TRANSKRIPT 🔗
00:00:00.133 --> 00:00:01.101
One more time with a
00:00:01.101 --> 00:00:02.002
little more energy.
00:00:02.002 --> 00:00:04.938
All right, more energy.
00:00:04.938 --> 00:00:06.172
Hello and welcome
00:00:06.172 --> 00:00:09.142
to this week’s epi…
00:00:09.142 --> 00:00:10.777
That was more energy, I felt it.
00:00:10.777 --> 00:00:11.244
You felt it?
00:00:11.244 --> 00:00:11.644
Go with it.
00:00:11.644 --> 00:00:14.647
All right.
00:00:26.993 --> 00:00:27.527
Hello
00:00:27.527 --> 00:00:28.661
and welcome to this week's
00:00:28.661 --> 00:00:30.764
episode of Inside Blackstone,
00:00:30.764 --> 00:00:31.931
where we bring together
00:00:31.931 --> 00:00:32.932
data and insights
00:00:32.932 --> 00:00:34.501
from across our portfolio
00:00:34.501 --> 00:00:35.702
to give you a fuller picture
00:00:35.702 --> 00:00:36.970
of what's happening on the ground
00:00:36.970 --> 00:00:38.538
in the real economy.
00:00:38.538 --> 00:00:39.739
I'm Christine Anderson,
00:00:39.739 --> 00:00:41.041
Global Head of Corporate Affairs,
00:00:41.041 --> 00:00:42.675
and here's the lineup.
00:00:42.675 --> 00:00:43.710
Our Monday Morning Meeting
00:00:43.710 --> 00:00:45.245
guest is my friend Gilles Dellaert,
00:00:45.245 --> 00:00:46.112
who leads our Credit
00:00:46.112 --> 00:00:47.480
& Insurance business.
00:00:47.480 --> 00:00:48.181
Then it's over
00:00:48.181 --> 00:00:49.115
to the Economic Weather
00:00:49.115 --> 00:00:50.884
Report with Winfield Sickles,
00:00:50.884 --> 00:00:51.518
and next
00:00:51.518 --> 00:00:53.153
my wide-ranging conversation
00:00:53.153 --> 00:00:54.254
with Eric Schmidt,
00:00:54.254 --> 00:00:56.322
CEO & Chair of Relativity
00:00:56.322 --> 00:00:59.092
Space, and former CEO of Google.
00:00:59.092 --> 00:01:00.093
And finally,
00:01:00.093 --> 00:01:01.594
a debrief with Jas Khaira,
00:01:01.594 --> 00:01:02.662
the Head of our AI
00:01:02.662 --> 00:01:04.798
investing platform, N1.
00:01:10.270 --> 00:01:11.071
Gilles, thank you so
00:01:11.071 --> 00:01:12.005
much for joining us.
00:01:12.005 --> 00:01:13.206
Thank you for having me.
00:01:13.206 --> 00:01:13.940
You and I spent
00:01:13.940 --> 00:01:14.541
a lot of time together
00:01:14.541 --> 00:01:15.408
at the beginning of this year
00:01:15.408 --> 00:01:16.209
when private credit
00:01:16.209 --> 00:01:17.243
was just in the
00:01:17.243 --> 00:01:17.777
news
00:01:17.777 --> 00:01:18.845
constantly
00:01:18.845 --> 00:01:20.080
driving the headlines.
00:01:20.080 --> 00:01:21.381
It seems to have quieted down
00:01:21.381 --> 00:01:21.815
quite a bit.
00:01:21.815 --> 00:01:22.916
What are you seeing?
00:01:22.916 --> 00:01:24.150
Yeah, the headlines were loud
00:01:24.150 --> 00:01:25.118
and they were everywhere
00:01:25.118 --> 00:01:26.052
at the start of the year.
00:01:26.052 --> 00:01:26.486
I think
00:01:26.486 --> 00:01:27.887
things have definitely quieted down.
00:01:27.887 --> 00:01:29.289
You're seeing less of it.
00:01:29.289 --> 00:01:30.256
I think it's in large part
00:01:30.256 --> 00:01:31.091
because the performance
00:01:31.091 --> 00:01:32.325
has held up a lot better
00:01:32.325 --> 00:01:33.760
than the kind of doom predictions
00:01:33.760 --> 00:01:35.161
that were being made at the time.
00:01:35.161 --> 00:01:37.297
Another dynamic playing out is AI.
00:01:37.297 --> 00:01:38.364
What are you seeing?
00:01:38.364 --> 00:01:39.432
Well, we're living through
00:01:39.432 --> 00:01:41.401
this unprecedented build-out
00:01:41.401 --> 00:01:42.368
across the globe.
00:01:42.368 --> 00:01:43.369
And with that
00:01:43.369 --> 00:01:44.838
come really large
00:01:44.838 --> 00:01:47.006
financing opportunities.
00:01:47.006 --> 00:01:49.242
The projects are substantial
00:01:49.242 --> 00:01:50.710
and they touch every part
00:01:50.710 --> 00:01:51.711
of the supply chain,
00:01:51.711 --> 00:01:52.645
if you will, all the way
00:01:52.645 --> 00:01:53.279
from the power
00:01:53.279 --> 00:01:53.780
that fuels
00:01:53.780 --> 00:01:55.014
the data centers to the equipment
00:01:55.014 --> 00:01:55.915
that goes in, the chips,
00:01:55.915 --> 00:01:57.417
the cooling, the services required
00:01:57.417 --> 00:01:58.852
and so we're active across
00:01:58.852 --> 00:02:01.020
all parts of that as a lender
00:02:01.020 --> 00:02:02.021
because a lot of financing
00:02:02.021 --> 00:02:03.056
dollars are required.
00:02:03.056 --> 00:02:05.692
Why chips and energy specifically?
00:02:05.692 --> 00:02:06.292
I would say
00:02:06.292 --> 00:02:09.295
power and compute right now are where
00:02:09.696 --> 00:02:11.131
things are most constrained
00:02:11.131 --> 00:02:12.132
and where
00:02:12.132 --> 00:02:13.032
things are most constrained,
00:02:13.032 --> 00:02:13.366
that's what
00:02:13.366 --> 00:02:14.701
we tend to run towards. Right.
00:02:14.701 --> 00:02:15.101
Because that's
00:02:15.101 --> 00:02:15.702
where the
00:02:15.702 --> 00:02:17.437
supply-demand imbalance is largest.
00:02:17.437 --> 00:02:18.338
That's where it creates
00:02:18.338 --> 00:02:19.973
opportunities for us to lean in
00:02:19.973 --> 00:02:21.374
and provide private solutions
00:02:21.374 --> 00:02:23.376
to borrowers to help fuel
00:02:23.376 --> 00:02:24.511
the CapEx build-out
00:02:24.511 --> 00:02:26.079
and the growth that will come. And
00:02:26.079 --> 00:02:27.280
I heard you say yesterday,
00:02:27.280 --> 00:02:28.715
it's not just an AI story,
00:02:28.715 --> 00:02:30.083
there's something broader going on.
00:02:30.083 --> 00:02:31.050
Yeah, I think it's broader.
00:02:31.050 --> 00:02:31.284
I think
00:02:31.284 --> 00:02:32.952
the global economy is growing.
00:02:32.952 --> 00:02:33.486
With that
00:02:33.486 --> 00:02:34.854
come needs from companies
00:02:34.854 --> 00:02:37.557
to finance that growth and the CapEx
00:02:37.557 --> 00:02:38.858
that is associated with it,
00:02:38.858 --> 00:02:40.193
as well as the strategic M&A
00:02:40.193 --> 00:02:41.661
Load full transcript
that might be associated with it.
00:02:41.661 --> 00:02:43.296
And we're active there
00:02:43.296 --> 00:02:44.531
with industrial companies,
00:02:44.531 --> 00:02:45.131
with companies
00:02:45.131 --> 00:02:46.733
that are in the aviation industry,
00:02:46.733 --> 00:02:47.233
the companies
00:02:47.233 --> 00:02:48.368
in the telecom industry.
00:02:48.368 --> 00:02:50.470
So, it's a much broader playbook.
00:02:50.470 --> 00:02:52.605
It's global and it's sizable.
00:02:52.605 --> 00:02:53.339
And all of
00:02:53.339 --> 00:02:54.474
it has one thing in common,
00:02:54.474 --> 00:02:56.643
which is we're backed by real assets
00:02:56.643 --> 00:02:57.076
that offer
00:02:57.076 --> 00:02:58.478
some form of inflation protection,
00:02:58.478 --> 00:02:59.913
which we really like in this
00:02:59.913 --> 00:03:01.247
economic environment.
00:03:01.247 --> 00:03:02.515
Well, exciting time in your business.
00:03:02.515 --> 00:03:03.650
Thanks for joining us. Very much so.
00:03:03.650 --> 00:03:04.684
Really appreciate it,
00:03:04.684 --> 00:03:05.885
and now over to Winfield
00:03:05.885 --> 00:03:07.353
and the Economic Weather Report.
00:03:11.357 --> 00:03:12.559
Thanks, Christine.
00:03:12.559 --> 00:03:14.460
It's Tuesday, September 29th,
00:03:14.460 --> 00:03:16.129
and there is a lot going on
00:03:16.129 --> 00:03:17.363
with major central bank
00:03:17.363 --> 00:03:19.566
meetings, the US-China Summit,
00:03:19.566 --> 00:03:21.034
and, as everybody
00:03:21.034 --> 00:03:22.635
that was stuck in traffic knows,
00:03:22.635 --> 00:03:23.970
the United Nations General
00:03:23.970 --> 00:03:25.605
Assembly last week.
00:03:25.605 --> 00:03:26.839
As I mentioned,
00:03:26.839 --> 00:03:27.774
when I was on air
00:03:27.774 --> 00:03:29.242
with Christine last week,
00:03:29.242 --> 00:03:30.176
the Fed recently
00:03:30.176 --> 00:03:31.377
delivered its first rate
00:03:31.377 --> 00:03:32.979
hike since 2023
00:03:32.979 --> 00:03:34.681
with a unanimous vote.
00:03:34.681 --> 00:03:35.815
In his press conference,
00:03:35.815 --> 00:03:36.883
Chairman Warsh cited
00:03:36.883 --> 00:03:38.418
robust economic growth,
00:03:38.418 --> 00:03:39.919
competition for capital,
00:03:39.919 --> 00:03:41.120
and geopolitics
00:03:41.120 --> 00:03:43.423
as the driver of long-term rates.
00:03:43.423 --> 00:03:44.257
Treasury yields
00:03:44.257 --> 00:03:45.358
continued to move higher
00:03:45.358 --> 00:03:46.593
last week on concerns
00:03:46.593 --> 00:03:48.394
of prolonged Middle East conflict
00:03:48.394 --> 00:03:49.395
with the associated
00:03:49.395 --> 00:03:50.730
commodity inflation,
00:03:50.730 --> 00:03:51.798
as well as strong
00:03:51.798 --> 00:03:53.800
economic data and continued
00:03:53.800 --> 00:03:56.169
AI infrastructure-related borrowing.
00:03:56.169 --> 00:03:57.103
The five-year US
00:03:57.103 --> 00:03:58.938
Treasury Yield moved above 5%
00:03:58.938 --> 00:03:59.606
for the first time
00:03:59.606
--> 00:04:00.940
in nearly 20 years,
00:04:00.940 --> 00:04:02.675
and then globally, government bonds
00:04:02.675 --> 00:04:04.677
reached an average yield of 4%,
00:04:04.677 --> 00:04:05.845
which was a post-global
00:04:05.845 --> 00:04:07.680
financial crisis high.
00:04:07.680 --> 00:04:09.182
Despite this volatility
00:04:09.182 --> 00:04:10.083
in treasuries,
00:04:10.083 --> 00:04:11.651
AI continues to underpin
00:04:11.651 --> 00:04:13.686
markets as adoption broadens.
00:04:13.686 --> 00:04:16.623
Meta shares jumped roughly 13% after
00:04:16.623 --> 00:04:17.590
launching its new
00:04:17.590 --> 00:04:18.958
AI agent, Muse,
00:04:18.958 --> 00:04:19.726
which already has
00:04:19.726 --> 00:04:20.860
millions of downloads
00:04:20.860 --> 00:04:21.794
and really highlights
00:04:21.794 --> 00:04:23.529
how AI is starting to move
00:04:23.529 --> 00:04:24.497
from the infrastructure
00:04:24.497 --> 00:04:27.233
build-out stage to real world usage.
00:04:27.233 --> 00:04:28.534
We're seeing that shift
00:04:28.534 --> 00:04:29.502
firsthand across
00:04:29.502 --> 00:04:31.137
multiple parts of the Blackstone
00:04:31.137 --> 00:04:31.971
ecosystem.
00:04:31.971 --> 00:04:33.573
Anthropic spend across
00:04:33.573 --> 00:04:35.041
our portfolio companies,
00:04:35.041 --> 00:04:36.943
borrowers, and GP Stakes
00:04:36.943 --> 00:04:37.644
PortCos
00:04:37.644 --> 00:04:39.679
grew approximately 27 times
00:04:39.679 --> 00:04:41.447
over the past 12 months.1
00:04:41.447 --> 00:04:43.016
This dynamic is creating
00:04:43.016 --> 00:04:45.451
winners across the AI value chain,
00:04:45.451 --> 00:04:46.886
with AMD recently
00:04:46.886 --> 00:04:47.920
becoming the fourth
00:04:47.920 --> 00:04:48.955
US chipmaker
00:04:48.955 --> 00:04:51.958
to surpass a $1 trillion market cap.
00:04:52.292 --> 00:04:54.093
Expanding use cases,
00:04:54.093 --> 00:04:54.861
this rising
00:04:54.861 --> 00:04:56.896
utilization and productivity
00:04:56.896 --> 00:04:58.164
gains continue
00:04:58.164 --> 00:04:58.931
to accelerate
00:04:58.931 --> 00:04:59.799
AI adoption
00:04:59.799 --> 00:05:01.000
and really reinforce
00:05:01.000 --> 00:05:02.902
this powerful investment cycle,
00:05:02.902 --> 00:05:05.171
as well as the deployment opportunity
00:05:05.171 --> 00:05:07.073
for so many of our businesses
00:05:07.073 --> 00:05:08.541
and funds.
00:05:08.541 --> 00:05:09.676
That’s it for this week’s
00:05:09.676 --> 00:05:11.077
Economic Weather Report.
00:05:11.077 --> 00:05:11.811
The week ahead
00:05:11.811 --> 00:05:13.112
brings jobs and inflation
00:05:13.112 --> 00:05:14.881
data and much more.
00:05:14.881 --> 00:05:16.249
Until next week,
00:05:16.249 --> 00:05:17.650
good luck out there.
00:05:17.650 --> 00:05:19.652
Now over to Christine’s conversation
00:05:19.652 --> 00:05:20.687
with Eric Schmidt.
00:05:27.493 --> 00:05:29.829
Eric Schmidt has had a front row seat
00:05:29.829 --> 00:05:31.898
to nearly every major technology
00:05:31.898 --> 00:05:33.966
shift of the last three decades.
00:05:33.966 --> 00:05:35.568
You know him from running Google
00:05:35.568 --> 00:05:37.003
and since then he's become
00:05:37.003 --> 00:05:38.271
one of the leading voices
00:05:38.271 --> 00:05:39.539
shaping the conversation
00:05:39.539 --> 00:05:41.407
around artificial intelligence.
00:05:41.407 --> 00:05:43.743
He is CEO of Relativity Space,
00:05:43.743 --> 00:05:44.544
giving him yet
00:05:44.544 --> 00:05:45.712
another vantage point on
00:05:45.712 --> 00:05:47.313
how breakthrough technologies
00:05:47.313 --> 00:05:49.816
move from idea to reality.
00:05:49.816 --> 00:05:52.819
Eric, welcome to Inside Blackstone. Oh,
00:05:52.852 --> 00:05:53.219
thank you.
00:05:53.219 --> 00:05:53.986
Thank you very much.
00:05:53.986 --> 00:05:54.187
Well,
00:05:54.187 --> 00:05:54.420
of course,
00:05:54.420 --> 00:05:55.988
I love the founder
00:05:55.988 --> 00:05:57.223
and the firm and everything
00:05:57.223 --> 00:05:57.924
you guys are doing,
00:05:57.924 --> 00:05:59.392
so. You're a technologist,
00:05:59.392 --> 00:06:00.660
you're a computer scientist.
00:06:00.660 --> 00:06:02.128
You give these incredible
00:06:02.128 --> 00:06:03.429
technical talks.
00:06:03.429 --> 00:06:04.831
I think your superpower
00:06:04.831 --> 00:06:06.165
is that you're able to break down
00:06:06.165 --> 00:06:07.667
complex ideas
00:06:07.667 --> 00:06:09.235
and make them more accessible.
00:06:09.235 --> 00:06:10.169
And that's ultimately
00:06:10.169 --> 00:06:11.204
what we're trying to do here. Well,
00:06:11.204 --> 00:06:11.604
thank you.
00:06:11.604 --> 00:06:12.739
It’s amazing
00:06:12.739 --> 00:06:13.873
the breadth of things
00:06:13.873 --> 00:06:14.974
that you've done in your career,
00:06:14.974 --> 00:06:16.342
that you are still doing.
00:06:16.342 --> 00:06:17.210
I'm not dead yet.
00:06:17.210 --> 00:06:18.611
You're not dead yet by any means.
00:06:18.611 --> 00:06:19.445
And I guess that's
00:06:19.445 --> 00:06:20.780
my question is,
00:06:21.114 --> 00:06:22.448
you're 71.
00:06:22.448 --> 00:06:23.816
When most people achieve
00:06:23.816 --> 00:06:25.051
the kind of success you have,
00:06:25.051 --> 00:06:26.085
or not even close,
00:06:26.085 --> 00:06:28.187
they spend the rest of their life
00:06:28.187 --> 00:06:29.155
sort of talking about it
00:06:29.155 --> 00:06:30.423
and what they accomplished.
00:06:30.423 --> 00:06:32.525
You keep reinventing yourself.
00:06:32.525 --> 00:06:34.827
I can't seem to stop.
00:06:34.827 --> 00:06:36.829
My own view is as you get older,
00:06:36.829 --> 00:06:38.765
you should take on more risk
00:06:38.765 --> 00:06:40.900
because you have less to lose, right?
00:06:40.900 --> 00:06:41.501
Right.
00:06:41.501 --> 00:06:43.102
And so, you might as well try
00:06:43.102 --> 00:06:44.137
the things that you thought
00:06:44.137 --> 00:06:45.004
were impossible.
00:06:45.004 --> 00:06:47.306
And if you fail, well, it's
okay.
00:06:47.306 --> 00:06:48.374
You know, you tried.
00:06:48.374 --> 00:06:49.342
And as long as you're
00:06:49.342 --> 00:06:51.611
trying in service of the ethics
00:06:51.611 --> 00:06:53.246
and the world that you care about.
00:06:53.246 --> 00:06:54.547
I care about democracy
00:06:54.547 --> 00:06:56.416
and I care about freedom,
00:06:56.416 --> 00:06:57.617
that's what I wanna work on.
00:06:57.617 --> 00:06:59.619
At this point in my own life,
00:06:59.619 --> 00:07:01.287
I'm interested in impact.
00:07:01.287 --> 00:07:03.289
Most of the problems in the world
00:07:03.289 --> 00:07:04.323
could be solved
00:07:04.323 --> 00:07:06.259
in Eric's little brain
00:07:06.259 --> 00:07:07.293
with an improvement
00:07:07.293 --> 00:07:09.028
or an adjustment in technology.
00:07:09.028 --> 00:07:10.863
If you look at sort of improvement
00:07:10.863 --> 00:07:12.265
in our lives, improvement
00:07:12.265 --> 00:07:13.433
in our health and so forth, it's
00:07:13.433 --> 00:07:14.033
all been related
00:07:14.033 --> 00:07:15.368
to technological progress.
00:07:15.368 --> 00:07:17.069
Starting with the invention of fire,
00:07:17.069 --> 00:07:19.505
which was extremely useful, right,
00:07:19.505 --> 00:07:20.139
way back when.
00:07:20.139 --> 00:07:21.007
But also dangerous,
00:07:21.007 --> 00:07:21.707
right? Of course,
00:07:21.707 --> 00:07:22.942
and everything is dual use
00:07:22.942 --> 00:07:24.343
and everything has to be managed.
00:07:24.343 --> 00:07:27.914
But just look at the dollar
00:07:28.347 --> 00:07:29.782
value of labor
00:07:29.782 --> 00:07:31.050
over 1,000 years,
00:07:31.050 --> 00:07:32.952
or 500 years, or 100 years,
00:07:32.952 --> 00:07:33.486
right?
00:07:34.587 --> 00:07:36.155
Human output, human hours
00:07:36.155 --> 00:07:37.290
in terms of productivity,
00:07:37.290 --> 00:07:39.058
are vastly improved.
00:07:39.058 --> 00:07:40.660
And because we've somehow
00:07:40.660 --> 00:07:41.694
collectively decided
00:07:41.694 --> 00:07:43.029
to have fewer children,
00:07:43.029 --> 00:07:44.163
not a good thing.
00:07:44.163 --> 00:07:45.965
I have four. Congratulations.
00:07:45.965 --> 00:07:46.666
Doing my part.
00:07:46.666 --> 00:07:48.468
Have as many as possible.
00:07:48.468 --> 00:07:49.769
More children are good and more
00:07:49.769 --> 00:07:51.304
grandchildren are good and so forth.
00:07:52.338 --> 00:07:53.940
We're going to have to automate.
00:07:53.940 --> 00:07:56.642
And so human wealth, human health,
00:07:56.642 --> 00:07:58.511
human wellbeing,
00:07:58.511 --> 00:07:59.545
one of the major
00:07:59.545 --> 00:08:01.414
determinants will be over
00:08:01.414 --> 00:08:02.381
the next 100 years,
00:08:02.381 --> 00:08:02.915
can we
00:08:02.915 --> 00:08:04.784
continue to accelerate progress
00:08:04.784 --> 00:08:05.284
in this?
00:08:05.284 --> 00:08:07.687
I profoundly believe we can.
00:08:07.687 --> 00:08:08.688
And I
00:08:08.688 --> 00:08:09.655
profoundly believe that
00:08:09.655 --> 00:08:10.690
long after I'm gone,
00:08:10.690 --> 00:08:12.692
humans will be much better.
00:08:12.692 --> 00:08:15.294
So, you tend to be way more optimistic
00:08:15.294 --> 00:08:18.231
on the benefits of AI than most.
00:08:18.231 --> 00:08:20.433
What gives you that confidence?
00:08:20.433 --> 00:08:21.067
Would you like
00:08:21.067 --> 00:08:22.301
most human diseases
00:08:22.301 --> 00:08:24.170
to be solved in the next 15 years?
00:08:24.170 --> 00:08:25.738
I would. I would. Sign me up,
00:08:25.738 --> 00:08:27.573
especially given my age.
00:08:27.573 --> 00:08:29.675
Would you like solutions to
00:08:29.675 --> 00:08:30.743
climate change?
00:08:30.743 --> 00:08:33.012
I would love that. Absolutely.
00:08:33.012 --> 00:08:35.781
Maybe we could have safer products.
00:08:35.781 --> 00:08:36.649
Maybe we could have better
00:08:36.649 --> 00:08:38.217
educational solutions.
00:08:38.217 --> 00:08:38.584
Maybe
00:08:38.584 --> 00:08:39.185
we could change the
00:08:39.185 --> 00:08:40.086
educational system
00:08:40.086 --> 00:08:41.587
so that kids are not bored,
00:08:41.587 --> 00:08:43.022
but rather they're engaged.
00:08:44.090 --> 00:08:46.893
The opportunities are significant.
00:08:46.893 --> 00:08:48.694
In the last month,
00:08:48.694 --> 00:08:51.697
10 major math problems were solved,
00:08:52.131 --> 00:08:53.666
including one called Navier-Stokes,
00:08:53.666 --> 00:08:55.434
which is historically important,
00:08:55.434 --> 00:08:56.569
by computers.
00:08:56.569 --> 00:08:57.537
Now you sit there and go, like,
00:08:57.537 --> 00:08:58.604
why do I care?
00:08:58.604 --> 00:09:00.907
Well, math is the foundation
00:09:00.907 --> 00:09:02.608
of that acceleration.
00:09:02.608 --> 00:09:03.509
For those
00:09:03.509 --> 00:09:04.443
listening to this podcast
00:09:04.443 --> 00:09:05.611
that are not technologists,
00:09:05.611 --> 00:09:07.980
what was the significance of that?
00:09:07.980 --> 00:09:09.315
Navier-Stokes is an equation
00:09:09.315 --> 00:09:11.417
that involves fluid flows. It’s
00:09:11.417 --> 00:09:13.419
one of the unsolvable math problems,
00:09:13.419 --> 00:09:15.821
until now. The specific
00:09:15.821 --> 00:09:17.823
conjecture had never been proven.
00:09:17.823 --> 00:09:18.791
And so, they use
00:09:18.791 --> 00:09:20.927
approximations for Navier-Stokes.
00:09:20.927 --> 00:09:21.928
I indeed funded
00:09:21.928 --> 00:09:23.829
some physics work in this area… Oh wow.
00:09:23.829 --> 00:09:24.730
…which I never understood,
00:09:24.730 --> 00:09:27.066
but it was really smart people.
00:09:27.066 --> 00:09:29.268
And the key thing is
00:09:29.268 --> 00:09:31.170
that these fluid flows model,
00:09:31.170 --> 00:09:33.139
for example, lift of an airplane,
00:09:33.139 --> 00:09:34.807
how do air conditioners work,
00:09:34.807 --> 00:09:36.542
there are many, many examples.
00:09:36.542 --> 00:09:39.545
So, being able to solve the problem
00:09:39.779 -->
00:09:41.113
in a particular way
00:09:41.113 --> 00:09:43.783
indicates better algorithmic answers.
00:09:43.783 --> 00:09:44.817
We can get a better
00:09:44.817 --> 00:09:46.385
and more accurate answer.
00:09:46.385 --> 00:09:47.620
Why do you care?
00:09:47.620 --> 00:09:49.155
The airplane will fly faster.
00:09:49.155 --> 00:09:51.123
The airplane will use less gas.
00:09:51.123 --> 00:09:52.825
Right? I can go on.
00:09:52.825 --> 00:09:54.694
The rocket will go faster.
00:09:54.694 --> 00:09:56.195
The fuel will be less.
00:09:56.195 --> 00:09:57.697
You'll get to Mars quicker.
00:09:57.697 --> 00:09:58.864
You know, what have you.
00:09:58.864 --> 00:10:01.734
All of these fluid flow things
00:10:01.734 --> 00:10:03.269
matter in a way
00:10:03.269 --> 00:10:03.903
that
00:10:03.903 --> 00:10:05.204
we sort of take for granted.
00:10:05.204 --> 00:10:06.372
Let's go back to the internet,
00:10:06.372 --> 00:10:07.506
because I think it's so interesting.
00:10:07.506 --> 00:10:08.774
You've been around
00:10:08.774 --> 00:10:10.109
the rise of the internet
00:10:10.109 --> 00:10:11.344
and other platform,
00:10:11.344 --> 00:10:11.711
you know,
00:10:11.711 --> 00:10:12.244
shifts,
00:10:12.244 --> 00:10:13.913
the rise at enterprise software,
00:10:13.913 --> 00:10:15.114
search, cloud
00:10:15.114 --> 00:10:16.015
computing, mobile,
00:10:16.015 --> 00:10:16.616
all of these
00:10:16.616 --> 00:10:19.185
major technological innovations.
00:10:19.185 --> 00:10:20.486
Why does this moment
00:10:20.486 --> 00:10:21.687
feel different to you?
00:10:21.687 --> 00:10:22.455
It's not different,
00:10:22.455 --> 00:10:24.323
it's just bigger. Bigger.
00:10:24.323 --> 00:10:25.191
The internet felt
00:10:25.191 --> 00:10:27.026
exactly the same as this.
00:10:27.026 --> 00:10:28.394
I started with mainframes,
00:10:28.394 --> 00:10:29.829
so that's how long
00:10:29.829 --> 00:10:30.696
I've been in this field.
00:10:30.696 --> 00:10:33.699
I've doing tech for 55 years,
00:10:34.233 --> 00:10:37.303
and when the PC revolution came
00:10:37.303 --> 00:10:39.305
along, everyone said, oh my god,
00:10:39.305 --> 00:10:40.406
this is this huge wave.
00:10:40.406 --> 00:10:41.874
It created Microsoft and Apple
00:10:41.874 --> 00:10:43.609
and so on and so on.
00:10:43.609 --> 00:10:46.512
This was 40 years ago, 45 years ago.
00:10:46.512 --> 00:10:48.648
So, these things take a while,
00:10:48.648 --> 00:10:50.983
and I think what, another example.
00:10:50.983 --> 00:10:53.085
Let's look at self-driving cars.
00:10:53.085 --> 00:10:54.453
Did you arrive at work today
00:10:54.453 --> 00:10:55.488
in a self-driving car?
00:10:55.488 --> 00:10:57.390
I did not. I suspect not, nor did I.
00:10:58.391 --> 00:11:00.259
And so, what do we think
00:11:00.259 --> 00:11:02.228
about that? When do you think we will?
00:11:02.228 --> 00:11:03.929
Well, it's beginning. Now
00:11:03.929 --> 00:11:04.497
self-driving
00:11:04.497 --> 00:11:06.065
cars… When do
you think New York City?
00:11:06.065 --> 00:11:06.532
New York City
00:11:06.532 --> 00:11:08.134
will be one of the last ones. Okay.
00:11:08.134 --> 00:11:08.768
But it'll come.
00:11:08.768 --> 00:11:09.301
It’ll come.
00:11:09.301 --> 00:11:10.002
But it's coming in
00:11:10.002 --> 00:11:11.103
the different cities,
00:11:11.103 --> 00:11:11.937
including obviously
00:11:11.937 --> 00:11:13.739
Waymo and other competitors.
00:11:13.739 --> 00:11:15.174
But why do I mention
00:11:15.174 --> 00:11:16.275
self-driving cars?
00:11:16.275 --> 00:11:17.510
They were largely,
00:11:17.510 --> 00:11:19.145
they were designed in the 90s.
00:11:19.145 --> 00:11:21.414
The first real test of self-driving
00:11:21.414 --> 00:11:23.416
cars was 2004.
00:11:23.416 --> 00:11:26.419
So that's 22 years and going.
00:11:26.852 --> 00:11:29.155
So, the diffusion rate is different
00:11:29.155 --> 00:11:30.856
from the invention rate.
00:11:30.856 --> 00:11:33.993
So, the faster the diffusion
00:11:33.993 --> 00:11:35.728
rate, the sort of broad adoption,
00:11:35.728 --> 00:11:37.763
the quicker the societal change.
00:11:37.763 --> 00:11:40.332
Why does it occur in what I do
00:11:40.332 --> 00:11:42.134
first? It's because we're already
00:11:42.134 --> 00:11:43.936
connected, right?
00:11:43.936 --> 00:11:45.738
It's essentially a softer problem.
00:11:45.738 --> 00:11:46.806
It's just connectivity.
00:11:46.806 --> 00:11:48.708
The marginal cost of connectivity, given
00:11:48.708 --> 00:11:49.075
we're
00:11:49.075 --> 00:11:50.242
so connected, is very low.
00:11:50.242 --> 00:11:51.711
The incremental cost of connecting
00:11:51.711 --> 00:11:53.913
you and me is essentially zero. Boom,
00:11:53.913 --> 00:11:54.847
it just happens.
00:11:54.847 --> 00:11:56.649
It's straightforward economics.
00:11:56.649 --> 00:11:58.617
Whereas getting self-driving
00:11:58.617 --> 00:12:00.186
cars is a great deal of capital.
00:12:00.186 --> 00:12:01.153
It's hard, and so forth.
00:12:01.153 --> 00:12:02.054
People are working on it.
00:12:02.054 --> 00:12:04.290
Why 20 years versus zero years,
00:12:04.290 --> 00:12:05.524
and that's the answer.
00:12:05.524 --> 00:12:07.526
You see the same, by the way, in AI.
00:12:07.526 --> 00:12:09.195
All of the gains right now
00:12:09.195 --> 00:12:10.062
are occurring in
00:12:10.062 --> 00:12:11.063
what is called,
00:12:11.063 --> 00:12:13.532
essentially, scale-free.
00:12:13.532 --> 00:12:14.266
And scale-free
00:12:14.266 --> 00:12:15.701
means that you can just keep doing
00:12:15.701 --> 00:12:16.635
it and get smarter.
00:12:16.635 --> 00:12:18.104
So, in math,
00:12:18.104 --> 00:12:19.138
I'm not a mathematician,
00:12:19.138 --> 00:12:20.372
but as best I can understand,
00:12:20.372 --> 00:12:22.074
they can just invent stuff,
00:12:22.074 --> 00:12:23.642
and they just keep inventing things,
00:12:23.642 --> 00:12:24.977
and they keep inventing things,
00:12:24.977 --> 00:12:26.245
and they're so clever.
00:12:26.245 --> 00:12:26.645
Well, now
00:12:26.645 --> 00:12:27.346
a computer
00:12:27.346 --> 00:12:28.814
can take all of those ideas,
00:12:28.814 --> 00:12:30.616
and if you just give it enough juice,
00:12:30.616 --> 00:12:32.218
it can do the same thing.
00:12:32.218 --> 00:12:33.819
Because it doesn't need data,
00:12:33.819 --> 00:12:35.221
it's self-contained.
00:12:35.221 --> 00:12:36.589
Unlocking all these solutions.
00:12:36.589 --> 00:12:38.057
Unlocking these solutions.
00:12:38.057 --> 00:12:39.258
In software,
00:12:39.258 --> 00:12:40.793
it turns out that once
00:12:40.793 --> 00:12:42.228
you start writing code
00:12:42.228 --> 00:12:43.562
and you start learning
00:12:43.562 --> 00:12:44.964
how well the code works,
00:12:44.964 --> 00:12:46.332
you can make it better.
00:12:46.332 --> 00:12:47.500
And then you can make it better
00:12:47.500 --> 00:12:48.267
and better.
00:12:48.267 --> 00:12:49.335
And there's what is called
00:12:49.335 --> 00:12:51.103
a recursive self-improvement loop,
00:12:51.103 --> 00:12:51.704
where it's
00:12:51.704 --> 00:12:53.606
getting smarter on its own.
00:12:53.606 --> 00:12:55.541
But this progress
00:12:55.541 --> 00:12:57.009
that I'm describing
00:12:57.009 --> 00:12:58.110
is happening faster
00:12:58.110 --> 00:13:00.312
than I have ever seen, right?
00:13:00.312 --> 00:13:01.847
What is true is each
00:13:01.847 --> 00:13:05.251
wave is bigger, but it's also faster.
00:13:05.284 --> 00:13:07.553
And we are not ready as a society.
00:13:07.553 --> 00:13:07.853
Eric,
00:13:07.853 --> 00:13:08.420
you gave me like
00:13:08.420 --> 00:13:09.388
10 things to unpack there,
00:13:09.388 --> 00:13:10.689
but let's try to do this.
00:13:10.689 --> 00:13:12.858
So recursive self-improvement, right?
00:13:12.858 --> 00:13:13.592
This is the thing
00:13:13.592 --> 00:13:15.694
that had the Anthropic researchers,
00:13:15.694 --> 00:13:17.029
two of them, making comments,
00:13:17.029 --> 00:13:17.630
one leaving
00:13:17.630 --> 00:13:18.264
making comments,
00:13:18.264 --> 00:13:19.832
the other sort of validating
00:13:19.832 --> 00:13:21.534
that there are serious concerns
00:13:21.534 --> 00:13:22.968
about the speed with
00:13:22.968 --> 00:13:25.070
which AI is learning. Right?
00:13:25.070 --> 00:13:26.906
So, by the way, every month
00:13:26.906 --> 00:13:28.107
there is a group of people
00:13:28.107 --> 00:13:29.341
who quit over this issue.
00:13:29.341 --> 00:13:30.409
Over the same topic,
00:13:30.409 --> 00:13:31.510
exactly.
00:13:31.510 --> 00:13:32.444
Silicon Valley,
00:13:32.444 --> 00:13:34.013
and in particular San Francisco,
00:13:34.013 --> 00:13:35.948
they're absolutely convinced
00:13:35.948 --> 00:13:38.384
that this is going to happen
00:13:38.384 --> 00:13:39.785
so quickly
00:13:39.785 --> 00:13:41.287
that it will threaten humanity
00:13:41.287 --> 00:13:42.087
in one way or the other.
00:13:42.087 --> 00:13:44.223
I personally do not agree with this. Okay,
why?
00:13:44.223 --> 00:13:46.192
Well, let me explain their argument.
00:13:47.793 --> 00:13:48.727
It's a relatively
00:13:48.727 --> 00:13:51.330
straightforward argument.
00:13:51.330 --> 00:13:54.333
As humans, we learn at a certain rate.
00:13:54.433 --> 00:13:55.734
If you have enough hardware
00:13:55.734 --> 00:13:56.902
and enough electricity
00:13:56.902 --> 00:13:58.737
and so forth and enough
00:13:58.737 --> 00:13:59.572
scale-free learning,
00:13:59.572 --> 00:14:01.073
which is what I'm talking about,
00:14:01.073 --> 00:14:02.541
it can just get faster
00:14:02.541 --> 00:14:04.910
and learn faster than humans can.
00:14:04.910 --> 00:14:06.745
And the way it works,
00:14:06.745 --> 00:14:09.748
I'll give you a simple example.
00:14:09.982 --> 00:14:11.817
I ask it to write a chapter
00:14:11.817 --> 00:14:12.818
and then it writes
00:14:12.818 --> 00:14:13.252
a chapter
00:14:13.252 --> 00:14:14.687
and I feed that chapter and say
00:14:14.687 --> 00:14:16.589
write the next chapter and so forth.
00:14:16.589 --> 00:14:19.124
Well, if I keep iterating like that,
00:14:19.124 --> 00:14:22.127
eventually it learns what I wanted
00:14:22.528 --> 00:14:23.863
with a process called reinforcement
00:14:23.863 --> 00:14:24.763
learning. It's
00:14:24.763 --> 00:14:25.965
easier to write the twelfth chapter
00:14:25.965 --> 00:14:26.832
than the first.
00:14:26.832 --> 00:14:28.500
And it gets smarter.
00:14:28.500 --> 00:14:29.768
And once it learns
00:14:29.768 --> 00:14:31.604
how to do it, it knows.
00:14:31.604 --> 00:14:33.539
It's like the way we learn. Right.
00:14:33.539 --> 00:14:35.341
But it can learn a lot faster.
00:14:35.341 --> 00:14:36.108
So,
00:14:36.108 --> 00:14:37.509
think of it as humans
00:14:37.509 --> 00:14:38.911
versus computers.
00:14:38.911 --> 00:14:39.945
I think
00:14:39.945 --> 00:14:41.213
the actual technical
00:14:41.213 --> 00:14:42.815
answer about this learning is it's
00:14:42.815 --> 00:14:43.849
not human learning,
00:14:43.849 --> 00:14:44.550
it's a different
00:14:44.550 --> 00:14:46.552
kind of intelligence.
00:14:46.552 --> 00:14:47.486
And so,
00:14:47.486 --> 00:14:49.455
the best way to state the fear
00:14:49.455 --> 00:14:51.056
is that we're inventing a different
00:14:51.056 --> 00:14:52.424
kind of intelligence
00:14:52.424 --> 00:14:55.494
that we cannot necessarily constrain.
00:14:55.861 --> 00:14:57.496
This is called the alignment problem.
00:14:58.464 --> 00:14:59.498
Also, in the past
00:14:59.498 --> 00:15:00.165
weeks, there
00:15:00.165 --> 00:15:03.168
was a very famous situation where
00:15:03.168 --> 00:15:04.770
there was an internal test
00:15:04.770 --> 00:15:06.939
going on at OpenAI.
00:15:06.939 --> 00:15:08.107
And this internal test,
00:15:08.107 --> 00:15:10.109
which was trying to check something,
00:15:10.109 --> 00:15:11.977
managed to create a set of bots,
00:15:13.012 --> 00:15:14.813
which are agents, if you will,
00:15:14.813 --> 00:15:16.115
they managed to collaborate
00:15:16.115 --> 00:15:18.284
to violate certain norms and laws,
00:15:18.284 --> 00:15:19.618
they managed to break into things, and
00:15:19.618 --> 00:15:21.353
they managed to, allegedly,
00:15:21.353 --> 00:15:22.454
do a little bit of harm.
00:15:22.454 --> 00:15:24.089
Now, nobody was hurt.
00:15:24.089 --> 00:15:25.124
But it's an example
00:15:25.124 --> 00:15:26.659
of what is possible.
00:15:26.659 --> 00:15:29.328
So, we must
00:15:29.328 --> 00:15:30.329
develop both
00:15:30.329 --> 00:15:31.630
recursive self-improvement
00:15:31.630 --> 00:15:33.232
which is the self-learning
00:15:33.232 --> 00:15:33.966
along with
00:15:33.966 --> 00:15:35.167
solving the alignment
00:15:35.167 --> 00:15:37.736
problem to human values.
00:15:37.736 --> 00:15:39.204
So, a simple way would be,
00:15:39.204 --> 00:15:40.873
I say to you,
00:15:40.873 --> 00:15:41.707
learn everything
00:15:41.707 --> 00:15:43.509
and you're a computer.
00:15:43.509 --> 00:15:44.576
That's not what I should say.
00:15:44.576 --> 00:15:45.978
I should learn everything,
00:15:45.978 --> 00:15:48.414
but, don't harm any humans,
00:15:48.414 --> 00:15:49.915
follow all laws, follow
00:15:49.915 --> 00:15:51.884
all laws, follow the US Constitution,
00:15:51.884 --> 00:15:53.085
make sure you don't get sued.
00:15:53.085 --> 00:15:53.552
You know,
00:15:53.552 --> 00:15:55.187
in other words, the set of assumptions. That
00:15:55.187 --> 00:15:56.488
you build into the model.
00:15:56.488 --> 00:15:57.256
These systems don't
00:15:57.256 --> 00:15:59.425
have a fear over the police.
00:15:59.425 --> 00:16:00.993
They don't feel feelings.
00:16:00.993 --> 00:16:02.995
And so, they have to be constrained.
00:16:02.995 --> 00:16:04.363
Come back to your optimism.
00:16:04.363 --> 00:16:06.065
It's in everything that
00:16:06.265 --> 00:16:07.766
you've written and you've said.
00:16:07.766 --> 00:16:09.268
What gives you optimism
00:16:09.268 --> 00:16:11.270
that we will be able
00:16:11.270 --> 00:16:12.938
to sort out a future where
00:16:12.938 --> 00:16:14.139
humanity wins?
00:16:15.140 --> 00:16:18.143
What's the alternative?
00:16:18.177 --> 00:16:19.845
I mean,
00:16:19.845 --> 00:16:20.779
the best way to get a
00:16:20.779 --> 00:16:22.348
headline is to say, in 10 years,
00:16:22.348 --> 00:16:23.615
we're all dead.
00:16:23.615 --> 00:16:25.017
I was on a panel
00:16:25.017 --> 00:16:27.052
with a guy who wrote a book
00:16:27.052 --> 00:16:28.153
which is entitled,
00:16:28.153 --> 00:16:29.755
‘If We Invent Superintelligence,
00:16:29.755 --> 00:16:30.723
We're All Dead.’
00:16:30.723 --> 00:16:32.057
It's very hard to argue
00:16:32.057 --> 00:16:33.659
with someone in that.
00:16:33.659 --> 00:16:34.994
The only problem with his arguments
00:16:34.994 --> 00:16:36.695
is it’s wrong, right?
00:16:36.695 --> 00:16:38.330
Because part of superintelligence
00:16:38.330 --> 00:16:38.697
is getting
00:16:38.697 --> 00:16:39.698
the super
00:16:39.698 --> 00:16:41.900
alignment problem solved.
00:16:41.900 --> 00:16:43.135
I would guess that your
00:16:43.135 --> 00:16:43.902
social network
00:16:43.902 --> 00:16:44.536
might be a little bit
00:16:44.536 --> 00:16:45.304
different than mine.
00:16:45.304 --> 00:16:45.738
What's the
00:16:45.738 --> 00:16:46.472
buzziest thing
00:16:46.472 --> 00:16:47.106
that you were hearing
00:16:47.106 --> 00:16:49.575
about this summer in the AI circles?
00:16:49.575 --> 00:16:51.510
The biggest and most important thing
00:16:51.510 --> 00:16:52.378
that has occurred in
00:16:52.378 --> 00:16:53.412
the last year
00:16:53.412 --> 00:16:56.415
is the arrival of long reasoning,
00:16:56.515 --> 00:16:57.950
deep reasoning.
00:16:57.950 --> 00:16:59.451
And when I say long reasoning,
00:16:59.451 --> 00:17:01.787
I mean eight hours of reasoning.
00:17:01.787 --> 00:17:02.621
So, what would happen
00:17:02.621 --> 00:17:04.056
is that you would start
00:17:04.056 --> 00:17:05.457
a year ago or two
00:17:05.457 --> 00:17:06.859
and it could kind of
00:17:06.859 --> 00:17:08.027
begin thinking
00:17:08.027 --> 00:17:08.994
and then it would start thinking
00:17:08.994 --> 00:17:10.329
about something else.
00:17:10.329 --> 00:17:12.131
Get distracted, like I do?
00:17:14.633 --> 00:17:16.035
We can make it feel like human,
00:17:16.035 --> 00:17:18.370
but somehow it takes a wrong path.
00:17:18.370 --> 00:17:18.971
It decides
00:17:18.971 --> 00:17:19.605
that instead of
00:17:19.605 --> 00:17:20.305
talking about math,
00:17:20.305 --> 00:17:21.073
we're going to talk about
00:17:21.073 --> 00:17:22.174
French literature.
00:17:22.174 --> 00:17:23.442
That's hilarious.
00:17:23.442 --> 00:17:24.410
Something which doesn't make any sense right?
00:17:24.410 --> 00:17:25.611
Right. Right.
00:17:25.611 --> 00:17:27.312
That problem was solved
00:17:27.312 --> 00:17:28.547
using essentially
00:17:28.547 --> 00:17:29.915
technical guardrails
00:17:29.915 --> 00:17:31.216
for what it kept thinking,
00:17:31.216 --> 00:17:33.952
so they said stay focused,
00:17:33.952 --> 00:17:34.787
stay focused,
00:17:34.787 --> 00:17:35.454
stay focused,
00:17:35.454 --> 00:17:36.889
is the way to understand it
00:17:36.889 --> 00:17:38.323
and that produced the ability
00:17:38.323 --> 00:17:39.258
to ask very,
00:17:39.258 --> 00:17:39.925
they're called chain
00:17:39.925 --> 00:17:41.126
of thought trains,
00:17:41.126 --> 00:17:41.727
where you can
00:17:41.727 --> 00:17:43.762
basically see its thinking
00:17:43.762 --> 00:17:45.297
how it solved a problem
00:17:45.297 --> 00:17:46.165
and these chains
00:17:46.165 --> 00:17:49.768
can be a thousand steps, two thousand steps.
00:17:50.135 --> 00:17:52.071
Now why am I mentioning this?
00:17:52.071 --> 00:17:52.938
I don't know about you,
00:17:52.938 --> 00:17:54.773
I can't think a thousand steps.
00:17:54.773 --> 00:17:55.407
I just don't,
00:17:55.407 --> 00:17:57.810
I'm just not that smart, but it can.
00:17:57.810 --> 00:18:00.012
So, it looks like deep reasoning
00:18:00.012 --> 00:18:00.979
will allow you
00:18:00.979 --> 00:18:02.648
to see deeper,
00:18:02.648 --> 00:18:04.983
as a human, than a human.
00:18:04.983 --> 00:18:05.651
The best
00:18:05.651 --> 00:18:07.486
AI is in service of humans
00:18:07.486 --> 00:18:07.986
doing things
00:18:07.986 --> 00:18:09.822
that humans are not so good at.
00:18:09.822 --> 00:18:11.924
So artificial superintelligence,
00:18:11.924 --> 00:18:13.392
how will we know we've achieved it
00:18:13.392 --> 00:18:14.226
and what do you think
00:18:14.226 --> 00:18:15.694
the timeline is?
00:18:15.694 --> 00:18:19.031
So, most people believe, right now,
00:18:19.298 --> 00:18:20.566
that we're at the beginning
00:18:20.566 --> 00:18:22.101
of superintelligence,
00:18:22.101 --> 00:18:24.369
which is broadly defined as humans
00:18:24.369 --> 00:18:25.904
plus super smart
00:18:25.904 --> 00:18:27.906
computers working together, right?
00:18:27.906 --> 00:18:29.975
There are more technical definitions.
00:18:31.110 --> 00:18:32.544
Now today,
00:18:32.544 --> 00:18:34.746
vision is better,
00:18:34.746 --> 00:18:36.849
data storage is better, retrieval
00:18:36.849 --> 00:18:37.950
is better
00:18:37.950 --> 00:18:39.351
and analytical stuff
00:18:39.351 --> 00:18:40.152
should be better
00:18:40.152 --> 00:18:41.787
and proofs are getting better.
00:18:41.787 --> 00:18:44.189
So, we're at the beginning of that.
00:18:44.189 --> 00:18:45.057
What is the thing
00:18:45.057 --> 00:18:46.825
that is not obvious how to do?
00:18:46.825 --> 00:18:49.228
And the answer is real discovery.
00:18:49.228 --> 00:18:50.729
So, the test that's used in
00:18:50.729 --> 00:18:53.732
the industry is, could you take
00:18:53.732 --> 00:18:54.566
all the information
00:18:54.566 --> 00:18:56.935
that's known in 1902,
00:18:56.935 --> 00:18:58.470
put it into a computer,
00:18:58.470 --> 00:19:00.906
and have it generate
00:19:00.906 --> 00:19:02.074
general relativity,
00:19:02.074 --> 00:19:02.975
special relativity,
00:19:02.975 --> 00:19:05.978
Einstein, 1902-1904?
00:19:06.211 --> 00:19:08.914
The consensus right now is no.
00:19:08.914 --> 00:19:10.315
And the consensus is no
00:19:10.315 --> 00:19:13.285
because it required a leap
00:19:13.385 --> 00:19:15.020
that was not obvious
00:19:15.020 --> 00:19:16.421
from any knowledge
00:19:16.421 --> 00:19:18.423
that existed at the time,
00:19:18.423 --> 00:19:19.324
okay?
00:19:19.324 --> 00:19:20.826
So, in the industry,
00:19:20.826 --> 00:19:21.326
we say, well,
00:19:21.326 --> 00:19:22.928
how would we solve that?
00:19:22.928 --> 00:19:24.663
Well, there's an obvious way,
00:19:24.663 --> 00:19:26.131
which is you take the computer
00:19:26.131 --> 00:19:28.433
and you just say, keep trying,
00:19:28.433 --> 00:19:28.667
you know,
00:19:28.667 --> 00:19:29.134
just it's
00:19:29.134 --> 00:19:30.769
the monkey with the keyboard
00:19:30.769 --> 00:19:32.204
eventually produces Mozart, but
00:19:32.204 --> 00:19:33.172
it takes a long time.
00:19:34.373 --> 00:19:35.274
I don't think that's
00:19:35.274 --> 00:19:35.507
gonna work
00:19:35.507 --> 00:19:36.742
because it's computationally
00:19:36.742 --> 00:19:38.310
not realistic.
00:19:38.310 --> 00:19:39.845
A better idea
00:19:39.845 --> 00:19:41.213
is to invent something
00:19:41.213 --> 00:19:43.348
which allows for analytical
00:19:43.348 --> 00:19:46.385
reasoning in non-coherent spaces.
00:19:46.451 --> 00:19:48.253
So, think of it as a leap.
00:19:48.253 --> 00:19:49.755
I happen to understand
00:19:49.755 --> 00:19:51.657
something in physics,
00:19:51.657 --> 00:19:53.125
and then you give me
00:19:53.125 --> 00:19:54.626
a completely unrelated problem,
00:19:54.626 --> 00:19:57.829
and somehow… Something clicks. …something clicks.
00:19:57.863 --> 00:19:59.364
You know, it's like in my case,
00:19:59.364 --> 00:20:00.065
in the morning
00:20:00.065 --> 00:20:00.666
you're in the shower,
00:20:00.666 --> 00:20:02.134
and you have this idea… Right.
00:20:02.134 --> 00:20:04.403
…that just showed up.
00:20:04.403 --> 00:20:06.171
Can we invent that?
00:20:06.171 --> 00:20:07.372
So, do you think these leaps
00:20:07.372 --> 00:20:09.141
are coming
00:20:09.141 --> 00:20:11.577
faster than we realize?
00:20:11.577 --> 00:20:13.111
The kind of leaps I'm talking to
00:20:13.111 --> 00:20:14.379
have not occurred yet.
00:20:14.379 --> 00:20:15.214
There's something called the San
00:20:15.214 --> 00:20:17.683
Francisco Consensus, by me anyway,
00:20:17.683 --> 00:20:18.951
and everyone in San Francisco
00:20:18.951 --> 00:20:20.886
thinks this will occur in two years.
00:20:20.886 --> 00:20:22.821
I don't. For one thing,
00:20:22.821 --> 00:20:23.222
I don't think
00:20:23.222 --> 00:20:24.089
there's enough computers,
00:20:24.089 --> 00:20:25.190
I don't think there's enough people,
00:20:25.190 --> 00:20:25.924
I don 't think that there's enough
00:20:25.924 --> 00:20:26.625
algorithms.
00:20:26.625 --> 00:20:28.227
I think maybe in my lifetime,
00:20:28.227 --> 00:20:29.461
maybe in a decade.
00:20:29.461 --> 00:20:30.896
That would be a big deal.
00:20:30.896 --> 00:20:32.331
But each of these moments,
00:20:32.331 --> 00:20:34.166
they’re happening so fast now.
00:20:35.434 --> 00:20:36.868
And by the way, it's fast
00:20:36.868 --> 00:20:38.503
because of the internet.
00:20:38.503 --> 00:20:39.938
It's because of what is called
00:20:39.938 --> 00:20:42.407
basically scalable computing.
00:20:42.407 --> 00:20:43.575
The supercomputers
00:20:43.575 --> 00:20:44.343
that are being used,
00:20:44.343 --> 00:20:45.444
so typical model
00:20:45.444 --> 00:20:46.745
takes three or four months
00:20:46.745 --> 00:20:48.714
costs a hundred million dollars to train,
00:20:48.714 --> 00:20:51.083
takes a few months to test,
00:20:51.083 --> 00:20:52.084
right?
00:20:52.084 --> 00:20:54.052
Thank God for the American
00:20:54.052 --> 00:20:54.886
financial industry
00:20:54.886 --> 00:20:55.654
that would be willing to
00:20:55.654 --> 00:20:57.055
finance these things.
00:20:57.055 --> 00:20:58.257
When we're seeing the creation
00:20:58.257 --> 00:20:59.391
of an entirely new
00:20:59.391 --> 00:21:00.225
financing model
00:21:00.225 --> 00:21:01.393
for this industry, right?
00:21:01.393 --> 00:21:02.828
If you want to achieve the vision
00:21:02.828 --> 00:21:04.663
I am describing,
00:21:04.663 --> 00:21:07.232
you have to do it at a world level.
00:21:07.232 --> 00:21:08.834
America is particularly good
00:21:08.834 --> 00:21:09.635
at building world
00:21:09.635 --> 00:21:10.702
standards and
00:21:10.702 --> 00:21:13.005
world-changing impacts at this scale.
00:21:13.005 --> 00:21:14.172
You're going to have to it
00:21:14.172 --> 00:21:16.308
in the US for regulatory reasons,
00:21:16.308 --> 00:21:16.875
and you're going
00:21:16.875 --> 00:21:18.243
have to find a way to finance it.
00:21:18.243 --> 00:21:19.411
You're also gonna have to wait
00:21:19.411 --> 00:21:20.012
to find a way
00:21:20.012 --> 00:21:21.513
solve the energy problem.
00:21:21.513 --> 00:21:23.849
The best energy solution is fusion.
00:21:23.849 --> 00:21:24.449
They look like
00:21:24.449 --> 00:21:25.617
they're going to get to fusion
00:21:25.617 --> 00:21:28.220
within two to three years.
00:21:28.220 --> 00:21:29.221
When that occurs,
00:21:29.221 --> 00:21:31.223
that's another massive moment
00:21:31.223 --> 00:21:32.391
in human history.
00:21:32.391 --> 00:21:34.259
I think it will occur in
00:21:34.259 --> 00:21:35.627
you know, three to five years.
00:21:35.627 --> 00:21:37.129
You've said you've sort of invested
00:21:37.129 --> 00:21:38.530
in everything when it comes to AI,
00:21:38.530 --> 00:21:39.131
but if you could
00:21:39.131 --> 00:21:41.466
only invest in one thing
00:21:41.466 --> 00:21:42.968
related to artificial intelligence,
00:21:42.968 --> 00:21:44.136
like one theme,
00:21:44.136 --> 00:21:45.671
not a single company, but one theme
00:21:45.671 --> 00:21:47.773
what would it be? Long reasoning.
00:21:48.740 --> 00:21:49.174
I don't
00:21:49.174 --> 00:21:50.842
think we appreciate
00:21:50.842 --> 00:21:53.145
when we have a thought partner
00:21:53.145 --> 00:21:53.979
that can
00:21:53.979 --> 00:21:56.548
think longer than we do.
00:21:56.548 --> 00:21:57.649
We can't think about how.
00:21:57.649 --> 00:21:59.818
So I'll give you an example.
00:21:59.818 --> 00:22:00.686
You're a CEO.
00:22:00.686 --> 00:22:02.854
You have 24-hour agents.
00:22:02.854 --> 00:22:04.056
You have one that's running,
00:22:04.056 --> 00:22:04.990
looking at security.
00:22:04.990 --> 00:22:06.224
Another one that's looking at power.
00:22:06.224 --> 00:22:07.459
Another one that’s looking at cash.
00:22:07.459 --> 00:22:08.827
Another one that’s looking at products.
00:22:08.827 --> 00:22:09.494
Another one that’s looking
00:22:09.494 --> 00:22:10.662
at global macro.
00:22:10.662 --> 00:22:12.264
And they're just running.
00:22:12.264 --> 00:22:12.864
And then your
00:22:12.864 --> 00:22:15.867
AI CEO every day when you get up says,
00:22:15.867 --> 00:22:17.169
does Eric need to know anything?
00:22:17.169 --> 00:22:18.704
Does Christine need to know anything?
00:22:18.704 --> 00:22:20.339
And they say, nah,
00:22:20.339 --> 00:22:21.340
nothing special,
00:22:21.340 --> 00:22:23.809
but it's always running.
00:22:23.809 --> 00:22:25.844
So, you wanna think about that
00:22:25.844 --> 00:22:27.579
in running a business,
00:22:27.579 --> 00:22:29.548
that those things are,
00:22:29.548 --> 00:22:30.582
you have to sleep,
00:22:30.582 --> 00:22:31.883
it doesn't.
00:22:31.883 --> 00:22:33.585
They can be disciplined into
00:22:33.585 --> 00:22:35.887
focus on this and not something else.
00:22:35.887 --> 00:22:37.689
Is that important? Absolutely.
00:22:37.689 --> 00:22:38.423
So, do you think there's
00:22:38.423 --> 00:22:41.093
a productivity boom that's coming?
00:22:41.093 --> 00:22:42.494
Well, I think it's already underway.
00:22:42.494 --> 00:22:43.328
It's already underway.
00:22:43.328 --> 00:22:45.664
An example would be my entire life,
00:22:45.664 --> 00:22:47.065
according to myself,
00:22:47.065 --> 00:22:48.467
I was a very, very good programmer
00:22:48.467 --> 00:22:49.868
when I was 22, right?
00:22:49.868 --> 00:22:50.769
I was one of the best,
00:22:50.769 --> 00:22:52.371
according to myself.
00:22:52.371 --> 00:22:55.540
My field is over, right?
00:22:55.807 --> 00:22:56.675
And yet you still tell
00:22:56.675 --> 00:22:57.843
people to learn coding.
00:22:57.843 --> 00:22:58.377
I do
00:22:58.377 --> 00:23:00.779
because I want them to be architects.
00:23:00.779 --> 00:23:02.514
So, I went from being able,
00:23:02.514 --> 00:23:04.249
I had to build the house.
00:23:04.249 --> 00:23:06.451
Now I can architect the house
00:23:06.451 --> 00:23:07.452
using the equivalent
00:23:07.452 --> 00:23:09.087
of Claude Code or
00:23:09.087 --> 00:23:11.456
OpenAI Codex or future products
00:23:11.456 --> 00:23:12.691
from Google,
00:23:12.691 --> 00:23:14.426
which they're working on,
00:23:14.426 --> 00:23:15.660
all of which
00:23:15.660 --> 00:23:17.462
allow me to say what I want
00:23:17.462 --> 00:23:18.663
and the code gets written.
00:23:18.663 --> 00:23:20.098
I look at the code it writes
00:23:20.098 --> 00:23:21.733
now. I can't write that code.
00:23:21.733 --> 00:23:23.969
I couldn't when I was 22, right?
00:23:23.969 --> 00:23:26.905
So, I got beaten, right? Hard.
00:23:27.873 --> 00:23:29.074
It's okay, right,
00:23:29.074 --> 00:23:30.809
I did fine.
00:23:30.809 --> 00:23:32.711
The next generation of people like me
00:23:32.711 --> 00:23:34.146
will be the architects
00:23:34.146 --> 00:23:35.947
and they'll be very, very smart
00:23:35.947 --> 00:23:37.215
and they know how to say,
00:23:37.215 --> 00:23:39.084
this is what I want.
00:23:39.084 --> 00:23:40.419
One of my friends,
00:23:40.419 --> 00:23:41.153
some of my friends
00:23:41.153 --> 00:23:41.987
are not so friendly
00:23:41.987 --> 00:23:42.954
and they're not so social,
00:23:42.954 --> 00:23:43.622
they're kind of,
00:23:43.622 --> 00:23:45.023
so we say asocial.
00:23:45.023 --> 00:23:46.324
No naming names.
00:23:46.324 --> 00:23:48.260
I'm not gonna name anybody.
00:23:48.260 --> 00:23:49.327
So, I said, what are you doing?
00:23:49.327 --> 00:23:51.062
And he said, I'm deleting names
00:23:51.062 --> 00:23:52.464
from my address book.
00:23:52.464 --> 00:23:53.865
And I said, why?
00:23:53.865 --> 00:23:54.833
I don't
00:23:54.833 --> 00:23:55.834
wanna talk to them anymore.
00:23:55.834 --> 00:23:57.035
I said, what do you do all day?
00:23:57.035 --> 00:23:59.070
Well, I get up in the morning,
00:23:59.070 --> 00:23:59.871
I have coffee,
00:23:59.871 --> 00:24:01.373
and then I check with my friends.
00:24:01.373 --> 00:24:02.574
And I say which friends?
00:24:02.574 --> 00:24:04.509
He said, my agent friends.
00:24:04.509 --> 00:24:05.243
Oh God. Okay,
00:24:05.243 --> 00:24:06.912
so, what do you do?
00:24:06.912 --> 00:24:08.613
And he spends all morning
00:24:09.815 --> 00:24:10.882
plugging and
00:24:10.882 --> 00:24:12.050
playing with his agents
00:24:12.050 --> 00:24:13.151
who are solving really
00:24:13.151 --> 00:24:14.686
interesting problems to him.
00:24:14.686 --> 00:24:15.487
And then he gives them
00:24:15.487 --> 00:24:18.290
a one-hour project for lunch.
00:24:18.290 --> 00:24:19.424
He has lunch
00:24:19.424 --> 00:24:20.192
and he comes back
00:24:20.192 --> 00:24:21.827
and he interacts with his agents.
00:24:21.827 --> 00:24:22.828
Now he's got a wife
00:24:22.828 --> 00:24:24.162
and a family and so forth.
00:24:24.162 --> 00:24:27.032
So, at five o'clock he gives it an overnight,
00:24:27.032 --> 00:24:28.233
you know, task. Right.
00:24:28.233 --> 00:24:28.900
And it goes to
00:24:28.900 --> 00:24:30.302
work. And it goes to work,
00:24:30.302 --> 00:24:32.170
and then it typically completes
00:24:32.170 --> 00:24:33.472
between two and four in the morning.
00:24:33.472 --> 00:24:34.573
And I said, do you wake up for that?
00:24:34.573 --> 00:24:35.474
He said, no, no, no, no,
00:24:35.474 --> 00:24:36.875
I wake up naturally.
00:24:36.875 --> 00:24:38.810
I know, with certainty,
00:24:38.810 --> 00:24:40.579
when I wake up, I'll have a result.
00:24:40.579 --> 00:24:41.980
I was with you on everything
00:24:41.980 --> 00:24:42.581
other than that
00:24:42.581 --> 00:24:44.015
these are his friends, right?
00:24:44.015 --> 00:24:44.282
I mean,
00:24:44.282 --> 00:24:44.649
I think
00:24:44.649 --> 00:24:46.585
they're amazing productivity tools,
00:24:46.585 --> 00:24:48.487
but I'm going to take us back now
00:24:48.487 --> 00:24:49.621
to data centers.
00:24:49.621 --> 00:24:51.256
The case for data centers was made
00:24:51.256 --> 00:24:51.990
long before
00:24:51.990 --> 00:24:53.058
artificial intelligence,
00:24:53.058 --> 00:24:55.093
for mobile, and cloud
00:24:55.093 --> 00:24:56.761
computing, and cybersecurity.
00:24:56.761 --> 00:24:57.896
All these things that
00:24:57.896 --> 00:24:59.364
it helps enable,
00:24:59.364 --> 00:24:59.865
yet now
00:24:59.865 --> 00:25:01.533
there's so much concern around
00:25:01.533 --> 00:25:02.267
data centers,
00:25:02.267 --> 00:25:03.568
what's your take?
00:25:03.568 --> 00:25:05.504
Well, if you're running one of these
00:25:05.504 --> 00:25:07.205
AI companies,
00:25:07.205 --> 00:25:08.874
your revenue is completely determined
00:25:08.874 --> 00:25:10.408
by your data centers.
00:25:10.408 --> 00:25:13.111
So, this is a massive change
00:25:13.111 --> 00:25:14.145
in what I do.
00:25:14.145 --> 00:25:15.013
It used to be
00:25:15.013 --> 00:25:16.348
that software was pretty high
00:25:16.348 --> 00:25:17.883
gross margin, you could sort of sell
00:25:17.883 --> 00:25:19.684
it, didn't have a lot of capital.
00:25:19.684 --> 00:25:22.053
Because you need the hardware,
00:25:22.053 --> 00:25:23.989
that's what's changed the economics.
00:25:23.989 --> 00:25:25.624
So, in the short
00:25:25.624 --> 00:25:27.259
term, it's the same thing
00:25:27.259 --> 00:25:28.693
as in the gold rush,
00:25:28.693 --> 00:25:29.861
the people who made all the money
00:25:29.861 --> 00:25:31.663
were the pickaxes and so forth
00:25:31.663 --> 00:25:32.063
and so on,
00:25:32.063 --> 00:25:33.265
the infrastructure. Picks
00:25:33.265 --> 00:25:34.299
and shovels,
00:25:34.299 --> 00:25:35.667
we have a whole thing on this, Eric.
00:25:35.667 --> 00:25:36.535
And this picks and
00:25:36.535 --> 00:25:37.869
shovels thing is real.
00:25:38.870 --> 00:25:39.371
So, from the
00:25:39.371 --> 00:25:41.306
standpoint of the economy,
00:25:41.306 --> 00:25:43.174
those are huge,
00:25:43.174 --> 00:25:44.609
they're real businesses.
00:25:44.609 --> 00:25:47.245
There's an estimate that 11% of US
00:25:47.245 --> 00:25:50.248
electricity demand in 2030 will be AI
00:25:50.248 --> 00:25:51.683
data-center-related.
00:25:51.683 --> 00:25:53.451
That's an extraordinary transition
00:25:53.451 --> 00:25:54.553
by any measure.
00:25:54.553 --> 00:25:56.588
Historically it was 2% or 3%.
00:25:56.588 --> 00:25:58.123
So, and that build-out
00:25:58.123 --> 00:25:59.491
is one of the primary drivers
00:25:59.491 --> 00:26:01.826
of economic growth in America today.
00:26:01.826 --> 00:26:03.061
So, thank goodness for that.
00:26:03.061 --> 00:26:03.862
If you could leave
00:26:03.862 --> 00:26:05.964
CEOs with one takeaway today,
00:26:05.964 --> 00:26:07.132
what would it be?
00:26:07.132 --> 00:26:08.867
The term that I've been using
00:26:08.867 --> 00:26:11.102
is being an AI-native CEO. So,
00:26:11.102 --> 00:26:12.137
what I'm trying to do in
00:26:12.137 --> 00:26:13.138
all my things
00:26:13.138 --> 00:26:14.973
is I'm tying to fully automate
00:26:14.973 --> 00:26:17.709
the execution of the businesses.
00:26:17.709 -->
00:26:20.312
In the factory floor in Relativity,
00:26:20.312 --> 00:26:21.346
I was walking through it
00:26:21.346 --> 00:26:22.147
one day and I thought,
00:26:22.147 --> 00:26:23.782
how many computers are here?
00:26:23.782 --> 00:26:24.416
So, I said,
00:26:24.416 --> 00:26:25.550
why don't we connect them all?
00:26:25.550 --> 00:26:26.651
So, they just connected
00:26:26.651 --> 00:26:27.686
all the computers.
00:26:27.686 --> 00:26:28.987
And so, the immediate thing
00:26:28.987 --> 00:26:31.022
that they can do, using AI,
00:26:31.022 --> 00:26:32.023
is they can look at,
00:26:32.023 --> 00:26:32.290
sort of,
00:26:32.290 --> 00:26:33.825
utilization, cross utilization
00:26:33.825 --> 00:26:35.327
times in new ways.
00:26:35.327 --> 00:26:37.028
It occurs as a natural byproduct
00:26:37.028 --> 00:26:37.395
of getting
00:26:37.395 --> 00:26:39.331
the digital systems connected.
00:26:39.331 --> 00:26:40.131
Furthermore,
00:26:40.131 --> 00:26:42.334
if you take generic data,
00:26:42.334 --> 00:26:44.069
unstructured data,
00:26:44.069 --> 00:26:44.869
and you connect it
00:26:44.869 --> 00:26:46.605
all, these systems are smart enough
00:26:46.605 --> 00:26:47.939
that they can interpolate
00:26:47.939 --> 00:26:49.074
between one data
00:26:49.074 --> 00:26:49.941
normalcy and another.
00:26:49.941 --> 00:26:50.375
In other words,
00:26:50.375 --> 00:26:52.577
they can combine disparate datasets
00:26:52.577 --> 00:26:54.412
and give you business insights.
00:26:54.412 --> 00:26:56.247
So how do I make more money?
00:26:56.247 --> 00:26:58.383
How do I increase customer retention?
00:26:58.383 --> 00:26:58.883
How do make
00:26:58.883 --> 00:27:00.251
sure I don't lose a customer?
00:27:00.251 --> 00:27:02.587
You can ask these questions.
00:27:02.587 --> 00:27:04.723
What I've been struck by
00:27:04.723 --> 00:27:07.058
is that when you take open-source
00:27:07.058 --> 00:27:09.127
data, that is public data,
00:27:09.127 --> 00:27:10.428
plus the proprietary
00:27:10.428 --> 00:27:11.963
data in your business,
00:27:11.963 --> 00:27:14.065
and you put it into a system,
00:27:14.065 --> 00:27:15.767
and you start asking questions,
00:27:15.767 --> 00:27:17.469
I go crazy over it.
00:27:17.469 --> 00:27:19.204
You can now talk across
00:27:19.204 --> 00:27:20.138
all of the data.
00:27:20.138 --> 00:27:21.272
You can get data fusion.
00:27:21.272 --> 00:27:22.941
You can actually figure stuff out.
00:27:22.941 --> 00:27:23.908
It drives me crazy
00:27:23.908 --> 00:27:24.542
a little bit, though,
00:27:24.542 --> 00:27:26.511
when I do have content given to me
00:27:26.511 --> 00:27:28.647
that is generated by the computer.
00:27:28.647 --> 00:27:29.981
Is it correct?
00:27:29.981 --> 00:27:31.182
Some is and some isn't.
00:27:31.182 --> 00:27:33.618
Sometimes I worry about the loss of
00:27:35.086 --> 00:27:37.422
original human thought.
00:27:37.422 --> 00:27:39.624
There's a whole bunch of concerns.
00:27:39.624 --> 00:27:40.892
I think the most one,
00:27:40.892 --> 00:27:42.260
the one that I've always been worried
00:27:42.260 --> 00:27:44.095
about, is the loss of deep reading,
00:27:44.095 --> 00:27:47.232
which I blame not on AI, but on
00:27:47.732 --> 00:27:48.566
social media.
00:27:48.566 --> 00:27:50.535
I used to read a book a week
00:27:50.535 --> 00:27:52.537
and now I'm too interrupt
00:27:52.537 --> 00:27:53.805
driven to do that.
00:27:53.805 --> 00:27:55.073
And my attention span
00:27:55.073 --> 00:27:56.675
is so much shorter. So much less.
00:27:56.675 --> 00:27:58.309
I've been watching young scientists,
00:27:58.309 --> 00:27:58.743
I fund
00:27:58.743 --> 00:28:00.779
a whole a bunch of young scientists.
00:28:00.779 --> 00:28:02.347
How do they do it?
00:28:02.347 --> 00:28:03.014
And the answer is
00:28:03.014 --> 00:28:04.282
they turn off the phone,
00:28:04.282 --> 00:28:07.152
they turn off, and they just do it,
00:28:07.152 --> 00:28:10.155
but it requires this huge strength
00:28:10.155 --> 00:28:12.190
to turn off the drug. Right?
00:28:12.190 --> 00:28:13.124
That interrupt drug,
00:28:13.124 --> 00:28:15.293
the serotonin thing that we have. Right.
00:28:15.293 --> 00:28:16.628
That's bad, in my view,
00:28:16.628 --> 00:28:18.229
that's bad for society.
00:28:18.229 --> 00:28:19.097
So, I have
00:28:19.097 --> 00:28:20.231
a bunch of young kids.
00:28:20.231 --> 00:28:21.366
I have two
00:28:21.366 --> 00:28:23.134
that are nearing their college years.
00:28:23.134 --> 00:28:23.935
What would you tell
00:28:23.935 --> 00:28:25.503
young people to study today?
00:28:25.503 --> 00:28:26.671
I used to say biology,
00:28:26.671 --> 00:28:28.273
now I say deep reasoning.
00:28:28.273 --> 00:28:28.807
Okay.
00:28:28.807 --> 00:28:31.242
If you're a non-technical person,
00:28:31.242 --> 00:28:33.511
you should figure out how to
00:28:33.511 --> 00:28:36.948
use these tools to make your dreams
00:28:36.948 --> 00:28:38.383
and your realities
00:28:38.383 --> 00:28:40.485
extraordinarily scaled.
00:28:41.586 --> 00:28:43.955
You want to be a global star,
00:28:43.955 --> 00:28:45.423
a global influencer,
00:28:45.423 --> 00:28:47.926
a global impactor, a global discoverer,
00:28:47.926 --> 00:28:49.594
a global singer,
00:28:49.594 --> 00:28:51.429
you want to use these tools,
00:28:51.429 --> 00:28:53.732
whatever it is that you want.
00:28:53.732 --> 00:28:54.532
Figure out a way to
00:28:54.532 --> 00:28:56.701
use them to amplify you
00:28:56.701 --> 00:28:57.769
and what you care about
00:28:57.769 --> 00:28:59.704
and your innate goodness.
00:28:59.704 --> 00:29:01.339
If you're a technical person,
00:29:01.339 --> 00:29:05.210
use the same tools to invent stuff
00:29:05.577 --> 00:29:06.611
and to invent stuff
00:29:06.611 --> 00:29:08.179
that changes the world.
00:29:08.179 --> 00:29:11.850
I've never seen the cost of entry
00:29:12.183 --> 00:29:13.151
to be so low
00:29:13.151 --> 00:29:14.886
and the availability of these ideas
00:29:14.886 --> 00:29:15.553
so great.
00:29:15.553 --> 00:29:17.355
The only thing that limits you
00:29:17.355 --> 00:29:19.190
is your
00:29:19.190 --> 00:29:20.191
curiosity,
00:29:20.191 --> 00:29:21.526
your willingness to take risks
00:29:21.526 --> 00:29:22.260
and so forth.
00:29:22.260 --> 00:29:23.995
So, get over it. And say,
00:29:23.995 --> 00:29:25.163
I want to dream,
00:29:25.163 --> 00:29:26.731
I want to use these tools
00:29:26.731 --> 00:29:29.501
to have this enormous impact, right?
00:29:29.501 --> 00:29:31.369
And some of you,
00:29:31.369 --> 00:29:32.537
not everybody,
00:29:32.537 --> 00:29:34.372
will have a huge impact.
00:29:34.372 --> 00:29:35.774
And it won't just be coming,
00:29:35.774 --> 00:29:37.609
being financially successful,
00:29:37.609 --> 00:29:39.444
you might become famous or important
00:29:39.444 --> 00:29:39.944
in something
00:29:39.944 --> 00:29:40.812
that you didn't even know
00:29:40.812 --> 00:29:41.746
was important,
00:29:41.746 --> 00:29:43.648
but you'll be so proud of yourself.
00:29:43.648 --> 00:29:45.150
I love the optimistic future
00:29:45.150 --> 00:29:45.817
that you've painted,
00:29:45.817 --> 00:29:47.218
and all that you taught us today.
00:29:47.218 --> 00:29:48.052
I really appreciate it.
00:29:48.052 --> 00:29:49.154
Thank you so, so much. Thank you.
00:29:53.925 --> 00:29:55.927
Joining us for The Debrief is Jas
00:29:55.927 --> 00:29:58.363
Khaira, our Head of N1, Blackstone's
00:29:58.363 --> 00:30:00.064
AI investing platform.
00:30:00.064 --> 00:30:00.331
Jas,
00:30:00.331 --> 00:30:01.332
you were the first person
00:30:01.332 --> 00:30:02.133
I thought of when I knew
00:30:02.133 --> 00:30:02.700
we were sitting down
00:30:02.700 --> 00:30:03.301
with Eric Schmidt.
00:30:03.301 --> 00:30:04.502
Thank you so much for joining us.
00:30:04.502 --> 00:30:05.370
Thank you, Christine.
00:30:05.370 --> 00:30:05.770
Alright,
00:30:05.770 --> 00:30:06.971
so, very
00:30:06.971 --> 00:30:08.473
wide-ranging conversation with Eric.
00:30:08.473 --> 00:30:10.775
He knows so much.
00:30:10.775 --> 00:30:11.743
No surprise,
00:30:11.743 --> 00:30:13.111
AI was front and center
00:30:13.111 --> 00:30:14.312
of that conversation.
00:30:14.312 --> 00:30:14.546
Yeah,
00:30:14.546 --> 00:30:15.413
you went from math
00:30:15.413 --> 00:30:16.815
algorithms to Hugging Face.
00:30:16.815 --> 00:30:19.818
It was definitely a broad discussion.
00:30:19.884 --> 00:30:22.020
I think Eric is an optimist,
00:30:22.020 --> 00:30:23.688
and we should all be bullish
00:30:23.688 --> 00:30:25.824
on optimism as an investment trend.
00:30:25.824 --> 00:30:27.525
But I think he also brought up
00:30:27.525 --> 00:30:28.927
what happened with Hugging Face
00:30:28.927 --> 00:30:29.928
and some of the challenges
00:30:29.928 --> 00:30:30.595
that we’re seeing in
00:30:30.595 --> 00:30:32.297
AI safety and ethics
00:30:32.297 --> 00:30:33.565
that is resulting in
00:30:33.565 --> 00:30:34.666
many of
the leaders of these
00:30:34.666 --> 00:30:36.167
AI companies focusing on that.
00:30:36.167 --> 00:30:37.569
There’s a good discussion and debate
00:30:37.569 --> 00:30:39.537
that’s happening right now.
00:30:39.537 --> 00:30:41.873
I loved Eric’s optimism.
00:30:41.873 --> 00:30:42.740
It was infectious.
00:30:42.740 --> 00:30:44.442
And it reminded me
00:30:44.442 --> 00:30:46.211
that if there is one long-term
00:30:46.211 --> 00:30:48.446
trend to be long, it’s optimism.
00:30:48.446 --> 00:30:49.948
I would say this idea
00:30:49.948 --> 00:30:51.049
of superintelligence
00:30:51.049 --> 00:30:52.083
and what it means,
00:30:52.083 --> 00:30:53.151
right, is something that
00:30:53.151 --> 00:30:54.786
everyone is still grappling with.
00:30:54.786 --> 00:30:56.187
What is your take on
00:30:56.187 --> 00:30:57.255
what superintelligence
00:30:57.255 --> 00:30:58.289
means for humanity?
00:30:58.289 --> 00:30:59.457
I think the current way
00:30:59.457 --> 00:31:00.525
to understand
00:31:00.525 --> 00:31:01.726
superintelligence
00:31:01.726 --> 00:31:04.062
is that the cutting-edge research
00:31:04.062 --> 00:31:05.330
within the labs
00:31:05.330 --> 00:31:08.066
intends to bring recursive work
00:31:08.066 --> 00:31:09.734
to the training itself.
00:31:09.734 --> 00:31:11.936
So, the training of the models
00:31:11.936 --> 00:31:13.771
now relies on compute
00:31:13.771 --> 00:31:15.473
in one virtuous circle
00:31:15.473 --> 00:31:16.741
to kind of enhance that.
00:31:16.741 --> 00:31:18.443
And you can imagine
00:31:18.443 --> 00:31:19.344
when you take that
00:31:19.344 --> 00:31:21.045
to its logical conclusion,
00:31:21.045 --> 00:31:22.347
you have superintelligence
00:31:22.347 --> 00:31:23.381
above that.
00:31:23.381 --> 00:31:24.849
I think
00:31:24.849 --> 00:31:25.783
we’ll have to see how
00:31:25.783 --> 00:31:27.185
that plays out in the real world.
00:31:27.185 --> 00:31:27.986
And yet it feels like
00:31:27.986 --> 00:31:29.354
the pace is really picking up,
00:31:29.354 --> 00:31:30.021
for sure.
00:31:30.021 --> 00:31:30.755
Things move
00:31:30.755 --> 00:31:31.689
faster than ever,
00:31:31.689 --> 00:31:32.223
and it’s
00:31:32.223 --> 00:31:33.424
one of the most dizzying parts
00:31:33.424 --> 00:31:34.859
about investing in this landscape,
00:31:34.859 --> 00:31:36.227
which is you really have to try
00:31:36.227 --> 00:31:37.028
to stay at the
00:31:37.028 --> 00:31:38.229
cutting edge of the frontier.
00:31:38.229 --> 00:31:39.230
Eric was also talking about
00:31:39.230 --> 00:31:40.698
how you need to stay an AI native.
00:31:40.698 --> 00:31:42.233
It's hard. And we're working
00:31:42.233 --> 00:31:44.102
actively with many of our CEOs
00:31:44.102 --> 00:31:45.503
to make them AI native
00:31:45.503 --> 00:31:47.272
and to bring workflows
00:31:47.272 --> 00:31:48.306
then change them with
00:31:48.306 --> 00:31:49.374
AI to help
00:31:49.374 --> 00:31:50.275
make our companies
00:31:50.275 --> 00:31:51.609
more efficient, improve them,
00:31:51.609 --> 00:31:53.544
have better quality, better products.
00:31:53.544 --> 00:31:54.312
Eric's doing that
00:31:54.312 --> 00:31:56.447
on the factory floor with Relativity.
00:31:56.447 --> 00:31:57.215
We're trying to do that
00:31:57.215 --> 00:31:58.983
across our 280 companies.
00:31:58.983 --> 00:31:59.250
Eric,
00:31:59.250 --> 00:32:00.184
you, and I were both talking
00:32:00.184 --> 00:32:00.652
about this.
00:32:00.652 --> 00:32:02.553
He had this line about,
00:32:02.553 --> 00:32:02.887
you know,
00:32:02.887 --> 00:32:04.088
if you're running one of these
00:32:04.088 --> 00:32:04.722
AI companies,
00:32:04.722 --> 00:32:05.690
and your revenue
00:32:05.690 --> 00:32:07.025
is completely determined
00:32:07.025 --> 00:32:08.126
by your data centers.
00:32:08.126 --> 00:32:09.460
We're certainly seeing at Blackstone
00:32:09.460 --> 00:32:11.162
massive data center demand
00:32:11.162 --> 00:32:12.697
that's just almost unmet.
00:32:12.697 --> 00:32:14.365
If one is able to generate
00:32:14.365 --> 00:32:16.668
$50 to $70 million of revenue
00:32:16.668 --> 00:32:17.769
per megawatt
00:32:17.769 --> 00:32:19.537
and your costs are $12 to $15
00:32:19.537 --> 00:32:21.139
all-in, including the chips
00:32:21.139 --> 00:32:23.141
depreciating over a shorter cycle
00:32:23.141 --> 00:32:24.108
and in the data center
00:32:24.108 --> 00:32:25.610
over a longer cycle,
00:32:25.610 --> 00:32:28.246
well, then what you should be doing
00:32:28.246 --> 00:32:30.415
is using all of your capital
00:32:30.415 --> 00:32:31.683
to purchase the next megawatt
00:32:31.683 --> 00:32:32.617
to serve that.
00:32:32.617 --> 00:32:32.984
And that's
00:32:32.984 --> 00:32:34.085
what's happening in the market.
00:32:34.085 --> 00:32:36.487
People are taking today's cash
00:32:36.487 --> 00:32:38.523
to purchase tomorrow's AI factory.
00:32:38.523 --> 00:32:38.923
That's great.
00:32:38.923 --> 00:32:39.824
Thank you so much for joining
00:32:39.824 --> 00:32:40.792
us, Jas. I appreciate it.
00:32:40.792 --> 00:32:41.826
Thank you, Christine.
00:32:41.826 --> 00:32:43.061
And thank you for joining us
00:32:43.061 --> 00:32:43.695
on this week's
00:32:43.695 --> 00:32:45.630
episode of Inside Blackstone.
00:32:45.630 --> 00:32:46.497
Follow us wherever
00:32:46.497 --> 00:32:47.498
you get your podcasts.