WEBVTT

00:00:00.080 --> 00:00:02.637
The thing that I want to just double
click on before we play this uh general

00:00:02.720 --> 00:00:05.677
AI clip is that he talks about how just

00:00:05.760 --> 00:00:07.917
more compute is not going to get us

00:00:08.000 --> 00:00:10.717
there. And I think that that's become

00:00:10.800 --> 00:00:14.477
less of a commonly held view in the

00:00:14.560 --> 00:00:17.997
let's get the newest Nvidia chips
and get 20,000 of them in one building and

00:00:18.080 --> 00:00:21.517
then push really hard. Meta in particular,
XAI in particular, have been

00:00:21.600 --> 00:00:25.597
investing a lot in building up their
compute power as a way to create more

00:00:25.680 --> 00:00:28.877
intelligent systems. So, it's funny
to hear him here say that that's not going

00:00:28.960 --> 00:00:30.317
to get us all the way to the mountain top.

00:00:30.400 --> 00:00:34.397
>> What a a helpful thing. That made
me feel less crazy because I feel like

00:00:34.480 --> 00:00:38.797
sometimes there is that assumption
that we don't need to like code things

00:00:38.880 --> 00:00:41.277
anymore. We just need to put enough

00:00:41.360 --> 00:00:46.237
chips in a room and then they will
figure it out for us. And to me, that's

00:00:46.320 --> 00:00:50.317
always I'm not a programmer, folks.
So, like maybe that's how it'll work. But

00:00:50.400 --> 00:00:52.317
that always felt very counterintuitive

00:00:52.400 --> 00:00:54.397
to me. like every other piece of

00:00:54.480 --> 00:00:57.677
software ever, we a human had to like

00:00:57.760 --> 00:01:01.597
figure out how it was going to work. And
now we're suddenly coming with this idea

00:01:01.680 --> 00:01:05.917
that well, if you just put enough
compute power together, it'll take care

00:01:06.000 --> 00:01:09.677
of all of that for us. And I've never
really bought that. So, it was it was

00:01:09.760 --> 00:01:14.557
nice to hear a technologist agree with me
in that way, like, oh, I'm not crazy.

00:01:14.640 --> 00:01:16.557
This is a big assumption for us to make.

00:01:16.640 --> 00:01:19.357
>> It'd be fun to hear what he has to
think about this now, but here's Alexander

00:01:19.440 --> 00:01:21.197
Wang. >> You believe
in general AI that we'll hit

00:01:21.280 --> 00:01:24.877
that at some point? AI that is just
generally smart can do anything a human

00:01:24.960 --> 00:01:28.477
can. >> I believe in some sense.
I believe in the sense that like for most

00:01:28.560 --> 00:01:31.997
technological things that humans can
conceive of that aren't physically

00:01:32.080 --> 00:01:34.717
impossible if humans survive they'll
happen at some point. Like I think

00:01:34.800 --> 00:01:38.637
humans are like infinitely creative,
infinitely ingenious, etc. I think it's

00:01:38.720 --> 00:01:42.637
very overblown the timelines that people
are talking about general AI happening.

00:01:42.720 --> 00:01:45.437
There's a lot of things that are wrong
about like the common arguments. One of

00:01:45.520 --> 00:01:49.117
which is people say that if Moore's
law keeps going then we'll have all this

00:01:49.200 --> 00:01:52.077
exponential compute. it's only a matter
of time before we produce these general

00:01:52.160 --> 00:01:55.997
AI morals laws is going to be dead and
then quantum computing is very far away

00:01:56.080 --> 00:01:59.677
despite recent press releases etc.
I think that leg of the argument is not

00:01:59.760 --> 00:02:02.557
actually that strong and then I also
think it's not even clear that if you

00:02:02.640 --> 00:02:05.917
have infinite compute you'll be able
to produce general AI. I think that's very

00:02:06.000 --> 00:02:10.077
unclear. So infinite compute helps
narrow AI because you're doing a number

00:02:10.160 --> 00:02:15.117
of scenarios and playing out every
scenario in go the the game many more

00:02:15.200 --> 00:02:18.157
permutations than poker many more
permutations than chess which is a

00:02:18.240 --> 00:02:20.797
finite data set. So yeah, more compute

00:02:20.880 --> 00:02:23.517
power on those things certainly gets you

00:02:23.600 --> 00:02:27.917
quicker ability or even just throwing
people into a random video game like

00:02:28.000 --> 00:02:32.237
open AI is sure >> definitely >>
but general AI taking somebody who

00:02:32.320 --> 00:02:36.397
mastered chess and then saying master
go and then master impressionist painting

00:02:36.480 --> 00:02:39.117
it's different >> it's very
different one of the arguments

00:02:39.200 --> 00:02:42.877
goes that once you have enough compute
you can create artificial life by

00:02:42.960 --> 00:02:45.837
basically simulating evolution
one of the more vogue arguments

00:02:45.920 --> 00:02:50.557
>> you have so much compute power that you
can say start with this tiny piece of

00:02:50.640 --> 00:02:54.157
bacteria, whatever, then grow it
and grow and grow an entire evolutionary

00:02:54.240 --> 00:02:58.477
system to the point at which there
is a human-like species and then grow that

00:02:58.560 --> 00:03:02.077
humanlike species in whatever number of
scenarios with a big brain into whatever

00:03:02.160 --> 00:03:04.557
comes after us. >> Even if
you just grow the humanlike

00:03:04.640 --> 00:03:06.957
species that's as intelligent,
then you're that's kind of good.

00:03:07.040 --> 00:03:08.957
>> That would be general
AI as well, right?

00:03:09.040 --> 00:03:11.517
Because general AI would normally most
people would define general AI as not

00:03:11.600 --> 00:03:14.237
even being smarter than us, but being
as smart. But somebody would have to code

00:03:14.320 --> 00:03:17.117
that and program that and build the
systems to do that. It's not just going

00:03:17.200 --> 00:03:20.077
to magically happen, right?
It's very unclear if that's even possible.

00:03:20.160 --> 00:03:21.997
>> That's an argument.
>> I honestly think that's the most

00:03:22.080 --> 00:03:25.117
plausible argument. It's very much so
science fiction in the sense that we're

00:03:25.200 --> 00:03:29.357
not close to being able to even validate
the hypothesis. So, I don't believe in

00:03:29.440 --> 00:03:33.357
general AI anytime soon. >> AI,
like everything else, the old saying

00:03:33.440 --> 00:03:35.517
is so true. Garbage in, garbage out.

00:03:35.600 --> 00:03:39.197
Without the right training data, you're
not going to get great results. But my

00:03:39.280 --> 00:03:43.037
pals at Uber, they're now working
with enterprises all over the world to

00:03:43.120 --> 00:03:47.757
source, label, evaluate, and scale
highquality real data for your startup.

00:03:47.840 --> 00:03:50.797
Uber is one of my favorite companies
on the planet because they demonstrated

00:03:50.880 --> 00:03:54.877
their ability to scale and to build
great products first in the ride share

00:03:54.960 --> 00:03:59.357
market. Then they figured out Uber Eats
and then they moved on to autonomous

00:03:59.440 --> 00:04:03.437
driving. And all of this was only
possible because of a deep understanding

00:04:03.520 --> 00:04:08.157
of how to collect, label, understand,
and analyze data. Now with Uber AI

00:04:08.240 --> 00:04:12.157
solutions, you can put the team
and tools that turned Uber into one of the

00:04:12.240 --> 00:04:13.997
world's best companies to work for you.

00:04:14.080 --> 00:04:18.477
partnering with top talent and experts
from Uber's global network. What better

00:04:18.560 --> 00:04:20.317
partner to help you scale your business

00:04:20.400 --> 00:04:25.837
than the company that organizes over 36
million rides every single day? Book a

00:04:25.920 --> 00:04:32.280
demo right now by going to uber.com/twist.
That's u ber.com/twist.