WEBVTT

00:00:00.080 --> 00:00:02.077
You have described that AI is a plat

00:00:02.160 --> 00:00:05.037
plat uh platform shift

00:00:05.120 --> 00:00:07.117
and what does what does that mean?

00:00:07.200 --> 00:00:09.677
Question one, where do you see that

00:00:09.760 --> 00:00:14.477
shift going in the next few years?
And importantly, the third part of my

00:00:14.560 --> 00:00:18.797
question would be fast forwarding
a few years, five years,

00:00:18.880 --> 00:00:20.157
what's going to seem obvious in

00:00:20.240 --> 00:00:23.757
hindsight that feels less clear today?

00:00:23.840 --> 00:00:26.397
You know, first of all, um it's great uh

00:00:26.480 --> 00:00:29.917
to be back here, uh Larry, and it's um I

00:00:30.000 --> 00:00:31.757
had a chance, in fact, yesterday when

00:00:31.840 --> 00:00:34.557
you put out the letter uh to kick off

00:00:34.640 --> 00:00:37.997
the forum and read it. And um and in

00:00:38.080 --> 00:00:41.437
there, you sort of had this one line of

00:00:41.520 --> 00:00:44.637
uh really I think when it comes to AI,

00:00:44.720 --> 00:00:46.477
the real question in front of all of us

00:00:46.560 --> 00:00:49.277
is how do you ensure that the diffusion

00:00:49.360 --> 00:00:51.677
of AI happens and happens fast? I mean I

00:00:51.760 --> 00:00:53.597
think you had that line of how do the

00:00:53.680 --> 00:00:56.637
models, the data and the infrastructure

00:00:56.720 --> 00:00:59.037
spread more evenly to create surplus

00:00:59.120 --> 00:01:00.717
everywhere. If you sort of think about

00:01:00.800 --> 00:01:04.477
it, the the way I come at this is not

00:01:04.560 --> 00:01:08.237
that um this has always been the arc of

00:01:08.320 --> 00:01:12.717
computation, right? You can sort of take
it in the last 30 years or the last 70

00:01:12.800 --> 00:01:14.877
years. It's always been about can you

00:01:14.960 --> 00:01:16.477
digitize

00:01:16.560 --> 00:01:20.317
artifacts on about people, places, and

00:01:20.400 --> 00:01:24.557
things and then build analytical and

00:01:24.640 --> 00:01:25.917
predictive [clears throat] power. Right?

00:01:26.000 --> 00:01:30.637
That's what the mainframes did. That's
what the mini computer did. That's what

00:01:30.720 --> 00:01:34.797
the client server error did. That's
what the web error did. The mobile cloud

00:01:34.880 --> 00:01:36.957
error did. So it like it depend

00:01:37.040 --> 00:01:40.157
irrespective of which paradigm or

00:01:40.240 --> 00:01:44.077
platform. It has been one continuous arc

00:01:44.160 --> 00:01:47.357
of saying let's make better sense of

00:01:47.440 --> 00:01:49.277
this world

00:01:49.360 --> 00:01:53.997
um by reasoning about it in digital form

00:01:54.080 --> 00:01:58.877
because in some sense once you have
these artifacts in digital form you can

00:01:58.960 --> 00:02:01.757
use a more malleable
resource like software

00:02:01.840 --> 00:02:06.317
>> right >> um which
doesn't have the same type of

00:02:06.400 --> 00:02:08.157
you know I'll call it [clears throat]

00:02:08.240 --> 00:02:11.197
marginal cost economics associated with

00:02:11.280 --> 00:02:13.997
it that allows us to then build more

00:02:14.080 --> 00:02:17.517
insight and more uh more capability and

00:02:17.600 --> 00:02:20.237
in that context AI I would say is of the

00:02:20.320 --> 00:02:23.117
same class at least like the web or the

00:02:23.200 --> 00:02:26.877
internet um or mobile or PC or the cloud

00:02:26.960 --> 00:02:31.037
or and maybe even greater and so to me

00:02:31.120 --> 00:02:34.317
right now where we are is you know let's

00:02:34.400 --> 00:02:36.557
take just what's happened with software

00:02:36.640 --> 00:02:39.197
engineering right which is one you know

00:02:39.280 --> 00:02:41.997
is knowledge work um you know you could

00:02:42.080 --> 00:02:44.237
say it's elite knowledge work

00:02:44.320 --> 00:02:47.197
>> it started off uh you know in fact my

00:02:47.280 --> 00:02:51.997
own belief in this generation
of AI and its capability

00:02:52.080 --> 00:02:56.317
uh really got built up when I first saw
GitHub copilot do code completions right

00:02:56.400 --> 00:03:00.877
so for the longest time we had the dream
that if you're a software developer can

00:03:00.960 --> 00:03:03.277
you predict the next sort of word or the

00:03:03.360 --> 00:03:06.957
next uh line of code uh and suddenly it

00:03:07.040 --> 00:03:09.597
started working with these models u then

00:03:09.680 --> 00:03:11.597
You said okay if I can do that then can

00:03:11.680 --> 00:03:14.797
I actually go and bring back you know

00:03:14.880 --> 00:03:19.117
the flow for a software developer
by going to a chat session and asking any

00:03:19.200 --> 00:03:23.517
question and it comes back with answers
that then uh you can use in your coding

00:03:23.600 --> 00:03:27.197
flow right so that was the next thing
then you said well if that's working can

00:03:27.280 --> 00:03:32.137
I assign it small tasks that was the
agent mode uh now you have complete

00:03:32.220 --> 00:03:35.357
[clears throat] autonomous agents where
you can give it your entire project

00:03:35.440 --> 00:03:40.077
right it can work uh you know >> 24/7
>> it can work for 24/7 7. I mean it's

00:03:40.160 --> 00:03:43.997
still we've got some ways to go
for these things to remain coherent long

00:03:44.080 --> 00:03:46.477
time but nevertheless it's getting

00:03:46.560 --> 00:03:49.117
better and better and interestingly

00:03:49.200 --> 00:03:51.917
enough you look at it uh the software

00:03:52.000 --> 00:03:56.397
developer still is got a lot of agency
in it right so that's why I kind of

00:03:56.480 --> 00:04:00.077
still think that you know going
and thinking of these as somehow living

00:04:00.160 --> 00:04:02.877
outside of the realm of human agency is

00:04:02.960 --> 00:04:07.197
probably not the right
way to think about it.

00:04:07.280 --> 00:04:12.097
If you like the video do like,
comment, share and subscribe.

00:04:12.180 --> 00:04:14.200
[music]