WEBVTT

00:00:00.240 --> 00:00:03.997
We're increasingly close to what I've
called a country of geniuses in a data

00:00:04.080 --> 00:00:07.037
center. A set of AI agents that are more

00:00:07.120 --> 00:00:09.737
capable than most humans at most things

00:00:09.820 --> 00:00:12.797
[music] and can coordinate at superhuman

00:00:12.880 --> 00:00:17.037
speed. That level of capability
is something the world has never seen

00:00:17.120 --> 00:00:19.597
before and brings a very wide range of

00:00:19.680 --> 00:00:22.157
both opportunities and concerns for

00:00:22.240 --> 00:00:24.717
humanity. On the positive side, we have

00:00:24.800 --> 00:00:27.437
the potential to cure diseases that have

00:00:27.520 --> 00:00:29.917
been incurable for thousands of years,

00:00:30.000 --> 00:00:35.037
to radically improve humans health,
and to lift billions out of poverty,

00:00:35.120 --> 00:00:39.357
including in the global south,
and create a better world for everyone. On

00:00:39.440 --> 00:00:41.597
the side of risks, I'm concerned about

00:00:41.680 --> 00:00:44.637
the autonomous behavior of AI models,

00:00:44.720 --> 00:00:49.117
their potential for misuse
by individuals and governments, and their

00:00:49.200 --> 00:00:51.517
potential for economic displacement.

00:00:51.600 --> 00:00:55.757
First, I want to thank Prime Minister
Modi for bringing us together. The

00:00:55.840 --> 00:00:58.317
energy and ambition in this room and

00:00:58.400 --> 00:01:00.557
across India are incredible. I've been

00:01:00.640 --> 00:01:03.117
spending the last few days meeting with

00:01:03.200 --> 00:01:05.917
Indian builders and enterprises, and the

00:01:06.000 --> 00:01:07.757
energy to build together here is

00:01:07.840 --> 00:01:11.117
palpable, unlike anywhere else. Um, this

00:01:11.200 --> 00:01:13.837
is the fourth AI summit we've held since

00:01:13.920 --> 00:01:16.637
the tradition was initiated at Bletchley

00:01:16.720 --> 00:01:19.837
Park back in 2023, which I still

00:01:19.920 --> 00:01:22.957
remember. And in those 2.5 years, the

00:01:23.040 --> 00:01:27.677
advances in the technology have been
absolutely staggering. Along with those,

00:01:27.760 --> 00:01:31.757
the advances in the commercial
applications and the societal and

00:01:31.840 --> 00:01:34.317
ethical questions around the technology

00:01:34.400 --> 00:01:37.197
have only grown more urgent. My

00:01:37.280 --> 00:01:40.397
fundamental view is that AI has been on

00:01:40.480 --> 00:01:42.797
an exponential for the la for the last

00:01:42.880 --> 00:01:46.397
10 years um and as part of a sort of

00:01:46.480 --> 00:01:48.957
Moore's law for intelligence and that we

00:01:49.040 --> 00:01:51.517
are now well advanced on that curve and

00:01:51.600 --> 00:01:53.757
there are only a small number of years

00:01:53.840 --> 00:01:56.637
for AI models surpassing the cognitive

00:01:56.720 --> 00:02:01.437
capabilities of most humans for most
things. We're increasingly close to what

00:02:01.520 --> 00:02:03.357
I've called a country of geniuses in a

00:02:03.440 --> 00:02:06.317
data center. a set of AI agents that are

00:02:06.400 --> 00:02:08.797
more capable than most humans at most

00:02:08.880 --> 00:02:12.397
things and can coordinate at superhuman

00:02:12.480 --> 00:02:16.637
speed. That level of capability
is something the world has never seen

00:02:16.720 --> 00:02:19.197
before and brings a very wide range of

00:02:19.280 --> 00:02:21.757
both opportunities and concerns for

00:02:21.840 --> 00:02:24.317
humanity. On the positive side, we have

00:02:24.400 --> 00:02:27.037
the potential to cure diseases that have

00:02:27.120 --> 00:02:29.517
been incurable for thousands of years,

00:02:29.600 --> 00:02:34.637
to radically improve humans health,
and to lift billions out of poverty,

00:02:34.720 --> 00:02:38.957
including in the global south,
and create a better world for everyone. On

00:02:39.040 --> 00:02:41.197
the side of risks, I'm concerned about

00:02:41.280 --> 00:02:44.237
the autonomous behavior of AI models,

00:02:44.320 --> 00:02:46.477
their potential for misuse by

00:02:46.560 --> 00:02:48.717
individuals and governments, and their

00:02:48.800 --> 00:02:51.357
potential for economic displacement.

00:02:51.440 --> 00:02:54.157
India has an absolutely central role to

00:02:54.240 --> 00:02:56.717
play in these questions and challenges

00:02:56.800 --> 00:03:01.437
both on the side of the opportunities
and on the side of the risks. As a sign

00:03:01.520 --> 00:03:03.677
of our commitment, we just this week

00:03:03.760 --> 00:03:06.397
opened an office in Bengaluru and hired

00:03:06.480 --> 00:03:10.527
Ireina Gos who has spent

00:03:10.610 --> 00:03:14.637
[applause] who has spent
three decades building

00:03:14.720 --> 00:03:17.197
businesses in India as our managing

00:03:17.280 --> 00:03:19.997
director for Anthropic India. We've also

00:03:20.080 --> 00:03:22.157
announced partnerships with major Indian

00:03:22.240 --> 00:03:25.117
enterprises this week including Infosys

00:03:25.200 --> 00:03:29.277
and others on the opportunities.

00:03:29.360 --> 00:03:31.597
One dynamic that we have observed is

00:03:31.680 --> 00:03:34.397
that technology and practices pioneered

00:03:34.480 --> 00:03:36.557
in India have historically set a

00:03:36.640 --> 00:03:39.677
standard for the global south and have

00:03:39.760 --> 00:03:44.077
helped to diffuse technology and
humanitarian benefits through the global

00:03:44.160 --> 00:03:48.237
south. We're therefore partnering with,
we have been partnering with for several

00:03:48.320 --> 00:03:50.797
months, nonprofits such as the Xstep

00:03:50.880 --> 00:03:53.757
Foundation, Prome and Central Square

00:03:53.840 --> 00:03:56.077
Foundation to use our models to advance

00:03:56.160 --> 00:03:58.637
digital infrastructure, education,

00:03:58.720 --> 00:04:00.957
agricultural efficiency, and health in

00:04:01.040 --> 00:04:03.437
the hopes of spreading AI's benefits

00:04:03.520 --> 00:04:06.237
across the global south, starting with

00:04:06.320 --> 00:04:10.637
India and diffusing out to the rest
of the global south. We're also partnering

00:04:10.720 --> 00:04:13.677
with Karia and the collective
intelligence project to build

00:04:13.760 --> 00:04:16.957
evaluations and metrics of our model

00:04:17.040 --> 00:04:19.277
claude's performance on India's many

00:04:19.360 --> 00:04:21.997
regional languages on practical and

00:04:22.080 --> 00:04:24.637
locally relevant tasks we'll benchmark

00:04:24.720 --> 00:04:27.357
like agriculture legal tasks and

00:04:27.440 --> 00:04:30.957
educational content on the risks India

00:04:31.040 --> 00:04:35.437
is the world's largest democracy and can
be a partner and leader in addressing

00:04:35.520 --> 00:04:39.837
the global security and economic risks
of the technology we'd like to work with

00:04:39.920 --> 00:04:42.237
India on testing and evaluation of

00:04:42.320 --> 00:04:44.637
models for safety and security risks in

00:04:44.720 --> 00:04:49.677
the tradition that was started
by many global and national AI security

00:04:49.760 --> 00:04:51.197
institutes that have been stood up

00:04:51.280 --> 00:04:54.157
around the world. Even more, we see a

00:04:54.240 --> 00:04:56.717
particularly strong opportunity to work

00:04:56.800 --> 00:05:01.277
with India on studying the economic
questions as part of the New Delhi

00:05:01.360 --> 00:05:03.597
Frontier AI impact commitments, which

00:05:03.680 --> 00:05:05.997
we're excited to join. As part of our

00:05:06.080 --> 00:05:08.317
anthropic economic futures program and

00:05:08.400 --> 00:05:10.957
anthropic economic index, we publish

00:05:11.040 --> 00:05:14.717
statistical insights into how AI impacts

00:05:14.800 --> 00:05:19.757
jobs in the economy. We're excited
to increasingly share this information,

00:05:19.840 --> 00:05:23.997
exchange information with the Indian
government to share insights and inform

00:05:24.080 --> 00:05:26.717
evidence-based policym convene meetings

00:05:26.800 --> 00:05:28.957
with economists, labor leaders and

00:05:29.040 --> 00:05:32.157
policy makers to adjust to to to adapt

00:05:32.240 --> 00:05:36.797
to the economic impacts of AI.
We believe that AI will greatly grow the

00:05:36.880 --> 00:05:41.037
economic pie including in India and
the global south. But that because it is

00:05:41.120 --> 00:05:45.037
happening so fast, it may lead to a time
of disruption and we need to work

00:05:45.120 --> 00:05:47.357
together between companies and the

00:05:47.440 --> 00:05:49.757
government to better manage that time of

00:05:49.840 --> 00:05:52.557
disruption and bring better prosperity

00:05:52.640 --> 00:05:55.757
smoothly to all. Um I and anthropic are

00:05:55.840 --> 00:05:59.997
very grateful to be part of all these
efforts and I'm honored to be here and

00:06:00.080 --> 00:06:04.757
working on these
questions with all of you.

00:06:04.840 --> 00:06:06.860
[applause]