I knew I wanted to do a startup because I felt like that was something. But you were just in a startup. Stripe was a startup. It's true. But I never I I felt like Stripe, the problem that we were solving was not my problem, right? It wasn't the problem I had grown up thinking about. It was an important problem that I I dedicated myself to that mission for a number of years, but I felt like it was going to succeed with or without me. And so then I had a first moment to really think about what is a mission that 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, right? If you could actually make a difference in how AI will play out in the world, like that would be a life well-lived.

Patrick had said, "Sam has seen lots of young people in your situation." And Patrick, I think 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?" And I said, "Well, 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."

And so the thing that I remember was a topic was, is it too late to start a lab with many of the best researchers? Is it possible? And this is what year? 2015, right? Because you think about just the degree to which Deep Mind 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 the reason it was actually impossible. And so Sam and I driving back to the city that night, I remember we looked at each other and we said, "We got to do this, right? Like we just have to." And so next day I was full-time on putting this together.

Initially, the set of people that I narrowed down to were actually Ilia, Dario, Amadai, Chrissa, and myself. That was going to be the team. It didn't quite come together. I asked Sam, "Okay, how do we break symmetry here? How do we actually get everyone to say, 'All right, we're joining.'" And Sam's suggestion was, "Invite people out for an off-site." So we set up a thing in Napa. And I actually made t-shirts. We came up with what I would really say is almost the technical plan that we have pursued for the past 10 years. Number one, solve reinforcement learning. Number two, solve unsupervised learning. And number three was gradually learn more complicated in quotes things.

At what point did you realize that like this nonprofit thing just wasn't going to work? 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? And we started to do the math on compute. and you start to realize that it's going to take big computer and we came across a company called Cerebras which was building a unique piece of computing hardware. you start to realize if we could buy a lot of those computers we could actually probably succeed at building an AGI if we could get exclusive access to librris that could give us an overwhelming advantage if we could buy very large data centers that could be something unique as well. And the thing about nonprofit fundraising is I think that there is essentially a cap to what is possible there. And so Elon, Sam, Ilia, 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.

Dota, we had our first big result, right? That really was like, "Wow, we can actually accomplish something when we put our mind to it. You can actually see all this compute coming together. You scale up the compute, you scale up the result." 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, right? The the algorithm we use called PO (PPO), you plan over every single time step. There's no hierarchy, right? And we just kept scaling PO and we exceeded the performance of the best humans. And that itself was the finding right that actually massive compute with simple algorithms, it works in practice. We can really make it happen and in this incredibly messy environment.

Is there a difference between reasoning and predicting? You mentioned sort of like predicting the next character, predicting the next word versus actually reasoning in first principles. I think they are connected in a deep way. So on the one hand, just predicting what comes next sounds like a pedestrian task, but if you really can predict the next word out of Einstein's mouth, you are at least as smart as Einstein. I think that those arguments fall flat because the point of prediction is not about 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.

I remember actually an early moment was the unsupervised sentiment neuron paper. 2017 and it's really the first time that we saw semantics arise from training on a language modeling objective. So you train on learn the next character, predict the next character, and then suddenly you get a neural net that understands sentiment. it can really learn the meaning of sentences. When you see something like GPD4, someone asked, "Why is this thing not an AGI?" Right? It's like actually really hard to put your finger on it because you can talk to it fluently in anything you want.

There's two steps to it. The first is unsupervised learning. You train a model just by having it predict what comes next. Then you do reinforcement learning, which is you basically have the AI learn on its own data, right? Fundamentally the technology we use to train during unsupervised stage and during the reinforcing stage they're exactly the same. You are just predicting but you've changed the structure of the data.

Take me back to the moment you found out that Sam had been fired. I got a text saying, "Can we hop on a video call?" I noticed that it was the board minus Sam who were on there. I was told that the board has decided that Sam would be removed. Also that I had been removed from the board, but would be staying with the company. It just wasn't right. Right after I hung up the call, I talked to my wife and I said, "Got to quit." And she said, "I agree."

That day when I quit, started to get all these messages, people saying, "I don't know what you and Sam are doing next, but I'm with you." I didn't really expect to get that kind of support. We all got together and we started to chart out what a new company could look like. That Sunday night the board replaced me as interim CEO with a new person and the company just rebelled. Everyone streaming out of the building and it was just like real chaos. Then this petition starts to circulate, so many people were trying to sign the petition at once it actually crashed Google Docs.

I woke up and I saw that Ilia had posted and had signed a petition, and it said that he wanted the company to come back together. And that was this real moment of relief. Throughout that weekend all the competitors were circling. People are getting offers and we actually did not lose a single person through that weekend. No one accepted a competing offer. They're not playing for money, they're playing for the person beside them.

You guys had the boldness to make that bet with a hundred billion dollars for data centers. That is the core of open AI is really encountering reality as it is. These are maybe the biggest machines that humanity creates and then you ask the question of why. It is because they have the potential to solve problems that matter for people, to solve, you know, come up with cures for cancer. But if we're computer constraint, how do you decide who to serve? This is going to be the most important question for society to answer. Where does the compute go? What problems are worthy?

The aspect of consumer that we're really dialed in on is solving goals. All of those people should have a personal AI, a personal AGI that's out there that knows them well, that has their personal context that is trustworthy. That level of having an AI that knows you and can help you achieve whatever it is you want to achieve and help you flesh out what are your goals. You should still set those goals. You should be in charge. ultimately they're the same technology.

What is iterative deployment and why do you do? One is you go for kind of build it in secret, but then at some point you push a button and you say deploy. And you've never deployed anything ever before. right? That's your first contact with reality and it's a very powerful system. But instead, what about if you take an approach where this is your hundredth system? And the world has also had a chance to adapt to them, to reconfigure around them. It was medical spam. Like advertising different drugs to people. not something we ever would have thought of as a problem, but we see it in front of our eyes and we get a chance to react and learn from it.

It's not just about the safety of the model. It's about the resilience of society. You think about engines, right? You build cars, but you also need seat belts. The same will be true for AI. Leading an AI is very critical for America because I think that this is how you can ensure that democratic values are protected and preserved.

My view of AI is it is about empowerment. It is about human agency. Instead of you doing work with your computers. Your computer actually does work for you. Everyone is going to be heading to a world where we're managers of agents and soon maybe the CEO of an autonomous AI corporation, right? Just imagine if you had the workforce of, you know, 100,000 person company all at your disposal.