WEBVTT

00:00:01.310 --> 00:00:03.330
[cheering]

00:00:08.280 --> 00:00:11.477
>> Thanks for having me.
>> Great to have you, Sam.

00:00:11.560 --> 00:00:14.477
>> It's interesting cuz
people have been talking about

00:00:14.560 --> 00:00:19.637
the potential downsides or serious
risks of AI for some time.

00:00:19.720 --> 00:00:23.117
And I think a lot of people have been
saying, "Why is this the moment where it

00:00:23.200 --> 00:00:25.997
sort of escaped the bubble and is now

00:00:26.080 --> 00:00:30.717
all of a sudden
an international topic of such

00:00:30.800 --> 00:00:34.597
intensity?" And I think a few
things have happened. One,

00:00:34.680 --> 00:00:40.397
the models have gotten so good. And people
for a long time looked at the models.

00:00:40.480 --> 00:00:43.997
Maybe they looked at the rate of progress,
but I don't think people really felt that

00:00:44.080 --> 00:00:47.797
the models were going
to come this far, this fast.

00:00:47.880 --> 00:00:52.797
We will get it right, by the way.
I am very confident in our

00:00:52.880 --> 00:00:56.837
I think the biggest thing that has changed
is this thing that our industry has been

00:00:56.920 --> 00:01:00.837
talking about for a while, which is no,
no, this technology, this approach is

00:01:00.920 --> 00:01:05.077
going to go very far. These models
are in some ways going to become more

00:01:05.160 --> 00:01:09.197
capable, smarter than people.
That finally happened. So, I think that

00:01:09.280 --> 00:01:13.397
was a big change to why now. And people
say, you know, my probability of doom is

00:01:13.480 --> 00:01:17.357
X or Y or whatever, it hits
in a different way. Technology

00:01:17.440 --> 00:01:20.877
to do. Somebody else wants
to build like a hacking bot, okay.

00:01:20.960 --> 00:01:27.680
The world can decide what they think
about that, but I think you also see um

00:01:28.080 --> 00:01:32.157
a real fear that some of these companies
or some companies developing AI could

00:01:32.240 --> 00:01:36.317
get too much power and be
able to sort of exert undue

00:01:36.400 --> 00:01:41.077
influence on the economy, push
a worldview out on people. And I think the

00:01:41.160 --> 00:01:45.557
world is right to be afraid of this. So,
we have these two big challenges. We have

00:01:45.640 --> 00:01:49.997
the potential of a loss
of control accident

00:01:50.080 --> 00:01:53.157
or some other serious thing that could
go wrong. too much power concentration

00:01:53.240 --> 00:01:55.917
and people fear that those developing AI

00:01:56.000 --> 00:02:00.397
could exert their worldview.
Um and there is a sort of narrow path we

00:02:00.480 --> 00:02:06.157
have to navigate with pragmatism, with

00:02:06.240 --> 00:02:09.397
you know, steadiness,
with the ability for the the

00:02:09.480 --> 00:02:13.077
world to like trust that we will
make dependable decisions, consistent

00:02:13.160 --> 00:02:17.197
decisions uh through this.
And I think people are saying, "Well,

00:02:17.280 --> 00:02:22.717
you know, this now seems pretty important.
Like let's let's get it right."

00:02:22.800 --> 00:02:25.277
I'm not going to get the days of the
week exactly right here, but something

00:02:25.360 --> 00:02:29.797
on like a Friday or a Saturday.
These models have gotten so good, so fast,

00:02:29.880 --> 00:02:31.957
that we need to treat the alignment and

00:02:32.040 --> 00:02:34.197
monitoring and security with a new level

00:02:34.280 --> 00:02:36.917
of rigor. And we have to be willing to

00:02:37.000 --> 00:02:39.077
pace our development of capabilities

00:02:39.160 --> 00:02:44.717
such that capabilities always So
such that alignment, safety, and

00:02:44.800 --> 00:02:48.157
monitoring are always
ahead of capabilities. And

00:02:48.240 --> 00:02:51.517
the Hugging Face incident triggered
a real reset, not just in our own company,

00:02:51.600 --> 00:02:53.237
but I think the whole industry.

00:02:53.320 --> 00:02:54.957
>> You know, we're
in the cyber area today,

00:02:55.040 --> 00:02:56.877
but maybe we move into another area. Is

00:02:56.960 --> 00:03:02.077
this a fair way to characterization,
or am I or is this unfair to characterize

00:03:02.160 --> 00:03:04.637
it like this? >> What I really
care about is that we have

00:03:04.720 --> 00:03:08.957
got a great culture of accident
reporting and learning. I think if we

00:03:09.040 --> 00:03:13.957
look at what happens with the FAA and the
NTSB and how safe airplanes have gotten,

00:03:14.040 --> 00:03:17.357
um as I understand it from talking
to people who have really studied,

00:03:17.440 --> 00:03:20.997
you know, new kinds of technology on the
order of magnitude of airplanes, one of

00:03:21.080 --> 00:03:24.797
the things that has gone so well there is

00:03:24.880 --> 00:03:26.637
is the culture of accident reporting.

00:03:26.720 --> 00:03:28.317
You know, accidents are going to happen.

00:03:28.400 --> 00:03:30.077
We're going to learn as much as we can

00:03:30.160 --> 00:03:35.597
and react from each one. Um and if,
you know, there's other kinds

00:03:35.680 --> 00:03:40.037
of technology where I don't think it's
gone I don't think society's had the same

00:03:40.120 --> 00:03:44.237
response, and I don't think
we have improved as much.

00:03:44.320 --> 00:03:47.397
to pull this thing together and present
you with the information to make a

00:03:47.480 --> 00:03:50.597
decision or to ble I think
this is one of the the

00:03:50.680 --> 00:03:54.797
profound changes to how we use technology

00:03:54.880 --> 00:03:57.917
that I've ever seen. I mean,
sci-fi movies, as you mentioned,

00:03:58.000 --> 00:04:01.637
have been hinting at this or playing
around with this for a long time, but

00:04:01.720 --> 00:04:04.557
this is one of those moments where it
feels like you're in the movie. Like the

00:04:04.640 --> 00:04:08.357
sci-fi thing came true.
You're talking to your personal super

00:04:08.440 --> 00:04:12.677
assistant, and you are working 10 times
faster or better or more creatively than

00:04:12.760 --> 00:04:16.557
you could have before. >> Sam,
you're invited back anytime you

00:04:16.640 --> 00:04:18.237
want to Dreamforce. Hold on, I'm

00:04:18.320 --> 00:04:20.997
>> Thank you very much. Thank you.

00:04:21.080 --> 00:04:24.280
>> Please thank Sam Altman.