Hard Fork Live, Part 1: Sam Altman and Brad Lightcap of OpenAI

I'm Kevin Roose, the tech columnist at The New York Times.

I'm Casey Newton from Platformer.

And this is Hard Fork.

Little behind the scenes. We were getting ready to go on and, if you were at the live show, you know that the show started with a marching band coming in that Kevin and I were leading. Kevin and I were marching down different staircases, trailed by a band of marching musicians. Kevin was loaded into his position, and I went down to the next door with my 3 musicians.

I went to open the door, and of course it was locked. They were already playing the cold open for the show, and we had a few seconds left. I was frantically trying to wave someone from the SFJAZZ Center, and he ran up with his keys. Fortunately, everything worked out. The music started on time, and—

Oh, I'm so glad. That was a near miss. I would've had to go out and march the band in myself.

Yeah. It's always interesting to me when they lock the audience into the theater. I wasn't sure exactly what was happening there.

Release the bees.

Well, look, when you do a 2-hour podcast taping, some people are going to try to leave, and you're going to want to have a plan for that.

It's true.

And we had a plan, and the plan was: You can't leave.

The best audience is a captive audience.

Absolutely.

This week it's Hard Fork Live. You'll hear our first-ever podcast taping in front of a live audience in San Francisco. We've got a special appearance from San Francisco Mayor Daniel Lurie, and the conversation that had everyone talking this week: our extended interview with OpenAI CEO Sam Altman and COO Brad Lightcap.

Got a little spicy, hmm. Well, Casey, Hard Fork Live is in the books. How are you feeling? Are you recovered?

I am floating. I think everyone should have the experience of starting a podcast and then having 700 people come to watch it. It really makes you feel good.

Yeah. We had such a great time. Thank you to everyone who came out. We had a packed house, and I have to say, it was so much fun.

It was so much fun. We had a cocktail hour afterward. We got to meet everybody and take selfies, and I was meeting folks who had flown down from Seattle, who had flown in from New York. We had one guest who came in from Switzerland. The resources that people put into coming to hang out with us just meant so much to us.

Yeah. People would tell me that they flew in from some other place, and I would think, “Really?”

You're like, “Do you have a conference here this week, or what actually brought you here?”

Is there a restaurant you wanted to go to? No, but people were so lovely, and we got to meet and talk with so many of our wonderful listeners after the show. It was a really fun experience.

And we also got to have some really great conversations onstage.

Well, Casey, Hard Fork Live is in the books. How are you feeling? Are you recovered?

I am floating. I think everyone should have the experience of starting a podcast and then having 700 people come to watch it. It really makes you feel good.

Yeah. We had such a great time. Thank you to everyone who came out. We had a packed house, and I have to say, it was so much fun.

It was so much fun. We had a cocktail hour afterward. We got to meet everybody and take selfies, and I was meeting folks who had flown down from Seattle, who had flown in from New York. We had one guest who came in from Switzerland. The resources that people put into coming to hang out with us just meant so much to us.

Yeah. People would tell me that they flew in from some other place, and I would think, “Really?”

You're like, “Do you have a conference here this week, or what actually brought you here?”

Is there a restaurant you wanted to go to? No, but people were so lovely, and we got to meet and talk with so many of our wonderful listeners after the show. It was a really fun experience.

And we also got to have some really great conversations onstage.

Yes. For those people who couldn't make it, the great news is that we recorded the entire show, and we're going to be bringing it to you on the podcast in 2 parts. Half of it we're going to post in this week's episode, and the other half we'll post in next week's episode. And if you can't wait and you want to watch the full thing right now, you can go over to our YouTube channel, youtube.com/hardfork, and find the full show there.

And let's say there's probably never been a better Hard Fork episode to watch than this one—

Yes.

—because it was visual as heck.

Yes, it was visual. There were costume changes. There were props. We took our pants off.

More than once.

Yes. We can't wait to bring you snippets from the show this week and next. And in the meantime, we're going to take a little vacation.

Yeah. It's been a great 6 months, but a couple times a year, we like to shut down the operation and give everybody a chance to rest, and now is that moment.

So thanks again. We hope you enjoy excerpts from Hard Fork Live this week and next, and we will have a special episode of a different show coming the week after that. We'll be back to our normal programming on July 18th.

See you then.

Have a great summer.

Have a good few weeks, and then we'll be back for most of the summer.

That's true.

Early summer.

Early summer.

Early summer.

Enjoy stone fruit season.

I'm Kevin Roose, a tech columnist at The New York Times.

I'm Casey Newton from Platformer.

And this is Hard Fork Live.

Wow. Should we take off our band leader jackets?

Let's take off the jackets.

Okay.

They've served their—

Give us a second.

They've served their purpose.

It's very hot.

That's all right.

These are not Amazon's finest.

Yeah.

Yeah.

Can I leave this with you? Thank you. Thank you.

Will you guard this with your life?

With your life.

Yes.

No ribbon. Yeah, probably.

The baton, too, probably. I don't think we need that for our interviews.

Tripping hazard. There you go. Thank you.

Wow. Thank you to Brass Animals. That's the band you just heard. They're incredible.

We love Brass Animals.

They will be back later. And thank you all for coming. What a surreal thing this is.

We record this podcast in a booth that's about 2 feet by 2 feet, and we just send it out there. We think people listen, and we hope that people listen, but it's really surreal to see it all in person.

Yeah, it's so much fun to have all the energy in this building. We've been talking as the hours have been counting down to this. In 2021, Kevin and I just started texting each other all the time. We felt like there was one era, one epoch, that was ending, and another that was about to begin, and we just wanted to talk about it to a bunch of people. It's a really long way from there to this moment right now.

Yeah, and one of the questions I get asked most frequently is, “What is a hard fork?” Casey, what is a hard fork?

Okay, so something a little embarrassing about Kevin and me is that we had a crypto phase. It happens.

Some people go goth.

Yeah. Talk to your teens about crypto.

Yes.

At the time, we thought, “Well, it's 2021. Our show will probably be about crypto for the rest of time. Hard fork is a really important concept in crypto. Let's build a show around it.”

Yes. But we're so excited for tonight, and Casey, you look great, by the way.

Oh, thank you, Kevin.

Doesn't he look great?

Kevin, very nice of you.

A couple weeks ago, you told me you were getting a new outfit for this show, and I thought, “Shit, I have to get a new outfit, too.” So I went on a little bit of an adventure trying to figure out what to wear tonight. As I do in times of crisis, I turn to AI. I made a little slideshow about my adventure. I hope it's okay if I show it.

A slideshow? Okay, great, yes.

Yeah, I love a slideshow. You know that.

Sure.

So I started with ChatGPT, where I put in the prompt, “Give me some ideas for a glow-up.” I said, “Do not change my face. Just give me a glow-up. Make me look a little better.” ChatGPT, with its infinite wisdom, came back with this. Okay.

Wait.

Very stylish, but as you'll notice if you look closely, not me.

Can I just ask, why did you choose the angle of the floor looking up at you?

I don't know.

Okay.

It was in the office.

All right.

Then I asked for another example, and it gave me that. Okay.

Great.

It's a hoodie. Still not me.

Very cool hoodie.

One more example I asked for. It put me in a nicer room, but again, if you zoom in, not my face. So ChatGPT really saw my request for a glow-up and said, “I can't help you there. We're going to need to involve plastic surgery.”

Yeah, the tech is only so powerful at this point.

So then I thought, “Okay, maybe it’s just ChatGPT. Maybe Gemini will do a better job.” So I said to Gemini the same prompt: “Don’t change my face.” Gemini said, “What if he looked like David Beckham? That would be good.”

But I didn’t give up there, because there’s an app out there called Doji. It’s sort of like a high-fashion thing where you can scan your body and your face, and it’ll render a 3D image of you and then tell you what to wear. So I put my photos into Doji, gave it these photos, and said, “Create my likeness and tell me what to wear.” It came back with a suggestion that looked like this. Thanks, Doji. What do you think?

I would have worn that.

Yeah, you could pull that off.

Okay.

I’m not sure I could. So, having failed at having AI dress me for tonight, I did what every straight man does in times of distress: I went to Uniqlo. That’s what I’m wearing tonight.

Good job. That’s a good job. Smart man.

Yeah.

Smart man. Well, I think it turned out great, Kevin.

Thank you.

And we have a really great show for you. I wanted to say, you guys sold out this building before we announced the guests, so I want to say thank you for that. Thank you for trusting us. We wanted to reward your trust in us with a really special show.

And before we get started, in grand Hard Fork tradition, we should do our disclosures, because we’re going to be talking a little bit about AI tonight.

Yeah.

So, Casey, you want to take it away?

Who’s excited to hear the disclosures? Amazing. Well, I’m proud to say my boyfriend works at Anthropic. He may even be here tonight. Love you, sweetheart.

And I work for The New York Times Company, which is suing OpenAI and Microsoft for alleged copyright violations related to the training of large language models. Did I get that right?

Yeah.

There we go.

The last thing I would say before we get started, Kevin, is that it’s also just a dream to be doing this show here in San Francisco. San Francisco is my home. It’s where we make the show every week, and it’s where so many of the changes we talk about every week are happening.

If I have any regrets, it’s just that I feel like San Francisco has been changing a lot all around us, and we’ve been so heads-down, it’s been hard to keep track of all the changes.

Yeah, it’s a really good point. If I could change one thing about tonight, I think we should have invited someone with a little bit of relevant expertise here, someone who really understands San Francisco politics. Who would that be?

Kevin, I told you to put the door on Do Not Disturb.

Let’s see who it is.

Who’s at the door? It’s the mayor of San Francisco—

Hey—

—Daniel Lurie. Hi, how are you? Nice to meet you.

Hi. Thanks for coming.

Oh, my goodness. Please have a seat. What a fun surprise. Thanks for stopping by, Mr. Mayor.

Thanks for having me. Thanks for bringing everybody to San Francisco.

Absolutely. I think they’re happy to be here. Are you happy to be here? Since you’re here, let’s toss a couple questions at you. You’ve been in office for just about 6 months now, and the tech community, I would say, has generally been very supportive of what you’re doing. You’ve even formed a council of tech advisers that includes our guest from later tonight, Sam Altman. What kind of advice are they giving you, and are you taking it?

Sam was on our transition committee, and now we have something that we started called the Partnership for San Francisco, where we have leaders from across business, arts, and culture giving us advice and helping to cheerlead for our city.

You all are seeing the revolution happening. There’s no better place in the world, in terms of an ecosystem, than San Francisco. There was a lot of talk for a number of years about how San Francisco was done. That was a bad bet, as everybody knows. I mean, the guests that you have tonight—you’d have to fly them in, but they’re living right here in San Francisco.

They were local.

It’s all happening right here.

Yeah.

You talked about Anthropic. You’ve got Dario, you’ve got OpenAI, you’ve got Salesforce, you’ve got Databricks. Cities across the globe would die to have one of those companies, and they’re all home-based right here in San Francisco.

So, I’m talking to them. I’m talking to arts and culture leaders, and we’re doing everything in our power to create the conditions for success. We’re off to a good start.

Now, we’re going to be talking a lot about AI tonight. I’m curious: Is AI helping you in your job at City Hall?

We are absolutely talking to all the companies, saying, “How can we get their help on synthesizing all the data that comes in?” We have 58 different departments at City Hall, and they don’t always talk to each other.

We have great intellectual horsepower here. We’ve got great universities, we’ve got these great companies, and we are engaging with them. They’re already starting to help, and you’ll see more in the coming years.

I feel like governments don’t always have a reputation for having state-of-the-art technology. Is there anything you wish that you had that you don’t have, or that the tech could do for citizens here that it can’t yet?

I think it’s just making sure that we’re crunching the data. We have over 34,000 city employees, and getting them to talk to each other and understand what they’re going through—

I’ll tell you a quick story. There were not staff meetings going on between our large agencies, and I instituted something every Tuesday morning. So this morning, at 9 a.m., we had the 20 large agencies get around the table.

This is not tech; this is old school, but it starts with old school.

Yeah, this technology is called the table.

Yes, that is right. But, by the way, tech doesn’t work if you don’t communicate with each other. That’s why I think everybody’s got to be back in the office. A lot of these AI companies are in the office 5 or 6 days a week. I went to OpenAI’s new office by Chase. They are in the office because they know it doesn’t work unless you’re communicating.

Our department heads are meeting with each other once a week, gaining knowledge from each other, seeing how they can help each other, and the tech then follows that. We’re hoping to lean in with all these great companies.

Now, I have to ask you, Mayor, about your social media presence. You are very active on short-form video apps such as Instagram Reels and TikTok. You post more Instagram Reels than a Gen Z clout chaser, and honestly, they’re pretty good.

I would say, grading on the curve of politicians, they’re great. I’m curious: Who does your social media? What’s the strategy there? Is it working?

I was told there was a review of my Instagram in the local—in The Chronicle—and it said that I had not yet made the camera my lover. So I have work to do.

That’s on OnlyFans.

Yes.

That’s—yeah, yeah, yeah.

So, listen, I’m having fun with it. We know there’s so much noise out there, and to break through and communicate with people like you—you all do so well with your show—we felt like we had to reach people directly, and it is taking off. You all can check it out, and you can learn what it's like to be mayor and to see all the small businesses that are amazing in San Francisco, the restaurants—the bars. I just went in honor of Pride. I just stopped by before to a bar that I've passed so many times. It's called The Sinch.

Gay bar on—with Pride Week. We went by there just now, and it's been there for years, and it is amazing. And, like, I wanna highlight what is so special about San Francisco, and that's what we're trying to do. And I usually just give the mic over to the restaurant owner or the arts leader and say, “What do you do?” The city gets to see it, and that’s what it’s all about.

It’s awesome. We’ve got to let you go, but before we go, I wanted to ask: Could we make a Reel with you? Would you make one with us?

Yeah, absolutely.

Just right now? Okay, let’s stand up.

All right, here we go.

Let’s do that.

Oh, but I don’t have my phone—oh, you got your phone.

Oh, I got it. I got it. Yeah.

Okay. Oh, boy.

I’ll do it.

Oh, boy.

So we can just turn it into selfie mode right here—

Yeah.

—and we can just go.

Yeah.

You are the maestro here, so you’ve got to direct it.

Okay.

Okay.

All right, I’m on Hard Fork right now. Hey, you two, tell us what’s going on here. This is your first live audience.

First live audience. We’re here for Hard Fork Live at SFJAZZ. Having a great time here with Mayor Lurie.

Woo!

Let’s go. What I’ve got to tell everybody—I’ve got to tell your audience—is that San Francisco, we are on the rise.

When we are at our best, there is no better city on the planet than San Francisco.

Period.

Let’s go, San Francisco.

And that’s on period. Thank you very much, Mayor Lurie, for stopping by.

Thank you, Mayor Lurie.

Thank you. Oh, my goodness. Kevin, we’ve already had so much show, and it’s literally just begun.

I didn’t even tell him that the most relevant thing was that you didn’t get your permit for your hot tub.

That’s right.

Mayor Lurie, you need to raise revenue. I know a guy.

When we come back, we’ll bring you our Hard Fork Live interview with Sam Altman and Brad Lightcap from OpenAI.

So, Casey, that was a really fun short interview with Mayor Daniel Lurie of San Francisco. Very grateful that he stopped by. And now we are going to bring you something different.

Yeah. And we want to give you the behind-the-scenes, because whether you were there at the show or you’re just about to listen, a lot happened backstage that you’re going to want to know before you hear this.

Yeah, so this was our interview with Sam Altman and Brad Lightcap from OpenAI. We’d invited them to come and talk with us about AI and the future, topics we talk about on this show all the time.

Yeah. And so, the way that this show works is we don’t see these folks before the show. The show just sort of starts, and they show up. Kevin and I are backstage. The most recent thing that has happened is that we did this amazing demo wearing these exoskeleton pants that help people who have mobility issues, and we need to remove the pants.

And so we go backstage, and we remove our pants in front of Sam Altman and Brad Lightcap. They were very cool about that, I would say. They didn’t make any comments or anything.

Yes, they were pointing and laughing.

Yes, as we feared that they might. And so we’re getting ready to go out on stage, and the thing that is supposed to happen is we’re going to have 5 or 6 minutes where 2 things happen. One is we shout out our families to thank them for being there, and then we want to set up the story of OpenAI in this moment. The company has had a lot going on, and you and I just wanted to banter back and forth a little bit to set up what was going to happen before Sam and Brad come on stage.

So I go up to Sam and Brad right before this happens, and I say to Brad, “Hey, thanks for being here,” and shake his hand. I go up to Sam, say, “Thanks for doing this,” and shake his hand. And then Sam says something like, “Hey, only ask interesting questions tonight.” Basically, “Come at me a little bit.”

And I say, “Okay, yeah, sure.” And I say, “If you want to troll me a little bit, make fun of me, go for it.” And what Sam says is, “Well, I don’t strike first, but I do strike back.” And I was kind of like, “Okay, well, I don’t think a lot of striking is going to be coming from me during the show,” but, “Okay, sure, whatever.”

And so then you and I head out on stage, and we had just started the bit that we had planned, sort of saying, “Okay, how’s everybody doing? Did you enjoy the first half of the show?” And I turn to my right, and I see Brad and Sam walking out on stage.

Yes, minutes before they were supposed to arrive on stage. They had our whole intro music playing. We were going to tee it up. They just kind of barged onto stage.

Yeah. And so when this happened, my thought was, “This is probably just a production mistake.” Someone backstage told them, “This is your moment,” and pushed them on stage. It’s a live show. These things happen. I think my first impulse was to say, “Hey, you guys want to give us a minute?”

But they just kind of advanced on us and sat down, and were basically like, “Okay, cool, what do you guys want to talk about?” And we were like, “Well, we kind of want to set up your segment.” In hindsight, though, Kevin, I think we realized that actually no one backstage had told them to come. This was kind of a power move that they were trying to do to get us flustered heading into what would happen next.

Yes, they were trolling us, and specifically they were trolling us about the lawsuit between The New York Times and OpenAI, which they know you and I are not involved in, right? They are not under any impression that we are part of the litigation team at The New York Times. But it is clear that they had something they wanted to get off their chest about that lawsuit, and I think just have a little fun with us.

Right.

Yeah. And so they came at us pretty hard. You will hear it in the interview, and you and I are trying to steer the conversation to stuff we can actually talk about. In fact, one of the things that was going to happen in the bit that we didn’t do was you saying, “Hey, by the way, about this whole lawsuit thing, I’m not involved in it, and I can’t talk about it.” And by walking out on stage, they sort of prevented that moment from happening.

Yeah. I will say I learned a lot about this interview, only some of it from the questions we asked. I learned a lot more about Sam Altman from this brief interaction before the segment actually was supposed to start.

Yeah, I mean, look, I think we got a lot out of just the questions that we asked, and we got into so many things that we’ve wanted to talk to him about: the business, the risks of job loss, and the risks of people using ChatGPT and having mental health breaks. But I do think you’re right, Kevin. You just learn something about people by observing them in public settings, how they behave, how they engage.

And so I think there’s just a lot for everyone to chew on. I was reflecting last night that the first time we had Sam Altman on the show, 2 days later he gets fired, and it sort of, in many ways, kicks off the moment that we’re living in now, and that was extremely surreal.

The evening that we had at Hard Fork Live was kind of a perfect sequel to that because now you have a person who is fully in control, who wants to bend reality to his will. And if there’s a couple of journalists he can just kind of kick in the shins on his way toward building God, he’s going to be happy to do it.

Yeah. And we should also say Sam did send us an email after the show apologizing for his behavior. He said he was, quote, “such an asshole,” and that he felt bad about it. So that tells you something. But yeah, what you’ll hear in this segment is the 2 of us being somewhat flustered that our planned introduction is just being interrupted by these 2 guys wandering onto the stage, and we’ll take it from there.

And I will also say I have no idea what’s going to happen in Sam Altman’s third appearance on Hard Fork, but the bar has been set really high.

Yes. All right, when we come back, we’ll have the interview with Sam Altman and Brad Lightcap from OpenAI.

We're about halfway through the show. I thought I would just check in. Are you guys having a good time?

Okay, good. Okay, good.

We're having a lot of fun.

Oh, boy.

They told us to come out.

Wait, wait.

They just pushed us out.

There's no way. Okay, great.

Do you want us to go back?

Well—

No, come on out.

Come out.

Sam. Hi, Brad.

Hey.

Wow.

Good to see you.

I love it. We're doing it live today, family. Okay.

You guys are learning a very important thing about our show, which is that it's edited.

Yeah. Kevin, do you have maybe one more thing you want to say before we get into the interview?

Yes. So we're here with Sam Altman and Brad Lightcap from OpenAI.

We'll just hang out. Do your thing.

You can check your email.

Go for it, yeah.

Well, what we were going to do is tee up your appearance a little bit by just giving a little lay of the land of what's been going on at OpenAI, which is a very busy company.

This is more fun, now that we're out here for this, though.

Yeah. No, jump in. You guys can be Statler and Waldorf over there.

We'll just provide commentary.

I had a list of headlines from the past couple of weeks, and if there's anything that just makes you want to roll your eyes, you can roll your eyes. Okay.

Are you going to talk about why you sue us because you don't like user privacy?

Okay.

Woo!

The last thing Sam said to me before he came on stage was, “I don't strike—”

I did say that. That's true.

Well, I would tee it up with the headlines.

Yeah. Do you want to say something, Kevin, about The New York Times?

Oh, yes. We should also give our disclosures, which is that we're just journalists. We are not involved in the lawsuit. I don't know.

Yeah.

We don't represent the company's views on the matter.

What do you think about it, though?

What do I think of the company's views? Are you trying to get me fired, Sam?

No, I just want to know.

I don't have any other skills.

It seems like you've got a podcast. A lot of things.

Yeah. Well, okay, great. So you said that.

I'm going to just pretend I didn't hear that.

You're pro the lawsuit.

What's that?

You're pro the lawsuit.

I think people should read the relevant filings and make up their own mind.

Yes, democracy! Democracy! We love it. We love it. We love it.

And what about your mind after reading them?

Well, it feels like you have something you want to say about the lawsuit.

No. Look, I like user privacy. I don't think companies—

You want to explain what you're talking about here?

Yeah, Sam, what do you mean?

Well, you guys are suing us.

I'm an independent contractor. I write a newsletter. It's called Platformer.

Yeah, don't drag him into this.

Yeah.

And one of the things that's happening is that your employer is The New York Times. Let's just say The New York Times.

Sure.

The New York Times. One of the great institutions, truly, for a long time, is taking a position that we should have to preserve our users' logs, even if they're chatting in private mode, even if they've asked us to delete them. We're happy to fight out the lawsuit, but we really think that privacy in AI is just this extremely important concept to get right for the future, and we care a lot about the precedent. We still love you guys, still love The New York Times, but that one we feel strongly about.

Well, thank you for your views. I'll just say it must be really hard when someone does something with your data that you don't really want them to. I don't know what that's like personally, but maybe someone else does.

Okay, let's get started with the questions.

I was recently told by a guest on stage that the singularity would be gentle. So I just wanted to point that out to you.

Yeah. Casey, read your headlines.

Speaking of, do we still want the headlines?

Let's go into the questions.

That's real.

These people know what's been happening with OpenAI.

Okay.

Yeah.

But I think it's important to give a sense of just the sheer volume of stuff you all are doing. We've been covering tech for a long time. I don't think either of us have ever seen a company that makes this much news this regularly in this many areas. You've got hardware stuff going on with Jony Ive. You've got ChatGPT continuing to grow. You're doing this defense contract, this $200 million defense contract, a deal with Mattel to make toys—

I think you were the first company to sign a deal with Mattel and the military in the same week.

I like that.

Stargate, your big data center project, your attempted conversion to a for-profit company. So there's just a lot going on in your worlds. Casey?

Well, we wanted to start with something that I thought you might have fun talking about, which is that rascal Mark Zuckerberg keeps coming after your employees.

I'm sure this is happening to you guys on some level all of the time, but I wondered if there's been any particularly funny or crazy moment over the past few weeks as they've really stepped this up.

Any that you think have been particularly funny?

Many.

Many.

Sure.

I don't know. I haven't slept in 4 years, so at this point, nothing fazes me.

One of the strangest things about the job is the amount of things that can go wrong by 11 o'clock on a Monday morning. It's just an astonishing diversity of stuff. So it's like, okay, Zuckerberg is doing some new insane thing. What's next?

Okay, I want to gossip for just 1 minute more.

Only 1?

Do you think Mark Zuckerberg—

Just 1? We're here for a lot of that.

We can do more gossip. Do you think Mark Zuckerberg actually believes in superintelligence, or do you just think he's saying that as a recruiting tactic?

I think he believes he's superintelligent.

Lightcap, off the ropes. Very good. All right. But it sounds like your confidence has not been shaken by the recent raid on your employees.

We're feeling good.

Yeah. All right, all right. So you recently wrote this essay I just mentioned, “The Gentle Singularity,” and you wrote, “We're past the event horizon. The takeoff has started.” People read that and thought, “Do these guys have a superintelligence that they're keeping in the basement?” I assume that's not true, but tell us a little bit about why you wrote that essay and when, in your mind, we hit this point of no return.

We don't have a superintelligence in the basement. But we have shipped a model that any of you can use, and I hope many of you do, that is quite smart relative to what you might have expected 5 years ago where the world would be with AI.

We've all adjusted to this. We've all just said, “Oh, this is the new world. We have PhD-level intelligence in our pocket, and we can use it, and we can talk to it all day. It can do all this stuff for us.” But it is kind of remarkable that this has happened, and this is the world now.

When you're living through moving history, you adapt so quickly that I think it's hard to get the perspective of, man, 5 years ago, most of the experts made fun of anyone who said AI, AGI might be even a plausible thing to work toward. And now here we are with this thing that has come quite a long way that we can use in all these ways.

We used to try to just say, “Hey, this AGI thing is coming. It might be a really big deal. It might be really important. You all should pay attention.” No one cared. And we shipped a product, and then people cared.

A thing we've learned again and again is that talking about it doesn't seem to break through. But if people can use it and feel it and see where it's good and where it's bad and integrate it with their lives, then they do.

And so now we see many years ahead of us of extreme progress that we feel is pretty much on lock, and models that will get to the point where they are capable of doing meaningful science and meaningful AI research. And we continue to feel a responsibility to tell the world about that. Most people won't listen. Maybe some more people will listen this time, but we'll ship products that expose these higher levels of intelligence that we'll build, and that is how I think people will really get their hands around what's happening.

Now, Brad, it's your job to manage the business of OpenAI. What does being past the event horizon toward superintelligence mean for OpenAI as a company? I imagine it makes lots of different kinds of decisions different, but how do you plan for a world like the one that Sam is describing as a person who runs a business?

Yeah. It's the fun part of what we do. We debate this internally a lot. We will wake up one day with this incredibly powerful thing, and will the world be different that day? I think what we've all agreed now is that it probably won't.

To Sam's point, these things really have to be integrated into people's lives. They have to be felt, and that change is more gradual. So we work really closely with companies, as much as we do with users, to figure out what that process will look like.

I do think businesses will look very different in the future. My personal metric for what a business in the superintelligence age means is that you've got 1 person who has a lot of agency and a lot of willpower, who has the capacity to start a company that can do billions of dollars in revenue. It's hard to imagine now. You think, “Okay, yeah, I need salespeople, and I need product people, and engineers, and accountants, and so on.”

But all of that stuff now can just be managed, right? It can be built into the system. And that just gives incredible agency to individual people.

Mm-hmm. So I want to get into the nitty-gritty of building this future. Right now, the agents that you all have built for coding are really extremely good. People can build a lot of amazing stuff with them. Outside that domain, we've seen less progress. Talk to me about what's going to happen in the next year that makes you feel like you can start knocking off 1 or more of these other domains.

Well, first of all, coding's pretty general-purpose. A system that can write code, just like a person who can write code, can make a lot of other things happen. We're beginning to see scientists be much more productive with this. We're seeing companies really change a lot of their workflows.

The thing, though, that I am most excited for is that most of the ways people use AI today are sort of like: send a request, get a response. You send ChatGPT a request. It might think for a second, it might not. It sends you something back. You're in one of those vibe-coding things, and you do something, and you get something back.

I think I'm very excited for a world where each of us has a copy of o3, or many copies of o3, that are just constantly running, constantly trying to say, “Oh, I see this happening, and now this, and I'm reading Slack, and I'm reading email, and I see this and this and this, and you asked about this yesterday. Here's a new idea.”

We have this team of agents, assistants, companions—whatever you want to call them—that are doing stuff in the background all of the time. And that, I think, will really transform what people can do and how we work, and maybe, to some extent, how we live our lives.

So I use o3 all the time. It is helpful to me as a journalist. It can fact-check stuff for me. It can edit stuff for me. When is the moment when it knows what I do and, in the morning, starts doing that stuff without me telling it?

That's what I'm talking about, except I don't think it should be without you telling it. But I would love it if I woke up every morning and there was a drafted response to every email that had come in overnight, and I could click and say, “I want to send this one. I want to edit this one. I want to send that one.”

If I could open ChatGPT and say, “Hey, here was the stuff you were working on yesterday that you didn't finish on your to-do list. Here are my attempts at that. Do you want me to take this action and that one and that one? And, by the way, here are these other things that happened overnight with a customer, in the world or whatever, and here's a set of stuff I could do for you.”

I have all of this ready to go, and I could just go through and say, “Okay, do that. Don't do that. Here's what should have been different here.” That I'm very excited about. But I don't want to go to sleep and have o3 just start taking actions for me.

All right. I use o3 a lot too, and I find it very useful. The thing I will say is that it lies. More than previous models, I feel like it is a crafty, shifty assistant that will once in a while make stuff up, and it seems like the hallucination rates on these newer models are staying about the same or maybe even getting worse. So do you have a theory on why that might be?

I think it did get a little bit worse from o1 to o3, and we'll make it much better in the next version. I think we're earlier in learning how to align reasoning models and also how people are using them in different ways, but I think we've now learned a lot, and I would suspect you'll be very happy with the next generation there.

So you made your largest acquisition to date this year with Johnny Ives IO. The first crop of AI hardware that we've seen has not been particularly successful. Brad and Sam, what do you guys feel like you're seeing that makes you feel like you can do something different here?

Every time you re-platform technology, there tends to be a corresponding set of things that get built that change how we interface with that technology. So I think the question here is: Is that going to happen again?

You miniaturize the PC, and you have the mobile phone. The PC itself was a miniaturization of the mainframe, and so on and so forth. I think this one has a different direction. I think this one is really going to be about this very aware, very contextual, almost companion-like system that is going to be less about a dependency on a screen.

I think there's a place for a screen in that world, but it's going to be really about an awareness of the ambient environment. What's going on?

Sam mentioned the trivial example of something that's looking at your email. You can build something that's really bad that does that today, but to get to the version of that that's transcendently good, there's a ton of context and a ton of awareness that you have to have about each situation so that it helps you craft exactly the right response.

Imagine that now in any arbitrary situation and wanting to have that with you all the time. So I think that's a very compelling direction for this type of hardware.

It sounds a lot like Alexa. Is it going to feel a lot different from Alexa?

Don't you just want to wait and be surprised and get some joy? It's been a long time since the world has gotten a fundamentally new kind of computer. Let us try.

Okay.

If it's Alexa, we're going to be really mad. I'm just saying that right now.

So will I.

And these people will remember this.

Yeah. No, I think we have a chance to do something truly great, but hardware's really hard, and it takes a while. I've always wanted to try to do a new kind of computer, but that hasn't worked most of the time, so we're really going to take our time and try to get it right.

Sam, a few years ago, you described your relationship with Microsoft and its CEO, Satya Nadella, as “the best bromance in tech.” The bromance has been feeling a little wobbly recently. OpenAI needs Microsoft's blessing for this for-profit conversion, and Microsoft is reportedly peeved at you about a bunch of things, including the terms of a planned acquisition.

Last week, The Wall Street Journal reported that things had gotten so tense that OpenAI executives were considering reporting Microsoft to the government for anticompetitive behavior.

Do you believe that?

Are you saying it's not true?

You know what I always think when I read that? “I hope I get to ask Sam Altman about it.”

So what is going on, and are you caught in a bad bromance?

I had a super nice call with Satya yesterday about many topics, including our hopefully very long and productive future working together. Obviously, in any deep partnership, there are points of tension, and we've certainly had those. But on the whole, it's been really wonderfully good for both companies. We're both ambitious companies, so we do find some flashpoints.

But I would expect that it is something that we find deep value in for both sides for a very long time to come. In a world where I do read these articles sometimes—“OpenAI and Microsoft are about to collapse”—my calls are like, “How do we figure out what the next decade together looks like?” And it just doesn’t.

It doesn’t.

Again, not to pretend there’s no tension.

Yeah.

There is. But there’s so much good stuff there, and I think there’s such a long horizon.

That’s what we in the business call a non-denial denial.

I don’t know.

Just kidding.

Let’s move on into some policy stuff. You’ve talked to President Trump. What does he think about AI? What are those conversations like? That was not intended to be a laugh line.

You want to take it first?

Yeah.

I’ll do it. No, that’s fine. That was not intended to be a laugh line either.

Yeah.

I think he really gets it. I think he gets the technology. I couldn’t say that about all presidents. I think he really understands the importance of leadership in this technology, the potential for economic transformation, the sort of geopolitical importance, and the need to build out a lot of infrastructure. They’re very productive conversations. He has done stuff that has really helped the whole industry. It is easier to permit data centers and new energy to run those data centers than it has, I think, ever been before, and that could have gone the other way.

Hmm.

Dario Amodei of Anthropic recently said that he thinks 50% of entry-level white-collar jobs could disappear due to AI in the next 1 to 5 years. Do you agree?

No. No, I don’t.

Why not, Brad?

We have no evidence of this, and Dario’s a scientist. I would hope he takes an evidence-based approach to these types of things. But we work with every business under the sun. We look at the problem and opportunity of deploying AI into every company on Earth. We have yet to see any evidence that people are wholesale replacing entry-level jobs.

I think there is going to be some change in the job market. I think it’s inevitable. Every time you get a platform shift, you get a change in the job market. In 1900, 40% of people worked in agriculture; it’s 2% today. Microsoft Excel has probably been the greatest job displacer of the 20th century, and if we knew a priori that Microsoft Excel was coming, and everyone was fretting about it, I think in retrospect we would have thought that was dumb.

But I think there will be change, of course. There’s no evidence of it today, and I think we will manage through it. We have a lot of empathy for the problem. We work with businesses every day to try and enable people to use the tools at the level of the 20-year-olds who come into companies, and use them with a level of fluency that far transcends anyone else at those organizations. We see it as our mission to make sure that people know how to use these tools and to drive people forward.

I have to say, we’ve had some listeners write into the show and say, “Hey, I’m a junior coder. I just got laid off. I’m not feeling really good about my prospects here.” So that’s been a pretty small sliver of the economy. But I hear you talk about what you want to do with o3. I think if it gets as good as you’re saying, it’s not just going to be the junior coders who are going to be affected by that, right? So I guess what I’m saying is, I feel like I’m seeing slivers of it now, and I’m curious what you make of those.

I do think there will be areas where some jobs go away, or maybe there will be some whole categories of jobs that go away. Any job that goes away, even if it’s good for society and the economy as a whole, is painful—very painful, extremely painful—in that moment. I totally get not just the anxiety, but that there is going to be real pain here in many cases.

In many more cases, though, I think we will find that the world is significantly underemployed. The world wants way more code than can get written right now. I think we are already seeing companies who said, “Oh, I’m going to need fewer coders,” now saying, “Paradoxically, I need more coders. They’re going to work differently, but I’m just going to make 100 times as much code, 100 times as much product with 10 times as many people, and we’ll still make 30 times as much money even if the price comes down.”

All of human history suggests that if you give people better tools, if technology keeps going, although there are always people who say, “We’re going to be working 3 hours a day and sitting on the beach, and we’re going to have run out of things to do,” human demand seems limitless. Our ability to imagine new things to do for each other seems limitless. We always seem to want more stuff, to play increasingly silly status games.

Our jobs would not have seemed like real jobs to people in the not very distant past. They’d be like, “You’re sitting around talking on stage, and you’re trying to make a piece of software, and you’re trying to do a podcast, and you’re trying to make people laugh.” That’s great, but that’s play. That’s not a job. You have plenty of food to eat. You have all this stuff to do. You have this unimaginable luxury.

Because I think human imagination and desire—demand, whatever you want to call it—are limitless, we will find incredible new things to do. Society will get way richer. Generally, as society gets richer, unemployment goes down, not up, and I’d expect to keep seeing that, even though people don’t talk about that very much. The entry-level people, I think, will be the people who do the best here. They’re the most fluent with the tools. They’re the most able to think of things in very new ways. They have this sort of large canvas.

There’s going to be real downside here. There’s going to be real negative impact. And again, any single job lost really matters to that person. The hard part about this is that I think it will happen faster than previous technological changes, but I think the new jobs will be better, and people will have better stuff. The take that half the jobs are going to be gone in 1 year or 2 years or 5 years or whatever, I think that’s not how society really works. Even if the technology were ready for that, the inertia of society, which will be helpful in this case, is—there’s a lot of mass there.

Yeah.

The thing we actually see empirically, if we want to talk about what we observe, is somewhat like what Sam’s describing. There’s actually a class of worker who is more tenured, more oriented toward a routine and a certain way of doing things in a certain role, who is not actually sophisticated in the use of these tools. They’re not adopting them. They tend to think that it’s not worth their time or whatever it may be, and I think there’s a lot of fear there.

A lot of what’s driving that fear are 20-somethings who are actually coming into the workforce, who have been using these tools for years and years and who’ve mastered them in a way that they look at these other jobs and they’re like, “Why would you waste your time doing that thing? I can do that much faster.” So the thing that companies actually worry about deeply is not the entry-level job. It’s really the job of the person who has been at the company for 30 years, who’s done something in a very rote and routine way, where there’s an urge on the side of management to really modernize the tool set. What do you do with that, right? That’s, I think, actually the addressable problem for us.

Sam, 2 years ago, you testified to Congress about the need for more AI regulation. More recently, you went back to Washington and testified again that you supported a light-touch regulatory regime. Earlier this year, you said you supported a federal preemption on state-level AI regulations, a version of which is now part of the Republican budget bill. What changed? Did you see the regulations that people were writing and think, “We wanted regulation, but not like that?”

No, I still think we need some regulation. But I would say a patchwork across the states would probably be a real mess and very difficult to offer services under. I also think that I have become more—I don’t know if “jaded” is quite the right word, but something in that direction—about the ability of policymakers to grapple with the speed of technology.

I worry that if we kick off a 3-year process to write something that’s very detailed and covers a lot of cases, the technology will just move very quickly. On the other hand, as these systems get quite powerful, I think we clearly need something. I think something around the really risky capabilities, and ideally something that can be quite adaptive—not a law that survives 100 years and says, “Here are exactly the things you can do and not do”—would be good.

But yeah, it's impossible for me to imagine a world where society doesn't decide we need some framework here.

Earlier this year, you adjusted GPT 4.0 after it inadvertently became more sycophantic than you intended. Since then, we've read more stories about how ChatGPT and other chatbots can destabilize people by sending them down conspiratorial rabbit holes, making them feel like they're having mystical experiences. Can that be stopped? Do you want it to stop?

Of course we want it to stop. We do a lot of things to try to mitigate that. If people are having a crisis that they talk to ChatGPT about, we try to suggest that they get help from a professional and talk to their family. If conversations are going down a rabbit hole in this direction, we try to cut them off or suggest to the user that they think about something differently.

But I think the broader topic of mental health and the way that interacts with overreliance on AI models is something we're trying to take extremely seriously and rapidly. We don't want to slide into the mistakes that I think previous generations of tech companies made by not reacting quickly enough.

Have you ever thought about literally putting a warning on it that says, “This is ChatGPT. You are not talking to God. You are not having a religious experience”?

The model will tell you things like that, and then users will write us and say, “You modified this,” and they change their custom instructions. But yes, there need to be a lot of warnings like that. However, for users who are in a fragile enough mental place, who are on the edge of a psychotic break, we haven't yet figured out how a warning gets through to them.

Well, we also have to be careful because there are an incredible number of use cases that I think probably, by sheer volume, outweigh some of the use cases you're describing, where people are really relying on these systems for pretty critical parts of their lives. These are things like almost borderline therapeutic use cases. I get stories of people who have rehabilitated marriages, rehabilitated relationships with estranged loved ones, things like that, where it's highly net positive and there's not a dependency.

It's the first time in their life that they've had something they feel like they can confide in, and it doesn't cost them $1,000 an hour, right? I was surfing in Costa Rica the other day, and a local Costa Rican guy paddled up to me and we were chatting. I asked him where he worked, and I said, “OpenAI.” He said, “Oh, you make ChatGPT.” He started crying. He was like, “ChatGPT saved my marriage. I didn't know how to talk to my wife, and it gave me tips on how to talk to my wife. I've learned that, and we're on a much better path.”

It sounds like a dumb and stupid story, but it's not. I was there.

That's great. We're back to even, because a chatbot tried to break up my marriage.

Well, not our chatbot, though.

It was your chatbot, but it was inside Bing, so we'll spread the blame around there.

Our model.

Now, Sam, you just had a kid. Congratulations.

Thank you.

Yeah.

Thank you.

Do you think over the course of their lifetime your kid will have more human friends or more AI friends?

More human friends, but AI will be, if not a friend, at least an important kind of companion of some sort.

Is that okay with you? If your kid, at one point when they're a little older, came home and said, “I've got an AI friend,” how would that make you feel?

If my kid felt like that was replacing human friends, I would have concerns about that. Again, there are edge cases. Any person who talks to hundreds of millions of people a day is going to meet a lot of edge cases, and in the sense that ChatGPT is talking to hundreds of millions of people a day, there are going to be some real edge cases in there.

But most people, much more than I was concerned they would, seem to really understand the difference between talking to a person and talking to ChatGPT. I still have a lot of concerns about the impact on mental health and the social impacts from the deep relationships that people are going to have with AI.

But so far, it has surprised me on the upside how much people really differentiate between “That's an AI, and I talk to an AI in some way and get something out of it,” and “That's a friend, and I talk to a friend, a person, in this other way and get a very different thing out of that.”

All right. Here's something I've always wanted to ask you. AI Twitter is still really active, even though Twitter doesn't exist anymore. You actively post there, share a lot of news there, and that's extremely helpful and good for Elon Musk, a man who is trying to destroy your company. Have you ever thought of just moving your posts somewhere else?

Where should I move them?

Well, you could create your own social app.

Don't go to Blue Sky. They don't like AI there. They're not going to be nice to you there.

It's a rough neighborhood.

Maybe a little. As a last thing, we wanted to know this too. We'd invite you both to answer this. Is there any part of your life that you feel like you want to wall off from AI a little bit? It's fun to talk about AI. We think it's all very useful. We're excited to keep building it. But this particular thing, we're going analog.

I've got to think about that one.

Surfing, presumably. Although maybe you're asking for tips.

Yeah, we're roboticizing that now too. It's unfortunate. But let me think about it.

I'm big on the analog stuff. I put my phone away and go for hikes every weekend, and I hang out with my family. I put my phone away, and I'm very happy not to have technology in the way for that.

Nice.

Yeah.

All right.

Brad and Sam—

That's Brad and Sam.

Thank you so much for joining us.

Thank you.

Thank you.

Thank you.

Thank you.

Big surprise.

Thank you.