Coatue Pt2. Open AI’s Kevin Weil Dives into All Things ChatGPT | BG2 w/ Bill Gurley & Brad Gerstner

You said today is the worst product you'll use. The model that you use today is the worst AI model that you'll ever use for the rest of your life.

1. Kevin Weil Open AI

Kevin, on this slide, in 5 years, OpenAI is the 9th most valuable company in the world. It's $300 billion in enterprise value today, as measured by the markets, and they have it going to $1.6 trillion.

I remember when you joined OpenAI and I asked you the question, “Can OpenAI get to $1 trillion in total value?” You said, “If I thought it was just $1 trillion, I wouldn't be nearly as excited as I am about going to OpenAI.”

When you think about that—and, by the way, less about market cap and more about the ambition, the opportunity, and the impact—what I want you to do for us today is blue-sky it. How long have you been at OpenAI now?

About a year.

Okay, so you're a year into this. A year in, do you think the opportunity is bigger than you thought when you started? If so, why? Walk us through that a little bit.

I think it's bigger, for sure, than I realized. I think it's been in Sam's head the whole time, so I'm not sure that there's been a revelation over the past year, but it's certainly opened my eyes.

2. User Growth

For one, we have grown faster than anything I've ever seen. I thought I had seen fast growth when I was at Instagram, for example. For those following at home, on slide 24 in the deck, they show the user growth, and we talked about it earlier, but it's just phenomenal. As we've grown, we're also deepening usage with the platform. People are using it more.

3. User Minutes

Right. Right. Right. They had a slide on that also. You said some comments on stage about how, as people use it more, the minutes go up.

Yeah. We look at weekly active users. That's sort of the metric that we gauge ourselves by from a growth perspective. Obviously, there are a lot of other things that we look at as well, but from a weekly active-user perspective, the reason we do that versus monthly is that we don't want someone who just comes once a month. Once a month means we're probably not adding all that much value to their life if they're coming once a month.

Over time, I think it's going to be more natural for us to measure daily active users because, if we're doing our job, then you're using ChatGPT on a daily basis—and hopefully multiple times per day—because it can help you multiple times per day with all the different things that you encounter in your life.

There's sort of the question: Are you a monthly user, a weekly user, or a daily user? Then, on a given day, how much time are you spending in the product?

4. Time Spent

You guys had a chart—your data, not ours—that showed that people were using ChatGPT a whole lot more time per day.

Yeah. And that's awesome. That's super exciting. It just means that we're actually helping people solve problems and they're not just using it as a novelty.

You go back a couple of years, when it launched, and it was really good at various sorts of writing. If you were going to do copywriting, or if you were going to have it summarize emails or things like that, you look at it today and it's so much broader. It can help in so many other areas.

People are uploading things. Actually, my own son had a health thing. He had to have minor surgery, which was not that big of a deal—at least it was not supposed to be that big of a deal. There was a tiny chance that there was something more serious.

He had the surgery. He's 8. The next day, he's playing a soccer game; he doesn't know any better. But we're still waiting for the biopsy of this thing to come back. We ended up getting a paper in the mail—a letter in the mail—with a whole bunch of complex medical stuff on it.

I'm halfway educated, but I couldn't make sense of it. I didn't know what it meant. I called the doctor and couldn't get hold of her. She was in surgery, I think, and she's busy; she's got a million things going on.

I'm looking at this, and there's a bunch of scary-looking medical terms. The first thing I do, of course, is take a picture of it, upload it to ChatGPT, and say, “Hey, should I be worried about this?”

It comes back immediately: “No, you're fine. This is completely benign. Don't worry about it. You're all good.” But I actually wasn't able to get hold of the doctor for 72 hours. That would have been a really bad 72 hours for me as a parent, sitting there stressing about this note.

5. How ChatGPT Expands

What you're describing, as someone who's gone from once a day to 20 times a day, is that you learn what it can do. You learn the kinds of things it can do. You can take a photo of an appliance and ask it how to set the clock or something.

One of the reasons I think it expands is that people come to understand all the things it can do and all the kinds of prompts. Another one is that I used to do a Google search, go to a web page, find a number, put that into a formula, and multiply it on a calculator. It can just do all that, right?

You can say, “Go get this and run this.” Now you can connect it to your email, your docs, and your calendar, and it can start to be useful in so many other ways. What you used to have to search through 20 docs for, now you can just ask for answers.

So it's so much more useful, and I totally agree. The way to understand it is that it's also constantly evolving, right? The models get step-changes better every 2 or 3 months. You just start using it.

The number-one advice I have for people when they ask what they should do with ChatGPT is just to start using it. When you start using it for things and take little, tiny risks—try it for new things that you're not sure it can do—you realize that, in most cases, it can.

Then it just becomes more and more helpful to you over time, and I think the numbers reflect that.

When you see—so, you're running product, and your goal is to drive engagement, drive utility, and thrill your customers—talk to us about the things you're actively doing to take somebody from monthly to weekly, and from weekly to daily.

I'm sure a lot of this is happening organically, as Bill describes, but I'm sure you're pressing on the flywheel. If we're at 1 billion monthlies today, how do you set a measure for success for you and your team? Do you have a big, audacious goal? “I want to get to 2 billion in 3 years.” How are you thinking about that?

We do fewer mechanical things around how to get a monthly user to a weekly user. We have an incredible research team—an absolutely world-class research team that's constantly iterating on models, making the models suddenly able to do things that computers have never been able to do before in the history of computers.

We think a lot more about how we stay really tight with the research team. It's just this incredible nucleus of creativity and discovery. How do we stay really close to them so that we can constantly be building products just on the edge of what the models can do?

Our feeling is that if we're building just on the edge of what the models can do, then we suddenly have a product that can do things that no one's ever seen a product do before. Even if it's not perfect when it starts, we believe in iterative deployment. We release early, and we release often. It's better to make lots of small mistakes and, collectively as a society, understand what AI is good at, what it's bad at, and how we work through it. We'd rather do that.

If you build a product that can do amazing things, even if it's only 70% right at that thing, in 2 or 3 months the next model is going to come along and it's going to be 95% good at that thing. All of a sudden, you've got something great. We've brought everybody along with us and learned from them at the same time—learned from the way that they're using it.

The thing that we've really tried to get right—and we're not perfect at this, but I think we're a lot better than we used to be—is the really tight loop between research and product. When we have that loop right, we're solving the problems that people have. Feedback goes back to research, and the research team says, “Oh, it can't do that very well? We can fix that.” Then the product gets better. That's when magic happens.

You said something prophetic on stage today. It's simple, but it really caught me. You said, “Today is the worst product you'll use. The model that you use today is the worst AI model that you'll ever use for the rest of your life.”

It's a simple thing, and it's obviously true when you think about it, but it changes the way you think about building products. If you think about it the right way, it makes you much more open to building products that only kind of work—whether you're building ChatGPT and other products, or whether you're an enterprise building some internal tool.

If the model can kind of do it, then it's going to be great at it in a few months.

What things are you most excited about that are on the horizon?

I'm really excited about this. In order for ChatGPT to be truly useful, you need to go from it just answering questions that you have to it actually doing things for you in the real world, ideally even proactively, to understand the things that you're going to need to do and help you do them or suggest them, queue up a bunch of actions for you before you even get there.

It's why we've been really focused on adding things like connections into Google Docs and the products and services that you use every day. We've been really focused on personalization and memory so that ChatGPT gets to know you and what your preferences are, how you like to interact with it. In my case, I've made sure that it knows about my wife and what she does—she's a seed investor—and my kids, how old they are.

Then when I ask it, “Hey, we have some free time. What should we go do?” or “We're going to go on this trip,” it doesn't just recommend random hotels. It recommends hotels that are great for my family. Each one is small, but it adds up, right?

Especially if you think about getting proactive and really getting to know you and helping with something that you're going to realize you need to do 4 hours from now, personalization is going to be a big deal.

No doubt. One quick question on the interactivity between different apps. There was some kerfuffle last week. I think Salesforce changed its terms of service in a way that would make it more difficult for other people to build models. How do you think that's going to play? Do you think people will put up walls or not? Will they be friendly with you running a widget on top of their app or not?

Yeah, it's a good question. I think we're going to feel our way there as an industry, as a society. It's one of the reasons that we believe so much in iterative deployment, where we release early, release often, because AI is going to change just about everything.

It's going to change the way the web works. It's going to change the way we interact with services, and we don't want to do that unilaterally. I think the way to do this smoothly is to iterate together on it.

Total sense. So, we figure the more we release and put this stuff out in the world, the more we all get a sense of where things are going. It's not like we know all the answers, right? We all get a sense of where things are going, and we can co-evolve in the right way.

I want you to have access to my contacts, my address book, my email, my texts, my everything.

Yeah.

That's your data. You should be able to take it to whatever AI you want to take your data to. You're paying for that data.

6. Regulatory Capture

On those lines, Kevin, one of the things that Bill and I have talked a lot about on this pod is our concern about regulatory capture and regulatory intervention. Folks who are alarmist and saying this is dangerous. We need to slow it down. We need to really regulate it.

Along those same lines, whether or not we're going to have open-source models or every model is going to be closed, it seems to me that OpenAI has been on this journey over the course of, let's call it, the last year or 2, where there was a belief, I think, that OpenAI was in the “go to Washington and regulate everything and kind of slow it down” camp. Now I think it's more clear, the distinction that's being drawn.

OpenAI is talking about launching an open-source model here shortly. You seem to be in the camp that it's too early to regulate: get out there, accelerate, make sure that we distribute around the world and America's models are doing well. Is that a fair characterization, you think, of where OpenAI is today?

I think no matter what, it's important to be engaging, right? As I said, AI is going to change everything, and I think we do better when we co-evolve. It's one of the reasons I was saying earlier that people should just be trying the models.

I think the same is true of our lawmakers and folks like that. They should just be using AI. You start using it and you're like, “Oh, this isn't so scary. This is helpful.” You understand the nuances. You understand what it's good at, what it's not good at, right? You get a sense for how it's improving.

I think that's really important. Context really matters. That's one of the things we do when we're in D.C. Part of it is just getting together and giving people a sense of, “Here's where it is today, and here's where we think it's going,” so that people can make informed decisions.

7. The Gentle Singularity

I don't know if you read Sam Altman's blog that was out this week called “The Gentle Singularity.” I'd encourage everybody to actually go read it. It's this idea, Bill, that it's not this scary thing that's all of a sudden going to show up, which I think a lot of people were thinking about with AGI—that there was going to be this moment that everything was going to change.

Sam's like, the world has already changed dramatically over the course of the last 24 months, but people aren't noticing it because it hasn't changed that much about our daily lives. We just switched from using Google to get 10 blue links to using ChatGPT to get answers to our questions.

I'm now uploading my blood test information to ChatGPT because I want it to tell me things that are useful about it. I still am at the point where I find those things pretty extraordinary. I can't believe that it can do these things and that I'm getting it for next to free.

But that level of profound change has come without that much societal disruption.

Yeah.

I think we definitely are in the camp that it is way too early to obstruct American AI with useless regulations. We don't even know what we should be regulating. It's like regulating the auto industry right out of existence early.

It took us 40 years to realize we needed seat belts and airbags, and that we needed speed limits and things like that. We'll need all those things for AI, but I think it's way too early to be doing that.

It's been a welcome development on our end to see OpenAI increasingly vocal about the need to make sure that we stay open.

Kevin, one thing—you didn't give any specifics, but I think you had a question from the audience about the hardware device, and you did share some philosophical thoughts. I think the audience would be thrilled to hear what you said earlier. This is about the acquisition of Jony Ive's business, correct? What do you guys intend to do in hardware?

Oh, I just said—and I think this is as true of hardware as it is of software—that AI is going to change everything about the way that we do our jobs, the way that we get stuff done in our personal lives. I think basically every product, service, device, et cetera, that we use will need to be reinvented.

Yes.

That doesn't mean that the incumbents can't reinvent it. I think sometimes they will, right? But I doubt in every case they will. Certainly history doesn't tell us that that would be true.

I think that means there's a huge opportunity for reinvention, and that's an opportunity for us in places where we think we can play a role. Where we think we have a perspective, we'll go compete. It also means great opportunities for startups.

Yes.

By the way, I do think it means good opportunities for incumbents if they can move quickly and sort of overcome the innovator's dilemma. That's a really hard thing to do.

No doubt. So I just think if you look at the products that you spend your time in on a daily basis and compare them to 5 years from now, I think they're going to be dramatically different. Is there any time window where we might find out more?

You'll have to find out.

Okay. We had you do a flyby today. We're going to have you come on the pod and we'll do a full-blown podcast together. We're here celebrating the 10th anniversary of East Meets West. We were saying earlier it's an event that we all look forward to coming to.

8. Level of Confidence

I'm going to ask you this final question, big picture. They forecasted out 5 years, right? What's possible for OpenAI? Skip the time window, whether it's 3 years or 5 years or 7 years. What's your degree of confidence that, given the value that OpenAI and ChatGPT are delivering into the world—and we haven't even gotten into really enterprise and everything you're doing with enterprise—you just announced a big deal with the government, the Defense Department, et cetera—what's your level of confidence that you guys will reach $100 billion of revenue in the future and kind of be that AI-defining company?

At $100 billion in revenue, it seems to me that this company could be on a path to being the most valuable company in the world, right? A $5 trillion to $10 trillion business in the fullness of time, which frankly would not surprise me for the winner in the age of AI—the winner in consumer and a big company in enterprise.

But now you've been here a year, right? You joined something that you thought was going to be the biggest thing of your career. You've been at extraordinarily large and successful companies. What's your level of confidence?

Look, I think the opportunity is all there in front of us. I've said 3 times now how I think AI is going to change everything.

Everything. And the path is there, and it's up to us to execute. Yeah, it's why—this place, OpenAI, moves faster than any company I've ever worked at in my life.

Wow.

9. Work Experience

And I thought I had worked at places that moved pretty fast—Twitter, Instagram.

Yeah, nothing compares.

I think it's because we have amazing people on the team. We really try to push decision-making and responsibility down into the teams. I think in an AI world, you need to do that because the AI models are so new. You don't know all of the capabilities top-down, upfront.

By the way, publicly, it's why we release early and often. We try and get it in people's hands because then they also get a chance to figure out what the models can be. So I have a lot of confidence in us. We've got to execute.

We've got to be on top of our game because we have a very serious set of competitors, and that's real.

But, man, I just feel good. When you look at this team, you've got Sarah Friar, Greg Brockman, Sam, et cetera. Even though the company's small, Bill, it's an extraordinarily deep bench of talented people who've done this in other places.

Not only does it come down to execution—as an investor, I of course want them to succeed—but in terms of defining the company that's going to do the right things in this moment, I feel good to have this company in the lead because these are great people that I think are going to do the right things. And so, it's great to have you helping lead the way.

You're nice to say it. I think the people you named tend to get unfair amounts of credit. I'm beyond excited for Sarah Friar to start. All the other people you named—we tend to get unfair amounts of credit.

I could give you 100 names of the people who are doing the real work and are actually the reason that OpenAI is successful. Hopefully, we can tell their stories over time, too.

We will. We will. All right. Thanks for being here, man.

Thanks so much. Appreciate it. Take care.