WEBVTT - OpenAI COO Brad Lightcap Talks What's Next For OpenAI

0:00:02.480 --> 0:00:08.720
<v Speaker 1>Bloomberg Audio Studios, podcasts, radio news. Well, we started today

0:00:08.760 --> 0:00:11.640
<v Speaker 1>with a big pop and shares of AMD, this after

0:00:11.680 --> 0:00:14.160
<v Speaker 1>it inked the landmark deal with open Ai to build

0:00:14.160 --> 0:00:17.400
<v Speaker 1>out infrastructure, giving the chip maker a chance to challenge

0:00:17.520 --> 0:00:20.680
<v Speaker 1>in video in the computing industry. Now, this comes on

0:00:20.720 --> 0:00:23.360
<v Speaker 1>the same day as open AI's annual Developers Day of

0:00:23.440 --> 0:00:27.520
<v Speaker 1>Developers Day, where we've seen a lot of stocks including Thigma, HubSpot, Cisco,

0:00:27.560 --> 0:00:30.960
<v Speaker 1>Mattel just to name a few, also moving significantly on

0:00:31.080 --> 0:00:33.839
<v Speaker 1>based on some of the comments coming out of that conference.

0:00:34.159 --> 0:00:36.880
<v Speaker 1>Joining us right now is Ed Ludlow. He is at

0:00:36.880 --> 0:00:39.400
<v Speaker 1>that conference and he's joined right now by the CEO

0:00:39.600 --> 0:00:41.320
<v Speaker 1>of open Ai, Brad Lightcap.

0:00:41.479 --> 0:00:45.920
<v Speaker 2>Ed. Yeah, and the headline is the ability to use

0:00:45.960 --> 0:00:49.440
<v Speaker 2>third party apps within chat GPT, and the profound impact

0:00:49.479 --> 0:00:52.720
<v Speaker 2>on the market was simply those name checked saw their

0:00:52.760 --> 0:00:55.959
<v Speaker 2>stocks move in a significant way. Delighted to speak once

0:00:56.000 --> 0:00:58.960
<v Speaker 2>again with Brad Lightcap that it's just simple as that,

0:00:59.040 --> 0:01:02.080
<v Speaker 2>as that really the ability to be within chat GPT

0:01:02.720 --> 0:01:07.000
<v Speaker 2>and access Spotify or Figma as two examples. When you

0:01:07.040 --> 0:01:10.319
<v Speaker 2>were discussing the idea of that and an API access

0:01:10.400 --> 0:01:13.960
<v Speaker 2>with those partners, was it a tough negotiation to get

0:01:13.959 --> 0:01:15.320
<v Speaker 2>them on board with the idea.

0:01:16.280 --> 0:01:18.959
<v Speaker 3>No, I mean that we've seen enthusiasm for this from

0:01:19.000 --> 0:01:21.720
<v Speaker 3>the beginning. You remember, way back when we were lost

0:01:21.720 --> 0:01:23.280
<v Speaker 3>something called plugins.

0:01:23.400 --> 0:01:24.800
<v Speaker 4>Back in the early days of chat ept.

0:01:24.959 --> 0:01:27.280
<v Speaker 3>This was one of our first attempts to start to

0:01:27.319 --> 0:01:30.119
<v Speaker 3>build an ecosystem around chat GPT so that chat ept

0:01:30.240 --> 0:01:32.479
<v Speaker 3>can start to engage with and interact with the applications

0:01:32.520 --> 0:01:34.160
<v Speaker 3>that are important to you in your personal life and

0:01:34.200 --> 0:01:36.759
<v Speaker 3>at work. And now we really have a.

0:01:36.720 --> 0:01:39.800
<v Speaker 4>Much richer surface through MCP.

0:01:39.520 --> 0:01:42.039
<v Speaker 3>And other protocols to be able to bring applications into

0:01:42.120 --> 0:01:44.840
<v Speaker 3>chat ept and for really to allow you to engage

0:01:44.880 --> 0:01:47.800
<v Speaker 3>with chat GPT in the kind of work around work right,

0:01:48.160 --> 0:01:51.160
<v Speaker 3>it's the contextual aspect of I'm doing X, or I

0:01:51.160 --> 0:01:53.200
<v Speaker 3>need why I'm on a road trip and I want

0:01:53.240 --> 0:01:55.000
<v Speaker 3>to know what playlist would go well with this in

0:01:55.040 --> 0:01:57.480
<v Speaker 3>the context of my broader trip planning. That allows you

0:01:57.520 --> 0:01:59.720
<v Speaker 3>now to kind of use chat GPT to solve that

0:01:59.800 --> 0:02:03.240
<v Speaker 3>high level task and then also integrate apps contextually to

0:02:03.280 --> 0:02:04.480
<v Speaker 3>solve those specific problems.

0:02:04.520 --> 0:02:06.559
<v Speaker 2>And we're in a place where chat GPT has become

0:02:06.640 --> 0:02:10.000
<v Speaker 2>more of an operating system. Whether that was your ambition

0:02:10.240 --> 0:02:14.240
<v Speaker 2>or not, is that where you want to take it

0:02:14.280 --> 0:02:18.760
<v Speaker 2>to be an OS and a developer driven platform. It's

0:02:18.760 --> 0:02:21.440
<v Speaker 2>almost like an app store, you know, on paper based

0:02:21.440 --> 0:02:22.880
<v Speaker 2>on what you announced this afternoon.

0:02:23.639 --> 0:02:26.480
<v Speaker 3>Well, we've always thought of chatgybt as like a super assistant.

0:02:27.040 --> 0:02:28.560
<v Speaker 4>We never set out to build a chatbot.

0:02:28.600 --> 0:02:31.400
<v Speaker 3>We always wanted to build something that was really true

0:02:31.400 --> 0:02:33.679
<v Speaker 3>to you and what your preferences.

0:02:33.160 --> 0:02:35.440
<v Speaker 4>Are, what your goals are. They could actually help you

0:02:35.480 --> 0:02:36.080
<v Speaker 4>achieve more.

0:02:36.480 --> 0:02:39.040
<v Speaker 3>And so I think part of that is CHATGIBD having

0:02:39.200 --> 0:02:42.560
<v Speaker 3>an appreciation and understanding of the applications in your life

0:02:42.560 --> 0:02:45.800
<v Speaker 3>that are important, and I think enabling that kind of

0:02:45.800 --> 0:02:49.640
<v Speaker 3>connectivity and interoperability makes Chattibt Richard also enables a lot

0:02:49.680 --> 0:02:51.560
<v Speaker 3>of pass through for people to be able to engage

0:02:51.600 --> 0:02:53.560
<v Speaker 3>with apps they love it as well as new apps.

0:02:53.600 --> 0:02:55.320
<v Speaker 5>The pass through bit is interesting.

0:02:55.639 --> 0:02:58.480
<v Speaker 2>Were their concerns with some of those technology companies you're

0:02:58.480 --> 0:02:59.200
<v Speaker 2>partnering with that.

0:02:59.200 --> 0:03:00.960
<v Speaker 5>It would take traff they away from there?

0:03:01.120 --> 0:03:04.760
<v Speaker 3>In domains now, I think mostly people are really focused

0:03:04.760 --> 0:03:06.240
<v Speaker 3>on building into new interfaces.

0:03:06.320 --> 0:03:06.480
<v Speaker 2>Right.

0:03:06.520 --> 0:03:09.080
<v Speaker 3>This is just like mobile in some sense, where you

0:03:09.120 --> 0:03:11.240
<v Speaker 3>have a new interface, you have a new form factor.

0:03:11.639 --> 0:03:14.320
<v Speaker 3>People are going to want to use mobile form factors

0:03:14.320 --> 0:03:16.080
<v Speaker 3>on the go and apps like.

0:03:16.080 --> 0:03:17.880
<v Speaker 4>Spotify in some ways exist.

0:03:17.800 --> 0:03:20.280
<v Speaker 3>Almost because they really nail mobile and so we think,

0:03:20.320 --> 0:03:23.280
<v Speaker 3>actually there's an opportunity for builders to create entirely new

0:03:23.320 --> 0:03:26.280
<v Speaker 3>applications that are even native to chat GPT, and of

0:03:26.320 --> 0:03:29.360
<v Speaker 3>course for services you love to be able to benefit

0:03:29.360 --> 0:03:29.680
<v Speaker 3>there too.

0:03:30.080 --> 0:03:33.160
<v Speaker 2>Is there a revenue sharing agreement with those third parties

0:03:33.240 --> 0:03:35.200
<v Speaker 2>whose apps are accessible to we chat GPT.

0:03:35.760 --> 0:03:37.640
<v Speaker 3>So we're going to figure out the economics of this

0:03:37.680 --> 0:03:41.000
<v Speaker 3>over time. You know, we're brand new here. Plugins was

0:03:41.080 --> 0:03:43.240
<v Speaker 3>the first version of this and that was even an experiment,

0:03:43.240 --> 0:03:45.240
<v Speaker 3>and so like everything at Opening Eye, we take this

0:03:45.320 --> 0:03:47.840
<v Speaker 3>very experimental mindset to making.

0:03:47.680 --> 0:03:48.440
<v Speaker 4>Sure we get it right.

0:03:48.520 --> 0:03:50.480
<v Speaker 3>But the idea is we do we do want to

0:03:50.480 --> 0:03:53.120
<v Speaker 3>build something that's useful for developers, and of course there's

0:03:53.160 --> 0:03:54.800
<v Speaker 3>going to be you know, have to be some exchange

0:03:54.840 --> 0:03:58.080
<v Speaker 3>in there of economics and value, and we'll have to

0:03:58.080 --> 0:03:59.040
<v Speaker 3>figure out to get that right.

0:03:59.360 --> 0:04:02.720
<v Speaker 2>You have hit eight hundred million active weekly users. You

0:04:02.760 --> 0:04:05.200
<v Speaker 2>announced on stage. Actually Greg Brockman told me that at

0:04:05.240 --> 0:04:08.400
<v Speaker 2>eight o'clock this morning, and maybe we missed it, but

0:04:08.560 --> 0:04:11.920
<v Speaker 2>it's a significant milestone. All the time, I'm asked by

0:04:11.920 --> 0:04:14.480
<v Speaker 2>all kinds of people, do we have any sense of

0:04:14.520 --> 0:04:17.640
<v Speaker 2>within that eight hundred million, how many are base level

0:04:17.760 --> 0:04:21.360
<v Speaker 2>free users and how many are premium level paid subscribers.

0:04:22.120 --> 0:04:25.080
<v Speaker 3>Yeah, we have a very healthy, you know, funnel of

0:04:25.120 --> 0:04:27.200
<v Speaker 3>people that choose to pay for chatchubut.

0:04:28.080 --> 0:04:30.200
<v Speaker 4>You know, it's surpassed for my expectations.

0:04:30.200 --> 0:04:33.599
<v Speaker 3>Frankly, were people have this kind of conception that consumers

0:04:33.920 --> 0:04:36.839
<v Speaker 3>tend to not pay for software and you know, similar

0:04:36.920 --> 0:04:39.000
<v Speaker 3>even to what I was saying before around how do

0:04:39.040 --> 0:04:42.560
<v Speaker 3>you co develop the product alongside the business model. Chatchbt

0:04:42.680 --> 0:04:45.120
<v Speaker 3>is a great example of that, where the subscription model

0:04:45.120 --> 0:04:47.320
<v Speaker 3>I think has been really a testament to how valuable

0:04:47.360 --> 0:04:50.640
<v Speaker 3>it is for more users than I think we expected

0:04:50.680 --> 0:04:52.680
<v Speaker 3>to be willing to pay for it. So we don't,

0:04:52.720 --> 0:04:54.960
<v Speaker 3>I think, just close the exact number, but it's a

0:04:55.320 --> 0:04:56.880
<v Speaker 3>healthy amount and more every day.

0:04:57.440 --> 0:05:01.919
<v Speaker 2>Open AI in the beginning went off the consumer for

0:05:02.080 --> 0:05:05.839
<v Speaker 2>uses aggressively. You are now very focused on the enterprise business.

0:05:06.360 --> 0:05:08.560
<v Speaker 2>What is the strategy for that and how do you

0:05:08.600 --> 0:05:10.320
<v Speaker 2>prioritize your enterprise business?

0:05:10.520 --> 0:05:12.360
<v Speaker 4>Yeah, I'm glad you asked about that.

0:05:12.560 --> 0:05:15.880
<v Speaker 3>So really today's announcements actually I think target what are

0:05:16.120 --> 0:05:18.760
<v Speaker 3>an important set of use cases for the enterprise things

0:05:19.000 --> 0:05:22.159
<v Speaker 3>we've been hearing enterprises ask us about now for some time,

0:05:22.320 --> 0:05:26.159
<v Speaker 3>so specifically, one is we now have an ability for

0:05:26.279 --> 0:05:29.360
<v Speaker 3>enterprises to build agents in a much more visual, much

0:05:29.360 --> 0:05:32.240
<v Speaker 3>more intuitive way. You've heard us say twenty twenty five

0:05:32.279 --> 0:05:34.360
<v Speaker 3>has been the year of agents. We think that's true.

0:05:34.720 --> 0:05:36.960
<v Speaker 3>Codex has been a great example for us of that.

0:05:37.320 --> 0:05:40.360
<v Speaker 3>Our coding agent now available through an API.

0:05:40.520 --> 0:05:42.719
<v Speaker 5>Also, the lead time to make software is a lot shorter.

0:05:42.960 --> 0:05:43.919
<v Speaker 4>It's gotten a lot shorter.

0:05:44.040 --> 0:05:46.280
<v Speaker 3>I think you saw today we demoed live demo, I

0:05:46.279 --> 0:05:48.360
<v Speaker 3>think three or four different things that we've built in

0:05:48.400 --> 0:05:48.880
<v Speaker 3>real time.

0:05:49.200 --> 0:05:51.240
<v Speaker 4>We expect that to continue to be to the trend.

0:05:51.720 --> 0:05:54.560
<v Speaker 3>Things like Agent Builder allow enterprises to be able to

0:05:54.560 --> 0:05:59.479
<v Speaker 3>build agentic experiences, powerful identic experiences on the go, iteratively

0:06:00.040 --> 0:06:03.120
<v Speaker 3>and connected into the tools and sources of information that

0:06:03.160 --> 0:06:03.839
<v Speaker 3>matter for the business.

0:06:03.920 --> 0:06:06.040
<v Speaker 2>The data point that jumps out me is your API

0:06:06.160 --> 0:06:09.880
<v Speaker 2>is handling more than six billion tokens per minute, and

0:06:09.920 --> 0:06:13.800
<v Speaker 2>that helps explain why the AMD deal, you know, which

0:06:13.800 --> 0:06:16.120
<v Speaker 2>is focused on inference. You are involved in all of

0:06:16.160 --> 0:06:19.120
<v Speaker 2>these domains of the company. I've already asked Greg, but

0:06:19.160 --> 0:06:21.359
<v Speaker 2>I've got to ask you, how are you going to

0:06:21.480 --> 0:06:24.880
<v Speaker 2>finance yet another infrastructure project? Like is there going to

0:06:24.920 --> 0:06:28.880
<v Speaker 2>be some debt here specific for the AMD capacity, and

0:06:28.920 --> 0:06:31.480
<v Speaker 2>how do you move quickly to get it online?

0:06:31.800 --> 0:06:34.480
<v Speaker 3>Yeah, well, the high level thing is we are tremendously

0:06:34.480 --> 0:06:37.359
<v Speaker 3>compute constrained. It feels like we're in this kind of

0:06:37.360 --> 0:06:40.039
<v Speaker 3>recurring theme of being compute constrained. And I think the

0:06:40.080 --> 0:06:42.120
<v Speaker 3>reason for that is the answer to the question you ask,

0:06:42.200 --> 0:06:45.000
<v Speaker 3>which is demand. Right, we see there are multiples of

0:06:45.000 --> 0:06:49.400
<v Speaker 3>demand that are lateent and untapped from what we have today.

0:06:49.400 --> 0:06:52.120
<v Speaker 3>And even today, obviously by any standard, demand in revenue

0:06:52.120 --> 0:06:56.679
<v Speaker 3>growth has been torrid in its pace, and so really

0:06:56.720 --> 0:06:57.920
<v Speaker 3>we have to invest ahead of that.

0:06:57.960 --> 0:06:59.080
<v Speaker 4>And I think that's going to be the.

0:06:59.040 --> 0:07:01.560
<v Speaker 3>Ray limitter for us to be able to go capture demand,

0:07:01.560 --> 0:07:03.880
<v Speaker 3>whether it's consumer or enterprise, and for us to be

0:07:03.920 --> 0:07:07.960
<v Speaker 3>able to build new models, paralyze more experiences, more product experiences,

0:07:08.120 --> 0:07:11.080
<v Speaker 3>and then enable users specifically to be able to use those.

0:07:10.920 --> 0:07:14.360
<v Speaker 4>Products more actively in their daily life at work and

0:07:14.400 --> 0:07:14.720
<v Speaker 4>at home.

0:07:14.800 --> 0:07:17.600
<v Speaker 3>And so, you know, even things like Sora, the app

0:07:17.640 --> 0:07:20.760
<v Speaker 3>we just launched, we wish we could invite more people

0:07:20.760 --> 0:07:22.840
<v Speaker 3>onto it now, but we just need more compute. So

0:07:22.880 --> 0:07:27.400
<v Speaker 3>the AMD deal we're excited about being you know, directionally

0:07:27.920 --> 0:07:28.640
<v Speaker 3>a way for us.

0:07:28.520 --> 0:07:28.960
<v Speaker 5>To do that.

0:07:29.800 --> 0:07:32.360
<v Speaker 2>I've got to ask about the report that open AI

0:07:32.520 --> 0:07:36.119
<v Speaker 2>closed secondary or the ability for employees to sell shares

0:07:36.120 --> 0:07:39.200
<v Speaker 2>at a five hundred billion dollar valuation. I already asked

0:07:39.200 --> 0:07:41.680
<v Speaker 2>you this question, but what is the metric we're supposed

0:07:41.680 --> 0:07:44.960
<v Speaker 2>to judge your success by the five hundred billion dollar valuation?

0:07:45.360 --> 0:07:48.440
<v Speaker 2>The six billion tokens per minute? To you, Frad, what

0:07:48.600 --> 0:07:48.800
<v Speaker 2>is it?

0:07:49.520 --> 0:07:52.720
<v Speaker 3>For me? It's it's actually kind of a metric that

0:07:52.880 --> 0:07:57.400
<v Speaker 3>we we talked about is tokens. It's you mentioned six

0:07:57.440 --> 0:08:01.880
<v Speaker 3>billion tokens per minute on our That is the purest

0:08:02.320 --> 0:08:02.600
<v Speaker 3>for me.

0:08:02.720 --> 0:08:05.880
<v Speaker 4>The kind of essence of utility is that consumption metric.

0:08:05.960 --> 0:08:09.760
<v Speaker 3>And so we've actively tracked that metric to see how

0:08:09.760 --> 0:08:12.800
<v Speaker 3>people's consumption of AI is growing over time. And you

0:08:12.880 --> 0:08:16.080
<v Speaker 3>see this happen in amazing ways. So things like Codex,

0:08:16.080 --> 0:08:19.400
<v Speaker 3>for example, we've seen grow ten x since August purely

0:08:19.440 --> 0:08:22.880
<v Speaker 3>on consumption of tokens around coding, and you start to

0:08:22.920 --> 0:08:25.880
<v Speaker 3>see that same pattern emerge across multiple lanes of use

0:08:26.040 --> 0:08:28.400
<v Speaker 3>and across multiple areas of work. And that's the metric

0:08:28.440 --> 0:08:30.280
<v Speaker 3>I look at because if that number is going up,

0:08:30.320 --> 0:08:32.080
<v Speaker 3>it means people are using us for more things, and

0:08:32.120 --> 0:08:32.920
<v Speaker 3>that's the ultimate Goal.

0:08:33.520 --> 0:08:35.920
<v Speaker 5>Brad Lightcap is open ai Coo.

0:08:36.280 --> 0:08:38.240
<v Speaker 2>There has been a fire hose of headlines and it

0:08:38.360 --> 0:08:39.839
<v Speaker 2>is moved markets all day long.