Good morning,

This Stratechery interview is technically another installment of the Stratechery Founder series; OpenAI is a startup, which means we don’t have real data about the business. At the same time, OpenAI is clearly one of the defining companies of this era, and potentially historically significant. To that end, I am putting this interview in the category of public company interviews and making it publicly available; Sam Altman is of much higher interest than me, and should not be the reason to subscribe to Stratechery (I previously interviewed Altman, along with Microsoft CTO Kevin Scott, in February 2023).

Still, it was my interview, so I focused on very traditional Stratechery points of interest: I wanted to understand Altman better, including his background and motivations, and what bearing those have had on OpenAI, and might have in the future. And, on the flipside, I care about business: is OpenAI a consumer tech company, and what should that mean for their tactical choices going forward?

We cover all of this and more, including: Altman’s background, the OpenAI origin story, and the genesis of ChatGPT, which changed everything. What does it mean to be a consumer tech company, and did that make the relationship with Microsoft untenable? Should OpenAI have an advertising product, and was the shift away from openness actually unnecessary, and maybe even detrimental to OpenAI’s business? Altman also gives some hints about what is coming next, including open source models, GPT-5, and an expanding consumer bundle married to an API-driven business with your OpenAI identity on top. Plus, at the end, some philosophical questions about AI, creation, and the best advice for a senior graduating from high school.

As a reminder, all Stratechery content, including interviews, is available as a podcast; click the link at the top of this email to add Stratechery to your podcast player.

On to the Interview:

An Interview with OpenAI CEO Sam Altman About Building a Consumer Tech Company

This interview is lightly edited for clarity.

The Path to OpenAI

Sam Altman, welcome to Stratechery.

SA: Thanks for having me on.

You’ve been on once before, but that was a joint interview with Kevin Scott and I think you were in a car actually.

SA: I was. That’s right.

Yeah, I didn’t get to do my usual first-time interviewee question, which is background. Where’d you grow up? How did you get started in tech? Give me the Sam Altman origin story.

SA: I grew up in St. Louis. I was born in Chicago, but I really grew up in St. Louis. Lived in sort of like a quiet-ish suburb, like beautiful old tree-lined streets, and I was a computer nerd. There were some computers in my elementary school and I thought they were awesome and I would spend as much time doing stuff as I could. That was like the glory days of computing, you could immediately do whatever you wanted on the computer and it was very easy to learn to program, and it was just crazy fun and exciting. Eventually my parents got me a computer or got us a computer at home, and I was always a crazy nerd, like crazy nerd in like the full sense just science and math and computers and sci-fi and all of that stuff.

Well, I mean, there’s not a large intervening period before you’re off at Stanford and you started Loopt as a 19-year-old, it was acqui-hired seven years later. I think it’s fair to characterize that as a failed startup — seven years is a good run, but it’s an acquihire. How does this lead to a job at Y Combinator and eventually running the place in short order?

SA: We were one of the first companies funded by Y Combinator in the original cohort, and I was both tremendously grateful and thought it was the coolest thing, it was this wonderful community of people, and I still think the impact that YC has had on the tech industry is underestimated, even though people think it’s had a huge impact. So it was just this hugely important thing to me, to the industry and I wanted to help out. So while I was doing that startup, I just would hang around YC.

It was really seven years of prepping to be at YC in some respects. You didn’t maybe realized it at the time.

SA: In some respect. Not intentionally, but a lot of my friends were in YC, our office was very close to the YC office, so I would be there and I would spend a lot of time talking to PG [Paul Graham] and Jessica [Livingston] and then when our company got acquired, PG was like, maybe you should come around YC next.

I really didn’t want to do that initially because I didn’t want to be an investor — I was sort of unhappy obviously with how the startup had gone, and I wanted to do a good company and I didn’t think of myself as an investor, I was proud not to be an investor. I had done a little bit of investing and it didn’t seem to me like what I wanted to focus on, I was happy to do it as like a part-time thing. It was kind of interesting, but I don’t want to say I didn’t respect investors, but I didn’t respect myself as an investor.

Was there a bit where PG got to be both an investor and a founder because he founded YC, and by virtue of taking it over, it was a very cool position, but you also weren’t the founder?

SA: I didn’t really care about being a founder, I just wanted an operational job of some sort, and what he said to me that finally was pretty convincing, or turned out to be convincing, was he’s like, you know, “If you run YC, it’s closer to running a company than being an investor”, and that turned out to be true.

I’m going to come back to that in a second, but I do want to get this moment. I think you started running YC in 2014 and around then or 2015, you have this famous dinner with Elon Musk at the Rosewood talking about AI. What precipitated that dinner? What was in the air? Did that just come up or was that the purpose?

SA: Did AI just come up?

Yeah, was the dinner, “Let’s have a dinner about AI”?

SA: Yeah, it was very much a dinner about AI. I think that one dinner, for whatever reason, one specific moment gets mythologized forever, but there were many dinners that year, particularly that summer, including many with Elon. But that was one where a lot of people were there and for whatever reason, the story got very captured. But yeah, it was super explicit to talk about AI.

There’ve been a lot of motivations attached to OpenAI in the years that have gone on. Safety, the good of humanity, nonprofit, openness, etc. Obviously the stance of OpenAI has shifted around some of those topics. When I look back at this history though, and it actually I think has been augmented just in the first few minutes of this conversation, I see an incredibly ambitious young man putting himself in position to build a transformative technology, even at the cost of one of the most coveted seats in Silicon Valley. I mean, this was only a year after you’d taken over YC Combinator. What was your motivation for doing OpenAI at a personal level? The Sam Altman reason.

SA: I believed before then, I believed then I believe now, that if we could figure out how to build AGI and if we could figure out how to make it a net good force in the world, it would be one of the most exciting, interesting, impactful, positive things anybody could ever do.

Scratch all those itches that were still unscratched.

SA: Yeah. But it’s just like, I don’t know. It’s gone better than I could have hoped for, but it has been the most interesting and amazing cool thing to work on.

Did you immediately feel the tension of, “Oh, I’m just going to do this on the side?”, at what point did it became the main thing and YC became the side?

SA: One of the things I was excited about to do with YC was, YC was such a powerful thing at the time that we could use it to push a lot more people to do hard tech companies and we could use it to make more research happen and I have always been very interested in that. That was the first big thing I tried to do at YC, which is, “We’re going to start funding traditional hard tech companies and we’re also going to try to help start other hard tech things”, and so we started funding supersonic airplanes and fusion and fission and synthetic bio companies and a whole bunch of stuff, and I really wanted to find some AGI efforts to fund.

Really there were one or two at the time and we did talk to them and then there was the big one, DeepMind. I’d always been really interested in AI, I worked in the AI lab when I was an undergrad and nothing was working at the time, but I always thought it was the coolest thing, and obviously the structure and plans of OpenAI have been some good and some bad, but the overall mission of we want to build beneficial AGI and widely distributed, that has remained completely consistent and I still think is this incredibly important mission.

So it seemed like a thing that YC should support happening and we had been thinking that we wanted to fund more research projects. One thing that I think has been lost to the dustbin of history is the incredible degree to which when we started OpenAI, it was a research lab with no idea, no even sketch of an idea.

Transformers didn’t even exist.

SA: They didn’t even exist. We were like making systems like RL [Reinforcement learning] agents to play video games.

Right, I was going to bring up the video game point. Exactly, that was like your first release.

SA: So we had no business model or idea of a business model and then in fact at that point we didn’t even really appreciate how expensive the computers were going to need to get, and so we thought maybe we could just be a research lab. But the degree to which those early years felt like an academic research lab lost in the wilderness trying to figure out what to do and then just knocking down each problem as it came in front of us, not just the research problems, but the like, “What should our structure be?”, “How are we going to make money?”, “How much money do we need?”, “How do we build compute?”, we’ve had to do a lot of things.

It almost feels like you dropped out of college to start Loopt and then you’re like, “Well you know what? The college hangout experience is pretty great, so I’m going to hang out at YC”, and then it feels like OpenAI is like, “You know what, just doing academic research and discovering cool things is also pretty great”, it’s like you recreated college on your own because you missed out.

SA: I think I stumbled into something much better. I actually liked college at the time, I’m not one of these super anti-college people, maybe I’ve been more radicalized in that direction now, but I think it was also better 20 years ago. But it was more like, “Oh wow, the YC community is always what I imagined like the perfect academic community to be”, so it was at least it didn’t feel like I was trying to recreate something. It was like, “Oh, I found this better thing”.

Anyway, with OpenAI, I kind of got into it gradually more and more, it started off as one of six YC research projects and one of hundreds of companies I was spending time on and then it just went like this and I was obsessed.

The curve of your time was a predictor of the curve of the entire entity in many regards.

SA: Yeah.

Maybe the most consequential year for OpenAI — well, I mean there’s going to be a lot of consequential years here — but from a business strategy nerd perspective, my sort of corner of the Internet — was 2019, I think. You released GPT-2. You don’t open source the model immediately and you create a for-profit structure, raise money from Microsoft. Both of these were in some senses violation of the original OpenAI vision, but I guess I struggle because I’m talking to you, not like OpenAI as a whole, to see any way in which these were incompatible with your vision, they were just things that needed to be done to achieve this amazing thing. Is that a fair characterization?

SA: First of all, I think they’re pretty different. Like the GPT-2 release, there were some people who were just very concerned about, you know, probably the model was totally safe, but we didn’t know we wanted to get — we did have this new and powerful thing, we wanted society to come along with us.

Now in retrospect, I totally regret some of the language we used and I get why people are like, “Ah man, this was like hype and fear-mongering and whatever”, it was truly not the intention. The people who made those decisions had I think great intentions at the time, but I can see now how it got misconstrued.

The fundraising — yeah, that one was like, “Hey, it turns out we really need to scale, we’ve figured out scaling laws and you know, we’ve got to figure out a structure that’ll let us do this”. I’m curious though, as a business strategy nerd, how are we doing?

Well, we’re going to get to that, I am very interested in some of these contextual questions. Just one more touch on the nonprofit — the story, you talk about the myth. The myth is this was, the altruistic reasons you put forward and also this was a way to recruit against Google. Is that all there is to it?

SA: Why be a nonprofit?

Yeah. Why be a nonprofit and all the problems that come with that?

SA: Because we thought we were going to be a research lab. We literally had no idea we were ever going to become a company. Like the plan was to put out research papers. But there was no product, there was no plan for a product, there was no revenue, there was no business model, there were no plan for those things. One thing that has always served me well in life is just stumble your way in the dark until you find the light and we were stumbling in the dark for a long time and then we found the thing that worked.

Right. But isn’t this thing kind of like a millstone around the company’s neck now? If you could do it over again, would you have done it differently?

SA: Yeah. If I knew everything I knew now, of course. Of course we would have set it up differently, but we didn’t know everything we knew now and I think the price of being on the forefront of innovation is you make a lot of dumb mistakes because you’re so deep in the fog of war.

The ChatGPT Origin Story

Tell me the ChatGPT story? Whose idea was it? What were the first days and weeks like? And we’ve talked offline before, you’ve expressed that this was just sort of a total shock. I mean is that still the story?

SA: It was a shock. Like look, obviously we thought that someday AI products were going to be huge. So to set the context, we launched GPT-3 in the API, and at that point we knew we needed to be a company and offer products, and GPT-3 was doing fine. It was not doing great, but it was doing fine. It had found real product-market fit in a single category. This sounds ridiculous to say now, but it was a copywriting, was the one thing where people were able to like build a real business using the GPT-3 API.

Then with 3.5 and 4, a lot of people could use it for a lot of things, but this was like back in the kind of dark ages and other than that, the thing people were doing with it was developers. We have this thing called The Playground where you could test things on the model, and developers were trying to chat with the model and they just found it interesting. They would talk to it about whatever they would use it for, and in these larval ways of how people use it now, and we’re like, “Well that’s kind of interesting, maybe we could make it much better”, and there were like vague gestures at a product down the line, but we knew we wanted to build some sort of — we had been thinking about a chatbot for a long time. Even before the API, we had talked about a chatbot as something to build, we just didn’t think the model was good enough.

So there was this clear demand from users, we had also been getting better at RLHF and we thought we could tune the model to be better and then we had this thing that no one knew about at the time, which was GPT-4, and GPT-4 was really good. In a world where the external world was used to GPT-3 or maybe 3.5 was already out, I don’t remember, I think it was by the time we launched ChatGPT — it was, but not 4. We’re like, “This is going to be really good”.

We had decided we wanted to build this chat product, we finished training GPT-4 in, I think August of that year, we knew we had this marvel on our hands, and the original plan had been to launch ChatGPT, well, launch a chatbot product with GPT-4, and because we had all gotten used to GPT-4, we no longer thought 3.5 was very good and there was a lot of fear that if we launched something with 3.5 that no one was going to care and people weren’t going to be impressed.

But there were also some people, and I was one of them, that had started to think, “Man, if we put ou