the thing that is so special about deep learning is
33
00:02:00.920 --> 00:02:02.959
what has led it to be so general, which is
34
00:02:03.000 --> 00:02:07.840
that we figured out this small set of researchers figured
35
00:02:07.840 --> 00:02:13.479
out this algorithm that could learn anything, and at scale
36
00:02:13.960 --> 00:02:16.560
it gets better and better. And then a whole set
37
00:02:16.560 --> 00:02:19.800
of people have worked very hard to deliver that scale
38
00:02:19.840 --> 00:02:22.800
on these models, and then the whole world has figured
39
00:02:22.840 --> 00:02:25.400
out how to build incredible products and services around that.
40
00:02:26.479 --> 00:02:28.879
But more than anything, this is a story of the scientists.
41
00:02:29.840 --> 00:02:32.000
And when you've the last time you were in India
42
00:02:32.000 --> 00:02:35.159
to now, a lot has changed in the world. But
43
00:02:35.199 --> 00:02:37.680
what do you think has changed in India with respect
44
00:02:37.680 --> 00:02:38.280
to AI.
45
00:02:40.120 --> 00:02:42.080
Is just a little over a year ago that I
46
00:02:42.120 --> 00:02:45.919
was here, and before talking about what's changed in India,
47
00:02:45.960 --> 00:02:47.960
I think it's worth talking about what's changed in Ai
48
00:02:48.000 --> 00:02:52.360
because this will ground this. A year ago AI was
49
00:02:52.719 --> 00:02:56.680
just able to do high school math. And this was incredible.
50
00:02:57.960 --> 00:03:01.080
AI could do it, you know, a very good, not
51
00:03:01.120 --> 00:03:03.719
an excellent, but a very good job at high school mathematics.
52
00:03:04.360 --> 00:03:07.560
And people couldn't believe it. And I was in India
53
00:03:07.800 --> 00:03:10.000
speaking an event like this, and I was talking about it,
54
00:03:10.520 --> 00:03:14.000
and people were it had just happened. People genuinely felt
55
00:03:14.000 --> 00:03:15.759
wonder about it. They're like, you know, this can do
56
00:03:15.840 --> 00:03:18.919
what an eleventh grader can do. This is amazing because
57
00:03:18.960 --> 00:03:21.439
only a couple of years before that, you know, AI
58
00:03:21.520 --> 00:03:24.280
couldn't really like do any math at all. It couldn't
59
00:03:24.280 --> 00:03:27.599
do grade school math. It was struggling with that. By
60
00:03:27.680 --> 00:03:33.599
last summer it was competing at the hardest mathematics competitions
61
00:03:33.639 --> 00:03:36.400
we had in the in the world and doing okay.
62
00:03:37.280 --> 00:03:40.759
And last week there was a new thing called first proof,
63
00:03:41.039 --> 00:03:45.919
where mathematicians put out ten research problems that weren't the
64
00:03:45.960 --> 00:03:48.680
solutions to which were not known, and took our best
65
00:03:48.680 --> 00:03:51.919
mathematicians the cutting edge of research new knowledge to figure
66
00:03:51.960 --> 00:03:55.159
out because it hadn't Ye've been shared and I believe
67
00:03:55.360 --> 00:03:58.000
our latest day I got seven of those problems right.
68
00:03:58.639 --> 00:04:02.199
So AI has gone from doing okay at high school
69
00:04:02.199 --> 00:04:05.479
math to being able to do new research level mathematics
70
00:04:05.479 --> 00:04:08.520
figure out new knowledge. It's also happening in physics, and
71
00:04:08.599 --> 00:04:11.680
this is an amazing change. In a year. We've gone
72
00:04:11.719 --> 00:04:14.439
from AI that could do what we expected a very
73
00:04:14.479 --> 00:04:16.959
smart high school student to do to pushing the edge
74
00:04:16.959 --> 00:04:21.560
of human knowledge the other So that's change number one.
75
00:04:21.959 --> 00:04:24.279
The other change is that AI has gone from being
76
00:04:24.319 --> 00:04:28.519
able to write a little bit of programming to completely
77
00:04:28.600 --> 00:04:31.879
changing what it means to be a computer programmer. Last
78
00:04:31.920 --> 00:04:33.680
time I was here, people were amazed at these sort
79
00:04:33.680 --> 00:04:36.839
of autocomplete tools for code. And now if you type
80
00:04:36.839 --> 00:04:39.360
in an idea, one year later, if you type in
81
00:04:39.399 --> 00:04:41.800
an idea is something like Codex, you can have an
82
00:04:41.959 --> 00:04:47.360
entire application created. The job of a programmer is now.
83
00:04:48.079 --> 00:04:50.079
I think in this last year changed more than it's
84
00:04:50.160 --> 00:04:55.639
changed in any year that I've been an adult. Codex
85
00:04:55.800 --> 00:04:58.199
is our fat India is our fastest growing market for Codex,
86
00:04:58.279 --> 00:05:00.639
So especially what's happening here get to the next part
87
00:05:00.639 --> 00:05:04.920
of the question, is really amazing. The biggest change that
88
00:05:04.959 --> 00:05:06.439
I feel from last time I was here is when
89
00:05:06.480 --> 00:05:08.079
I was here last time, I felt like India was
90
00:05:08.120 --> 00:05:11.639
a consumer of AI. People were using these services, people
91
00:05:11.680 --> 00:05:16.680
were doing other stuff. And now the builder energy in
92
00:05:16.680 --> 00:05:18.079
India's off the charts.
93
00:05:18.600 --> 00:05:21.759
And you, in fact said something a little controversial last time.
94
00:05:21.839 --> 00:05:24.199
Now it's controversial because at that time you said, you know,
95
00:05:24.240 --> 00:05:26.199
you had painted a picture of despair that you know,
96
00:05:26.240 --> 00:05:28.920
a ten million dollar front would not be enough to
97
00:05:29.480 --> 00:05:33.040
you know, to build an NLM, a meaningful LM out
98
00:05:33.040 --> 00:05:35.480
of India. And now yesterday you're talking about a full
99
00:05:35.480 --> 00:05:37.600
stack coming out of India and India can actually lead
100
00:05:37.600 --> 00:05:37.920
an AI.
101
00:05:38.079 --> 00:05:41.519
So yeah, that comment, I mean, that got taken out
102
00:05:41.560 --> 00:05:45.360
of context. That was a few years ago. But the
103
00:05:45.360 --> 00:05:48.879
the comment was not that India couldn't do an amazing
104
00:05:49.319 --> 00:05:51.600
job with a ten million dollar company and a ten
105
00:05:51.639 --> 00:05:54.240
million dollar model. I think that clearly is the case.
106
00:05:54.680 --> 00:05:57.600
It was just could you make a very frontier model
107
00:05:57.600 --> 00:06:00.240
for ten million dollars? I think you couldn't. And I
108
00:06:00.240 --> 00:06:01.879
think it's even more true now that you can't.
109
00:06:02.759 --> 00:06:05.079
It's even it's even more true now that you can't.
110
00:06:04.839 --> 00:06:07.680
Make a frontier model for ten Yeah, that's they've gone
111
00:06:07.839 --> 00:06:13.199
very expensive. Yeah, but if India went with enough funding,
112
00:06:13.240 --> 00:06:15.920
for sure an India company could do that. And the
113
00:06:16.000 --> 00:06:18.759
smaller narrow models that India's doing are incredible.
114
00:06:19.480 --> 00:06:21.759
So is that is that surprising to you? Is anything
115
00:06:21.759 --> 00:06:23.800
that you surprise you between your last trip and this
116
00:06:23.879 --> 00:06:27.120
trip about India and AI that you didn't expect In
117
00:06:27.199 --> 00:06:27.680
terms of.
118
00:06:29.759 --> 00:06:31.759
That stat that I mentioned a couple of minutes ago
119
00:06:31.759 --> 00:06:33.839
that this is the fastest growing market for Codex in
120
00:06:33.879 --> 00:06:39.759
the world, I'm very happy. I guess I shouldn't be surprised,
121
00:06:39.759 --> 00:06:41.680
but it's it's great to see. I was at I
122
00:06:41.839 --> 00:06:44.279
T Deli this morning and the energy from people building
123
00:06:44.279 --> 00:06:47.480
this new generation of startups. That is a place where
124
00:06:47.480 --> 00:06:49.800
I expect India to surprise the world on the.
125
00:06:49.839 --> 00:06:52.959
Upside and the obviously, the other side of that coin
126
00:06:53.120 --> 00:06:54.920
is the disruption and the job especially in the it
127
00:06:55.600 --> 00:06:59.240
IT enterprise service sector. About eight percent of our GDP
128
00:06:59.319 --> 00:07:02.920
comes from that one space. And a lot has been
129
00:07:02.920 --> 00:07:06.759
said about this in the last ten days because of Codex.
130
00:07:06.839 --> 00:07:09.759
Before Codex was another claude and it just showed how
131
00:07:09.920 --> 00:07:11.879
fast software it's kind of just moved out. I know,
132
00:07:11.879 --> 00:07:13.360
a lot have been said about in the last week,
133
00:07:13.360 --> 00:07:16.160
But I still want you to reflect on that. How
134
00:07:16.240 --> 00:07:18.279
much of that is a threat for how shouldn't you
135
00:07:18.319 --> 00:07:18.759
think about that?
136
00:07:20.040 --> 00:07:21.879
I do think that is going to change a lot,
137
00:07:21.920 --> 00:07:25.079
and there's going to be a big impact there, and
138
00:07:26.120 --> 00:07:29.519
I think it is never helpful to pretend otherwise. However,
139
00:07:31.639 --> 00:07:35.800
what I expect to happen, as has happened with every
140
00:07:35.800 --> 00:07:39.360
other step forward in computer programming, is people will operate
141
00:07:39.439 --> 00:07:42.360
at a higher level of abstraction. They will be able
142
00:07:42.399 --> 00:07:47.560
to do more. There will still be a need for it. Broadly,
143
00:07:49.160 --> 00:07:52.319
the amount of product and code that's produced will increase
144
00:07:52.360 --> 00:07:55.439
a lot. The expectations will go up. But I think
145
00:07:55.480 --> 00:07:58.319
as long as the country and the companies that do
146
00:07:58.399 --> 00:08:01.720
this adapt to this new world, there will be plenty
147
00:08:01.759 --> 00:08:05.399
of new things to do. I'm I'm not a jobs
148
00:08:05.439 --> 00:08:07.000
doomer in the sense that I don't think there will
149
00:08:07.040 --> 00:08:08.920
be jobs for people in the future. I think there
150
00:08:08.920 --> 00:08:12.600
will be a lot of jobs. Every technological revolution has
151
00:08:12.720 --> 00:08:16.040
panicked about jobs going away, and every technological revolution has
152
00:08:16.040 --> 00:08:20.040
found new jobs. On the other side, the promised leisure
153
00:08:20.079 --> 00:08:22.360
has never happened. You know, this idea that we all
154
00:08:22.360 --> 00:08:25.399
get to go on permanent vacation that's never quite worked out.
155
00:08:25.399 --> 00:08:27.879
I don't think it will. I hope we get more
156
00:08:27.920 --> 00:08:32.279
in that direction. But our desire to be useful to
157
00:08:32.320 --> 00:08:35.080
each other, our desire for more and better stuff, our
158
00:08:35.080 --> 00:08:38.240
desire to express our creativity, we never kind of reach
159
00:08:38.320 --> 00:08:40.039
the end of that. I think there is no natural
160
00:08:40.080 --> 00:08:42.240
limit to that. There's a whole big universe out there
161
00:08:42.240 --> 00:08:43.960
to go explore if we can figure out, you know,
162
00:08:44.000 --> 00:08:45.720
space travel or whatever. I don't know where it's going to.
163
00:08:45.759 --> 00:08:47.440
Go, but.
164
00:08:50.039 --> 00:08:52.759
But I do think clearly things are going to evolve
165
00:08:52.960 --> 00:08:55.080
fast and you have to be on the right side
166
00:08:55.080 --> 00:08:55.320
of that.
167
00:08:55.960 --> 00:08:58.879
And you know that that statement about jobs is a
168
00:08:58.960 --> 00:09:01.279
very low di stabm in India because you know five
169
00:09:01.360 --> 00:09:06.879
hundred million people under the age of thirty. Uh, you
170
00:09:06.919 --> 00:09:09.799
understand the gravity of that of jobs and that and
171
00:09:09.840 --> 00:09:12.840
that that that anxiety that that India has. Do you
172
00:09:12.879 --> 00:09:14.679
discuss this when you when you meet the Prime minister,
173
00:09:14.759 --> 00:09:17.159
when you meet people who matter in policy, how much
174
00:09:17.200 --> 00:09:18.919
of that comes up and what are the what are
175
00:09:18.919 --> 00:09:21.759
the answers you have to that anxiety?
176
00:09:21.879 --> 00:09:24.679
Yeah, the main thing that comes up with political The
177
00:09:24.720 --> 00:09:30.840
main things that come up with political leaders are infrastructure, jobs,
178
00:09:32.120 --> 00:09:35.919
sort of fair distribution and benefits and safety broadly speaking,
179
00:09:38.080 --> 00:09:41.639
and those are the proportion of those is different in
180
00:09:41.679 --> 00:09:44.720
different countries, and also the level of focus on the
181
00:09:44.799 --> 00:09:48.279
upsides versus the potential downsides is different in different countries.
182
00:09:51.200 --> 00:09:57.399
We talk about jobs impact a lot. We one version
183
00:09:57.440 --> 00:10:00.360
of this that comes up in it feels like almost
184
00:10:00.360 --> 00:10:04.440
every conversation I have socially is what should my kids
185
00:10:04.480 --> 00:10:07.679
study for the future? Should my kid studies what's it
186
00:10:07.720 --> 00:10:09.600
going to look like? Like? This is this is a
187
00:10:09.679 --> 00:10:11.960
version of asking that question. You know, where where is
188
00:10:12.000 --> 00:10:18.039
the economy going to go? It's really hard to answer
189
00:10:18.120 --> 00:10:20.600
that specifically. If you study history. One thing I love
190
00:10:20.679 --> 00:10:23.600
to do is to read about the history of technology,
191
00:10:23.960 --> 00:10:26.720
and if you look at, say the primary source material
192
00:10:27.240 --> 00:10:31.600
of people that were experiencing the industrial evolution, there was
193
00:10:31.639 --> 00:10:33.799
a lot of panic about jobs. There were a lot
194
00:10:33.840 --> 00:10:35.840
of predictions about what the jobs were going to be,
195
00:10:36.519 --> 00:10:37.799
a lot of fear there was going to be no jobs,
196
00:10:37.799 --> 00:10:39.399
and a lot of prediction about what the new jobs
197
00:10:39.440 --> 00:10:42.320
would be. They were sort of shockingly wrong, and none
198
00:10:42.360 --> 00:10:43.759
of them were like I'm going to be the CEO
199
00:10:43.799 --> 00:10:47.399
of an AI company. Certainly none of them were like
200
00:10:47.399 --> 00:10:51.039
I'm going to be a YouTube influencer. So I think
201
00:10:51.080 --> 00:10:53.960
this is very hard to predict. But the skills that
202
00:10:54.039 --> 00:11:00.240
will work no matter what fluency with AI tools, resilience, adaptability,
203
00:11:00.320 --> 00:11:03.480
figuring out what you know people are gonna want, and
204
00:11:03.559 --> 00:11:06.000
how to be useful to them, how to work with
205
00:11:06.039 --> 00:11:08.639
other people. These these are all very good things to learn.
206
00:11:09.200 --> 00:11:14.200
And uh yeah, I think it's useless to pretend there
207
00:11:14.240 --> 00:11:16.799
won't be a big change. The change won't be as
208
00:11:16.840 --> 00:11:20.879
fast as some people in the AI industry predict because
209
00:11:21.039 --> 00:11:23.960
society has always got more inertia. It always takes longer.
210
00:11:24.480 --> 00:11:29.080
But eventually the change will be huge and we'll find
211
00:11:29.080 --> 00:11:30.200
all sorts of new things to do.
212
00:11:31.200 --> 00:11:32.519
I want to come back to that a little later
213
00:11:32.559 --> 00:11:34.720
about where that resistance you think will come from that
214
00:11:34.759 --> 00:11:37.440
inertia of society. Does it come more from the developed economies,
215
00:11:37.559 --> 00:11:41.759
from the from the underdeveloped economies? But before that, a
216
00:11:41.799 --> 00:11:44.639
little bit more about India and AI. That five layer
217
00:11:44.759 --> 00:11:49.600
cake that Jensen spoke about, the whole AI stack, the energy,
218
00:11:49.600 --> 00:11:52.440
the data centers, that chips, the models, and the applications
219
00:11:52.480 --> 00:11:55.279
that entire five layer cake. Which areas do you think
220
00:11:55.320 --> 00:11:58.120
India has the ability and the right to win in
221
00:11:58.679 --> 00:12:01.159
and which areas do you think it's a very risky
222
00:12:01.200 --> 00:12:02.879
bed for India to kind of make.
223
00:12:05.600 --> 00:12:10.200
I think India should you know, whether you subscribe to
224
00:12:10.240 --> 00:12:12.080
this it's a five layer cake or a seven layer cake.
225
00:12:12.120 --> 00:12:14.759
People to find it a little bit differently. But I
226
00:12:14.759 --> 00:12:18.159
think India should play at all of those levels. I
227
00:12:18.200 --> 00:12:24.639
think that there are there are important advantages to vertical
228
00:12:24.639 --> 00:12:29.320
integration and being able to work across all those levels
229
00:12:29.360 --> 00:12:31.879
for an economy of the size and importance of India's
230
00:12:32.159 --> 00:12:35.159
is a good thing to do. And you know, I
231
00:12:35.240 --> 00:12:37.200
talked with the Prime Minister just before coming here. He
232
00:12:37.279 --> 00:12:40.720
is certainly motivated to play at all of those levels.
233
00:12:40.759 --> 00:12:41.639
And I think it's important.
234
00:12:42.200 --> 00:12:43.960
And if we do play at all those levels, and
235
00:12:44.080 --> 00:12:46.360
you know, you're talking about more than seventy percent of
236
00:12:46.360 --> 00:12:49.000
the population already on the Internet. I don't know what
237
00:12:49.000 --> 00:12:52.519
percentage of that you estimate is doing artibision intelligence, but
238
00:12:52.879 --> 00:12:55.080
assuming we want a lot of that to move into
239
00:12:55.080 --> 00:12:57.840
the AIS domain, and that's a big change. Last time
240
00:12:57.960 --> 00:12:59.519
we heard a lot of fear and anxiety. But ye
241
00:12:59.559 --> 00:13:01.519
now we're about let's embrace it and let's kind of
242
00:13:01.840 --> 00:13:06.000
run with it. Does the world have the computing power
243
00:13:06.360 --> 00:13:09.559
it needs for India to become an AI first society.
244
00:13:10.759 --> 00:13:13.600
Not yet, but the world is rapidly going to have
245
00:13:13.639 --> 00:13:17.039
to work together to solve that problem. A question I'd
246
00:13:17.039 --> 00:13:20.600
love to ask people is how many GPUs would you
247
00:13:20.720 --> 00:13:23.200
like working for you all the time? How many GPUs
248
00:13:23.240 --> 00:13:26.840
would you personally like thinking about your problems, helping you
249
00:13:26.879 --> 00:13:29.960
with your work coming, you know, implementing your ideas, writing
250
00:13:30.000 --> 00:13:33.200
the software you want, operating the you know, humanoid robots
251
00:13:33.240 --> 00:13:35.279
you'll have in your home and garden and everything else,
252
00:13:35.320 --> 00:13:37.759
and you know, building new houses for you whatever, Like,
253
00:13:38.000 --> 00:13:41.480
how many do you want? And people give different answers.
254
00:13:42.039 --> 00:13:45.279
No one ever says less than one. Some people say
255
00:13:45.320 --> 00:13:49.759
a thousand. I would like a thousand. If you go
256
00:13:49.919 --> 00:13:52.919
multiply that out, I say, there's eight billion people in
257
00:13:52.960 --> 00:13:57.240
the world. We have no way to deliver eight trillion
258
00:13:57.279 --> 00:14:01.159
GPUs anytime soon. I mean, that just seems diculous. We're
259
00:14:01.159 --> 00:14:02.559
not going to do that, to be clear, We're not
260
00:14:02.559 --> 00:14:06.679
going to do that on Earth at least, But I
261
00:14:06.720 --> 00:14:09.320
do think it shows the ass It points to the
262
00:14:09.399 --> 00:14:11.720
level of ambition we need to have about how much
263
00:14:11.759 --> 00:14:14.879
compute capacity we're going to build out. And this will
264
00:14:14.919 --> 00:14:20.759
be the most expensive complex infrastructure project the world has
265
00:14:20.840 --> 00:14:24.759
ever collectively taken on for all of our welfare. So
266
00:14:25.080 --> 00:14:27.919
we're going to need a lot of compute capacity. The
267
00:14:28.000 --> 00:14:30.799
good news it will be impossible to build this out
268
00:14:30.840 --> 00:14:33.559
the old fashioned way, but we'll have AI and robots
269
00:14:33.600 --> 00:14:36.399
helping us do this, and you know, we'll figure it out.
270
00:14:37.200 --> 00:14:40.039
And is that why space keeps coming up nowadays? There's
271
00:14:40.039 --> 00:14:43.200
another space race, but that I keep yearning about.
272
00:14:43.399 --> 00:14:48.799
I honestly think the idea, with the current landscape of
273
00:14:48.879 --> 00:14:53.320
putting data centers in space is ridiculous. It will make
274
00:14:53.399 --> 00:14:58.559
sense someday, but if you just do like the very
275
00:14:58.639 --> 00:15:03.440
rough math of of launch costs relative to the cost
276
00:15:03.440 --> 00:15:05.679
of power we can do on Earth, to say nothing
277
00:15:05.679 --> 00:15:07.480
of how you're going to fix a broken GPU and
278
00:15:07.519 --> 00:15:11.159
space and they do break a lot still. Unfortunately we
279
00:15:11.200 --> 00:15:14.679
are not there yet. There will come a time. Space
280
00:15:14.759 --> 00:15:17.320
is great for a lot of things. Orbital data centers
281
00:15:17.320 --> 00:15:20.399
are not something that's going to matter at scale this decade.
282
00:15:21.080 --> 00:15:23.919
But then all this infrastructure build out sam it feels
283
00:15:24.000 --> 00:15:27.320
like it kind of forces you organizations like yours to
284
00:15:27.360 --> 00:15:30.639
have a very intimate relationship with government. And is that
285
00:15:30.720 --> 00:15:32.960
something that's kind of changing with AI? This whole power
286
00:15:33.000 --> 00:15:36.679
structure that Now. At one point, big tech was a
287
00:15:36.799 --> 00:15:41.399
very a very privatized force with a very complicated relationship
288
00:15:41.440 --> 00:15:43.320
with government. Now it seems to be that you need
289
00:15:43.360 --> 00:15:45.559
to have a good relationship government, whether it's Indian companies
290
00:15:45.559 --> 00:15:48.639
with Indian government or companies like yours with the US government.
291
00:15:49.480 --> 00:15:50.759
Is the government to be enabled.
292
00:15:50.480 --> 00:15:53.320
To Yeah, I think it's going to be really important,
293
00:15:53.360 --> 00:15:55.639
not just for building our infrastructure, but just given the
294
00:15:56.080 --> 00:15:58.360
level of impact this is going to have on society
295
00:15:58.799 --> 00:16:01.960
and the need to true dem pocratize this technology, governments
296
00:16:01.960 --> 00:16:03.879
are going to have to be involved and companies like
297
00:16:03.879 --> 00:16:05.440
ours are going to have to partner with governments.
298
00:16:05.720 --> 00:16:07.759
And how was that? I mean, have you ever thought
299
00:16:07.759 --> 00:16:10.600
of paused and thought about the evolution of that relationship
300
00:16:10.639 --> 00:16:14.080
between big tech and government? By they can I call
301
00:16:14.159 --> 00:16:15.480
you big tech or not? Really?
302
00:16:17.159 --> 00:16:20.399
I don't think most I don't feel like big tech. Okay,
303
00:16:21.320 --> 00:16:22.200
I don't think we're there.
304
00:16:22.120 --> 00:16:29.840
Yet, Okay, but tech companies like yours and government. I mean, I.
305
00:16:31.600 --> 00:16:39.000
Think the tech industry started out is this extremely libertarian,
306
00:16:39.240 --> 00:16:41.159
you know, we don't need the government, the government doesn't
307
00:16:41.200 --> 00:16:44.200
need us sort of the view that has changed a
308
00:16:44.240 --> 00:16:47.279
lot even before I like the last couple of decades
309
00:16:47.279 --> 00:16:49.519
as the companies got bigger and more central to the
310
00:16:49.559 --> 00:16:54.039
economy and the way the whole world works. That's changed significantly.
311
00:16:56.480 --> 00:16:59.360
But maybe never before has it been this important, just
312
00:16:59.440 --> 00:17:02.440
given the scale of the infrastructure that needs to happen.
313
00:17:02.799 --> 00:17:04.880
And how do you feel about that? Because we you know,
314
00:17:04.920 --> 00:17:06.720
one of the theories, one of the things we keep
315
00:17:06.720 --> 00:17:10.000
hearing is that this White House is very close to
316
00:17:10.000 --> 00:17:13.240
Silicon Valley. Silicon Valley was very close sponsor of JD. Vans.
317
00:17:13.319 --> 00:17:16.839
That whole relationship is a very intimate one has is
318
00:17:16.920 --> 00:17:19.519
that Is that true? Is that good? Is that bad?
319
00:17:20.680 --> 00:17:23.160
I would say close in some ways and not close
320
00:17:23.200 --> 00:17:26.119
in others. You know, there are some tight ties. And
321
00:17:26.160 --> 00:17:29.200
then also this administration has had some big criticisms of
322
00:17:29.240 --> 00:17:36.160
tech and is that good or bad? I think a
323
00:17:36.279 --> 00:17:40.240
close cooperation between tech companies and the government is going
324
00:17:40.279 --> 00:17:44.000
to become increasingly important over time. It obviously won't be
325
00:17:44.039 --> 00:17:46.839
a perfectly smooth relationship, but the better it can be,
326
00:17:46.960 --> 00:17:48.240
I think, the better for all of us.
327
00:17:49.079 --> 00:17:52.079
And what about government to government relationships. You know, what
328
00:17:52.079 --> 00:17:57.599
what God announced today was very interesting partnership of several
329
00:17:57.799 --> 00:17:59.920
countries who are thinking about AI and a very advanced
330
00:18:00.920 --> 00:18:04.920
Is it all these countries versus China. Is that what's
331
00:18:04.960 --> 00:18:08.799
kind of playing out? Or is it? Are we all
332
00:18:08.880 --> 00:18:11.119
kind of in a race on somewhere China is also
333
00:18:11.119 --> 00:18:11.759
in that race.
334
00:18:12.079 --> 00:18:14.440
I suspect it'll turn out to be way more complex
335
00:18:14.480 --> 00:18:18.279
than that. I think it will shift over time. They're
336
00:18:21.200 --> 00:18:25.720
I do sort of think that AI will become one
337
00:18:25.759 --> 00:18:28.000
of the most important political issues in the world, one
338
00:18:28.000 --> 00:18:32.799
of the highest order bits of geopolitical tension and cooperation.
339
00:18:34.279 --> 00:18:37.559
But I don't think it'll be a fixed thing. I
340
00:18:37.559 --> 00:18:40.640
think as it develops, and as I mean this has
341
00:18:40.680 --> 00:18:43.359
always happened that sort of political alliances shift over time.
342
00:18:43.759 --> 00:18:47.519
I expect them to shift more in the future, not less,
343
00:18:47.519 --> 00:18:48.440
as things move faster.
344
00:18:49.400 --> 00:18:52.839
But what is How do you think about China and
345
00:18:52.880 --> 00:18:54.880
all of this? I mean, you know, we keep hearing
346
00:18:54.920 --> 00:18:58.160
that China's leap years ahead, and China has got all
347
00:18:58.200 --> 00:19:00.279
this energy, and China has got all this renewal and energy,
348
00:19:00.279 --> 00:19:02.599
and China's got all these robotics and all this technology.
349
00:19:04.079 --> 00:19:06.799
Are we hearing the right thing? And why are they
350
00:19:06.920 --> 00:19:08.079
leapers ahead? What have they done?
351
00:19:08.119 --> 00:19:10.319
I would say very ahead in some areas and not
352
00:19:10.400 --> 00:19:12.839
ahead in others. I think this is like I mean
353
00:19:12.880 --> 00:19:15.960
that sounds like a boring statement. It's probably always always
354
00:19:16.039 --> 00:19:19.079
true in the world, like in terms of manufacturing physical robots,
355
00:19:19.400 --> 00:19:23.880
clearly ahead, a big edge on things like electric motors, magnets,
356
00:19:23.880 --> 00:19:30.839
stuff like that, clearly ahead on energy built out. So
357
00:19:30.920 --> 00:19:34.119
there are places where China is significantly ahead of other countries,
358
00:19:34.119 --> 00:19:36.359
and then there's places where I think we're ahead of them.
359
00:19:36.960 --> 00:19:41.039
And my guess is that that's sort of always what
360
00:19:41.079 --> 00:19:42.799
it's been like and what it will continue to be. Like,
361
00:19:43.039 --> 00:19:45.359
it's hard to be ahead on everything or behind on everything.
362
00:19:45.519 --> 00:19:47.519
Maybe if you have like the only super intelligence in
363
00:19:47.559 --> 00:19:49.200
the world, you could do it. I think that would
364
00:19:49.200 --> 00:19:50.480
actually be bad for other reasons.
365
00:19:51.160 --> 00:19:54.920
But is it a race, Sam, I mean it, we
366
00:19:55.039 --> 00:19:58.720
keep framing it as a race sort of. Is it
367
00:19:58.720 --> 00:20:01.319
a collaborative journey? Is it a yeah?
368
00:20:01.319 --> 00:20:03.200
I mean both. That's like the hard part of this, right,
369
00:20:03.200 --> 00:20:06.559
there's always there are always ways in which the global
370
00:20:06.559 --> 00:20:09.759
economy is interdependent and very collaborative, and then there's always
371
00:20:09.799 --> 00:20:15.759
ways in which it's a competition. And it's so tempting
372
00:20:15.799 --> 00:20:17.640
to say it's all one or the other. Reality is
373
00:20:17.680 --> 00:20:22.920
always messier than that. But I do think it's important
374
00:20:23.000 --> 00:20:27.599
to look at the net differences in power and the
375
00:20:27.640 --> 00:20:31.880
impact that could have. And I think it's you know,
376
00:20:32.559 --> 00:20:35.640
really important that democracy leads with AI.
377
00:20:36.640 --> 00:20:38.279
And when you say democracy leads with the air, you
378
00:20:38.279 --> 00:20:39.680
mean China should not lead with the A.
379
00:20:43.319 --> 00:20:45.839
It was a little bit more nuanced of a comment
380
00:20:45.880 --> 00:20:47.839
than that, but if you want to reduce.
381
00:20:47.559 --> 00:20:50.519
It to that, no, because in your sphys today you
382
00:20:50.559 --> 00:20:53.839
mentioned authoritarian you know, a couple of times. So I
383
00:20:53.920 --> 00:20:55.680
was just kind of, you know, how do we think
384
00:20:55.720 --> 00:20:57.880
about China, how do we unpack China as What I.
385
00:20:57.839 --> 00:20:59.839
Meant more is, I don't think there should be any
386
00:21:00.039 --> 00:21:02.880
single superintelligence in the world. I don't think there should
387
00:21:02.880 --> 00:21:04.839
be any one person, or any one country, or anyone
388
00:21:04.880 --> 00:21:08.039
company in charge of superintelligence, including the United States. I
389
00:21:08.039 --> 00:21:12.400
think that would also be bad. I think the world
390
00:21:12.440 --> 00:21:16.359
is at its best when power is widely distributed, when
391
00:21:16.400 --> 00:21:19.400
people have a lot of different ideas, and when there's
392
00:21:19.599 --> 00:21:23.640
enough of a balance in power that we can all
393
00:21:24.039 --> 00:21:27.240
we countries, individuals, companies, whatever, sort of keep each other
394
00:21:27.279 --> 00:21:30.359
in check and all have input and over time, even
395
00:21:30.400 --> 00:21:32.880
though we'll do some dumb things along the way, you know,
396
00:21:32.920 --> 00:21:34.920
people kind of get to pick the best ideas and
397
00:21:35.000 --> 00:21:38.680
let those rise up So what I meant by authoritarian
398
00:21:38.759 --> 00:21:42.880
here is you don't literally want one AI in charge
399
00:21:42.920 --> 00:21:45.640
of the world, no matter who has it. I think
400
00:21:45.640 --> 00:21:46.240
that would be bad.
401
00:21:46.519 --> 00:21:48.519
Yeah. And do you think if you compare to the
402
00:21:48.519 --> 00:21:52.440
big tech eral of seven years ago, AI is going
403
00:21:52.440 --> 00:21:55.000
to fragment power more or is it going to concentrate
404
00:21:55.039 --> 00:21:56.880
it more? Because it seems to be that big tech
405
00:21:56.920 --> 00:22:01.039
the old big tech is equally sort of staying. Is
406
00:22:01.119 --> 00:22:01.680
keeping space?
407
00:22:03.279 --> 00:22:05.680
That's a good, that is at some point, it's in
408
00:22:05.680 --> 00:22:07.680
some sense, I think one of the most important questions
409
00:22:07.680 --> 00:22:10.799
in front of us right now. You can totally imagine
410
00:22:10.839 --> 00:22:15.240
a world where AI massively concentrates power, where say a
411
00:22:15.279 --> 00:22:17.920
single company or country is able to hold onto AI
412
00:22:18.640 --> 00:22:22.640
and use that to a mass a gigantic amount of
413
00:22:22.640 --> 00:22:25.680
power and wealth in the world. You can imagine another
414
00:22:25.759 --> 00:22:29.839
extreme where everybody on Earth has a superintelligence with no
415
00:22:30.000 --> 00:22:33.759
rules whatsoever, and you know, some pretty terrible things in
416
00:22:33.839 --> 00:22:39.160
chaos happen. You can imagine many worlds there in the middle.
417
00:22:40.039 --> 00:22:43.960
I personally think something way more towards the democratized version
418
00:22:44.000 --> 00:22:46.720
of that is good. We of course will need some regulation,
419
00:22:46.839 --> 00:22:49.440
We of course will need some guardrails, But I think
420
00:22:49.480 --> 00:22:51.759
putting this technology in the hands of lots of people
421
00:22:52.240 --> 00:22:58.440
is possible, and that will be a decentralization of economic power.
422
00:22:58.799 --> 00:23:03.519
The clearest example of this you can see is in
423
00:23:03.559 --> 00:23:08.119
the last six and especially three months, the power that
424
00:23:08.160 --> 00:23:10.559
a one or two or three person company can have.
425
00:23:11.960 --> 00:23:16.119
This just didn't happen a few years ago. It was impossible.
426
00:23:17.079 --> 00:23:19.920
In this new world of things like Codex and general
427
00:23:19.960 --> 00:23:23.480
purpose knowledge working agents, we're going to have to get
428
00:23:23.559 --> 00:23:27.400
used very quickly to one person startups that have enormous
429
00:23:27.440 --> 00:23:31.000
amounts of success or two person, three person startups. And
430
00:23:31.240 --> 00:23:32.039
I think this is great.
431
00:23:32.359 --> 00:23:36.240
Yeah, And how competitive is how competitive is all of this?
432
00:23:36.880 --> 00:23:38.880
From your seat? I mean, you know, we saw a
433
00:23:38.880 --> 00:23:45.359
little awkward moment on stage yesterday. There's several memes about
434
00:23:45.359 --> 00:23:46.880
it going around, but you.
435
00:23:46.920 --> 00:23:51.039
Got to give the internet something to laugh at. It
436
00:23:51.119 --> 00:23:55.839
is definitely competitive, But I will say that it's.
437
00:23:55.640 --> 00:23:58.160
Also very incestuous. I mean, you guys, because all you know,
438
00:23:58.160 --> 00:24:00.119
a lot of people were building out stuff for work
439
00:24:00.200 --> 00:24:02.119
with you at one point, and everybody knows everybody, and
440
00:24:02.119 --> 00:24:04.240
everybody's like in videos, investing in you, You're buying in
441
00:24:04.279 --> 00:24:06.000
video chips, and then you know it's.
442
00:24:05.880 --> 00:24:08.599
Just this is it is a weirdly small world. For sure.
443
00:24:10.920 --> 00:24:14.279
I think it's very competitive commercially, but almost all of
444
00:24:14.319 --> 00:24:18.839
the efforts that have a very advanced model, I think
445
00:24:18.960 --> 00:24:21.920
fuel the gravity of what's happening, and they're very committed
446
00:24:21.960 --> 00:24:26.759
to get in safety in alignment right and willing to cooperate.
447
00:24:26.799 --> 00:24:30.200
There a little bit more about what happened yesterday, if
448
00:24:30.200 --> 00:24:32.480
you want to share, this is a good opportunity.
449
00:24:32.880 --> 00:24:34.519
And I don't really have that much more to add.
450
00:24:35.440 --> 00:24:37.440
I want to push it, okay. You know, there's just
451
00:24:37.440 --> 00:24:39.359
so much to cover, Sam, and I learned so much.
452
00:24:39.480 --> 00:24:41.680
I'm very grateful for you because I learned so much
453
00:24:42.240 --> 00:24:44.759
to prep for this conversation, I was like three days
454
00:24:44.759 --> 00:24:46.839
of just like knead deep into it. So I thought
455
00:24:46.839 --> 00:24:48.720
I'd just kind of break it up into a game. Great,
456
00:24:48.799 --> 00:24:50.200
and we can get you know, I can just try
457
00:24:50.200 --> 00:24:52.400
to cover more and more stuff rather than having a
458
00:24:52.400 --> 00:24:55.039
conversation about it, and we'll see what we kind of do.
459
00:24:55.440 --> 00:24:58.240
So it's not a rapid fire, because it's just too
460
00:24:58.279 --> 00:24:59.839
long to be a rapid fire. And I try to
461
00:24:59.839 --> 00:25:04.279
make could show up, but I couldn't. Is there any
462
00:25:04.319 --> 00:25:07.119
one thing you admire about Google's catch up in the
463
00:25:07.160 --> 00:25:09.519
AI race? Because ten years ago, they were out of
464
00:25:09.519 --> 00:25:10.039
the water.
465
00:25:11.240 --> 00:25:14.160
Well ten years ago. Ten years ago, Google was the
466
00:25:14.200 --> 00:25:20.839
only serious AI effort. Five years ago, maybe three years
467
00:25:20.839 --> 00:25:24.359
ago when we launched chatipt, they were then way out
468
00:25:24.359 --> 00:25:27.440
of it. But the first thing I admire about Google
469
00:25:27.960 --> 00:25:31.000
is demis and the Google team started working on AI
470
00:25:31.160 --> 00:25:33.759
before anyone else in the modern era with a lot
471
00:25:33.799 --> 00:25:37.920
of conviction, and uh, you know, I think without their inspiration,
472
00:25:38.039 --> 00:25:41.519
we certainly wouldn't be here. So that's one thing I admire.
473
00:25:41.640 --> 00:25:46.640
And then a second thing, more recently is their relentless
474
00:25:46.680 --> 00:25:49.960
focus and execution and ability to really scale model after
475
00:25:50.039 --> 00:25:52.480
being pretty far behind. It is quite impressive.
476
00:25:52.640 --> 00:25:56.599
Okay, were you surprised when Apple partnered with Google for Siri?
477
00:25:58.400 --> 00:26:02.240
Not particularly Okay, the one country that is by and
478
00:26:02.319 --> 00:26:05.680
large on the right path for regulation on AI.
479
00:26:09.519 --> 00:26:12.319
I don't think any of us know the answer to that. Yet.
480
00:26:12.519 --> 00:26:15.160
A thing that I'm happy about is countries are trying
481
00:26:15.160 --> 00:26:18.119
different approaches, and I think this is a good thing
482
00:26:18.119 --> 00:26:20.759
about many sovereign powers in the world. We will get
483
00:26:20.759 --> 00:26:24.240
to observe over the next few years very different approaches
484
00:26:24.759 --> 00:26:28.640
and see what works and what doesn't, and pretty quickly
485
00:26:28.640 --> 00:26:30.920
I believe the world will move towards more of what works.
486
00:26:30.960 --> 00:26:34.160
But I'm grateful for the experimentation and different countries trying
487
00:26:34.160 --> 00:26:34.759
different things.
488
00:26:34.960 --> 00:26:38.640
Okay, the short game that I'll give you some criticisms
489
00:26:38.680 --> 00:26:40.720
of AI and you give me a short defense of
490
00:26:40.759 --> 00:26:41.640
much of these criticisms.
491
00:26:41.640 --> 00:26:44.680
So the first might not have one For some, that's fine.
492
00:26:44.799 --> 00:26:47.000
And there's so many questions that you can always duck
493
00:26:47.680 --> 00:26:51.720
or pass. Too much concentration of power.
494
00:26:53.440 --> 00:26:56.759
There unless we push super hard to democratize, and I
495
00:26:56.759 --> 00:26:59.279
think everyone in the world needs to hold countries and
496
00:26:59.319 --> 00:27:02.759
companies to this threshold of You don't get to say
497
00:27:03.079 --> 00:27:05.119
concentration of power in the name of safety. We don't
498
00:27:05.119 --> 00:27:07.240
want that trade. It's got to be democratized.
499
00:27:07.640 --> 00:27:11.200
Okay. The amount of natural resources that are going into
500
00:27:11.240 --> 00:27:15.359
the data centers, the amount of water. The amount of water.
501
00:27:15.200 --> 00:27:18.480
Is totally fake. It used to be true. We used
502
00:27:18.519 --> 00:27:20.880
to do evaporative coolin in data centers, but now that
503
00:27:20.920 --> 00:27:22.960
we don't do that. You know, you see these like
504
00:27:23.000 --> 00:27:26.039
things on the internet where don't use Chatgypt it's seventeen
505
00:27:26.079 --> 00:27:29.200
gallons of water for each query or whatever. This is
506
00:27:29.279 --> 00:27:34.720
completely untrue, totally insane, no connection to reality. What is fair, though,
507
00:27:35.119 --> 00:27:38.720
is the energy consumption, not per query, but in total,
508
00:27:38.759 --> 00:27:41.359
because the world is now using so much AI is
509
00:27:41.680 --> 00:27:46.240
real and we need to move towards nuclear or wind
510
00:27:46.279 --> 00:27:47.319
and solar very quickly.
511
00:27:47.880 --> 00:27:50.240
There was a you know, we had Bill Gates last
512
00:27:50.319 --> 00:27:52.359
year and I asked him this question and here very
513
00:27:52.359 --> 00:27:55.559
interesting so today. At that time, apparently we calculated that
514
00:27:55.559 --> 00:27:58.519
there were ten iPhone worths of battery would go for
515
00:27:58.559 --> 00:28:01.960
every chatgipitquery, and now it's come down to one iPhone
516
00:28:01.960 --> 00:28:03.920
and one and a half iPhone battery life's worth of
517
00:28:04.319 --> 00:28:06.279
energy for every chagipity query. Is that correct?
518
00:28:06.279 --> 00:28:07.839
There's no way it's anything close.
519
00:28:07.640 --> 00:28:10.079
To that much. It's much less than that, way, way
520
00:28:10.119 --> 00:28:15.160
way less. Okay, all right. But his theory was that
521
00:28:15.160 --> 00:28:17.799
that the ais will learn from human evolution to be
522
00:28:17.839 --> 00:28:19.880
more efficient and how how much energy they consume.
523
00:28:22.960 --> 00:28:25.720
One of the things that is always unfair in this
524
00:28:25.759 --> 00:28:28.759
comparison is people talk about how much energy it takes
525
00:28:28.799 --> 00:28:32.079
to train an AI model relative to how much it
526
00:28:32.160 --> 00:28:35.039
costs a human to do one inference query. But it
527
00:28:35.079 --> 00:28:36.880
also takes a lot of energy to train a human.
528
00:28:36.920 --> 00:28:39.400
It takes like twenty years of life and all of
529
00:28:39.440 --> 00:28:42.319
the food you eat during that time before you get smart,
530
00:28:42.400 --> 00:28:44.880
and not only that, it took like the very widespread
531
00:28:44.920 --> 00:28:46.960
evolution of the one hundred billion people that have ever
532
00:28:47.039 --> 00:28:49.640
lived and learned not to get eaten by predators and
533
00:28:49.720 --> 00:28:52.279
learned how to figure out science and whatever to produce you.
534
00:28:52.559 --> 00:28:54.720
And then you took whatever you you know you took.
535
00:28:55.000 --> 00:28:58.839
So the fair comparison is if you ask Chatchiptia question
536
00:28:59.279 --> 00:29:01.319
how much energy does it take once its model is
537
00:29:01.319 --> 00:29:05.839
trained to answer that question versus a human, and probably
538
00:29:05.839 --> 00:29:08.359
AI has already caught up on an energy efficiency basis
539
00:29:08.359 --> 00:29:09.000
measured that way.
540
00:29:09.279 --> 00:29:13.279
Well, okay, fantastic AI is making my kids dumble.
541
00:29:17.559 --> 00:29:29.799
True for some kids. Look when I when I hear
542
00:29:29.920 --> 00:29:33.799
kids talk about AI, there are definitely some kids who
543
00:29:33.799 --> 00:29:37.519
are like, this is great. I cheated my way through
544
00:29:37.559 --> 00:29:40.200
all of high school. I never did any homework, thank you.
545
00:29:40.839 --> 00:29:42.880
And I'm like, what's your plan for like the rest
546
00:29:42.880 --> 00:29:45.799
of your life? And they're like, well, I assume I
547
00:29:45.799 --> 00:29:48.440
can like still use chattuby to do my job. This
548
00:29:48.519 --> 00:29:52.960
is very bad. We absolutely have to still teach our
549
00:29:54.079 --> 00:29:56.279
kids to learn and to think and to be creative
550
00:29:56.279 --> 00:29:59.599
and to use these tools. That's not what most kids say.
551
00:29:59.680 --> 00:30:04.519
Though most kids say, I can't believe what I can accomplish. Now,
552
00:30:05.599 --> 00:30:07.759
look at this thing that I've just made. I've built
553
00:30:07.759 --> 00:30:11.519
these incredible new workflows. You know, sure like I may
554
00:30:12.000 --> 00:30:14.400
use CHATGBT and the way you use Google when you
555
00:30:14.440 --> 00:30:17.160
were in high school to help with your homework. But
556
00:30:17.359 --> 00:30:19.519
now that I have this new tool, I'm doing these
557
00:30:19.519 --> 00:30:22.440
amazing things. I think we will need to find new
558
00:30:22.519 --> 00:30:25.400
ways to teach and evaluate in school to make sure
559
00:30:25.440 --> 00:30:29.039
every kid is brought along. But the potential of this technology,
560
00:30:29.400 --> 00:30:32.440
the ability to learn more, do more, I have no
561
00:30:32.519 --> 00:30:35.480
doubt about that. When I was in school, Google had
562
00:30:35.480 --> 00:30:40.640
just come out, and like my middle school teacher was like,
563
00:30:40.680 --> 00:30:42.880
this is the worst thing ever. You know, there's no
564
00:30:43.240 --> 00:30:46.640
point to teach in anymore. Why do you have to memorize,
565
00:30:46.759 --> 00:30:48.839
you know, the date that someone in history was born
566
00:30:49.039 --> 00:30:50.839
if you could just look it up on Google. And
567
00:30:50.880 --> 00:30:53.039
my answer was, I think it's a complete waste of
568
00:30:53.079 --> 00:30:55.160
time to memorize what year someone was born. I will
569
00:30:55.200 --> 00:30:56.640
just go look that up on Google if I ever
570
00:30:56.720 --> 00:30:59.279
need to know it again, which usually you don't that often.
571
00:30:59.759 --> 00:31:02.759
And then I watched over the next few years teachers
572
00:31:02.759 --> 00:31:06.240
come to peace with this, the education system evolve, and
573
00:31:06.839 --> 00:31:13.200
like always happens with new tools, the potential goes up
574
00:31:13.240 --> 00:31:15.799
and expectations go up, and we'll have to teach people
575
00:31:15.799 --> 00:31:18.640
to think harder, create more. And you know, I'm pretty
576
00:31:18.640 --> 00:31:21.119
sure that like a kid born today, when they're eighteen,
577
00:31:21.160 --> 00:31:22.920
graduating in high school will be able to do things
578
00:31:22.920 --> 00:31:25.720
that no one today can. And I think that's great.
579
00:31:26.640 --> 00:31:29.079
The other criticism is that it's just not a democratic enough,
580
00:31:29.119 --> 00:31:32.000
as in the people who want to pause AI don't
581
00:31:32.039 --> 00:31:35.400
have a voice. And I want to take this double
582
00:31:35.480 --> 00:31:38.680
click on this question, is that where do you see
583
00:31:38.680 --> 00:31:41.559
the resistance coming from. Is it coming from the people
584
00:31:41.559 --> 00:31:44.880
who have experienced technology at its highest levels and then
585
00:31:44.920 --> 00:31:46.880
they say we're nervous, you want to pause? Or do
586
00:31:46.880 --> 00:31:48.759
you see the resistance coming from those who have never
587
00:31:48.839 --> 00:31:51.720
experienced and are nervous to experience and then they feel
588
00:31:51.720 --> 00:31:52.920
the vulnerability.
589
00:31:55.319 --> 00:31:58.279
I mean it kind of comes from parts of it
590
00:31:58.319 --> 00:32:01.400
come from everywhere. Although as more people use the technology,
591
00:32:01.960 --> 00:32:04.960
I think there's less of a let's totally posit people
592
00:32:05.039 --> 00:32:08.720
like it, and there's more of a what is this
593
00:32:08.839 --> 00:32:10.799
going to mean? Can we have more of a voice?
594
00:32:11.200 --> 00:32:12.519
Can this go a little more slowly?
595
00:32:13.119 --> 00:32:17.400
Okay? The one thing Silicon Valley should learn from Chinese.
596
00:32:17.039 --> 00:32:19.000
Technology move fast.
597
00:32:19.839 --> 00:32:23.359
Yeah, the one thing Silicon Valley can learn from Indian technology,
598
00:32:24.400 --> 00:32:29.880
move faster. The one real and one imagined fear of
599
00:32:30.000 --> 00:32:31.359
Chinese dominance in AI.
600
00:32:36.240 --> 00:32:39.400
I think the imagined fear is that, you know, the
601
00:32:39.480 --> 00:32:42.400
Chinese military is going to send a billion humanoid robots
602
00:32:42.400 --> 00:32:44.920
marching through the streets and overpower the US army or
603
00:32:44.920 --> 00:32:49.519
something like that. And the real fear is and I
604
00:32:49.519 --> 00:32:52.440
think that's like an example of fighting the last war
605
00:32:52.480 --> 00:32:54.400
and thinking it's going to be the same. I think
606
00:32:54.440 --> 00:32:56.319
the real fear is a new kind of war between
607
00:32:56.400 --> 00:32:59.279
nations that gets fought over the Internet, where you influence
608
00:32:59.319 --> 00:33:01.799
people and hack into a bunch of critical infrastructure, and
609
00:33:01.880 --> 00:33:03.079
that feels totally possible.
610
00:33:03.559 --> 00:33:07.359
That's a very scary reality. What we say surveillance state,
611
00:33:07.799 --> 00:33:09.359
all those kind of things you don't.
612
00:33:09.559 --> 00:33:16.480
Super worried about that. I think increasingly so people are
613
00:33:16.599 --> 00:33:21.039
using fear of AI going wrong to justify a surveillance state.
614
00:33:22.079 --> 00:33:26.119
And I don't think that people doing that have fully
615
00:33:26.160 --> 00:33:27.880
thought through the downsides of the surveillance state.
616
00:33:28.240 --> 00:33:33.279
Ye, okay, you urgently need to look at You urgently
617
00:33:33.359 --> 00:33:35.799
need to look look something up. Your phone is discharged.
618
00:33:35.839 --> 00:33:38.519
You borrow my phone. The letter T on my type.
619
00:33:39.079 --> 00:33:41.200
My touchpad doesn't work, so you can only key in
620
00:33:41.279 --> 00:33:42.960
Grock Claude, Gemini, or deep Seek.
621
00:33:47.559 --> 00:33:49.160
I guess if I have to look something up, I
622
00:33:49.160 --> 00:33:49.799
picked Gemini.
623
00:33:50.200 --> 00:33:51.160
Okay, all right.
624
00:33:53.480 --> 00:33:54.960
For other things, I'd pick other models.
625
00:33:56.119 --> 00:33:58.440
The most compelling reason you have moved from a not
626
00:33:58.559 --> 00:34:01.519
for profit to a cap profit to pretty much any revenue,
627
00:34:01.519 --> 00:34:03.240
including ads.
628
00:34:06.000 --> 00:34:08.880
The need to democratize AI and the need to stay
629
00:34:08.920 --> 00:34:11.800
at the forefront of research. Both of those require huge
630
00:34:11.800 --> 00:34:12.599
amounts of capital.
631
00:34:13.719 --> 00:34:17.400
And do you think your research first company or a
632
00:34:17.440 --> 00:34:18.400
product first company?
633
00:34:18.559 --> 00:34:22.079
Research first company. The coolest thing about AI, I think
634
00:34:22.159 --> 00:34:25.400
is that the great majority of making a good product
635
00:34:25.480 --> 00:34:27.880
is doing good research. We make a great model, the
636
00:34:27.920 --> 00:34:31.480
product is pretty good. The early version of CHATGBT was
637
00:34:31.519 --> 00:34:34.239
almost no product at all. It was like a textbox
638
00:34:34.239 --> 00:34:35.960
that fed into a model and then put the output
639
00:34:36.039 --> 00:34:38.920
on the screen. So research to continue to push what
640
00:34:38.960 --> 00:34:41.719
a model is capable of is what enables everything else.
641
00:34:42.239 --> 00:34:46.199
Okay. The one criticism about you that bothers you when
642
00:34:46.199 --> 00:34:46.960
you hear it.
643
00:34:50.800 --> 00:34:54.199
I've heard basically like every possible criticism, and at some
644
00:34:54.320 --> 00:34:56.360
point none of them bother you because you're just like this,
645
00:34:56.519 --> 00:35:02.639
gotten so insane, so detached from truth. I guess the
646
00:35:02.679 --> 00:35:07.559
one that makes me the most sad is that like
647
00:35:07.840 --> 00:35:09.639
he's just in it for the power. He doesn't really
648
00:35:09.679 --> 00:35:11.480
care about sort of trying to do something that's gonna
649
00:35:11.480 --> 00:35:12.280
be helpful to the world.
650
00:35:13.119 --> 00:35:15.480
What what were you thinking when you when you decided
651
00:35:15.519 --> 00:35:18.639
not to take any equity and opening What was going
652
00:35:18.639 --> 00:35:19.159
on in your mind.
653
00:35:19.039 --> 00:35:21.159
To that that was truly one of the dumbest things.
654
00:35:21.159 --> 00:35:22.880
As I say, this is someone who's made a lot
655
00:35:22.880 --> 00:35:27.199
of dumb decisions. The conspiracy theories that that brought about,
656
00:35:27.840 --> 00:35:30.480
like the one I just mentioned, it was totally not
657
00:35:30.559 --> 00:35:33.440
worth it. We were like a nonprofit. I you know,
658
00:35:33.519 --> 00:35:35.559
there was like a kind of quirk of it where
659
00:35:35.840 --> 00:35:38.679
to be on the board of a nonprofit you had
660
00:35:38.679 --> 00:35:42.320
to like be sort of disinterested. And I had been
661
00:35:42.360 --> 00:35:44.360
like very fortunate to have a successful career, and I
662
00:35:44.360 --> 00:35:46.280
was like, I don't really care here either way. But
663
00:35:46.320 --> 00:35:47.159
it's a very dumb thing.
664
00:35:47.400 --> 00:35:49.639
And if somebody today were to offer and figure out
665
00:35:49.639 --> 00:35:51.079
a way for you to get into the cap table
666
00:35:51.119 --> 00:35:52.239
with cecutity, would you take it.
667
00:35:54.119 --> 00:35:56.920
I mean, at this point, I feel like so tired
668
00:35:56.960 --> 00:35:59.079
of the whole conversation and so trapped. I'm not sure
669
00:35:59.079 --> 00:36:01.360
what to do. It's like a lose lose thing for me.
670
00:36:01.440 --> 00:36:05.000
Okay, So one thing, in spite of all the differences
671
00:36:05.039 --> 00:36:07.519
you have with Elon Musk, the one thing you the
672
00:36:07.519 --> 00:36:10.280
one thing you admire about him, is there any one
673
00:36:10.320 --> 00:36:11.880
thing you can put all the different now I'm.
674
00:36:11.760 --> 00:36:17.000
Going to think of something, but give me a minute.
675
00:36:17.920 --> 00:36:24.360
He's extremely good at physical engineering and also extremely good
676
00:36:24.360 --> 00:36:27.079
at getting people to perform incredibly well at their jobs.
677
00:36:27.840 --> 00:36:32.800
Okay, super okay. Chagibiity is not known for its opinions.
678
00:36:32.840 --> 00:36:35.880
It's known for giving facts and factful answers back. So
679
00:36:35.920 --> 00:36:37.880
I'm going to ask your opinion on the news. There's
680
00:36:37.880 --> 00:36:39.519
been a lot of stuff in the news last week,
681
00:36:39.880 --> 00:36:42.840
so we want your views on would you support the
682
00:36:42.840 --> 00:36:45.960
government ordered ban on social media for kids under sixteen.
683
00:36:49.639 --> 00:36:51.559
I don't know if a total ban is right. Maybe
684
00:36:51.559 --> 00:36:55.760
it is. I think at least heavy restrictions would be good.
685
00:36:56.119 --> 00:37:01.119
I have a young child. I'm more interested in the
686
00:37:01.480 --> 00:37:04.280
what I think are the very negative impacts of infinite scroll.
687
00:37:05.920 --> 00:37:09.159
I would like my kid not to just be one
688
00:37:09.159 --> 00:37:11.320
of those iPad kids going like this all the time,
689
00:37:12.159 --> 00:37:16.239
So that seems very important. But yeah, probably social media
690
00:37:16.280 --> 00:37:18.960
for kids is something we have to be very careful.
691
00:37:18.719 --> 00:37:23.360
With Okay, it seems the clouds of war are again
692
00:37:23.519 --> 00:37:27.840
upon us. A lot of talk about in Iran if
693
00:37:27.960 --> 00:37:32.320
government were to ask to use AI companies like yours
694
00:37:32.760 --> 00:37:36.840
to help figure out how to get into war, how
695
00:37:36.880 --> 00:37:39.119
do you? How would you? How do you respond? Is
696
00:37:39.159 --> 00:37:42.119
Spentagon just another client for you? And you know they're
697
00:37:42.159 --> 00:37:44.159
you're the they're the customer. You do what they ask?
698
00:37:44.639 --> 00:37:45.920
Or do you draw a line somewhere?
699
00:37:49.800 --> 00:37:53.519
I don't think AI systems should be used to make
700
00:37:55.320 --> 00:37:57.559
AI systems should be used to make war fighting decisions.
701
00:37:57.599 --> 00:37:59.519
I don't think they're at a level of sophistication and
702
00:37:59.519 --> 00:38:02.679
reliability where this is a good idea that said, like,
703
00:38:02.920 --> 00:38:06.280
we certainly want to support the government, and there's a
704
00:38:06.320 --> 00:38:09.320
lot of things we can do already. Someday there will
705
00:38:09.360 --> 00:38:13.159
be I think applications of this really important applications of
706
00:38:13.239 --> 00:38:17.400
this to defense, but right now the models have clear limitations.
707
00:38:18.320 --> 00:38:20.840
It was used in for Maduro's capture. I believe that
708
00:38:21.000 --> 00:38:21.800
is that true? Is that.
709
00:38:23.199 --> 00:38:25.480
No? I just don't know, Like I'm sure it was
710
00:38:25.559 --> 00:38:28.079
used in some ways. There are things that AI can
711
00:38:28.320 --> 00:38:30.719
do a great job of today, Like I think using
712
00:38:30.800 --> 00:38:34.320
AI to analyze a huge amount of intelligence reports. Probably
713
00:38:34.320 --> 00:38:36.079
a great use of AI and maybe it was used
714
00:38:36.079 --> 00:38:36.639
some way like that.
715
00:38:36.880 --> 00:38:41.199
Yeah, your views on claud safety chief Nunk Sherma's dramatic resignation.
716
00:38:41.199 --> 00:38:42.519
I don't know if you followed that, but it made
717
00:38:42.519 --> 00:38:43.119
a lot of news.
718
00:38:43.239 --> 00:38:47.000
I heard something about it, like they've resigned in my
719
00:38:48.400 --> 00:38:50.159
it's all useless or something.
720
00:38:50.360 --> 00:38:51.800
Yeah, it was going to go to hell. I want
721
00:38:51.800 --> 00:38:54.840
to go write poetry. It was that that direction, and
722
00:38:55.199 --> 00:38:58.159
I lot a lot of safety people are getting I
723
00:38:58.159 --> 00:39:00.719
think overwhelmed. I mean that's the real piece I wanted
724
00:39:00.760 --> 00:39:01.360
you to reflect on.
725
00:39:04.719 --> 00:39:07.800
The part of it I agree with is the inside
726
00:39:07.880 --> 00:39:10.880
view at the companies of looking at what's going to happen,
727
00:39:11.000 --> 00:39:13.639
like the world is not prepared. We're going to have
728
00:39:13.760 --> 00:39:17.000
extremely capable models soon. It's going to be a faster
729
00:39:17.119 --> 00:39:20.960
takeoff than I originally thought. And that is stressful and
730
00:39:21.039 --> 00:39:21.920
anxiety inducing.
731
00:39:22.400 --> 00:39:25.719
Okay, reflect a bit on the future, the future of
732
00:39:25.800 --> 00:39:28.599
love and relationships. We keep hearing of people using AI
733
00:39:28.639 --> 00:39:34.519
agents to open up conversations on dating apps. How much
734
00:39:34.559 --> 00:39:37.039
of authenticity are we losing with love and relationships? So
735
00:39:37.039 --> 00:39:38.119
how much of does it come Atay?
736
00:39:39.000 --> 00:39:41.039
My debt would be that in the future we value
737
00:39:41.119 --> 00:39:44.280
human relationships much more. I think we are wired to
738
00:39:44.320 --> 00:39:46.360
do that. There will be some people who will like
739
00:39:47.559 --> 00:39:49.440
fall in love with an AI, or talk about their
740
00:39:49.440 --> 00:39:51.880
AI boyfriend or girlfriend, or have AI like you know,
741
00:39:52.199 --> 00:39:54.519
send messages on dating apps for them. Now, there will
742
00:39:54.559 --> 00:39:57.280
be some small percentage of the world that does that,
743
00:39:57.519 --> 00:39:59.639
and there will be a lot of breathless articles written
744
00:39:59.639 --> 00:40:02.760
about the end of society and all of that. But
745
00:40:02.800 --> 00:40:06.039
I think for almost everyone in a world with more abundance,
746
00:40:06.079 --> 00:40:07.920
in a world where you can kind of have anything,
747
00:40:08.679 --> 00:40:12.280
human connection, human attention, human warmth will be one of
748
00:40:12.360 --> 00:40:15.840
the most valuable commodities, and we will I bet we
749
00:40:15.840 --> 00:40:18.159
will care much more about that in a world with AI.
750
00:40:18.400 --> 00:40:21.480
Okay, you've invested in over four hundred companies. Is that
751
00:40:21.519 --> 00:40:22.800
I don't know if that's accurate.
752
00:40:23.039 --> 00:40:25.079
I mean County and y Common are like three thousand.
753
00:40:25.519 --> 00:40:28.199
Wow. Okay, if you take all the y Combinator investments.
754
00:40:28.239 --> 00:40:31.280
Fair But given the speed of AI growth and the disruption,
755
00:40:31.920 --> 00:40:34.480
where would you advise people in this room to invest today?
756
00:40:34.840 --> 00:40:36.360
Is the one place that you think they should.
757
00:40:39.239 --> 00:40:41.400
The cool thing about a moment like this, when the
758
00:40:41.440 --> 00:40:45.840
ground is shaken, is sort of everything is up for grabs.
759
00:40:46.599 --> 00:40:50.280
Things are going to change so fast. Things are changing
760
00:40:50.360 --> 00:40:53.480
so fast, as we talked about earlier, you can have
761
00:40:53.480 --> 00:40:55.760
these one person companies doing the things that used to
762
00:40:55.800 --> 00:40:58.760
take much bigger companies, but they can move faster, much
763
00:40:58.760 --> 00:41:01.400
more efficiently. So there will be all of the do
764
00:41:01.440 --> 00:41:04.119
the existing things better, and then there's a whole set
765
00:41:04.159 --> 00:41:08.679
of new things we just couldn't imagine before. It feels
766
00:41:08.719 --> 00:41:11.280
like the investment landscape is more open than it's ever
767
00:41:11.320 --> 00:41:15.039
been in any of my time as an adult. There
768
00:41:15.039 --> 00:41:16.719
were many times before where it felt like there was
769
00:41:16.760 --> 00:41:18.880
only one thing to invest in, there's only one frontier,
770
00:41:19.599 --> 00:41:21.320
and now it feels like you can kind of do
771
00:41:21.360 --> 00:41:22.639
whatever whatever you want.
772
00:41:23.400 --> 00:41:26.880
Okay, how far are we from Asi? You said Agi
773
00:41:27.000 --> 00:41:29.320
were a few years away? How far are we from Asi?
774
00:41:30.360 --> 00:41:32.280
No? I said from Asi were a few years away?
775
00:41:33.280 --> 00:41:34.679
So Agi how far then?
776
00:41:35.119 --> 00:41:38.559
I mean? Agi feels pretty close at this point. Like
777
00:41:38.639 --> 00:41:43.800
if you had asked me, I think, if you had
778
00:41:43.800 --> 00:41:48.159
asked most people six years ago, what would you think
779
00:41:48.280 --> 00:41:51.599
if we had systems that could do new research on
780
00:41:51.639 --> 00:41:53.840
their own? What would you think if we had systems
781
00:41:53.840 --> 00:41:56.440
that could make an entire complex computer program on their own,
782
00:41:56.840 --> 00:41:59.480
that could do pretty sophisticated knowledge work in all these
783
00:41:59.519 --> 00:42:01.880
different FeAs fields. You know, you could have one system
784
00:42:01.880 --> 00:42:05.480
that could act as an AI doctor, lawyer, computer scientist.
785
00:42:05.880 --> 00:42:08.800
Would say, Okay, that sounds pretty general and pretty intelligent.
786
00:42:10.239 --> 00:42:14.840
We get used to whatever we have, but just watching
787
00:42:14.880 --> 00:42:18.199
how much the technology we already have is accelerating us internally,
788
00:42:18.880 --> 00:42:22.559
I would say it's pretty close. And given what I
789
00:42:22.639 --> 00:42:24.920
now expect to be a faster takeoff, I think super
790
00:42:24.960 --> 00:42:26.400
intelligence is not that far off.
791
00:42:27.199 --> 00:42:38.440
The one thing you never ask CHADGPT, I.
792
00:42:38.440 --> 00:42:41.960
Think I would never ask it like how to be happy?
793
00:42:43.519 --> 00:42:45.360
I would rather ask like a wise person.
794
00:42:45.400 --> 00:42:51.199
Then that's a really interesting because that's really interesting because
795
00:42:51.880 --> 00:42:55.159
one of the most users off AI is for companionship,
796
00:42:55.360 --> 00:43:01.280
for you know, for that categorization of to the extent
797
00:43:01.360 --> 00:43:04.440
of therapy, and you know, it's yeah.
798
00:43:04.400 --> 00:43:06.960
I think like for sort of things like therapy or
799
00:43:07.000 --> 00:43:10.000
life advice, it can be pretty good. I think for
800
00:43:10.079 --> 00:43:13.079
like life philosophy, I'm still not gonna take it.
801
00:43:13.119 --> 00:43:18.000
Okay, okay, wonderful. So which is less likely to happen
802
00:43:18.239 --> 00:43:22.639
that the t s MC loses its monopoly over worldship
803
00:43:22.760 --> 00:43:26.840
manufacturing or musk? Can you become friends again? Which one
804
00:43:26.880 --> 00:43:31.039
is which one is less likely? More likely?
805
00:43:33.559 --> 00:43:35.039
I think, musk, can I become in your friends again
806
00:43:35.159 --> 00:43:36.880
is less likely. I feel like I have more control
807
00:43:36.880 --> 00:43:40.639
over that one, but both very unlikely things.
808
00:43:41.199 --> 00:43:42.840
What is it about d s MC that you know,
809
00:43:42.880 --> 00:43:48.159
the semiconductor semiconductor co operation? What is it about it
810
00:43:48.199 --> 00:43:50.320
that makes it so? I mean, the whole world is
811
00:43:50.400 --> 00:43:50.800
kind of.
812
00:43:50.719 --> 00:43:54.119
Just it's they have just this relentless focus on getting
813
00:43:54.159 --> 00:43:57.440
better at what they do. This is this incredibly complex
814
00:43:57.480 --> 00:44:01.400
process and they're just a machine at how they optimize
815
00:44:01.599 --> 00:44:03.280
and get better and learn it every level.
816
00:44:03.559 --> 00:44:07.000
Okay, the most recent update you can share with us
817
00:44:07.000 --> 00:44:09.559
about your partnership with Johnny. If you're talking about these
818
00:44:09.599 --> 00:44:12.400
new cool devices, it's going to change the form factor
819
00:44:12.440 --> 00:44:15.119
of a cell phone, how you interact with machines, anything
820
00:44:15.119 --> 00:44:16.119
you can share with us about that.
821
00:44:23.280 --> 00:44:25.760
We have basically used computers for the same way in
822
00:44:25.960 --> 00:44:28.639
for like fifty years. You know, I think the work
823
00:44:28.639 --> 00:44:31.400
that Xerox Park did to invent all the concepts of
824
00:44:31.440 --> 00:44:34.480
windows and pointing devices and everything that we now take
825
00:44:34.519 --> 00:44:37.719
for granted is amazing. Putting it on a phone and
826
00:44:37.760 --> 00:44:41.760
adding multi touch was also amazing, But fundamentally it's been
827
00:44:42.400 --> 00:44:45.320
you know, a similar kind of idea, and then AI
828
00:44:45.400 --> 00:44:48.079
came along, and AI is stuck in that same form factor.
829
00:44:48.079 --> 00:44:51.159
But AI is like a quite different thing. AI. You
830
00:44:51.199 --> 00:44:53.440
can now talk to a computer and natural language and
831
00:44:53.440 --> 00:44:56.079
it can do incredibly complex things for you. It can
832
00:44:56.159 --> 00:44:59.239
understand a huge amount of context, and the form factor
833
00:44:59.280 --> 00:45:02.719
of computers not quite work for this. You know, I want,
834
00:45:03.000 --> 00:45:05.559
I want to use a piece of technology that is
835
00:45:05.599 --> 00:45:07.800
observing my whole life and that has all the context
836
00:45:07.880 --> 00:45:10.280
and is maximally useful to me, and that has the
837
00:45:10.880 --> 00:45:13.480
that is in my way the least, and that can
838
00:45:13.639 --> 00:45:15.559
not just be honor off but these sort of subtle
839
00:45:15.559 --> 00:45:20.679
states in between, and we may fail at this. This
840
00:45:20.800 --> 00:45:22.960
is very hard, but I think we have a chance
841
00:45:23.000 --> 00:45:25.800
to make a new kind of products, a new family
842
00:45:25.800 --> 00:45:28.599
of products that we're really designed around AI, that are
843
00:45:28.679 --> 00:45:31.519
kind of participating in your life and not in the
844
00:45:31.519 --> 00:45:32.000
way of it.
845
00:45:32.639 --> 00:45:34.480
So that's the theory. But what is the product. I mean,
846
00:45:34.599 --> 00:45:38.159
you'll have to wait and see do we get when
847
00:45:38.159 --> 00:45:40.840
can we see this? Because it's sounding very revolutionary.
848
00:45:40.920 --> 00:45:46.079
So I mean, we hope to talk and be able
849
00:45:46.119 --> 00:45:48.880
to talk about it late this year. But okay, to
850
00:45:49.039 --> 00:45:50.039
hardware is hard.
851
00:45:50.199 --> 00:45:55.719
Okay, the one thing that governments should regulate and one
852
00:45:55.719 --> 00:45:57.079
thing they should avoid regulated.
853
00:46:02.239 --> 00:46:05.239
This is not a competent answer. Generally speaking, I think
854
00:46:05.239 --> 00:46:08.400
it's probably a good idea for governments to focus on
855
00:46:08.519 --> 00:46:13.599
regulating the really potentially catastrophic issues and being more lenient
856
00:46:13.840 --> 00:46:17.199
on the less important issues until we understand them better.
857
00:46:17.679 --> 00:46:21.519
Okay. Is there any one introspection or learning from your
858
00:46:21.679 --> 00:46:25.719
professional disagreements with any of the co founders you've had.
859
00:46:33.719 --> 00:46:44.320
It's hard to reduce that to a single SoundBite. One
860
00:46:44.320 --> 00:46:46.400
thing that Greg and I have disagreed on a lot
861
00:46:46.480 --> 00:46:49.119
over the years is how much the company should focus
862
00:46:49.199 --> 00:46:51.639
versus do a lot of things. And I think I've
863
00:46:51.639 --> 00:46:53.800
been wrong about trying to do so many things, and
864
00:46:53.840 --> 00:46:57.800
I've really learned from him the importance of extremely narrow
865
00:46:57.920 --> 00:46:58.599
deep focus.
866
00:46:58.840 --> 00:47:02.639
Okay, interesting one mistake, the biggest mistake you have seen
867
00:47:02.679 --> 00:47:04.760
corporates making in adopting AI.
868
00:47:12.480 --> 00:47:15.320
I was in a meeting yesterday yesterday with the big
869
00:47:15.320 --> 00:47:24.119
company who was planning to spend twenty twenty six strategizing
870
00:47:24.599 --> 00:47:27.519
twenty twenty seven getting the company ready in twenty twenty
871
00:47:27.519 --> 00:47:35.440
eight deploy And that may work for other kinds of technology. Apparently,
872
00:47:35.480 --> 00:47:38.119
if you do like a giant erp migration. That's the
873
00:47:38.199 --> 00:47:41.239
kind of timeline it takes. Doing that for AI will
874
00:47:41.239 --> 00:47:47.239
be a catastrophic mistake. The nimbleness required, the speed, the
875
00:47:47.320 --> 00:47:49.239
commitment required is just totally different.
876
00:47:49.559 --> 00:47:54.360
Okay, Okay, get to the last piece here. Following heads
877
00:47:54.360 --> 00:47:56.480
of state want your opinion on one thing they need
878
00:47:56.519 --> 00:47:59.599
to do for their people's future in this new AI
879
00:47:59.639 --> 00:48:05.119
content chi Jinping Putant Trump, what's the one thing you
880
00:48:05.159 --> 00:48:06.400
tell each of them that they should do.
881
00:48:10.079 --> 00:48:11.880
I would say the same thing to everyone, which as
882
00:48:11.920 --> 00:48:14.000
you've got to democratize this technology and put it in
883
00:48:14.039 --> 00:48:16.119
the hands of people. They're not all going to listen
884
00:48:16.159 --> 00:48:16.840
to that. Obviously.
885
00:48:21.199 --> 00:48:23.440
It's a fair answer, of course, but it's a safe answer.
886
00:48:23.440 --> 00:48:25.159
But is it Is it a different thing that you know,
887
00:48:25.280 --> 00:48:27.239
like a young country should do to an old country
888
00:48:27.239 --> 00:48:29.880
from a from a head of state perspective, I mean.
889
00:48:31.719 --> 00:48:37.480
I don't. I mean I feel increasingly radicalized on this point.
890
00:48:37.519 --> 00:48:39.599
I don't think any other strategy is going to work.
891
00:48:39.800 --> 00:48:45.320
Okay. The one if I can ask you to describe
892
00:48:45.360 --> 00:48:48.400
your relationship with Satia Nadella because he calls you, he
893
00:48:48.440 --> 00:48:52.039
calls Opening Eye a friend of me. Uh, I would.
894
00:48:51.880 --> 00:48:55.480
Just say friends Okay.
895
00:48:54.239 --> 00:48:58.320
Cool, your most expensive higher Ever, you don't have to
896
00:48:58.320 --> 00:49:00.639
tell me who, but how much money did it? Of course,
897
00:49:00.679 --> 00:49:01.880
because we.
898
00:49:02.559 --> 00:49:04.119
Are no comparison to matter here.
899
00:49:08.360 --> 00:49:11.960
Okay, okay, which is my last question? Now? Which are
900
00:49:11.960 --> 00:49:15.320
these four statements that you have you have made and
901
00:49:15.360 --> 00:49:19.039
you have now upgraded or revised. Do you regret Statement
902
00:49:19.079 --> 00:49:22.079
one that India building a foundation model with ten million
903
00:49:22.079 --> 00:49:25.119
dollars is hopeless? Statement two that you will remain a
904
00:49:25.159 --> 00:49:26.519
not for profit? Sorry, canarify that?
905
00:49:26.639 --> 00:49:30.480
Yeah? I said anyone building that point was anyone foundation model?
906
00:49:30.480 --> 00:49:32.840
Okay, okay, so that's Stapment one. Statement too, you will
907
00:49:32.880 --> 00:49:35.679
remain a not for profit. Statement three too much air
908
00:49:35.719 --> 00:49:39.199
regulation can stifle innovation. And statement for open a I
909
00:49:39.320 --> 00:49:40.599
won't accept advertising.
910
00:49:42.480 --> 00:49:45.239
I don't think I ever said we won't accept advertising. Sorry,
911
00:49:45.320 --> 00:49:47.320
I don't think I ever said we won't accept advertising.
912
00:49:47.320 --> 00:49:49.360
I think guys, I have misgivens. I guess I'll pick
913
00:49:49.400 --> 00:49:50.320
the nonprofit one.
914
00:49:50.400 --> 00:49:53.400
Okay, all right, super Sam, you're completely out of time. Time.
915
00:49:53.440 --> 00:49:55.119
Thank you so much, Thank you so much for your time,
916
00:49:55.159 --> 00:49:58.840
Thank you for being here. It's it's really really a
917
00:49:58.880 --> 00:50:00.000
privilege to have you with us.
918
00:50:02.320 --> 00:50:05.880
You were listening to Sam Altman on Express Conversations. If
919
00:50:05.880 --> 00:50:08.400
you like the show, then do subscribe to us wherever
920
00:50:08.480 --> 00:50:11.119
you get your podcasts. You can also recommend the show
921
00:50:11.159 --> 00:50:13.320
to someone you think we like it, share it with
922
00:50:13.360 --> 00:50:16.119
a friend or someone in your family. It's the best
923
00:50:16.199 --> 00:50:18.639
way for people to get to know about us. You
924
00:50:18.679 --> 00:50:21.559
can also tweet us at Express Podcasts and write to
925
00:50:21.679 --> 00:50:27.320
us at podcasts at Internet Express dot com.