Greg Brockman on Building OpenAI, the Board Crisis, and the Race to AGI

Greg Brockman

It's true, but I felt like Stripe, the problem we were solving was not my problem. It wasn't the problem I'd grown up thinking about. It was an important problem and I dedicated myself to that mission for a number of years, but I felt like it was going to succeed with or without me. So then I had a first moment to really think about what is a mission I want to dedicate myself to, where I would spend the rest of my life working on this problem just to see it play out in a slightly better way. And it was very clear to me that top of the list was AI. If you could actually make a difference in how AI will play out in the world, that would be a life well lived.

Greg Brockman

Patrick had said Sam has seen lots of young people in your situation, and Patrick really hoped that Sam would convince me to stay. A few minutes of talking to Sam, he's like, 'Okay, you clearly have already decided. It is very obvious.' And so he asked, 'Well, what are you planning on doing next?' I said, 'I'm thinking about doing an AI company.' And he said, 'I'm also thinking about doing something in AI. We should keep in touch.' I talked to Sam maybe one more time after I was leaving Stripe, and he asked, 'Are you still thinking about doing something in AI?' I said yes. He said, 'I'm also starting to get more details and putting together this dinner in July.' I flew out for the dinner. And the topic I remember was: is it too late to start a lab with many of the best researchers? Is it possible?

Greg Brockman

2015. You think about the degree to which DeepMind had all the researchers, all the capital, all the data — it just felt like, is it even possible to get something off the ground still? People came up with all sorts of reasons it was hard. No one could come up with a reason it was actually impossible. Sam and I were driving back to the city that night, and I remember we looked at each other and said, we've got to do this. We just have to. The next day I was full-time on putting this together. It was tough because it was very ill-defined. We had a mission, a vision — we think we can build human-level AI, make it be something positive for the world, make the benefits be distributed broadly. But how? And how do you get people to actually leave their jobs to come and join this thing?

Greg Brockman

Initially, the set of people I narrowed down to were Ilya, Dario Amodei, Chris Olah, and myself. That was going to be the team. We spent a lot of time together talking about potential visions for the lab, potential ways things would work. It didn't quite come together — partly a question of, will this have enough momentum? Dario felt he needed to go and establish a name for himself, and he wasn't sure if this was really going to be it. Meanwhile, I was starting to get John Schulman interested. Dario and Chris ended up deciding to go to Google Brain. So it was really just Ilya, me, and John starting to be, maybe, a few others.

Greg Brockman

I had a group of about 10 people, many of them saying, 'I'm interested, but who else is in?' I asked Sam, how do we break symmetry here? How do we actually get everyone to say, all right, we're joining? Sam's suggestion was: invite people out for an offsite. So we set up a thing in Napa, and I actually made t-shirts. This is before they had joined — there were no official offers, no one had joined, we didn't have a structure, we had nothing. We just had an idea, a vision, a mission. We flew people out, we drove up to Napa together, and it was an amazing day. The ideas were flowing. We came up with what I would really say is almost the technical plan we've pursued for the past 10 years. Number one, solve reinforcement learning. Number two, solve unsupervised learning. Number three, gradually learn more complicated, in quotes, things. After that offsite, I sent offers to everyone and said, 'We want to get started in the next two to three weeks. Please let me know if you're in.'

Greg Brockman

Google DeepMind was the 10,000-pound gorilla in the field. They just had lots of capital, the track record. This is before AlphaGo — AlphaGo came out a couple months later, but it wasn't a surprise. The momentum was very clearly there. So the question of whether it was really possible to build something independent and new, it wasn't obvious.

Greg Brockman

In 2017 we started to think very hard about, first of all, how do we really achieve the mission? How do we actually build an AGI? What will that look like? We started to do the math on compute, and you realise it's going to take a big computer. We came across a company called Cerebras, which was building a unique piece of computing hardware, and the kind of computer they were promising, we realised was going to be far advanced of where our compute calculations looked. You start to realise that if we could buy a lot of those computers, we could probably succeed at building an AGI. If we could get exclusive access to Cerebras, that could give us an overwhelming advantage.

Greg Brockman

The thing about nonprofit fundraising is that there is essentially a cap to what is possible there. And so Elon, Sam, Ilya, and I all agreed that the only path forward for OpenAI, the only path to achieve the mission, was to create a for-profit entity associated with OpenAI of some form. We were committed to that direction, and that is something we knew was the only way to achieve the mission.

Greg Brockman

The way OpenAI works is it's a series of moments where you realise it's real now. And every time you think you understand it, that it's really settled in for you, you realise there's a new horizon you had not yet appreciated. There was the initial launch — 'Wow, we actually got a team together, now we can pursue this mission.' But you show up at the office the next day and you're like, 'Well, what do we do?' We didn't even have a whiteboard. Ilya and John wanted to write something on a whiteboard, and I was like, 'I will get a whiteboard. That's something I can do.'

Greg Brockman

DOTA, we had our first big result. That really was, 'Wow, we can actually accomplish something when we put our mind to it.' You can see all this compute coming together — you scale up the compute, you scale up the result. There were multiple moments with the GPT series, and an early one was the unsupervised sentiment neuron paper. That one's interesting because it's 2017, and it's really the first time we saw semantics arise from training on a language-modelling objective. You train on predict the next character, and then suddenly you get a neural net that understands sentiment, understands if something is positive or negative. That was a moment where you realise we are building machines that can learn semantics — not just where the commas are and where the nouns and verbs are, but the meaning of sentences.

Greg Brockman

And then I remember we were playing with GPT and someone asked, 'Why is this thing not an AGI?' It's actually really hard to put your finger on it, because you can talk to it fluently in anything you want. It clearly wasn't an AGI. It was lacking something. But if you'd described your criteria for AGI two months prior, it probably wouldn't be compatible with what GPT-4 was. There are many moments along the way where you feel like it's real now, it's going to really happen — the economy is going to transform into this compute-powered world. And I think those moments are not yet at the end. We have many more breakthrough moments ahead where you realise the next stage is possible.

Greg Brockman

That was something very compelling about it. And the ironic thing is we'd actually set out with DOTA to develop new methods, because the reinforcement learning at the time was clearly not going to scale. The algorithm we use is called PPO — you plan over every single time step, there's no hierarchy. Whereas a human, that's not how you plan your day. We knew this algorithm was incredibly flawed, would never scale, had all these problems. But you've got to start somewhere. You've got to push your baselines to reach the wall, so you actually see the limits of what good looks like with what you have, and then you can bring to bear a new algorithm. We just kept scaling PPO and we exceeded the performance of the best humans. And that itself was the finding: massive compute with simple algorithms — that works not just in theory but in practice, in an incredibly messy environment where you cannot program it, you cannot look ahead, you cannot do a search. You just need this almost human-like intuition. And by the way, the neuron we used, tiny little insect brain, similar number of synapses as a truly an insect brain. And you realise, wait, what if you had the same computational approach but scaled it up to something much more human-brain scale? What would that be like? Very evocative question.

Greg Brockman

I think they're connected in a deep way. On the one hand, just predicting what comes next sounds like a pedestrian task. But if you can really predict the next word out of Einstein's mouth, you are at least as smart as Einstein. You can make arguments — 'oh, well' — but I think those arguments fall flat. There's something false there, because the point of prediction is not being able to predict what is known. The point is you put yourself in a new situation you've never seen before and predict what comes next. There's something deeply connected between intelligence and prediction — there's a long story of academic literature about how you think about this, compression, they're all kind of part of the same thing.

Greg Brockman

Now, these reasoning models — the thing I think is very interesting is that we train them with reinforcement learning. Back to the original OpenAI plan, there are two steps. The first is unsupervised learning: you train a model just by having it predict what comes next, and there it's much more static, observational data. It's data it's never seen before, situations never seen before, but it's a situation that has already happened. Then you do reinforcement learning, which is you basically have the AI learn its own data. It makes its own action — 'here's the action I'm going to take' — you get an observation from the world, and you learn from that. The way you actually train it is still predicting: it's trying to predict, if I take this action, what's the thing that's likely to happen, and you reinforce that depending on how good a job you did. The beauty of that is it's now an AI that has this background knowledge and has real-world experience. But fundamentally, the technology we use to train during the unsupervised stage and during the reinforcing stage is exactly the same. You are just predicting, but you've changed the structure of the data.

Greg Brockman

The thing about OpenAI is that if you truly believe in the mission, if you truly believe in the possibility of creating machines that have the intelligence level of humans, it means the stakes always feel very high. The question of who's making the decision, what are the values that go into those decisions — these things that are maybe mundane in a typical company, that are much more like office politics, start to take on this existential weight. That has coloured a lot of OpenAI's more high-profile conflicts. Even just the question of who gets credit for a particular thing suddenly takes on this existential weight.

Greg Brockman

That is the overwhelming dynamic I have observed in this field. It's not just about OpenAI, actually. One observation I had early on is that this technology is by nature very fragmentary. When you have a lot of pressure, you can get a diamond or you can get cracks. Often you'll see diamonds form in pockets — teams of people that really work together, that have a lot of high trust, that know how to operate. But sometimes you see them splinter off and go their own way. Within AI, we've gotten some real benefits out of diversity of approach, different groups really pushing each other, in order to bring this technology in a more beneficial way — how to think about all the thorny questions around safety, what it means to be safe, what it means to actually deploy this technology, how to mitigate but also how to maximise those benefits. There's a lot of very healthy debate. It's always gone on within OpenAI's walls, and now it's starting to happen in the world, and that's something we as a society really benefit from.

Greg Brockman

I was at home. I got a text saying, 'Can we hop on a video call?' So I hopped on the video call, and I noticed it was the board minus Sam who were on there.

Greg Brockman

No. I inferred something was up.

Greg Brockman

I was on the board at that point, yes. I was told the board had decided that Sam would be removed. Effectively, the message I got was the same messaging that was in the public post. I asked if I could have any more information. I was told no, not right now — maybe another time. Again I was told nothing more to share. And then I was told, 'Wait, there's more.' I had also been removed from the board, but would be staying with the company because I was very critical to the company and the mission. I asked, again, any reasons, any feedback? Told no. Towards the end I was told, 'Hey, in this new setup, hopefully you can get feedback in this new configuration.' And that was the conversation.

Greg Brockman

It just wasn't right.

Greg Brockman

No. I felt like I understood what had happened. To some extent it comes down to a lack of communication — you realise there are all these different things that have buffered. Approximately, I kind of knew. For every person there, I have a pretty good model of why they acted the way they did. But it wasn't what was most important to me in the moment. I just knew that this wasn't right. Right after I hung up the call, I talked to my wife and said, 'Got to quit.' And she said, 'I agree.'

Greg Brockman

That day, I started to get all these messages of people saying, 'I don't know what you and Sam are doing next, but I'm with you. I want to go start something with you.' That was a real, honest surprise. I didn't really expect that kind of support, that kind of outpouring. There were a few of my close collaborators who quit that day as well — Jakub, Szymon, Aleksander — and the five of us, those people plus Sam, we all got together and started to chart out what a new company could look like. I remember feeling that first day, 'Okay, there's a 10% chance that we actually get the company back.'

Greg Brockman

The next day we set up a meeting at Sam's house. A bunch of people from the company came by and we showed the vision we'd been sketching out. We spent time over that weekend also negotiating with the board and the company, trying to figure out if there's a path back together that makes sense. That Sunday night, the board replaced Mira as interim CEO with a new person, and the company just rebelled. We'd actually been at the office, we thought we were close to a deal to come back. Then the board made that change, and suddenly it was everyone streaming out of the building — real chaos. I was on video calls with many of the people who'd been interested in coming to this new company, saying it's going to be okay, we have a plan. We'd been building this little life raft for the small set of people we expected to want to come, and suddenly no one wanted to be associated with that entity. People wanted to stand up for what they viewed as right.

Greg Brockman

Sam talked to Satya, who we'd been talking to about being a funder to support this new endeavour. I was like, 'Actually, could we expand from the small life raft? Could we take everyone?' And they said, 'All right, we'll figure it out somehow.' This is right before Thanksgiving — a lot of people were supposed to be flying home, and instead they cancelled their flights, and the office was packed. Everyone was at the office just to be there, to be part of it, even if they couldn't contribute to any of these conversations. They just wanted to be there as this history was made. Then this petition starts to circulate. So many people were trying to sign it at once, it actually crashed Google Docs — you had to have certain people designated as the person you go to, to put your name on the document, so you didn't have too many editors at once. That was a statement that was really heard loudly. I probably got home around 5am, went to sleep, woke up 45 minutes later, checked Twitter, and I saw that Ilya had posted and had signed the petition, saying he wanted the company to come back together. That was a real moment of relief. I felt so much gratitude — it just felt like, okay, we can put this back together. We can get back to a good track.

Greg Brockman

It was tough. That was definitely a very close relationship — you'd been the officiant at my civil ceremony. We'd been through extremely tough times together, and like any relationship, you always have your ups and downs. We spent a lot of time afterwards really talking things through, trying to understand and articulate some of the things we had let build up or left unsaid between us. Through that process I think we got to a really good place, and for me it felt like we got to closure on everything that had happened.

Greg Brockman

Deeply grateful. It was never something I would have asked for, something I never would have expected. The way I operate, I'm very much an in-the-trenches kind of leader — try to lead from the front. And sometimes when I do that — sorry, I'm getting a little emotional — I don't always look back to see if everyone's following. I just run right in. And when people do come and really help to build the thing, it makes me feel so grateful for them, to feel like they have exceeded my expectations in every way.

Greg Brockman

It was not guaranteed, because throughout that weekend all the competitors were circling. Just imagine this feeding frenzy — people were getting offers. And we actually did not lose a single person through that weekend. No one accepted a competing offer.

Greg Brockman

Yeah, that was a diamond moment.

Greg Brockman

That was an intense experience to go through, an intense experience to come back to. Honestly, one of the hardest moments for me at OpenAI was when Ilya left. It was maybe the only moment in OpenAI's history where I felt like I didn't want to do it anymore. I think I needed some time to find my way back to remembering why I was doing this, and why it was so important, and why it was worth the pain.

Greg Brockman

I trained language models on DNA sequences — for the Arc Institute. It was a very great experience. I took the skills I had and applied them in a very different domain, one that's very personally meaningful to both me and my wife. She has a lot of health conditions, and we think about what AI can do for her health, what it can do for the health of animals. It felt like maybe we could help in this very different way from how I've been pursuing this technology. That was a very positive part of the experience.

Greg Brockman

I think I've learned to just keep going for something that's worth it. If you have a mission that matters, the fact of you keep going through the ups and the downs — there are going to be moments where it's all over, moments where it's all back — and you just can't let those moments pull you off course. The degree of personal resilience you have to grow during these times, because if you're leading, people look to you for that steadiness, for that direction. A lot of what I've tried to grow is being able to both understand the details of what we're doing, and also be decisive. There have been moments where I approached OpenAI through a lens of uncertainty, feeling like I don't know what the right answer is, but there are lots of very smart people here who have very strong opinions. So I've tried to understand all those opinions and figure out how to put them together. Sometimes that's the right thing. Sometimes you realise the opinions are mutually contradictory — they can't all be true at once — and you do just have to pick. That means there's going to be someone who's upset, someone who quits, someone who feels slighted. A lot of what I've tried to do is have a stronger sense of self, and a stronger sense of when there's conviction that we need to act on it.

Greg Brockman

When I reflect on OpenAI, Stripe, even rewinding to college, the way I tend to operate is I both really love the day-to-day activity — the individual contribution, the software, thinking through the problem — but I also really care about the environment in which these things are done. I'm willing to give up on type-one fun, the quick hit of getting to build the thing, for something more like type-two fun: it's painful in the moment, but it's worthwhile. What you do is create an environment where everyone else can do the IC work, the great thing. It's not always the easiest — you really do have to be willing to take on great personal pain. In the words of Ilya, you have to suffer. If you're not suffering, you're not building value. And I think there's deep truth to it.

Greg Brockman

Ilya has a particular way of talking that's very unique to him — there's always deep inspiration in the words he chooses. This picture of suffering was something we thought about throughout OpenAI. We had so much uncertainty from the beginning. Is this thing going to work? There are many reasons why it might not, why it should not, why you could even say it cannot — how do you get the people, how do you pursue the technology, how do you get enough capital, how do you keep people motivated, how do you make the right decisions? Each of these is extremely hard, extremely uncertain. It's easy to just sweep the problems under the rug and blindly say go. That is the negative side of Silicon Valley culture — you just blindly do the thing, you do a reality distortion. But I don't think that works in AI, and I don't think that's how we've operated ever. The way we've always operated is to encounter the hard truth, to see the reality as it is. That's contributed to the successes we've had — thinking about the problems differently. Even in the early days we thought, okay, if we just write some papers and publish them, we'll get citations, we'll be the coolest people at these conferences — but will we achieve the mission? How is it that you do that activity and then AGI goes better for the world? They're not connected. Maybe it's a foundation, maybe a step, but it's not sufficient. And then you start thinking about the bigger questions: what would it take to build an AGI? It's not pleasant, because you realise there's no path. You realise you need dollars, and you don't have any mechanisms to raise them. You could raise a hundred million, maybe five hundred million, but a billion? Pretty hard. And you look at what OpenAI has accomplished with the resources we've raised — there truly would be no other way to do it besides having leaned into the suffering and trying to understand the truth of what we're trying to accomplish.

Greg Brockman

Make the hard decision. Have the hard conversation.

Greg Brockman

From my Harvard freshman writing class: keep cutting words, in order to be clear and communicate well.

Greg Brockman

I read a lot. Triage aggressively.

Greg Brockman

Gauss and Descartes — people who were incredibly thoughtful, very much ahead of their time, visionaries who came up with real breakthroughs that transformed how we think and how we live.

Greg Brockman

That it's going to be a force for good in their personal life, that they'll benefit from it, and it will help advance science, medicine, and really lift up everyone.

Greg Brockman

People don't understand how focused I am on this mission, in a way that's been very personally painful at many turns. I just believe this technology can help empower people and benefit everyone, and I really want to help make that happen.

Greg Brockman

That one I can't tell you.

Greg Brockman

We are in this phase where you apply AI to its own development process, and it's going to get faster and faster. That's been happening certainly since ChatGPT — we use ChatGPT to make our development process 10%, 20% faster. Now we have these amazing coding tools which have truly revolutionised how software engineering is done. Most of what we do in the production of models is bottlenecked by software — implementing these systems, scaling them up, managing these massive computers. We're going to be hitting a phase soon where the AI also comes up with its own research ideas and tests those out, runs experiments. The speed of iteration and innovation is going to continue to increase as a result of what we're producing.

Greg Brockman

There's certainly a lot of attempts to distil models, and that comes from companies in the US and from all over the world. But it misses the core point, which is that the way this technology is developing, it is on an exponential. Any time we have a model, we've already moved on to the next one. We're already moving to the next level. We put in a lot of effort to protect against distillation, make it harder — especially with things like chain of thought and other parts of the model that are not really necessary to get the outputs to someone. But the core advantage we have, the strength we're building up over time, is really about not just any one model. It's about the machine that makes the models.

Greg Brockman

That is part of it. There are two reasons. One is distillation. The second, in some ways more important, is that we had this insight when we first developed the reasoning paradigm — it gives us an interpretability mechanism we had not been anticipating, because you can really read the model's thoughts. You can see exactly how it got to an answer, interpret what was actually motivating it. Now, the problem is, if you train the model to have a chain of thought that looks good, then you lose all the faithfulness — the model knows that part of the desired answer is for the chain of thought to look a certain way, so it may not be representative of how it actually arrived at that answer anymore. We made an early decision to avoid any temptation to train these chains of thought to look favourable, to look like something you could present to a user. That made us lean out, for competitive and for safety reasons, from showing these intermediate thoughts.

Greg Brockman

It's hard to know what percentage of the code is not written by AI. It's a vanishing fraction. The actual writing of code, currently the AI is much better than humans — given the right context, the right structure. There are parts of the structure of the code that our human experts are still much better at — thinking about how the modules should be laid out, how the pieces should work, the definition of certain kinds of interfaces. But the actual writing of code is essentially all AI now.

Greg Brockman

We're getting close. In chip design — in the design of our own chip last year — we applied our technology to shrinking the area used by the circuits. The optimisations the model produced were actually on our list. It didn't come up with something novel that no human ever would have, but it implemented it faster in a way we wouldn't have had time to accomplish. If you look at math and physics, we now are solving open math problems, open physics problems. We've resolved a particular physics problem recently in quantum physics in the opposite way the community expected, with a beautiful, elegant formula. New ideas from these models are extremely doable — we're starting to see it in some of these domains. Applying it in harder domains, ones that require more real-world context, we have line of sight for how to accomplish it, but we have a lot of work to do.

Greg Brockman

We put a lot of effort into neutrality for our models, to have them represent truth. You can see exactly the values that go into our models on our website — we have a publicly published spec which defines, and you can give feedback on, the different ways we want our model to behave. We've spent a lot of effort to get to this neutral point of view. When you see these screenshots on Twitter, they're not always fully honest about where they came from — sometimes there are memories behind the scenes that tweak the answer, or hidden instructions, or previous parts of the conversation. And sometimes there's just no right answer. You can have a question that says 'answer in one word', and no matter which one you say, you're going to get some claim of bias. The core of it, in my mind, is that we care a lot about truth and about having an AI that really represents you.

Greg Brockman

We've gone through an evolution of how we train the models to user preferences. At one point, last year, the models really did start to lean into telling you what you wanted to hear — 'oh, that's such a great answer.' We reacted to that. We said this is not how we want our models to operate, and we made changes. Because the true thing we want the models aligned to is helping you solve your goals, your long-term goals. In the moment it feels good to be told, 'that was a great question, best question anyone's ever asked' — but that's not what you actually want. We've made great technological improvements to make sure our AI training does not result in what's called hacking the grader. We really want a good signal that's about the goal, not just your short-term quick hit. That, to me, is maybe the most important part of the vision for where our personal AGI is going to take us — alignment with your long-term well-being, your long-term goals. That's what will most empower people, put you in the driver's seat, because you'll have this entity operating on your behalf 24/7. You're asleep, and it's out there trying to figure out, what is it that Shane wants? How can I do it better? And is actually able to accomplish it.

Greg Brockman

We're certainly in a global AI renaissance, and the dynamics between countries are not yet fully defined. We have this concentration of where the breakthrough algorithms come from — the US, Western companies. There's clearly a lot of innovation happening around the world, but the balance of dynamics, which countries rely on which providers, all of that is still being determined.

Greg Brockman

I do think leading in AI is very critical for America, because that's how you can ensure democratic values are protected and preserved. Every country is starting to realise they need some sort of sovereign AI strategy. If this is becoming the basis of economic security, of national security, they need to participate somehow. If you look at the US efforts around chip exports, technology exports — if you lean too far out, then everyone else has to develop their own competitor or rely on someone else building this. If you lean too far in, maybe you lose your advantage. The question is how you balance those, how you maintain leadership. But leadership is not just about being ahead. Leadership is about also bringing along the world with you.

Greg Brockman

We in general are heading to a compute-constrained world. If you think about the amount of value these models can produce, it's extreme — it's not just answering a quick question anymore. It's going deep, spending a lot of tokens to put together a bunch of different data sources, searching through your enterprise knowledge base to solve a hard problem, writing software that's better than a human would. If you look at the progress between GPT-5 to 5.1 to 5.2 to 5.3 Codex to 5.4, it's been extreme — these models are getting extremely better at understanding your intent, moulding to what you want to accomplish. And we put them in surfaces like Codex that make them very usable, so that you as a developer can really fly. All of this is powered by compute fundamentally, and there's not enough. If you wanted one GPU for every person in the world, you're talking about 8 billion GPUs. We're not on a trajectory to build anywhere near that level — it's hundreds of thousands of GPUs, millions coming up. It's not surprising that there's too little compute in the world and we're going to need much more to bring this technology to everyone.

Greg Brockman

It's going to give us an advantage — not just for the business, but for actually delivering on the mission of bringing this technology to everyone.

Greg Brockman

Who's laughing now? I think our competitors are not having a good time on compute. Let me put it that way.

Greg Brockman

That is the core of OpenAI — really encountering reality as it is. Really thinking about the implication of what we'll accomplish in the next six months, the next 12 months, the next 10 years. That's true for the grand mission, it's true for day-to-day how we design pieces of our software, and it's true for scaling up compute. We are deeply motivated by bringing this technology to everyone, and we think about lots of different mechanisms for how to do that well and safely.

Greg Brockman

Yes. This kind of thing happening this year is not out of the question. Have you been to any of these data centres? It is a very different experience to walk amongst these racks, to walk down the rows, and you look at the cables that are all perfectly exactly the right length. You realise that what a data centre is, is a massive machine. These are maybe the biggest machines humanity creates. And then you ask, why do we build these machines? It is because they have the potential to solve problems that matter for people — cures for cancer, helping people run businesses, sometimes maybe mundane queries. The purpose, in my mind, is really about how you deliver value, how you deliver on people's goals. The opportunity presented by these massive machines targeting one problem is something we have not yet really internalised.

Greg Brockman

This is going to be the most important question for society to answer. Where does the compute go? What problems are worthy? There are lots of worthy problems, but you need to prioritise them, because you only have so much compute. One thing we really believe is that everyone is going to need access to compute — that's why we have a free tier of ChatGPT. We've put effort into making sure people are able to use this technology, because we believe that is core to what we're doing. Putting this technology in people's hands empowers them, lets them achieve goals, helps them understand the technology and shape how it slots in. You could take a very different approach and say it's all about the ivory tower — just solve the problem and then distribute the breakthroughs somehow. There's merit to that, but that's not where I'd put the balance of what we do. We do want to make great strides on specific problems, but that should be in service of wanting the benefits of this technology to be broadly distributed.

Greg Brockman

A lot of what I've been thinking about recently is focus. This field is opportunity incarnate — you can take AI and apply it to any problem, anything you want to build is now on the table. The problem is there's only so much compute. Where do you want to put it? So you need synergies — multiple things happening that all add up. One plus one equals ten. That's the dream. For this next phase of OpenAI, very clearly enterprise, because the economy is becoming this compute-powered economy before our very eyes. We've seen it with software engineering, and it's going to happen with every single field of work people do with a computer. Rather than you doing work with your computer, your computer is going to do work for you. We need to be there to help people deploy these models, figure out how to get the most benefit.

Greg Brockman

There's also going to be a blurring of the line between enterprise and consumer, because entrepreneurship is going to become far easier than ever. One of my friends was describing that his sister wished someone had created this app — she had a picture of exactly what she wanted. And he, in the meanwhile, was typing into Codex, and a few hours later he shows her the app. She's like, 'Where did this thing come from? Who built this?' And he said, 'You did.' Anyone can be a builder. Codex is for everyone — it's not just for software engineers. Everyone now can be a software engineer if they have a vision, this agency, a thing they want to accomplish. On the consumer side, 'consumers' is too broad a term — there's entertainment, self-expression, and there's also solving goals. The aspect we're really dialled in on is solving goals. About 4 billion people use smartphones, and all those people should have a personal AI, a personal AGI, that knows them well, has their personal context, is trustworthy, that they can ask for advice. If your favourite musician is in town, it just goes and proactively purchases tickets — and maybe it knows, oh, I should check in before doing this, or maybe I have prior approval. You should still set your goals — you should be in charge — but that's something we want to create, and I think the whole planet, 8 billion people, is going to really benefit from and need access to a personal AGI.

Greg Brockman

I think we're going to have data centres everywhere. Data centres in space have many technical problems associated with them. Even the data centres we build today are very finicky — massive machines with very breakable, very expensive components. We've had many issues where the cables were just too taut — literally too tight — and then you get signal integrity issues and the computer doesn't work. Figuring out how to maintain systems today: people go and physically do it, and it'll probably move to robotics. Space feels like a grand challenge, but we're going to have such need for compute that we need to be thinking about all options.

Greg Brockman

Iterative deployment is one of the core pillars of how OpenAI has approached getting this technology to benefit people. I think I was probably the person who articulated those two words. It emerged as we thought about our first product deployment. There were two different routes. One is you build it in secret — you don't deploy anything, you have a lot of time to polish it, get it right, and then at some point you push a button and say deploy. I remember thinking, could I sign up for that strategy? Could I be accountable for it? Do you want to be sitting in a room thinking, okay, we ran all our tests, are we ready to deploy — and you've never deployed anything ever before? That's your first contact with reality, with a very powerful system that's going to really change the world. That is a very tough problem set. Instead, what about an approach where this is your hundredth system — you've had to solve this problem 99 times before, with systems of increasing power, and the world has also had a chance to adapt and reconfigure around them?

Greg Brockman

We learned very early with GPT-3 what it's like to deploy something. We spent a lot of time thinking about all the misuses of GPT-3, the ways it could go wrong — misinformation, these grand pictures. And do you know what the number one misuse of GPT-3 was? It was medical spam — advertising different drugs to people. Not something we ever would have thought of as a problem, but we saw it in front of our eyes and got a chance to react and learn from it. Iterative deployment is the idea that we bring intermediate versions of this technology. It's not an excuse to blindly deploy — you still need to think at every step about all the ways this might be misused, the downsides, the risks, and mitigate those. But you get to see it, see if you're right, learn from reality, and do better the next time.

Greg Brockman

At various points in OpenAI's history, we've had some hope that, hey, there are people who have deployed transformative technologies before, maybe they can tell us the answers. And it's never been so simple. They do have wisdom and insights, which we've incorporated, but we realised we're the closest ones to this technology. By virtue of creating it, we have an understanding of the ways we could shape it that is hard for someone who isn't so close to opine on. The right choices are extremely specific to the facts of the technology. There are different pressures exerted by cell phones versus mainframe computers versus AI versus electricity — each has its own unique proclivities and problems. The individuals doing it matter too — the dynamics between different humans have been hugely impactful for how AI is playing out today. A lot of what we spend our time doing is dreaming, thinking about all the implications of what we might do. One thing I've observed is that we haven't really been surprised by some moments along the way, but we have been surprised by when they arrive, how hard they are to accomplish, exactly the order in which we see them. The world we're moving towards is, in many ways, more wonderful and awe-inspiring than many of the ones we anticipated.

Greg Brockman

We have found that safety is actually a core product feature. No one wants a model that is not aligned with them. You want a model you can trust, that does the right things in any circumstance. We have invested — possibly far more than people perceive, possibly more than any other lab — in safety. We have in ChatGPT the broadest deployment of these language models in the world, used by the most people. We have to care. I don't think there's a sustainable state where the people building this technology and having successful products are not also investing super hard in safety. The challenge is that some aspects of delivering safety are not necessarily short-term — you have to think long-term, not just for your business but for what you're creating.

Greg Brockman

One thing people miss is that it's not just about the safety of the model, it's about the resilience of society. If you look at how transformative technologies enter the world, society builds around their strengths and their risks. You build cars, but you also need seat belts, you need roads, and you reorient cities around how the technology works. You think about electricity — you have safety standards, rules about where you're allowed to put the electric poles and high-voltage lines. The same will be true for AI. It's not just about the model itself — it's about how they integrate into a world with a society that is resilient. The OpenAI Foundation has this as one of its key focuses, helping society invest in and build a resilient layer for AI.

Greg Brockman

One important piece is that we need to ultimately ensure this technology benefits people. It is very clear that institutions, jobs, life paths people thought would be stable — those assumptions may not hold anymore. We need to make sure we provide support, that we're all there to support each other as this technology rolls out. From a regulatory perspective, whether it's things like everyone should have access to compute — how do we ensure that's true? How do we ensure that as this technology generates more economic value, it doesn't accrue to just one place? This should be something everyone is benefiting from, that people feel in their daily lives.

Greg Brockman

A good example is the number of people who say their life, or the life of a loved one, was saved through the use of ChatGPT. That's something that should be supported and protected. You can do that through regulation by thinking about privacy and privilege. You talk to a doctor, you talk to a lawyer — those are privileged conversations, you feel comfortable sharing them, and there are certain guardrails, well defined in the law, on when a healthcare provider would have to provide that information to law enforcement. We don't have anything like that for AI right now. But people are using these tools, and they should, because they're so important for giving people access to information they wouldn't get otherwise — and they should have the appropriate protections there too. There's a lot of leaning into how these models insert into people's lives, how we make sure we can continue to innovate while the benefits flow broadly, how we ensure America remains a leader. You think about robotics, where I think we are not the leader. For AI, we have to make sure we continue with this remarkable position.

Greg Brockman

You think about data centres — there's clearly been concern about whether they drive up electricity prices, and we have a commitment to ensure they do not. Each of these things can be achieved through many different mechanisms — sometimes regulation, sometimes commitments from the company, sometimes just people understanding the facts. A good example is data centres and water usage. That's something people talk about a lot, but actually our data centres use incredibly little water. That's misinformation, that they use a lot.

Greg Brockman

It is, because it's a closed loop. You basically fill up a giant swimming pool of water and circle it around. It's a fixed amount that's not very large. But people really understanding the why — why are we building these things, why is it worthwhile, how does it benefit me — and being able to give people that empowerment, whether it's helping them feel they can be an entrepreneur now, build a business, create something. All of that we have to solve for. We have to make sure people feel it in their daily lives.

Greg Brockman

This technology, it is uncertain exactly how it will play out — and surprising how it will play out. The AIs we have right now, the world we have right now, is not really something anticipated by science fiction. It's just different. Some inevitable conclusions turn out to not quite look the same way when they come to pass. It's always easiest to see what you lose. The change is coming — there's no denying that. But it's much harder to see, a priori, what you gain. Think about Uber. If you described it to someone in 1950, you have to think about computers, mobile phones, GPS — and it's all so you can get a car to appear where you are in three minutes. That's actually crazy, that level of technological investment for that kind of use case. But it really happened, and it didn't just happen for that one use case — it happened for millions of other use cases.

Greg Brockman

My view of AI is that it is about empowerment, about human agency. That does mean some of these institutions and jobs will turn out not to be as stable as we thought. It will affect people. But the question to lean into is: what do you gain, and how do you benefit? Now you can be a builder — anything you can imagine can become real. So, what do you imagine? How do you build that skill? One thing I've observed is that across multiple generations of this technology, the people who get the most benefit are the people who did it for the previous one. The more you build the skill — and at the core of it is agency, having a vision, having ideas — because now the barrier to entry to trying them out is lower than ever. There will be new opportunity created. The world does need to think about how we support everyone through this moment of uncertainty. The economy will be a compute-powered economy — it will be different, but there will be a place for everyone to contribute.

Greg Brockman

Leaning into this technology is going to be a critical skill — really understanding how you get the most out of AI. Because we're all going to be heading to a world where we're managers of agents, and soon maybe the CEO of an autonomous AI corporation. Imagine if you had the workforce of a 100,000-person company all at your disposal, operating on your behalf, 24/7 — as long as you've got the tokens, the compute to power it. Which again, I think everyone needs access to compute; that's so critical for the world to get right. Because then you can point that at any problem, and the number of problems humanity could want to solve are boundless. The more people lean into this technology, figure out how to combine these technologies in new ways, how to interact with our agents to manage them, and think about what is it that I want, what is my purpose, what do I want to see in the world — it is going to be easier than ever to accomplish that. What we gain, I think, is going to be almost unimaginable in its upside.

Greg Brockman

One interesting thing about how technology has played out to date is that it's really been about contorting ourselves to the machine. Think about how many people work where you have this box and you're typing away, getting your carpal tunnel, your shoulders hunched — all of those things that were not natural, not what we're designed for. We're going to be moving to a world where it's not just that you do work with your computer, but your computer does work for you. That presents opportunities and risks. One core thing is that if you have machines that help people actualise their goals, sometimes people have conflicting goals — how do you resolve that? How do you decide the bounds on what an AI will help you with? How do you make sure the benefits don't just go to one corporation, one set of people, but actually lift up everyone? We need to raise the floor, so everyone has access to a great life and this technology, and I think it'll correspondingly lift the ceiling. There should be something — I don't know if the right word is safety net — that really makes sure everyone gets brought along, and then we're going to be able to accomplish so much more.

Greg Brockman

Think about access to medical care. We should be in a world, if we do our job right, where everyone has a doctor in their pocket that is better than any team of doctors today. The world's best doctors, there for you, actually reading your charts, thinking 24/7 about how to help with this condition. It's disruptive. It's not going to come for free in terms of how this technology interacts with the world, and we've already seen the beginning errors of it. But even over the next two years, I think it will be a force for good — though we have to acknowledge all the ways it could go wrong, the risks of it, in order to achieve those upsides.

Greg Brockman

Achieving the OpenAI mission of ensuring that artificial general intelligence benefits all humanity.

Greg Brockman

This was a great conversation, man. I had a great time.