WEBVTT

00:00:06.520 --> 00:00:12.360
Reid, it's my pleasure to be here with you
at Collège des Bernardins in Paris.

00:00:12.360 --> 00:00:14.800
You are a legend in Silicon Valley.

00:00:14.800 --> 00:00:18.560
You've worked closely with Peter Thiel and Elon Musk at PayPal,

00:00:18.560 --> 00:00:21.200
and you've been one of the founders of LinkedIn.

00:00:21.200 --> 00:00:23.960
Then, you've been an early investor in OpenAI.

00:00:23.960 --> 00:00:25.880
Regarding AI,

00:00:25.880 --> 00:00:29.600
each of us has had our “Aha!” moment using AI or GenAI.

00:00:29.600 --> 00:00:31.840
What's been yours, Reid?

00:00:31.840 --> 00:00:36.640
Well, I've had a number, but probably the very first one

00:00:36.720 --> 00:00:41.400
was the ability to start doing something
similar within the research arena.

00:00:41.400 --> 00:00:45.240
I had the fortune and the privilege

00:00:45.240 --> 00:00:48.240
to have very early access to GPT-4,

00:00:48.480 --> 00:00:51.240
and I realized what some of its superpowers were.

00:00:51.240 --> 00:00:54.680
So, what I started doing was saying, okay,

00:00:54.760 --> 00:00:57.520
I'd like to think about

00:00:57.520 --> 00:01:00.000
how AI’s development in what is

00:01:00.000 --> 00:01:03.000
known as a Ministry of Experts format

00:01:03.120 --> 00:01:06.440
could possibly parallel some aspects of economics,

00:01:06.640 --> 00:01:09.840
game theory, and the prisoner's dilemma.

00:01:09.920 --> 00:01:12.240
And so, I went in and started typing

00:01:12.240 --> 00:01:16.520
those questions, and it gave me great research briefs,

00:01:16.520 --> 00:01:19.320
things to research, things to look at, and so forth.

00:01:19.320 --> 00:01:25.560
It was like, ah, imagine if every

00:01:25.640 --> 00:01:27.240
information question that

00:01:27.240 --> 00:01:31.000
you might start with, which could
of course include writing a column,

00:01:31.080 --> 00:01:36.880
could be addressed by an immediate, on-demand
research assistant, who wouldn't always be right,

00:01:36.880 --> 00:01:40.520
but could always give you something
compelling to start with.

00:01:40.600 --> 00:01:46.720
And so, your learning is only bounded
by your ability to ask questions.

00:01:46.800 --> 00:01:49.760
When I was writing "Impromptu,"
the first book on AI,

00:01:49.760 --> 00:01:53.080
co-written with AI, published
last year,

00:01:53.160 --> 00:01:55.960
I created thousands of prompts and answers,

00:01:55.960 --> 00:01:59.680
and then I selected the ones
that were most interesting for readers.

00:01:59.760 --> 00:02:03.240
Because sometimes you go: well, that
first question, that first prompt wasn't very good,

00:02:03.240 --> 00:02:08.960
but when you asked,
"Answer this question from the perspective of a brand marketer

00:02:09.040 --> 00:02:11.960
who loves Apple," or

00:02:11.960 --> 00:02:15.400
"May as a brand,
now answer this question,"

00:02:15.400 --> 00:02:17.760
suddenly, you get
a much more interesting question,

00:02:17.760 --> 00:02:19.000
and then that becomes helpful.

00:02:19.000 --> 00:02:22.000
And that's part of what you learn
through the process

00:02:22.200 --> 00:02:24.600
of beginning to experiment
with these tools.

00:02:24.600 --> 00:02:30.120
Thank you for this, Reid, for your thoughtful
perspectives on the technology.

00:02:30.120 --> 00:02:33.800
And thank you for being with me
here at Collège des Bernardins today.

00:02:33.920 --> 00:02:34.520
My pleasure.