Where AI goes from here

Note: Transcripts are automatically generated from episode audio, and are not fully corrected for spelling, grammar, and formatting.

SARAH FRIAR: Trust is earned. It doesn’t just fall down on you like manna from heaven. It is earned, so it takes time, and it’s grindy. About a third of those one point two billion people around the globe ask ChatGPT a health or wellness question every week. And if you go look in health deserts, that number spikes. It can be upwards of eighty, ninety percent of all users using it for health or wellness. We just announced this morning a hundred thousand advertisers on the platform. It’s important because those one point two billion people, ninety percent of them can get AI for free because we’re building a durable business model behind it. And free is important because our mission is AGI for the benefit of humanity, not AGI for the benefit of humanity who can pay for it.

JEFF BERMAN: Hi, folks. Jeff Berman here. Masters of Scale Summit is happening right now in San Francisco. You will hear more of the phenomenal conversations from the summit stage on Masters of Scale, Rapid Response, and Pioneers of AI in the weeks ahead. But we wanted you to hear this one right away. In today’s episode, our very own Reid Hoffman sits down with OpenAI Chief Financial Officer Sarah Friar. They discuss the frontier AI company’s role in this extraordinary moment and what is coming next. Stay tuned.

[THEME MUSIC]

FRIAR: Okay, we get a little hug.

HOFFMAN: Yes, exactly.

FRIAR: Since we’re in a hearts conference, we get a little hug.

HOFFMAN: Yes, exactly. It’s super smart and super wonderful, I think, is the right way to describe you.

FRIAR: I will give you the check in the back.

HOFFMAN: Yes, exactly.

FRIAR: Thank you.

Copy LinkWhy AI feels like a new renaissance

HOFFMAN: So let’s start with what kind of age this is, the age of AI.

FRIAR: Wow, that is a big question.

HOFFMAN: Yes.

FRIAR: Well, thank you for letting us be here. We are in an incredible moment to be alive. A lot of people talk about the age of AI as another industrial revolution, right? We laid the rail tracks. We put up electricity wires. We gave people the first taste of what it was like to get an extra hour of light in their homes. I think we’re in that early, early zone in the age of AI. But I think it’s so much more than that. Sam has used this term, a renaissance of creativity and innovation. We announced yesterday the acquisition of a company called Orchestra Bio, and what is that? It’s a platform for AI agents and scientists working together to speed up drug discovery and drug delivery out into the world.

Beyond that, if you look at our platform today at OpenAI, what we’re seeing is people going from asking, the kind of ChatGPT moment, to doing, the agentic moment, to actually becoming really proactive. Everybody in the world can now have a virtual coworker. And what does that mean for one point two billion people who use ChatGPT every week? That’s a fifth of the world’s population. They might be asking something fairly prosaic, “What do I make for dinner tonight for my picky kid?” Or literally doing something that is life-changing. That’s what we’re talking about.

And a couple more comments on what it means for OpenAI. We are seeing an incredible inflection in our business too. For example, in ads, we just announced this morning a hundred thousand advertisers on the platform, now offering in eighty-nine different countries. But why is that important? It’s important because those one point two billion people, ninety percent of them can get AI for free because we’re building a durable business model behind it. And free is important because our mission is AGI for the benefit of humanity, not AGI for the benefit of humanity who can pay for it —

HOFFMAN: Yes.

FRIAR: Or who live inside a corporation, right? Literally everyday people getting to benefit. And we’re going to talk about it, but it’s not without risk either.

These models—because I remember when you called me about DALL-E, and you were talking about a panda riding a bicycle, and I was like, “Why is Reid Hoffman excited about a panda riding a bicycle?” But we can talk about that later.

HOFFMAN: Hey, it’s great.

FRIAR: But they’re very powerful in areas like cyber. The bad guys are getting access to very powerful Chinese models. We need to make sure the good guys, the defenders, have the tools. This morning, the Journal just ran an article about PHP, where we’ve been working with them since the end of the summer, when we discovered, using actually an N minus one model, five, six, all a set of vulnerabilities in PHP, which really means the internet. About seventy percent of websites would have been impacted. It would have brought down probably a lot of the internet. So we’ve been working in a very secret way to get those vulnerabilities fixed before the offense discovers them. And I’m so happy to announce today that we’ve done that work. Is the work done? Definitely not. Is the bar going to keep getting raised? Absolutely. But yeah, the age of AI, I mean, I’m grateful—

HOFFMAN: Yeah.

FRIAR: To be alive.

Copy LinkHow OpenAI makes space for big questions

HOFFMAN: Yeah, what a time to be alive. Last year we had Brené Brown on, and part of the thing for leadership, and leadership in these times where we’re redefining the scope of how society and industry operate, what it means to be human, where we’re going—it’s important to have the cycles to think about the big questions, to make the space for that. So how do you do that? How does OpenAI do that, given that the questions here are like the invention of electricity, the invention of the printing press? How do you make that space for these important, very big thoughts?

FRIAR: Yeah, and the printing press or electricity happened at a much slower pace.

HOFFMAN: Yeah.

FRIAR: Although I bet if you lived then, you thought it was an incredible pace.

HOFFMAN: Yes, yes.

FRIAR: By the way, to make the space, we’ve used a phrase called pacing the frontier. Pacing the frontier is really about, as we think about safety and alignment, giving our researchers, our whole company, and the world to some degree, time to digest the implications. Right now, we have paced the frontier most recently by actually holding back our most powerful model, partially because we want to make sure its safety and alignment are at a place where our researchers have time to really think it through.

That model is the model that solved the Navier-Stokes millennium math problem, and also about 350 more theorems have certainly moved along from a theoretical proof perspective. What I think a lot about internally, where I try to make the space, is that theories are great, and we need to work closely with the scientific community on that. But what does it actually mean in practice?

I was a material scientist, so I actually remember Navier-Stokes coming up, only because I think Stokes was Irish, so I was very taken by that. I’m Northern Irish, in case you wonder why I’m going down that track, hence the green shoes. But what does it really mean when you discover something in fluid dynamics from a theoretical math problem? It means now we actually can think about how to make heart valves and artificial hearts more efficient, more effective. It means data center cooling actually can become more efficient, which means that when we go into communities and talk about water use, we now can talk about less water being required.

It’s about two Olympic swimming pools’ worth, by the way, for a data center, particularly one with a closed-loop cooling system that uses about 300 households’ worth of water per year. So it’s not a big amount of water, but if we can make that 100 households, suddenly it’s a much better conversation with the communities.

So to me, finding the time is about allowing that pacing of the frontier. If you saw us at work, we really try to make space for that. We just had an offsite last week at Sam’s house. It’s Friday and Saturday, and you give up all your time to sit down and have these much deeper conversations. Unlike a lot of corporate America, where offsites are about, “What’s our revenue going to look like next year or next quarter, and how are we going to think about headcount?” I would say that shrinks back a lot. That’s still my job.

HOFFMAN: Yes.

FRIAR: To make sure we have a strong financial footing. But most of the space is for these much bigger conversations. And of course, it’s about how we engage with the world writ large as we have those conversations and help people understand the pace at which we’re moving.

Copy LinkHow trust gets earned through real outcomes

HOFFMAN: Yeah. One of the key things is trust, because trust in all institutions is at an all-time low.

Right? It’s actually disastrous for society. It’s disastrous for us functioning, and obviously it makes business itself difficult. But given AI, and given that most people are like, “What’s going on?” what’s the work, what’s the effort, to try to build some understanding and trust? What are you doing? What is OpenAI doing? What could people do to help understand that?

FRIAR: Yeah. First of all, trust is earned. It doesn’t just fall down on you like manna from heaven. It is earned, so it takes time, and it’s grindy because you’re filling the bucket. Things will happen where you take stuff out of the bucket, but you’ve got to fill the bucket.

I would start by saying something like health care is actually a very strong footing for us in a world where people do trust AI and health care together. Of those 1.2 billion people around the globe, about a third ask ChatGPT a health or wellness question every week. And if you go look in health deserts, like West Texas, for example, that number spikes. It’s not a third. It can be upwards of 80, 90 percent of all users using it for health or wellness. So that’s kind of a starting point where consumers feel it.

Then we can create tools for physicians, which we’ve done, and to me that’s important because if you look at AI in the world of medicine, there’s a lot we can talk about: better, faster drug discovery. At Boston Children’s Hospital, they gave us genetic data, diagnostic data, and scientific literature, and today I think they gave us something like 300-plus not-yet-resolved diseases to try to resolve, and today we’ve gotten to 40. There’s still a lot to get, but what does that 40 mean? It means answers for families who didn’t have an answer before.

For physicians, or nurses across the whole gamut, a lot of their work is paperwork. My brother is a doctor in the UK, in the NHS. He’s an HDU doctor. What that means is if you show up in the ER, his job is to triage you right away. If you’re having a heart attack, you go to cardio. If you’ve got a gunshot wound, you go to trauma. But mostly, you don’t know, and you do not want to show up in front of my brother. Not because he’s not a great doctor, but because it means not-good things are happening for you.

HOFFMAN: Yes. You’re in trouble.

FRIAR: Do you think that man wants to do paperwork in that moment? “Let me fill out a form.” He’s like, “No, I’m in. What can I do?” When I talk to him about how much time he spends on paperwork, we can shrink that.

So if you think about how we work with enterprises, we work with some of the largest hospital chains in the world. There, we just did a proof of concept that showed HCA, which is the largest in the US, that we could lower the time a patient spends in a hospital by 25 percent. What that means is those patients get out of the hospital faster into community medicine, and they’re not getting sicker, because often if you hang around in a hospital too much, you will actually just catch the next disease.

So it’s a way to talk about the unified platform that OpenAI is building, from consumers to individuals, to a dentist’s office, a doctor’s office, to a large enterprise. And it’s important that we’re touching all three, which comes back to your question of trust, because then you start earning trust because you’re showing up, creating outcomes, and you’re not just telling the world, “Trust us, AI is great.” You’re actually showing what these magical outcomes can be.

Copy LinkHow OpenAI balances mission with risk

HOFFMAN: So one of the things that kind of throws people off, and I think it’s good for Sam to have said things like this, is that you have to accept some of the bad things, some of the risks, et cetera, because that is what the path looks like in order to get to the good things. One of your primary jobs is also navigation of risk.

How do you both, from your perch and at OpenAI, think, “OK, here’s how we navigate risk in order to get these good things”?

FRIAR: First of all, it’s deeply embedded at the company level through our mission: AGI for the benefit of humanity. If we don’t manage risk, we will not be able to achieve the mission.

That actually is a very strong alignment function for us internally, because every decision we make, from small ones to really big ones, has to come back to, “Is it mission-aligned?” In terms of managing risk, actions always speak louder than words. In the case of models that are very cyber-capable, we have, number one, held back models until we’re sure. That’s not necessarily great for business in the short run, but it’s absolutely mission-aligned, and I believe good for business in the long run.

Second is to do the work. With something like what we saw with agentic swarms, it’s about making sure we feel it’s safe to now start repacing the frontier. And then I think the third thing is going out into the world. There’s risk inside our business, but frankly, I feel like I’ve been on a personal world tour to remind everyone where we are with cyber and the need to patch those vulnerabilities, because it’s starting to happen.

Last week I was in Europe, and one of the news reports was about a supermarket chain in France where hackers had gotten into their systems and turned up the temperature on their freezers, spoiling all of the frozen goods. That is the sort of hacking that we have not really seen historically. Before, we’ve thought about ransomware and trying to break into banks to steal money. But this is going to start happening in many, many other ways.

I started my career actually covering InfoSec, so I’ve seen the original viruses, Nimda, Code Red. So we’re on a continuum of that. But again, the speed is ticking up, and I think part of earning trust is showing people, “This is where we’re going,” and then providing them frontier tools to be on the defense.

Copy LinkHow industry and governments keep pace

HOFFMAN: So one of the things is the risk is going to have to be managed not just by OpenAI, but by the industry, by governments, by banks, by folks with frozen-goods freezers, right? How should people think about their responsibilities? What is your hope for the various ways to work together, whether companies with companies, industries, or government? How do we collectively navigate these risks?

FRIAR: Yeah. First of all, at the industry level, it’s just getting started, right? We are seeing some really major shifts happen, and we have a phenomenal economic research team. Some of the data they put out in September showed that if you look at our platform, and we get to see a broad swath of what companies are doing today, the frontier firms that we define as the top 10 percent of companies in the world, the number of agentic tokens they’re using today versus even just an average firm is about 8x. And that number’s gone from about 2x back in January.

So the pace of companies using technology has really started to accelerate for the frontier. And if you’re average, or God forbid below average, you cannot get left behind. I’ll give you another stat. At OpenAI, internally, I often say we have this benefit of living in the future. Let’s not waste it.

So if you were to look at our agentic token use, it’s 33x the frontier. My belief is that a year from now, two years from now, the frontier firms will have caught up to us. The traditional firms will have caught up to the frontier. So that growth is happening, that transformation is happening. Of course, what does that really mean inside companies? That means you’re probably starting with things like coding tools. You’ve now started to deploy cyber. You’re starting to deploy knowledge work, like Dots, the product we just rolled out, so the virtual assistant.

Mine is called Una. She has a fine Irish name, because Una was the wife of Fionn mac Cumhaill. If you know any Northern Irish mythology, he was the giant, but his wife was the brains of the operation. She was Una. She is my brains of the operation.

HOFFMAN: Yes.

FRIAR: Female power.

HOFFMAN: Yes.

FRIAR: But they’re starting to roll out that knowledge work where people literally have that virtual coworker, and then they’re starting to transform based on their vertical, like a life sciences company, a chip design company, a bank, or by horizontal. What I’m seeing in the office of the CFO is, “I’m trying to do finance like no one’s ever done it before. We are doing finance like no one’s ever done it.” So that’s kind of one way: How do we talk to companies and make it real? By the way, we just announced this morning that we’re live on GCP, on Google, so we now can meet our customers on every cloud that exists — all the big ones: Amazon, Azure, GCP, Oracle. That’s going to be a huge tailwind into 2027.

On the government front, it’s also about going to meet governments where they are. The reason I was in Dublin is that Ireland right now has the presidency of the EU. I wanted to spend time with the Taoiseach because I’m probably one of the few people who can say his title correctly.

HOFFMAN: Yes, but do tell them what it is, because while I know what Taoiseach is, I’m not —

FRIAR: The Taoiseach is the prime minister, effectively —

HOFFMAN: Yes.

FRIAR: Or president, whatever your title, but Taoiseach. An Taoiseach. We really want to make sure continents like Europe don’t get left behind. In that top 10 percent of frontier firms that we see, only 10 percent of them are European-based. So what’s that going to mean for GDP growth in Europe? You do not want to get left behind again like you did, to some degree, in the internet revolution.

So it’s both bringing governments along, but also recognizing that we are talking about having regulation that makes sense for frontier labs. Today we work with organizations like CASI here in the US, AISI in the UK, but you have to find people who can actually keep pace with you. Those two organizations do. I think those are some of the main ways: inside organizations, meeting them where they are, solving their actual problems, not some sort of made-up “Here’s what you could do.” And then, with governments, showing up and making sure we’re part of the conversation.

Copy LinkWhy AI can expand human agency

HOFFMAN: So our theme for Masters of Scale Summit this year is our agency. It’s not just because I wrote this book, Superagency, which I know you have a copy of.

FRIAR: I do.

HOFFMAN: Because I sent you one.

FRIAR: No, no. I bought it off the shelf.

HOFFMAN: Yes.

FRIAR: Read it five times.

HOFFMAN: Good. We’ve actually talked about it, so it wasn’t meant to be quite so self-deprecating, but it is entertaining. One of the reasons I did this is because it’s really important, and part of the reason why we’re doing the summit this way, is to realize that agency is, to some degree, a mindset.

People worry about AI because they worry about it taking their agency from them. It takes my job, it takes my cognition. It’s like, no, that’s a question of mindset — how you approach it, how you engage it. So how do you think about how the world at large, not just the builders of these amazing AI systems, should think about their agency? And how should they embody their agency with AI?

FRIAR: First of all, I find extremist conversations so frustrating. There’s so much nuance and so many shades between white and whatever color we’re going over here. The reality is, if we go to the world of jobs to start with, jobs are changing, but work is changing inside of jobs.

Again, what we see in our research is that if you look at someone’s job title and then how they use ChatGPT, they only use it for about 40 percent of what their job title says. Let me make that a little more real. As a finance person, if you show up with a finance job title today on our platform, you do about 25 percent of your tasks on ChatGPT that are truly financial. Like, “Run me a DCF” — or, God, now I’ve started to talk in finance-speak. It’s scary.

HOFFMAN: Yes.

FRIAR: OK, do a reconciliation of treasury or whatever. Twenty-five percent are engineering tasks. What does that mean? It means everyone has now become a builder. I am set free to now build sites, to build whether it’s for fun stuff or professional stuff, in a way that I’ve done more coding in the last two months than way back when I actually was an engineer.

HOFFMAN: Yeah.

FRIAR: And now the coding language for me is English. I just talk to my computer. I don’t even tippy-tap anymore. So 25 percent is engineering. Twenty-five percent is marketing, which gladdens my heart, because I tell finance people all the time, “If you don’t realize you’re in sales, you are losing, because you need to go sell. You need to sell an annual plan. You need to go sell a change to a number.”

HOFFMAN: Budget.

FRIAR: You need to sell a budget. You really need to sell a budget. Particularly at OpenAI, keeping people to a budget is a sales motion in and of itself. So people’s jobs are really changing.

Again, sorry, because Dublin is very fresh in my mind. I did an event on Monday night. You know I have a nonprofit on the side that I do with an amazing co-founder called Ladies Who Launch.

HOFFMAN: Yes.

FRIAR: So it was about 250 to 300 women in the room, and just this excitement onstage, them talking about how empowered they felt, how much agency —

HOFFMAN: Yes.

FRIAR: They had gotten back, and they were talking about, “Oh, my goodness, when I did my first startup, I used to be awake till the wee small hours creating the pitch deck because I had my two kids. I had to get them to bed, and then I would start the work. Then I would walk into these meetings to fundraise, and people would blind me with DCF-type terms, and I didn’t know what any of that meant. Now I just sit there, and I ask Chat, and I feel so empowered.” I was talking about it earlier in the green room, too. One of my best examples was Linda.ai. This is a company built by two sisters in Dublin. Their dad is a dentist. He had a receptionist, which is their word, called Linda. Their whole lives, it was, “How do we clone Linda?” because Linda was so good.

They have cloned Linda in Linda.ai. Linda.ai worked with Linda to call the six hundred people he had in his contacts to say, “Hey, you should come in and get your teeth cleaned or looked at, or whatever.” Three days later, because Linda.ai doesn’t stop, Linda.ai had two hundred and fifty new scheduled patients coming in. So it’s great for the dentist, it’s great for the people, and it’s great for teeth. And that, to me, is agency coming back into the world in really optimistic ways. I am super optimistic for the builders of the world —

HOFFMAN: Yeah.

FRIAR: Because they are empowered like never before.

Copy LinkHow local trust shapes AI infrastructure

HOFFMAN: A hundred percent. We only have a couple minutes left. By the way, thank you for the gymnastics I know you did on your calendar in order to get here and be with us.

FRIAR: For Reid, it’s always worth it.

HOFFMAN: No, but I actually think it’s worth sharing: Where are you going next, and why?

FRIAR: Okay, so I’m running out the door. For Reid, I will make everything work. But today I’m supposed to be in Savannah, Georgia, which I’m super excited to go to. We have a new powered land site that’s going to become a large data center, and I want to go spend some time with the community. Because data centers can clearly be a contentious conversation for communities, they can also be really good. They bring jobs, like really high-skill jobs. Like Abilene, Texas, in West Texas, that I was just talking about, they bring electricians, HVAC. We do a lot in the community to actually give people those skill sets, and so we find a lot of the unions are actually quite excited by what’s coming.

In Georgia, what we’ve done is given the community an eighty million dollar grant that they can deploy over five years, but they get to decide what it means. What do they want to do? The very first ask that came to me was, “We would like to pay for free school dinners or free school lunches.” And I was like, “Absolutely. How much is it?” It’s like four point five million dollars. I’m like, “That is an easy decision.” And since we did that, the uptake on this free lunch has gone up fifty percent. It’s not to give us a huge amount of kudos. I’m actually saying it because I think this is where you earn trust. You go, you listen, you hear concerns, you talk about how you’re mitigating, you bring communities along, but most importantly, you show up and you don’t tell them what they need to do, you ask them. And I think this is the important part about reindustrializing America right now.

In totality, AI is probably adding about a point to GDP, depending on how you cut it. That is growth. Growth begets better outcomes, I think, for the nation writ large, and this is why I want to see growth not just here, but I’d like to see it back in my homeland. I’d like to see it back in my home continent.

HOFFMAN: Definitely.

FRIAR: I would like to see it around the world. And I think, like we said at the beginning of this talk, there’s never been a better time to be alive.

HOFFMAN: Yeah.

FRIAR: We need to figure out how to embrace it while managing the risks, but how cool—

HOFFMAN: Yeah.

FRIAR: That we get to do this.

HOFFMAN: So you can see why I wanted to open with Sarah. Let’s give her a hand.

FRIAR: Thank you. Thank you.