WEBVTT - OpenAI CEO Sam Altman Talks IPO Planning, DevDay

0:00:02.520 --> 0:00:07.070
<v Speaker 1>Bloomberg Audio Studios. Podcasts. Radio. News.

0:00:07.910 --> 0:00:10.230
<v Speaker 2>In just a moment, our own Ed Ludlow sitting down

0:00:10.270 --> 0:00:12.430
<v Speaker 2>with OpenAI's Sam Altman. We're going to take you to

0:00:12.450 --> 0:00:14.700
<v Speaker 2>the West Coast as soon as that begins. This is

0:00:14.710 --> 0:00:19.700
<v Speaker 2>happening at OpenAI's Dev Day, Developer's Day. A company targeting

0:00:19.700 --> 0:00:24.220
<v Speaker 2>$ 30 billion in funding at a $ 1. 4 trillion valuation. And

0:00:24.280 --> 0:00:27.010
<v Speaker 2>that is what their goal was, to get to that valuation.

0:00:27.170 --> 0:00:31.480
<v Speaker 3>Yeah. That, though, is like, it's not overshadowing, but that's not,

0:00:31.520 --> 0:00:33.380
<v Speaker 3>you know, Dev Day is supposed to be about DOTS, right?

0:00:33.420 --> 0:00:38.779
<v Speaker 3>This new Muse challenger to Meta. So, yeah, okay. Let's

0:00:38.800 --> 0:00:41.300
<v Speaker 3>go now to San Francisco and to Ed Ludlow, who's

0:00:41.320 --> 0:00:45.370
<v Speaker 3>standing by with Sam Altman, the CEO of OpenAI.

0:00:45.550 --> 0:00:49.050
<v Speaker 1>Sam, do you know what's interesting in the presentation is

0:00:49.090 --> 0:00:53.650
<v Speaker 1>how internally OpenAI staff talk to DOTS? Do you get

0:00:53.670 --> 0:00:55.950
<v Speaker 1>what I mean? I think I do, but tell me.

0:00:57.400 --> 0:01:01.150
<v Speaker 1>It isn't treating it like AI, like, hey, ChatGPT question.

0:01:01.290 --> 0:01:03.240
<v Speaker 1>It's a conversation. Yeah, yeah, yeah. It's not.

0:01:03.660 --> 0:01:06.400
<v Speaker 4>I don't think it's like you have to primorify this.

0:01:06.700 --> 0:01:08.330
<v Speaker 4>We don't quite talk to it like a person, but

0:01:08.390 --> 0:01:11.310
<v Speaker 4>there is a different style to it than traditional ChatGPT.

0:01:11.550 --> 0:01:12.690
<v Speaker 1>The reason I wanted to get to that first is,

0:01:12.950 --> 0:01:14.220
<v Speaker 1>who's DOTS for, really?

0:01:14.880 --> 0:01:17.590
<v Speaker 4>It started as a premium product. We know it's, you know,

0:01:17.630 --> 0:01:19.270
<v Speaker 4>we're putting our best model, Astra, in this.

0:01:19.550 --> 0:01:24.170
<v Speaker 1>Premium being prosumer, sorry to interrupt, or it's an enterprise product?

0:01:24.970 --> 0:01:26.589
<v Speaker 4>More like for getting work done, which I think is

0:01:26.650 --> 0:01:28.490
<v Speaker 4>largely enterprise, but there's a lot of people that are

0:01:28.510 --> 0:01:30.050
<v Speaker 4>going to build their own startups that are sort of

0:01:30.110 --> 0:01:32.880
<v Speaker 4>in the traditional prosumer category. But this is like a

0:01:33.060 --> 0:01:36.700
<v Speaker 4>very capable model and it's a ton of usage. So

0:01:37.200 --> 0:01:38.460
<v Speaker 4>we're starting for something at the top end of the

0:01:38.500 --> 0:01:40.459
<v Speaker 4>market and you will see us over time make a

0:01:40.520 --> 0:01:41.619
<v Speaker 4>mass market consumer version.

0:01:42.260 --> 0:01:45.900
<v Speaker 1>How much do you... and the team, various people at

0:01:46.180 --> 0:01:50.300
<v Speaker 1>OpenAI benchmark against the agents that are already out there.

0:01:50.340 --> 0:01:53.020
<v Speaker 1>So how much do you use a Grok bot, a Muse,

0:01:53.080 --> 0:01:56.580
<v Speaker 1>an Instinct from the product perspective? Because they've been out

0:01:56.610 --> 0:01:58.690
<v Speaker 1>there a while and people are actually using them. I've

0:01:58.890 --> 0:02:02.130
<v Speaker 1>tried them. I don't use them regularly. I have not.

0:02:02.980 --> 0:02:05.520
<v Speaker 4>Found a place for them in my life, but I

0:02:05.540 --> 0:02:08.910
<v Speaker 4>try to try every major new AI product. This is

0:02:08.940 --> 0:02:12.090
<v Speaker 4>a totally different category. The amount of work, the level

0:02:12.110 --> 0:02:14.330
<v Speaker 4>of capability, the level of intelligence this can take on,

0:02:15.270 --> 0:02:16.730
<v Speaker 4>I think this is just a completely new thing.

0:02:17.690 --> 0:02:20.570
<v Speaker 1>Why is it different? Literally, are we talking about its

0:02:21.110 --> 0:02:24.059
<v Speaker 1>capabilities or is this an economic discussion of how you've

0:02:24.160 --> 0:02:26.800
<v Speaker 1>priced it or where you've made it available in existing tiers?

0:02:27.389 --> 0:02:30.410
<v Speaker 4>Well, I mean, it is priced higher than other things

0:02:30.430 --> 0:02:33.109
<v Speaker 4>in the market. But the reason for that is because

0:02:33.150 --> 0:02:36.750
<v Speaker 4>of the capability and the amount of compute we're putting

0:02:36.790 --> 0:02:38.680
<v Speaker 4>into this, which is why I think it's useful in

0:02:38.720 --> 0:02:40.040
<v Speaker 4>a new kind of way at a new level.

0:02:40.400 --> 0:02:45.579
<v Speaker 1>You've launched GBT 6.1 Sol at Astra capability. And a

0:02:45.600 --> 0:02:47.620
<v Speaker 1>fifth of the price. But the price is so different.

0:02:47.660 --> 0:02:50.040
<v Speaker 1>And I've been trying to understand what is the big

0:02:50.080 --> 0:02:54.190
<v Speaker 1>picture story. It seems like... we're moving towards cheaper inference

0:02:54.610 --> 0:02:56.810
<v Speaker 1>at a time where everyone's much more willing to pay

0:02:57.210 --> 0:03:00.690
<v Speaker 1>for it. How did you arrive at the tiers that

0:03:00.720 --> 0:03:01.130
<v Speaker 1>you've done?

0:03:01.160 --> 0:03:03.100
<v Speaker 4>I mean, we've talked for a long time about our

0:03:03.139 --> 0:03:07.260
<v Speaker 4>goal is to make increasingly better intelligence and to drive

0:03:07.300 --> 0:03:10.280
<v Speaker 4>the price down and down. And the reason that I

0:03:10.300 --> 0:03:12.730
<v Speaker 4>think both of those are true is, The better you

0:03:12.770 --> 0:03:14.350
<v Speaker 4>make AI, the more people want to use it. So

0:03:14.389 --> 0:03:16.650
<v Speaker 4>people are clearly willing to spend more on AI than

0:03:16.690 --> 0:03:20.660
<v Speaker 4>ever before. But they want to do so much with it.

0:03:20.750 --> 0:03:23.290
<v Speaker 4>So our belief is that if we continue to drive

0:03:23.310 --> 0:03:26.370
<v Speaker 4>the quality up and the price down and really lead

0:03:26.410 --> 0:03:30.269
<v Speaker 4>the Pareto frontier at every point on that curve, people

0:03:30.330 --> 0:03:33.809
<v Speaker 4>will use our tools in tremendous ways. And just even

0:03:33.850 --> 0:03:36.530
<v Speaker 4>here today, hearing from developers about what they're building with

0:03:36.550 --> 0:03:39.150
<v Speaker 4>this and what they can do if we continue to

0:03:39.170 --> 0:03:40.930
<v Speaker 4>drive the price down is really quite remarkable.

0:03:41.440 --> 0:03:44.940
<v Speaker 1>What is OpenAI's business momentum right now? How did you

0:03:45.000 --> 0:03:46.580
<v Speaker 1>end the third quarter? What does that look like?

0:03:46.900 --> 0:03:48.910
<v Speaker 4>I don't think we put our specific numbers out.

0:03:48.940 --> 0:03:51.070
<v Speaker 1>There have been lots of numbers in the newspapers this morning.

0:03:51.490 --> 0:03:53.090
<v Speaker 1>It is our....

0:03:55.030 --> 0:03:56.790
<v Speaker 4>I mean, it's like a very steep growth period for us.

0:03:57.230 --> 0:04:00.550
<v Speaker 4>I think what people love are models. We have really

0:04:01.250 --> 0:04:03.660
<v Speaker 4>been able to make new products and new tools for developers,

0:04:03.680 --> 0:04:12.050
<v Speaker 4>for enterprises. ChatGPT is growing phenomenally well. So across the board, consumers, developers, enterprises,

0:04:12.190 --> 0:04:13.930
<v Speaker 4>all this new stuff that we're launching at Dev Day.

0:04:14.770 --> 0:04:17.390
<v Speaker 1>We see a lot of business momentum. The reported number

0:04:17.410 --> 0:04:21.770
<v Speaker 1>was ARR 70 billion, I think as a 3Q number.

0:04:22.310 --> 0:04:22.729
<v Speaker 4>I don't know.

0:04:22.750 --> 0:04:23.770
<v Speaker 1>Is that meaningful to you?

0:04:23.910 --> 0:04:27.550
<v Speaker 4>I don't know where these reported leaks or reported numbers

0:04:27.589 --> 0:04:30.500
<v Speaker 4>come from, but I am obviously not going to comment

0:04:30.690 --> 0:04:33.660
<v Speaker 4>either way on that, but I will say like very pleased.

0:04:33.500 --> 0:04:36.740
<v Speaker 1>With the growth. The interesting thing about DOT as a

0:04:36.779 --> 0:04:41.300
<v Speaker 1>product and agent but also 6.1 Sol, is you made

0:04:41.320 --> 0:04:45.200
<v Speaker 1>the decision to release those. This week, you held back

0:04:45.339 --> 0:04:49.060
<v Speaker 1>the release of Astra 6.1. And last week, you paused

0:04:49.140 --> 0:04:53.380
<v Speaker 1>training on a separate model. Explain how you come to

0:04:53.400 --> 0:04:58.330
<v Speaker 1>that decision. So Sol 6.1, okay. We held back Astra 6.1.

0:04:58.310 --> 0:05:02.270
<v Speaker 4>We've been talking about this for a few months now,

0:05:02.310 --> 0:05:08.150
<v Speaker 4>but we're clearly getting into a very steep part of

0:05:08.170 --> 0:05:11.330
<v Speaker 4>the capabilities curve. And given that our whole mission is

0:05:11.390 --> 0:05:15.960
<v Speaker 4>to build this for people and to empower people, and

0:05:16.000 --> 0:05:17.720
<v Speaker 4>part of that's safety, part of that's why they're distributing

0:05:17.740 --> 0:05:21.500
<v Speaker 4>the benefits, when we need to make a decision about

0:05:22.380 --> 0:05:25.540
<v Speaker 4>slowing down our training or our model release so that

0:05:25.580 --> 0:05:27.680
<v Speaker 4>we can make more progress and put more of our

0:05:27.720 --> 0:05:34.510
<v Speaker 4>attention into safety, alignment, monitoring, security, we will happily do that. Now,

0:05:34.589 --> 0:05:37.070
<v Speaker 4>of course we're going to continue to make progress, but

0:05:37.570 --> 0:05:41.330
<v Speaker 4>the world should always have confidence in our safety cases,

0:05:41.850 --> 0:05:44.710
<v Speaker 4>in our safety claims, and that we are going to

0:05:44.770 --> 0:05:47.469
<v Speaker 4>be responsible with this technology. And, and we don't want

0:05:47.490 --> 0:05:49.500
<v Speaker 4>to do this thing where like, Oh, we need, you know,

0:05:49.560 --> 0:05:51.640
<v Speaker 4>regulation or we need our competitors to do X, Y,

0:05:51.640 --> 0:05:53.400
<v Speaker 4>or Z, or we won't be safe. We are going

0:05:53.440 --> 0:05:55.540
<v Speaker 4>to do this responsibly. We want the world to have

0:05:55.560 --> 0:05:58.060
<v Speaker 4>that confidence. Um, now there are still things that we

0:05:58.100 --> 0:06:01.300
<v Speaker 4>can very reasonably release. So like a faster version of

0:06:01.380 --> 0:06:05.310
<v Speaker 4>Astra or a cheaper version of Astra, um, six we

0:06:05.350 --> 0:06:09.030
<v Speaker 4>can totally do of these wonderful new products. But, you know,

0:06:09.050 --> 0:06:12.190
<v Speaker 4>I think like, I understand why some people are just

0:06:12.230 --> 0:06:15.440
<v Speaker 4>like full Yolo pedal to the metal go, but we're

0:06:15.460 --> 0:06:17.320
<v Speaker 4>gonna have to be responsible as we push towards these

0:06:17.360 --> 0:06:18.419
<v Speaker 4>new levels of capabilities.

0:06:18.580 --> 0:06:20.740
<v Speaker 1>There are some difficult questions posed. I'm very conscious of

0:06:20.779 --> 0:06:23.659
<v Speaker 1>the developer audience and also like the users of the technology.

0:06:24.080 --> 0:06:26.080
<v Speaker 1>You said that the two main things that they want

0:06:26.140 --> 0:06:30.580
<v Speaker 1>are speed and performance. and you released new mechanisms to

0:06:30.620 --> 0:06:33.880
<v Speaker 1>achieve that with the existing generation of model. Why is

0:06:33.920 --> 0:06:37.839
<v Speaker 1>that better as a technique than releasing a more capable model?

0:06:38.100 --> 0:06:40.220
<v Speaker 4>Well, I mean, they want that too. We will also

0:06:40.240 --> 0:06:44.960
<v Speaker 4>do that. So I think all of those things are good. Speed, price,

0:06:45.060 --> 0:06:48.120
<v Speaker 4>and capability are certainly three of the things people ask

0:06:48.140 --> 0:06:50.940
<v Speaker 4>for the most often. And I think the answer is

0:06:51.290 --> 0:06:55.430
<v Speaker 4>people want all of them in different use cases. That said, ultra-fast,

0:06:55.610 --> 0:06:57.710
<v Speaker 4>I think, is a really special thing. I did not

0:06:57.730 --> 0:06:59.370
<v Speaker 4>appreciate how much I was going to like it and

0:06:59.390 --> 0:07:00.930
<v Speaker 4>how hard it is for me to go back. I

0:07:00.950 --> 0:07:03.610
<v Speaker 4>should have known because we all want everything to be faster, but.

0:07:03.550 --> 0:07:06.490
<v Speaker 1>It is awesome. Is it an economic proposition for OpenAI?

0:07:06.650 --> 0:07:08.010
<v Speaker 1>Are the economics a bit good for you?

0:07:10.470 --> 0:07:12.650
<v Speaker 4>Well, it's much more expensive to run, but we do

0:07:12.690 --> 0:07:14.890
<v Speaker 4>charge more for it too. So we tried to make

0:07:14.930 --> 0:07:16.450
<v Speaker 4>it something that, you know, it's like a.

0:07:16.870 --> 0:07:19.050
<v Speaker 1>We can talk to Sarah about the balance between that.

0:07:19.070 --> 0:07:20.850
<v Speaker 4>Yeah, I mean, it's like, you know, I think the...

0:07:21.540 --> 0:07:24.150
<v Speaker 4>market will decide, but I think people will be excited

0:07:24.190 --> 0:07:25.890
<v Speaker 4>about it at this price and it's reasonable for us.

0:07:26.230 --> 0:07:28.950
<v Speaker 1>You were just asked again about an IPO and what

0:07:28.990 --> 0:07:32.030
<v Speaker 1>you said is you want to be public one day. Yes.

0:07:32.230 --> 0:07:35.470
<v Speaker 1>But essentially the duties of being a public company are

0:07:35.510 --> 0:07:39.160
<v Speaker 1>incompatible with a dedication or focus on safety. Is that?

0:07:39.300 --> 0:07:39.780
<v Speaker 1>Not quite.

0:07:40.320 --> 0:07:42.540
<v Speaker 4>To be a newly public company and for us to

0:07:42.740 --> 0:07:47.080
<v Speaker 4>understand how to live life as a public company with,

0:07:47.180 --> 0:07:52.700
<v Speaker 4>you know, pressure from investors and sort of just all

0:07:52.720 --> 0:07:55.020
<v Speaker 4>of the additional weight that comes with that while we

0:07:55.060 --> 0:07:57.490
<v Speaker 4>are going through this period of adjusting to this new

0:07:57.520 --> 0:08:00.790
<v Speaker 4>level of capability and the new safety requirements for that.

0:08:01.170 --> 0:08:03.450
<v Speaker 4>We just want to get our feet under us, make

0:08:03.590 --> 0:08:06.490
<v Speaker 4>sure we understand how to operate in this new way,

0:08:06.550 --> 0:08:08.390
<v Speaker 4>be able to make some of these decisions in front

0:08:08.410 --> 0:08:10.750
<v Speaker 4>of us without the pressure of being a newly public company.

0:08:11.410 --> 0:08:14.170
<v Speaker 4>And then I think we'll be public at some point.

0:08:14.530 --> 0:08:16.690
<v Speaker 4>There will be more new levels that come, but it

0:08:16.750 --> 0:08:17.590
<v Speaker 4>won't be all at the same time.

0:08:18.170 --> 0:08:21.990
<v Speaker 1>Do you feel support from the existing cap table to

0:08:22.030 --> 0:08:24.010
<v Speaker 1>just hold off 2028? Does it matter?

0:08:25.670 --> 0:08:27.990
<v Speaker 4>Our investors seem like very happy with us and very patient.

0:08:28.210 --> 0:08:30.830
<v Speaker 1>Let me ask you this to conclude. It's very rare

0:08:30.940 --> 0:08:33.819
<v Speaker 1>people have said for you and Debra Amodei and Elon

0:08:33.860 --> 0:08:37.160
<v Speaker 1>Musk to agree on something, this idea of pacing the frontier.

0:08:37.600 --> 0:08:41.859
<v Speaker 1>And OpenAI has disclosed quite clearly like internal frameworks of

0:08:41.920 --> 0:08:44.140
<v Speaker 1>what you plan to do. I think people are still

0:08:44.160 --> 0:08:46.710
<v Speaker 1>trying to understand what the collective effort looks like. If

0:08:46.730 --> 0:08:50.340
<v Speaker 1>you collaborate with your peers or you work with the administration,

0:08:50.840 --> 0:08:52.239
<v Speaker 1>what does that actually result in?

0:08:53.050 --> 0:08:56.189
<v Speaker 4>Well, I think one thing that would be good would

0:08:56.230 --> 0:09:00.610
<v Speaker 4>be shared standards for safety cases about when, you know,

0:09:00.650 --> 0:09:04.429
<v Speaker 4>what you need at each level to proceed without undue risk,

0:09:04.929 --> 0:09:08.150
<v Speaker 4>and then some sort of shared methods for ensuring that

0:09:08.230 --> 0:09:10.890
<v Speaker 4>everybody is doing a good job in implementing those. And

0:09:10.929 --> 0:09:14.580
<v Speaker 4>that could be independent evaluators, that could be government review,

0:09:14.620 --> 0:09:16.480
<v Speaker 4>that could be companies checking each other's works. There have

0:09:16.500 --> 0:09:18.760
<v Speaker 4>been many proposals there, but I think those two things.

0:09:18.640 --> 0:09:21.760
<v Speaker 1>Would be good. How do you feel about the disclosure

0:09:21.820 --> 0:09:24.059
<v Speaker 1>of the risk. You talked to the United Nations about

0:09:24.100 --> 0:09:26.720
<v Speaker 1>it doesn't matter the percentage risk, it's still too great.

0:09:27.140 --> 0:09:32.140
<v Speaker 1>The news cycle had, in anthropics prospectus, the existential threat

0:09:32.160 --> 0:09:35.870
<v Speaker 1>to humanity disclosed in black and white. Is that incumbent

0:09:35.929 --> 0:09:39.250
<v Speaker 1>on you as a frontier lab? I appreciate you felt confidentially,

0:09:39.270 --> 0:09:40.910
<v Speaker 1>but do you just need to put it out there?

0:09:40.929 --> 0:09:43.650
<v Speaker 4>I think this has nothing to do with going public.

0:09:43.670 --> 0:09:45.929
<v Speaker 4>This is just, I think, our responsibility for our mission

0:09:46.050 --> 0:09:51.220
<v Speaker 4>and what we're we see in the sort of shifting

0:09:51.260 --> 0:09:52.240
<v Speaker 4>landscape as things evolve.

0:09:53.100 --> 0:09:56.250
<v Speaker 1>Sam Altman, OpenAI CEO, having us here at Dev Day.

0:09:56.330 --> 0:09:58.910
<v Speaker 1>Thank you very much. Thank you for coming. Indeed. Romain,

0:09:58.970 --> 0:09:59.330
<v Speaker 1>back to you.

0:09:59.350 --> 0:10:03.100
<v Speaker 3>All right. there you have it. OpenAI CEO Sam Altman

0:10:03.200 --> 0:10:06.320
<v Speaker 3>out there in San Francisco. It's Dev Day for OpenAI.

0:10:06.360 --> 0:10:09.880
<v Speaker 3>That's Sam Altman sitting down with Bloomberg Tech's Ed Ludlow.