WEBVTT - OpenAI COO Brad Lightcap Talks GPT-5

0:00:02.759 --> 0:00:10.240
<v Speaker 1>Bloomberg Audio Studios, podcasts, radio news. We want to welcome

0:00:10.360 --> 0:00:14.080
<v Speaker 1>our global audience across Bloomberg Radio and television Open AI.

0:00:14.320 --> 0:00:17.919
<v Speaker 1>It's released GPT five. It's most advanced model yet. The

0:00:17.920 --> 0:00:22.639
<v Speaker 1>company says it offers key improvements in major areas like reliability, accuracy,

0:00:22.800 --> 0:00:24.800
<v Speaker 1>and there's the strongest generator of AI model yet in

0:00:24.880 --> 0:00:26.160
<v Speaker 1>coding and writing and health.

0:00:26.760 --> 0:00:27.160
<v Speaker 2>For more.

0:00:27.280 --> 0:00:30.639
<v Speaker 1>We bring in Brad lightcap Open AI COO and this

0:00:30.720 --> 0:00:34.319
<v Speaker 1>feels so primed for enterprise adoption. Brad, when I think

0:00:34.360 --> 0:00:36.280
<v Speaker 1>of writing, when I think of coding, what is the

0:00:36.280 --> 0:00:37.160
<v Speaker 1>opportunity there?

0:00:38.560 --> 0:00:42.680
<v Speaker 3>Yeah, well, good morning, thanks for having me. GBT five

0:00:42.760 --> 0:00:45.760
<v Speaker 3>is a significant step forward in a few different domains.

0:00:45.800 --> 0:00:49.080
<v Speaker 3>So you mentioned coding, you mentioned writing in health for

0:00:49.200 --> 0:00:53.760
<v Speaker 3>consumers and medical professionals, and we think that opportunity unlocks

0:00:54.040 --> 0:00:55.880
<v Speaker 3>an amazing set of things in the enterprise that now

0:00:55.880 --> 0:00:56.680
<v Speaker 3>become possible.

0:00:57.600 --> 0:00:58.920
<v Speaker 4>It's a much more reliable model.

0:00:58.960 --> 0:01:01.640
<v Speaker 3>So it's better at things like calling tools, it's better

0:01:01.640 --> 0:01:05.320
<v Speaker 3>at things like structured thinking and reasoning, problem solving. And

0:01:05.360 --> 0:01:07.360
<v Speaker 3>what we see in the enterprise is when you make

0:01:07.400 --> 0:01:11.080
<v Speaker 3>these core capabilities better, the number of use cases enterprises

0:01:11.120 --> 0:01:12.240
<v Speaker 3>can adopt these models for.

0:01:12.280 --> 0:01:15.080
<v Speaker 4>Increases and so coding being significant.

0:01:15.520 --> 0:01:19.120
<v Speaker 3>It really is the language of computers and that was

0:01:19.120 --> 0:01:21.560
<v Speaker 3>a significant area of demand for us when we were

0:01:21.560 --> 0:01:23.559
<v Speaker 3>talking to customers about what they wanted in this model.

0:01:23.760 --> 0:01:26.479
<v Speaker 1>I mean, PhD level is what many are calling it,

0:01:26.560 --> 0:01:28.800
<v Speaker 1>well what Sam is calling it, and I'm sure yourself.

0:01:29.160 --> 0:01:32.160
<v Speaker 1>What's interesting is when you've got seven hundred million weekly

0:01:32.280 --> 0:01:34.920
<v Speaker 1>users of chatchipt, how much is that a funnel a

0:01:34.959 --> 0:01:36.920
<v Speaker 1>read across into enterprise? How much you could get the

0:01:36.959 --> 0:01:39.360
<v Speaker 1>inbound because ultimately the people in the old workforce are

0:01:39.400 --> 0:01:40.039
<v Speaker 1>already using it.

0:01:42.080 --> 0:01:44.520
<v Speaker 3>Well, really early on when we launched chat GPT, what

0:01:44.560 --> 0:01:47.680
<v Speaker 3>we found is I think, you know, a number of

0:01:47.720 --> 0:01:50.360
<v Speaker 3>months after we launched it, I think something like ninety

0:01:50.400 --> 0:01:53.880
<v Speaker 3>two percent of the Fortune five hundred we're actively using

0:01:53.920 --> 0:01:55.840
<v Speaker 3>chatchipt or people at ninety two percent of the Fortune

0:01:55.880 --> 0:01:57.920
<v Speaker 3>five hundred were actively using it. And so it was

0:01:58.040 --> 0:01:59.600
<v Speaker 3>very obvious for us we needed to go build a

0:01:59.600 --> 0:02:02.880
<v Speaker 3>work product because I think chatchapt as a product is

0:02:03.280 --> 0:02:06.200
<v Speaker 3>as useful in an enterprise environment, in a work environment

0:02:06.600 --> 0:02:10.359
<v Speaker 3>as it is in your personal life. It's an amazing

0:02:10.400 --> 0:02:14.240
<v Speaker 3>companion if you do anything from marketing to software engineering

0:02:14.919 --> 0:02:17.519
<v Speaker 3>to data analysis and research and I think there was

0:02:17.560 --> 0:02:20.120
<v Speaker 3>a lot of organic adoption when people discovered the tool,

0:02:20.400 --> 0:02:23.080
<v Speaker 3>realizing that it could make people much better at their

0:02:23.160 --> 0:02:25.600
<v Speaker 3>jobs and able to do more. And so we really

0:02:25.680 --> 0:02:27.800
<v Speaker 3>leaned in with that and we're trying to build the

0:02:27.800 --> 0:02:28.560
<v Speaker 3>best product we can.

0:02:29.320 --> 0:02:32.480
<v Speaker 1>It's like one tenth of the planet using chatchypt Brad.

0:02:32.480 --> 0:02:36.120
<v Speaker 1>But I'm interested in some analysis that Meno Ventures has done.

0:02:36.280 --> 0:02:38.919
<v Speaker 1>They've analyzed the LLLM market, particularly in the enterprise space,

0:02:39.040 --> 0:02:40.840
<v Speaker 1>and they've just tried to push back saying, look, you

0:02:41.000 --> 0:02:43.519
<v Speaker 1>lost the market share. Open Ai went from fifty percent

0:02:43.520 --> 0:02:45.720
<v Speaker 1>in the enterprise market share down to twenty five percent,

0:02:46.160 --> 0:02:49.000
<v Speaker 1>and funny enough, the company that they back, which is anthropic,

0:02:49.240 --> 0:02:51.919
<v Speaker 1>took the lead. What do you say to those of statistics,

0:02:52.000 --> 0:02:53.840
<v Speaker 1>Is it something you're seeing within your own numbers.

0:02:55.919 --> 0:02:58.120
<v Speaker 3>It's hard to measure these things. You know, you can

0:02:58.160 --> 0:03:01.520
<v Speaker 3>find you can find measurements that that's the opposite. But

0:03:01.560 --> 0:03:03.519
<v Speaker 3>what we really focus on is value for customers, Like

0:03:03.560 --> 0:03:05.720
<v Speaker 3>we've got to deliver the absolute best models and then

0:03:05.760 --> 0:03:10.280
<v Speaker 3>the absolute best products for developers, for startups, for enterprises

0:03:10.760 --> 0:03:13.520
<v Speaker 3>large and small, and that's that's been our focus. I

0:03:13.520 --> 0:03:15.440
<v Speaker 3>think you know, our API is a good example of

0:03:15.480 --> 0:03:18.240
<v Speaker 3>where we've really invested lately. We've got over four million

0:03:18.280 --> 0:03:21.640
<v Speaker 3>developers now actively using the API every day to build

0:03:21.680 --> 0:03:25.560
<v Speaker 3>new products. We support thousands and thousands of startups that

0:03:25.600 --> 0:03:28.400
<v Speaker 3>are building with us, that we work deeply with on

0:03:28.680 --> 0:03:31.040
<v Speaker 3>trying to improve our product so that they can ultimately

0:03:31.040 --> 0:03:33.639
<v Speaker 3>build a better product. And in the enterprise, I think,

0:03:33.680 --> 0:03:36.200
<v Speaker 3>you know, the demand that we see there is really unabated.

0:03:36.240 --> 0:03:39.400
<v Speaker 3>We grew chat GPT enterprise seats from three million seats

0:03:39.440 --> 0:03:42.040
<v Speaker 3>to five million seats in a matter of two months,

0:03:42.040 --> 0:03:45.840
<v Speaker 3>and so that growth is accelerating and we're just starting

0:03:45.880 --> 0:03:47.600
<v Speaker 3>to scratch the surface I think on the impact that

0:03:47.600 --> 0:03:50.480
<v Speaker 3>we can have both for developers and for enterprises. So

0:03:50.760 --> 0:03:52.720
<v Speaker 3>we see it as a long game and you know,

0:03:52.720 --> 0:03:54.600
<v Speaker 3>we're here to just do our best for customers.

0:03:55.480 --> 0:03:58.040
<v Speaker 1>For that, BOYD, you need infrastructure.

0:03:58.640 --> 0:04:00.880
<v Speaker 2>Tell us a little bit about the costs of training

0:04:00.880 --> 0:04:03.680
<v Speaker 2>this model and how you're looking to expand with stargate

0:04:03.720 --> 0:04:06.040
<v Speaker 2>the project, and just been talking about how SoftBank's been

0:04:06.040 --> 0:04:08.320
<v Speaker 2>teaming up with fox Con, for example, to take over

0:04:08.560 --> 0:04:12.920
<v Speaker 2>an ohio ev plant. How is that continuing to meet

0:04:12.960 --> 0:04:15.400
<v Speaker 2>your demands or not meet them as the case may be.

0:04:17.040 --> 0:04:21.440
<v Speaker 3>Well, yeah, we've seen demand for AI increase at just

0:04:22.160 --> 0:04:25.320
<v Speaker 3>a torrid pace. Obviously, at the root of that is

0:04:25.440 --> 0:04:29.200
<v Speaker 3>getting right the infrastructure equation, and we think ultimately that's

0:04:29.200 --> 0:04:32.000
<v Speaker 3>going to be a critical input to the US and

0:04:32.040 --> 0:04:33.200
<v Speaker 3>its allies.

0:04:33.040 --> 0:04:34.279
<v Speaker 4>Being competitive in this area.

0:04:34.320 --> 0:04:36.720
<v Speaker 3>And so Stargate for US was a five hundred billion

0:04:36.760 --> 0:04:40.400
<v Speaker 3>dollar project to invest here in the United States to

0:04:40.480 --> 0:04:44.960
<v Speaker 3>build AI infrastructure for open AI and ultimately for the country.

0:04:45.560 --> 0:04:47.400
<v Speaker 3>We're working with a lot of great partners on that

0:04:47.440 --> 0:04:50.320
<v Speaker 3>project and hope to bring in more and I think,

0:04:50.440 --> 0:04:52.640
<v Speaker 3>you know, that's just the beginning. We're going to continue

0:04:52.640 --> 0:04:55.520
<v Speaker 3>it to invest aggressively. We've always found ourselves somewhat on

0:04:55.560 --> 0:04:59.000
<v Speaker 3>the wrong side of the demand curve for AI, you know,

0:04:59.080 --> 0:05:01.960
<v Speaker 3>despite the investment, significant investment to date, and so for

0:05:02.000 --> 0:05:03.280
<v Speaker 3>as long as we see demand, we're going.

0:05:03.200 --> 0:05:04.440
<v Speaker 4>To continue to invest aggressively.

0:05:05.000 --> 0:05:07.599
<v Speaker 3>AI is interesting in that the more you invest and

0:05:07.640 --> 0:05:09.200
<v Speaker 3>the more you make it available, the more you make

0:05:09.240 --> 0:05:14.240
<v Speaker 3>it you know, cost cost approachable for enterprises and for consumers,

0:05:14.440 --> 0:05:16.240
<v Speaker 3>the more people want to use it. So it's an

0:05:16.279 --> 0:05:18.920
<v Speaker 3>amazing trend and we'll continue to invest behind it.

0:05:19.160 --> 0:05:22.480
<v Speaker 1>We're speaking with Brad Lightcap of open Ai, the CEO

0:05:22.600 --> 0:05:25.280
<v Speaker 1>for our radio and TV audience is Brad. How has

0:05:25.320 --> 0:05:28.240
<v Speaker 1>the rollout ultimately been Do you think because so many

0:05:28.240 --> 0:05:30.560
<v Speaker 1>people wanted to use the app that maybe we hit

0:05:30.640 --> 0:05:32.320
<v Speaker 1>limits quicker than some anticipated.

0:05:33.880 --> 0:05:36.479
<v Speaker 3>Well, we're trying our best to keep up with demand.

0:05:37.240 --> 0:05:40.520
<v Speaker 3>Serving infrastructure at scale at seven hundred million users and

0:05:40.520 --> 0:05:43.839
<v Speaker 3>then millions of developers and many many billions of tokens

0:05:43.880 --> 0:05:47.000
<v Speaker 3>per minute that we process is not for the faint

0:05:47.000 --> 0:05:48.920
<v Speaker 3>of heart. Thankfully, I don't have to do that part

0:05:48.920 --> 0:05:51.080
<v Speaker 3>of it. But we're doing our best to make sure

0:05:51.080 --> 0:05:54.320
<v Speaker 3>the rollout is successful and we're hoping that by the

0:05:54.360 --> 0:05:56.200
<v Speaker 3>end of the week here everyone gets access.

0:05:56.400 --> 0:05:59.120
<v Speaker 1>Your job description is more about building the enterprise relationships

0:05:59.120 --> 0:06:01.159
<v Speaker 1>and partnerships. I also think about the partnership you have

0:06:01.200 --> 0:06:03.560
<v Speaker 1>in Microsoft and Sati and Adela is out there really

0:06:03.600 --> 0:06:06.279
<v Speaker 1>talking about integrating already and how excited he was for

0:06:06.320 --> 0:06:09.120
<v Speaker 1>the product, but there's a tension in the relationship there.

0:06:09.560 --> 0:06:12.680
<v Speaker 1>I'm interested in what you think the progress being made.

0:06:12.680 --> 0:06:16.040
<v Speaker 1>Sam Alman was on Networks talking about progress being made

0:06:16.080 --> 0:06:18.560
<v Speaker 1>with a relationship going forward. Microsoft, can you give us

0:06:18.560 --> 0:06:21.080
<v Speaker 1>a timeline about when you think a deal will be

0:06:21.120 --> 0:06:24.480
<v Speaker 1>done in the future of how they interact with your product?

0:06:24.600 --> 0:06:25.520
<v Speaker 2>And were broadly how.

0:06:25.520 --> 0:06:27.760
<v Speaker 1>Much ownership they continue to have in the business as

0:06:27.800 --> 0:06:28.760
<v Speaker 1>a for profit one.

0:06:30.320 --> 0:06:32.800
<v Speaker 3>Yeah, Well, we feel really positive about the relationship with

0:06:32.839 --> 0:06:35.359
<v Speaker 3>Microsoft and they've they've been a great partner throughout the history.

0:06:35.200 --> 0:06:35.680
<v Speaker 4>Of open Ai.

0:06:35.800 --> 0:06:38.640
<v Speaker 3>They've been with us from the beginning really since before

0:06:38.720 --> 0:06:43.080
<v Speaker 3>chat Ept obviously have been a huge infrastructure partner for

0:06:43.160 --> 0:06:44.880
<v Speaker 3>us with Azure, and.

0:06:44.880 --> 0:06:47.040
<v Speaker 4>So we continue we expect that to continue.

0:06:47.040 --> 0:06:50.000
<v Speaker 3>We see we see no future that you know of

0:06:50.040 --> 0:06:52.520
<v Speaker 3>open ai that doesn't include Microsoft in a significant way.

0:06:53.440 --> 0:06:55.599
<v Speaker 3>We've got to work on what that future looks like together.

0:06:55.640 --> 0:06:57.600
<v Speaker 3>We're in that process with them right now. We feel

0:06:57.640 --> 0:07:00.840
<v Speaker 3>very good about it. But we think also ultimately there's

0:07:00.960 --> 0:07:03.560
<v Speaker 3>you know, the world is really big, and the demand

0:07:03.600 --> 0:07:06.240
<v Speaker 3>for these systems and these models in the enterprise and

0:07:06.360 --> 0:07:09.359
<v Speaker 3>consumer is significant, and so they represent not only an

0:07:09.400 --> 0:07:12.080
<v Speaker 3>infrastructure partner for us, but a great partner to be

0:07:12.080 --> 0:07:13.880
<v Speaker 3>able to help distribute and bring the benefits of the

0:07:13.880 --> 0:07:16.440
<v Speaker 3>technology to the world given the size of their footprint.

0:07:16.480 --> 0:07:18.880
<v Speaker 4>So, you know, more to work through there, But we

0:07:18.920 --> 0:07:19.680
<v Speaker 4>feel good about it.

0:07:19.720 --> 0:07:21.760
<v Speaker 3>And like I said, I think you know, when we're

0:07:21.760 --> 0:07:23.320
<v Speaker 3>standing at the finish line of all this, they'll be

0:07:23.320 --> 0:07:23.800
<v Speaker 3>there with us.

0:07:24.560 --> 0:07:27.640
<v Speaker 1>Meanwhile, the stuff of Suliman over at Microsoft is busy

0:07:27.640 --> 0:07:30.400
<v Speaker 1>in the tussle for talent, so to plenty of other

0:07:30.520 --> 0:07:34.040
<v Speaker 1>rivals that we understand, Mark Zuckerberg busy. And what's interesting

0:07:34.360 --> 0:07:37.920
<v Speaker 1>is while you, as one of those long tenured employees

0:07:38.080 --> 0:07:41.160
<v Speaker 1>and staff over at open Ai, remain committed because of

0:07:41.240 --> 0:07:44.120
<v Speaker 1>innovations such as this, But what about the liquidity that

0:07:44.120 --> 0:07:47.679
<v Speaker 1>we're talking about bringing to some of your well people

0:07:47.720 --> 0:07:50.160
<v Speaker 1>that you work alongside. How is that going? We understand

0:07:50.200 --> 0:07:52.200
<v Speaker 1>they might be even a five hundred billion dollar valuation

0:07:52.360 --> 0:07:55.280
<v Speaker 1>involved in what is a secondary sale of your shares

0:07:55.520 --> 0:07:56.920
<v Speaker 1>to the likes of Thrive Capital.

0:07:58.840 --> 0:08:02.480
<v Speaker 3>Yeah, well, nothing there to share here, But we continue

0:08:02.520 --> 0:08:06.080
<v Speaker 3>to see very healthy demand for on the investors side

0:08:06.120 --> 0:08:08.120
<v Speaker 3>for wanting to be I think part of the Opening

0:08:08.120 --> 0:08:09.960
<v Speaker 3>Eye journey and mission and we're very grateful for that.

0:08:10.600 --> 0:08:12.160
<v Speaker 3>And on the talent side, look, I think you know,

0:08:12.240 --> 0:08:14.440
<v Speaker 3>Opening Eye was founded as a nonprofit. It was founded

0:08:14.440 --> 0:08:16.640
<v Speaker 3>as a mission driven company to be able to build

0:08:17.320 --> 0:08:20.400
<v Speaker 3>you know, general intelligence that's beneficial for all of humanity,

0:08:20.920 --> 0:08:23.200
<v Speaker 3>and we haven't strayed from that mission. I think ultimately

0:08:23.240 --> 0:08:26.080
<v Speaker 3>that's what attracts talent is people want to work for

0:08:26.120 --> 0:08:28.880
<v Speaker 3>a project that's bigger than themselves and something that's going

0:08:28.920 --> 0:08:31.320
<v Speaker 3>to be impactful for us and for you know, for

0:08:31.520 --> 0:08:34.480
<v Speaker 3>for for humanity and our species. And so I think

0:08:34.480 --> 0:08:37.960
<v Speaker 3>that's the thing that ultimately attracts people to where they work. Obviously,

0:08:38.000 --> 0:08:40.480
<v Speaker 3>like it's a competitive market and we continue to compete.

0:08:40.679 --> 0:08:42.040
<v Speaker 3>But at the end of the day, when we ask

0:08:42.120 --> 0:08:43.760
<v Speaker 3>people what it is that keeps them at opening, I

0:08:43.840 --> 0:08:44.400
<v Speaker 3>it's the mission.

0:08:44.800 --> 0:08:45.320
<v Speaker 4>The mission.

0:08:45.720 --> 0:08:49.440
<v Speaker 1>At the moment, you've got something that's generally intelligence, but

0:08:49.440 --> 0:08:51.400
<v Speaker 1>it's not artificial general intelligence.

0:08:51.440 --> 0:08:52.679
<v Speaker 4>Brad, when do you get there?

0:08:52.720 --> 0:08:57.520
<v Speaker 3>Briefly, you know, I've sworn off making predictions in AI.

0:08:57.600 --> 0:09:00.680
<v Speaker 3>It's too hard, the curves are too steep. But I think,

0:09:00.720 --> 0:09:02.680
<v Speaker 3>you know, we feel really good about the rate of progress.

0:09:02.760 --> 0:09:05.600
<v Speaker 3>GPT five is a great representation of how we start

0:09:05.640 --> 0:09:08.480
<v Speaker 3>to make progress on little things, you know, things like

0:09:08.520 --> 0:09:11.080
<v Speaker 3>for example, being able to have the model dynamically reason

0:09:11.440 --> 0:09:14.520
<v Speaker 3>and decide how much it wants to think about the

0:09:14.559 --> 0:09:17.200
<v Speaker 3>problem that you ask it to solve. That's something that

0:09:17.240 --> 0:09:20.560
<v Speaker 3>we do natively as humans that previously our models couldn't do.

0:09:20.679 --> 0:09:22.600
<v Speaker 4>So it's these little steps forward that we.

0:09:22.520 --> 0:09:25.080
<v Speaker 3>Think accumulated and ultimately get us to something that will

0:09:25.080 --> 0:09:26.199
<v Speaker 3>be truly remarkable.

0:09:26.800 --> 0:09:29.840
<v Speaker 1>I'd like cap talking about the latest GPT five. We

0:09:29.880 --> 0:09:32.360
<v Speaker 1>thank you so much, CEO of open Ai,