Sam Altman & Brad Lightcap: Which Companies Will Be Steamrolled by OpenAI? | E1140

1. 0:00 – 0:55

   Intro

   1. SASam Altman0:00

      There are two strategies to build on AI right now. There's one strategy which is assume the model is not going to get better, and then you kind of, like build all these little things on top of it. There's another strategy which is build assuming that OpenAI is going to stay on the same rate of trajectory, and the models are going to keep getting better at the same pace. It would seem to me that 95% of the world should be betting on the latter category. But a lot of the startups have been built in the former category. When we just do our fundamental job because we like have a mission, we're going to steamroll you.

   2. HSHarry Stebbings0:28

      Ready to go? Guys, I'm so excited for this. I've wanted to do this for a long time. Also, this is the first time that you've done an interview together.

   3. SASam Altman0:45

      I think it is, yeah.

   4. BLBrad Lightcap0:46

      I think that's right.

   5. HSHarry Stebbings0:47

      This is going to be the most unique interview then that you've done together. So this is very exciting. I want to start, I spoke to many mutual friends before and they said, "We've got to start with

2. 0:55 – 3:15

   Building OpenAI 7 Years Ago

   1. HSHarry Stebbings0:55

      context." Sam, what gave you the conviction to- to do this seven years ago?

   2. SASam Altman0:59

      I think there were two things that seemed... Well, I've been interested in AI since I was a little kid. Um, but, and I studied it at college and nothing was working. But when we started, there were two things that seemed really important. One, deep learning seemed to actually legitimately be working. And two, it got better with scale. We didn't know how predictably it got better with scale at the time, but it was clear that like bigger was better. And that seemed like a remarkable set of things. And the confusing thing to us at the time was like, "Why does everybody else not see this and why is e- everybody else not jumping on it?" But they weren't. And so we wanted to do it.

   3. HSHarry Stebbings1:33

      Can I ask, when there were those moments of doubt from everyone else, which there were across those years, what gave you the conviction to stick at it when bluntly very few others had that same confidence?

   4. SASam Altman1:44

      It just seemed to us like it was going to work and we kept making progress. Like it- we- it was not- it was- I wouldn't call- I would not call it blind faith, although there is some amount of you just, you know, you got to believe you can do a hard thing. But it- it felt really important to us t- to do this, that if we could do it, it would be, you know, hugely meaningful, um, to the world in some way and that it might work. Like we had an attack vector we believed in, we had, ah, and then we had continued data that the approach was working. Of course the specifics took a long time to figure out. Uh, you know, we did not start off doing language models obviously. We kind of knew that if we could keep doing things that we previously thought were impossible, that was somehow a good sign for progress. And we had this like fundamental conviction on the approach and the attack vector at a very high level for a very long time. And the details took a long time to work out and many brilliant discoveries by our colleagues. There was never any doubt that AI would be a big deal if we could do it. So that's helpful. Like it's going to be really valuable. Um, the approach we got successively more confident in, although it- it did take some wandering in the jungle for a while, or the desert, whatever that phrase is.

   5. HSHarry Stebbings2:59

      (laughs)

   6. SASam Altman3:01

      Um, and then, you know, it's like if you believe something with high conviction and everybody else doubts it, it's like slightly motivating.

   7. HSHarry Stebbings3:08

      Yeah.

   8. SASam Altman3:08

      It's definitely kind of annoying, but it's slightly motivating.

   9. HSHarry Stebbings3:10

      I mean as a VC that would be contrarian, which is not what we do because we're sheep. But

3. 3:15 – 11:34

   Origins of the Unique Partnership

   1. HSHarry Stebbings3:15

      I- I do want to start on- on actually the relationship that we have here because it is such a unique partnership. And again, we said this is the first time you've been interviewed together. How did the partnership come to be? Brad, why don't you tell me that?

   2. BLBrad Lightcap3:26

      Sure. Well, Sam and I have worked together a long time. Um, and we actually, we spent a lot of time at YC looking at, um, this batch of companies that were starting to hit the growth stage that were these really deeply technical projects. Nuclear fusion reactors, quantum computers, self-driving cars, satellites, things like that. And I was kind of focused on those, uh, on those- those companies from an investment perspective and OpenAI was kind of the first company I saw that I was like, "You know what? This one's kind of unique because it kind of just seems to be getting better over time. It's not this kind of binary risk." Um, and I remember pointing that out to Sam and saying, "I think there's something that's going to be different about this company as compared to some of the other companies that we were looking at, at the- at the time." Um, and I ended up s- spending more time with Greg and Ilya, um, and, uh, the properties that Sam describes of these systems just getting better with scale, at first kind of unpredictably and then more predictably, I thought that was just so unique. And, um, and I think we kind of saw the same thing maybe somewhat from different angles. Uh, I saw it mostly from an investment perspective of if that's true, this is going to be really important. Um, just as an investment outcome, just as something that's going to have real impact on the world. Um, and so I really felt that kind of conviction early on and I just wanted to help any way I could.

   3. HSHarry Stebbings4:39

      Did you have that plan that you wanted to join full time? Like when did that come into fruition that you wanted this to be your mission for the next multi-decade?

   4. BLBrad Lightcap4:47

      It- it wasn't at first. I actually was mostly just trying to help Sam recruit a CFO. (laughs)

   5. HSHarry Stebbings4:51

      (laughs)

   6. SASam Altman4:51

      Brad actually worked at OpenAI full time before I did.

   7. HSHarry Stebbings4:54

      Wow.

   8. BLBrad Lightcap4:54

      That's true. I beat him there. Um, first time I've beaten Sam on anything.

   9. HSHarry Stebbings4:56

      (laughs)

   10. BLBrad Lightcap4:57

       But, uh...

   11. HSHarry Stebbings4:58

       Just take it as a win.

   12. BLBrad Lightcap4:59

       Yeah, exactly. Um, but, uh, no, I was trying to- to help him recruit and at the time no one wanted the job. Um, I asked probably 25 people, uh, if they would want to be CFO of OpenAI, which at the time was just a- a small kind of non... sleepy nonprofit. And I went 0 for 25. And honest to God, the reason I'm here is because I was so embarrassed to come back 0 for 25 that I said, "You know what? Why don't I just help out, uh, nights and weekends." And, uh, and then that turned into full time very quickly.

   13. HSHarry Stebbings5:26

       I had no idea about that.

   14. BLBrad Lightcap5:27

       Yeah.

   15. SASam Altman5:27

       Yeah. I was sort of doing like half my time on OpenAI, half on YC.

   16. HSHarry Stebbings5:30

       Wow.

   17. SASam Altman5:31

       Yeah.

   18. HSHarry Stebbings5:31

       When did you-

   19. SASam Altman5:31

       And there were all kinds of broken things. When did you get full time YC then, Sam? Uh, I started f- doing OpenAI full ti- it was kind of like a gradualish process, but I think like by the spring or summer of 2019.

   20. HSHarry Stebbings5:45

       Okay. So- so Brad beat you to OpenAI. I- I think that great partnerships are about complementary skill sets.

   21. SASam Altman5:50

       That's for sure.

   22. HSHarry Stebbings5:52

       And so I wanted to hear from each of you, like an all star Mr. and Mrs., like what is Brad amazing at that the world doesn't know?

   23. SASam Altman6:00

       Look, I think one of the... One sign of like a good partnership, I'm thankful to have this with like a lot of the key-... people at OpenAI, certainly with Brad, is like, um, if you can't do each other's job, uh, you know, maybe Brad could do my job for a week. I certainly could not do Brad's job for a week. Um, and I, I think that ability to divide up as a team, um, and have a very high bandwidth communication channel with each person and all together as a leadership team is super important. Brad is good at a lot of things. Uh, and I'll talk about just two here in the interest of time. One is adaptability. Uh, Brad joined to do finance obviously, and now does something, I guess it's like in the sphere of finance, but very, very different. Um, we didn't have a business at all, or, uh, we didn't have an appreciable business until very recently. And when it became clear that we were gonna have a very fast-growing business, um, I kind of like looked around and was like, "We really need somebody. You know, we gotta, we gotta get someone to do this." And, uh, I kind of like looked around the room and I asked Brad to do it, and it was, and he was just like, "Okay. I'll figure it out." Like, "I'll, you know, uh, just like, you know, I might need like a little bit of time to get up to speed, but this is, you know, I've done like business S-ish stuff before and can go like build all this out." So the, the willingness to just like take on new challenges at each level of company scale and figure it out as you're going, Brad is super great at. And then the other one is, um, well, I'm like financially illiterate, so all of that seems amazing to me.

   24. HSHarry Stebbings7:34

       (laughs)

   25. SASam Altman7:35

       Uh, but, but, but to build out a new product category and go-to-market function around that takes a very wide array of skills and a great deal of patience, um, and sort of, uh, like a customer obsession, uh, from a product to a business model to a how we're gonna deal with customer support and everything else that goes around that. Um, and Brad's ability to see the whole picture of that and how it comes together so that, you know, a company that... We're here today at this enterprise sales event. I think if you had said a year ago, "We're gonna be like a great organization," or not yet a great organization, "We're gonna even be a very good organization at doing, uh, you know, an enterprise go-to-market function," I would've said, uh, very low chance that that's gonna happen. And now we have a pretty good one.

   26. HSHarry Stebbings8:27

       I mean, we're gonna discuss it later 'cause I think the go-to-market that you've built is incredible. Um, if we flip the tables though, what would you say is Sam's biggest strength that not many people consider or know?

   27. SASam Altman8:40

       Well, I s- some people know this, but I think-

   28. HSHarry Stebbings8:43

       You can say none. That's fine. (laughs)

   29. SASam Altman8:45

       Um, I'll, I'll say two things. Uh, they're interrelated. One is I think at any given point in a company's life, there's only like one to three things that really matter at that point. Those things change, but there, there's almost never 10 things that really matter.

   30. HSHarry Stebbings9:00

       Mm-hmm.
4. 11:34 – 12:45

   Challenges Slowing OpenAI's Innovation

   1. SASam Altman11:34

   2. HSHarry Stebbings11:34

      What are the biggest things that would prevent or slow down the velocity of OpenAI's decision-making innovation?

   3. SASam Altman11:42

      I think we have the best researchers and best research culture that I'm aware of in the world. Um, if we lost either of those things, that would be really bad. Not having enough compute resources would be, uh, really bad. And I think, you know, we, we love doing cool research 'cause scientific advancement is like the coolest, most exciting thing in the world, but really, we're here to, like, do useful stuff for other people. And if we did the best research in the world and then...... we make it as efficient as we can, but we still don't have enough compute to provide it to everybody on Earth who's- wants to use it, and is gonna want to use it so much more as these models get way better. That would get in the way. That'd be really bad. So, uh, the second thing I was going to say for priorities, uh, is think about how we get enough compute to fulfill the demand of people who want to use these.

   4. HSHarry Stebbings12:27

      How do you think about answering that? I know it's a Holy Grail question.

   5. SASam Altman12:30

      Uh, that one I probably won't answer in front of a camera, but I am optimistic.

   6. HSHarry Stebbings12:34

      (laughs)

   7. SASam Altman12:34

      By, by treating that as a whole system problem, um, I am optimistic we will really surprise the world on the upside.

5. 12:45 – 15:49

   Collaborative Decision-Making Process

   1. HSHarry Stebbings12:45

      Good. Can I ask on the decision-making, how do you guys make decisions between the two of you? How do you determine what to get- delegate versus what not to?

   2. SASam Altman12:52

      I think it comes back to just being really aligned on what is most important. And you'll probably just hear me repeat that phrase, but, um, things that are kind of specific to or even tangential to the most important things, we really spend a lot of time on, uh, as an executive team, as a leadership team, trying to make t- the right decision around. Sometimes it's obvious, sometimes it's not. Everything else gets, uh, gets delegated. So, um, I probably make ten decisions a day that don't go to Sam 'cause they're not the most important thing. Um, but we will spend an entire executive team meeting on one thing, uh, and then we'll spend the next meeting on that one thing, uh, if it's really the most important thing.

   3. HSHarry Stebbings13:27

      Do you agree with the saying that it's, like, one or two decisions a year define a company, or do you agree with the you make ten decisions a day and actually it's all about the incremental little decisions that add up to the progress of the company? I'm always stuck between both mindsets.

   4. SASam Altman13:42

      I very much think it's both.

   5. HSHarry Stebbings13:43

      Yeah.

   6. SASam Altman13:43

      Um, I think there are... Uh, one of the things that I loved about being an investor was that job is really a job about one or two decisions a year, or maybe one or two decisions a decade. An operator role is definitely not my natural... This is not my natural place in the world, by the way, but in an effort to, uh, get slightly better at it, one of the things I have learned is that it is true that there are only a handful of strategic decisions. It feels more like one or two a month than one or two a year, but it's not, like, that many, like, big, like, here is the- here is the what decisions. But the, like, the how decisions are... There are a lot of those, and I think people who claim there are not a lot of those have not tried to run a complex company before. Because it would be ridiculous to say that any CEO makes one or two decisions a year or a month. Um, it is really nonstop. But there's a difference between, like, the big, like, "We're gonna do ChatGPT or we're not gonna do ChatGPT." And then the, like, to make that successful along the way in the spirit of making that one decision a successful one, here are the 10,000 little things you have to do along the way.

   7. HSHarry Stebbings14:54

      Why do you think you're not an operator?

   8. SASam Altman14:56

      I mean, I'm manifestly not. Like, I- I- I was, I was very happy. Well, I had a lot of fun being an investor. Um, it- it's, it's not a fulfilling- it was not a fulfilling job for me. Um, but it's a very fun one. And, and I kind of like, you know, all of the, like, things that people say to make fun of investors-

   9. HSHarry Stebbings15:17

      Mm-hmm.

   10. SASam Altman15:17

       ... are somewhat true of, like, for a quality of life job, it's a great, great trade-off.

   11. HSHarry Stebbings15:21

       (laughs)

   12. SASam Altman15:22

       Um, but yeah. With no false humility, I'm just not a operator by nature. Uh, I'm happy to do it 'cause I, like, really love OpenAI and I think AGI will be the most important thing I ever touch, but this is not my natural fit.

   13. HSHarry Stebbings15:37

       It's just funny to hear when you think about OpenAI being, you know, the fastest-growing company-

   14. SASam Altman15:40

       Brad would agree, I'm sure. I would agree. Yeah.

   15. HSHarry Stebbings15:42

       (laughs)

   16. SASam Altman15:44

       I would definitely agree.

   17. HSHarry Stebbings15:44

       That's one where you'll, like, decline to comment.

   18. SASam Altman15:46

       (laughs)

   19. HSHarry Stebbings15:46

       No, no comment on that one. I love that.

6. 15:49 – 18:52

   Balancing Marginal Revenue & Cost in LLM Products

   1. HSHarry Stebbings15:49

      Um, can I ask, we, we mentioned kind of the compute element. In terms of, like, marginal cost versus marginal revenue, how do we think about when, like, marginal revenue exceeds marginal cost? I think that's one that a lot of people suggested that we talk about today, especially with LLM-based products, obviously. How do we think about that? And that could be on both sides.

   2. SASam Altman16:08

      I mean, truly, I think of all the things we could talk about, that is the most boring. No offense, that is the most boring question I could imagine. We will-

   3. HSHarry Stebbings16:13

      Really?

   4. SASam Altman16:14

      I-

   5. HSHarry Stebbings16:14

      Why is that boring?

   6. SASam Altman16:15

      All you have to believe is that the price of compute will continue to fall and the value of AI as the models get better and better will go up and up. And, like, the equation works out really easily. There's ways it can go wrong, like if the price of compute, if we don't make enough compute in the world and the supply-demand thing gets out of balance and we choose for compute, or by factor of bad p- planning we cause compute to be really expensive, then sure, maybe that's, um, the way it goes. But I think we can drive the cost of a very high quality of intelligence to very near zero. And that will just be phenomenal for most things in the world. Not everything, there will be some negatives, but I think, I think the cost of intelligence is about to get really, really cheap.

   7. HSHarry Stebbings16:59

      How does open source and the rise of open source further enable that or impact that?

   8. SASam Altman17:03

      There will be a place for open source models in the world. P- some people will want them. Um, some people will want managed services. Some people, a lot of people use both. I, I kind of think all of these are details that are, like, quite interesting in some sense, but miss the bigger picture, which is we are in the midst of a legitimate and pretty big technological revolution where intelligence is going from this very limited thing, which is, you know, smart humans have it, but if you, like, wanna do something that requires a lot of intelligence, you gotta get a lot of smart people to do something. Like, if you want to make an- a thing like OpenAI, you need a ton of smart people, a ton. If you think about everything in the stack, not just people who work at OpenAI, but the people who make chips and build data centers and all of that, to something where one person will be able to access abundant and very inexpensive intelligence to do just amazing things.

   9. HSHarry Stebbings18:01

      Do you think we overestimate adoption in a year and underestimate it in 10?

   10. SASam Altman18:05

       I mean, probably, 'cause I- I think that's, like, actually a very deep insight on the way that technology...... gets adopted in general, because no matter how amazing something is, societal inertia is just a big deal. But you only ever get a lot of adoption for something amazing, but also it takes a while to get going. And so that's ... I think you do for something cool, you get the one-year, 10-year thing.

   11. BLBrad Lightcap18:28

       Yeah.

   12. SASam Altman18:28

       So probably.

   13. BLBrad Lightcap18:29

       I think we'll have a very fast inversion of expectation and reality. I think right now expectations are extremely high. Reality still pretty bad, honestly.

   14. SASam Altman18:37

       Yeah.

   15. BLBrad Lightcap18:37

       These models are not that good.

   16. SASam Altman18:38

       Yeah.

   17. BLBrad Lightcap18:38

       Uh, I think very quickly expectations will start to come down as people come into contact with today's models. But then very quickly also these models will get really, really good and they'll, you'll see this inversion of expectations reality where all of a sudden then expectations have to

7. 18:52 – 20:48

   Navigating Model Commoditization in AI

   1. BLBrad Lightcap18:52

      catch up.

   2. HSHarry Stebbings18:52

      My, my own question is, you kind of mentioned kind of actual model quality maybe not being as good as can be, and like expectation and reality. The other cool question, which might be a little bit boring, but it's just the commoditization of models. And I, I've never seen them before where you have like Mistral one week so hyped, and then you have, you know, whatever Bard the next week, and it's like the transience of different players being preceded in the media as kind of winning, so to speak, is so moving every week. Is this a game of commoditization?

   3. SASam Altman19:24

      There was a time when there were like more than a hundred car companies in the US, I believe, or at least close to that. And if you go, like look at some of the old media at the time, it was like, "No, there's this better car. No, there's this better one. No, there's this better one." And I think that same thing holds true for most new industries. I think it's fine. I mean, I think it's probably good. Uh, but I don't think that's where the enduring value will be. I think eventually it will shake out. There will be a small number of providers, just a relatively small number, you know, dozen, something like that, doing models at big scale, and it'll be extremely complex, extremely expensive. And the differentiation, and I hope we all continue to push each other to make the models better, cheaper, faster, and commoditize in that sense. And the long-term differentiation will not be, I don't think, the base model. Like that's just, you know, intelligence is just like some emergent property of matter or something. Uh, the, the long-term differentiation will be the model that's most personalized to you, that has your whole life context, that plugs into everything else you want to do, that's like well-integrated into your life. Um, but for now, the curve is just so steep that the right thing for us to focus on is just make that base model better and better.

8. 20:48 – 26:03

   AI Startup Strategies for Model Progress

   1. SASam Altman20:48

   2. HSHarry Stebbings20:48

      Mm-hmm. Can I ask you, you mentioned obviously your time investing and, you know, Brad you obviously engage with so many large enterprises around the world today. For me as an investor, I see so many AI companies and I'm not investing in any application on AI companies because respectfully we've seen OpenAI come out with products and it's like, "Well, that killed a whole industry." Um, I don't-

   3. SASam Altman21:09

      You know, I, I think fundamentally there are two strategies to build on AI right now, or startups doing with AI. There's one strategy which is assume the model is not gonna get better, and then you kind of like build all these little things on top of it. Um, and then there's another strategy which is build assuming that OpenAI is gonna stay on its same rate of trajectory, and the models are gonna keep getting better at the same pace. Um, it would seem to me that 95% of the world should be betting on the latter category. But a lot of the startups have been built in the former category. And then when we just do our fundamental job, which is make the model and its tooling better with every crank, then you get the OpenAI killed my startup meme.

   4. HSHarry Stebbings21:53

      Mm-hmm.

   5. SASam Altman21:54

      Um, if you're building something on Open- on GPT-4 that a reasonable observer would say, "If GPT-5 is as much better as GPT-4 over GPT-3 was, not because we don't like you, but just because we, like have a mission, we're gonna steamroll you." But there's a giant set of startups where you benefit from GPT-5 being way better. And if you build those and AI progress keeps going the way that we think it's gonna go, I think on the most part you'll be really, you'll be really h- for the most part you'll be really happy.

   6. HSHarry Stebbings22:30

      As an investor looking for an investment thesis that will actually last, what are those that will not be steamrolled that I can invest in, Sam, versus those that could be?

   7. SASam Altman22:39

      Um, ask the company whether, uh, 100X improvement in the model is something they're excited about. It's actually, we can tell pretty well 'cause we know the companies that come to us saying, "We want the next model. When is it coming out? When is it coming out? I wanna be the first to try it. It's gonna be the best thing for my company." And then there's a lot of companies that we don't hear from on that, in that regard. Um, and I think that's like a pretty good delineation, um, is if there's a clear path to how better intelligence, better underlying intelligence accelerates that product and that company, um, they should... The c- most companies can tell that story really clearly.

   8. HSHarry Stebbings23:12

      And so like Klarna would be an example of that?

   9. SASam Altman23:14

      Klarna's a good example.

   10. HSHarry Stebbings23:15

       Because for Klarna, I mean, the numbers are astonishing.

   11. SASam Altman23:18

       And think how much better that gets if the next model is as good as we hope it's gonna be. I talked, uh, just this morning to an AI, like, medical advisor I guess they would call it. Um, and they were like, you know, "Here's the places the model's underperforming. It's still pretty useful for, like, these kinds of things. But if the model could just get, like, this much better on these metrics, um, we'd have all these other businesses. So, like can y'all do that faster? And then we can have, like, you know, this, like, thing that'll save all these lives and give people who have not had access to medical care, like, some access. And, you know, how soon can we get that? And, you know, here's how many people are dying every day you delay." It was an effective pitch, actually.

   12. HSHarry Stebbings23:58

       There were questions beforehand that I was like, "I've never asked that. That's like a terrible question." And I'm gonna proceed to ask most of them-

   13. SASam Altman24:04

       Great.

   14. HSHarry Stebbings24:04

       ... so I'm sorry for this. But we mentioned kind of model improvement there. Like, how do we see the rate of model improvement? Is it, like linear? Is it like... does it plateau at points? Obviously now it's accelerated faster than ever in the last whatever time period we want to call that. How do we see that rate of improvement in models?

   15. SASam Altman24:19

       It feels very punctuated externally.... which means I think we've done a suboptimal job on one of our core beliefs. We have this idea that iterative deployment, um, is important, and what you don't want is to go build AGI in secret in a lab. This is, like, the limit case, toil away for a couple of decades and then push a button and all at once the world has to, like, contend with AGI. And better than that to us, it seems, is to put, uh, you know, a model out into the world, let people have some time to think about that, react, figure out how they want to use it, what they'd like it to do differently, what they'd not like it to do, what guardrails society wants or doesn't want, and then, you know, build up sort of more, um, societal engagement with it. And I think in some sense one of the most important decisions we ever made was this one, and that includes things like deploying ChatGPT into the world and getting the world to take advanced AI seriously, which we tried to talk about for a long time and didn't really work and, you know, deploying that really did. But as we think about future models, uh, I- I think we underestimated, because we've, like, lived with these models for so long and because we watched them get better and better little by little, uh, we underestimated how much even with our strategy of iterative deployment a lurch forward some of these things would be. So as we think about the next models, we're trying to find a way to make that even smoother, um, so that it feels closer to the smoothness we feel internally, uh, to the external

9. 26:03 – 29:09

   Challenges of Iterative Deployment as OpenAI Scales

   1. SASam Altman26:03

      world.

   2. HSHarry Stebbings26:03

      Do you think the strategy of iterative d- iterative deployment will still be possible moving forward as you get bigger and bigger? You see obviously like Farin Llama released some on, like, uh, medical, uh, scientific writing and he got terrible blowback and had to pull it away. Bard obviously did theirs and they got an 8% reduction in share price. Uh, as you get bigger and bigger and bigger, releasing an imperfect product can have such ramifications. Is that iterative deployment still possible over time?

   3. SASam Altman26:29

      I think expectations setting matters a lot, but with the right expectation setting I think it is possible.

   4. BLBrad Lightcap26:33

      Yeah. I would agree with that. I think, um, we learn a lot also and so when we released Sora for example, um, we get an incredible amount of feedback from the creative community, from media, from, you know, from industry, and we actually started now to kind of incorporate that feedback into how we think about our research roadmap for that, you know, for that specific modality. And so in a way, like, we- we kind of start with expectations really low. Um, we just try and learn, uh, and we really kind of just listen to the world and then we try and incorporate that as best we can so that by the time we actually have something we want to share, it's something that really feels useful and people have kind of natural familiarity with it. Um, and it almost feels like it was kind of built more for them. Um, and I think that's, like, kind of the- the mode that we'll- we'll operate in somewhat here is, uh, is this... I- it is really iterative, um, and it really is this kind of more co-development what w- what the world, maybe more than the world appreciates.

   5. HSHarry Stebbings27:25

      Can I ask one final thing and then I do want to go onto GTM. But you mentioned obviously the medical advisor earlier; I hear you've got a passion for how, funnily AI can solve cancer and specifically, uh, certain medical-

   6. SASam Altman27:35

      Well, it's more like I have a passion for how AI can help, I don't want to say solve, help like greatly increase the rate of scientific progress. Um, and curing cancer would be a great example of that. But I- I do generally believe, and this is like, you know, there's definitely just a personal element of excitement, but I think science is awesome, but I genuinely believe that scientific progress is like the highest order bit of progress for society; economic growth, quality of everyone's lives, all of that. And if AI can help people meaningfully increase the rate of scientific progress, which I believe it will, uh, I think that will be a triumph.

   7. HSHarry Stebbings28:15

      What do you think is the biggest barrier to that happening?

   8. SASam Altman28:17

      I think the models are just not smart enough which sounds like a annoying low information kind of cop out answer but I think it's like deeply fundamentally true. Like the models just aren't smart enough, you fix that one thing all these other things get better. There will be all these ways that we have to figure out how to integrate tools into people's workflow and, you know, model ability in different areas will- will matter a lot but if you zoom out, you know, doing scientific research with the help of GPT-2 would have seemed fairly laughable.

   9. HSHarry Stebbings28:50

      Mm.

   10. SASam Altman28:51

       With GPT-4 people do use it just in very- to help them do science just in extremely primitive and limited ways and with GPT-6 I think people will say "Hey this is like helping me as a general purpose tool in all these ways" and then with GPT-8 maybe people are like "You know what this can do some limited, maybe not so limited tasks for me."

10. 29:09 – 31:21

    Secrets to OpenAI's Efficient Scaling

    1. SASam Altman29:09

    2. HSHarry Stebbings29:09

       Can I move to the company scaling because I think it's really important to cover. I mean this is the most unprecedented company scaling really in history especially when you look at speed of revenue growth. Um, Brad you've been at the forefront of that w- (laughs) it's- it's a terrible question in many respects but how have you scaled so fast so efficiently and what's the secret to that and things seemingly not breaking?

    3. BLBrad Lightcap29:33

       Uh, well, things (laughs) it's always messy behind the scenes.

    4. HSHarry Stebbings29:35

       (laughs)

    5. BLBrad Lightcap29:37

       Um, but I appreciate you saying that on the outside at least it doesn't seem like things are breaking. Um, eh, well look we, I think we- we found a moment with ChatGPT that it- people kind of, it was the first like really human experience people have had with technology and we hear stories all the time of like where people use it and it's continues to amaze us actually how diverse these stories are. It's like on the one second you're hearing like a research scientist at a company talk about how productive it's made them and the next is like this thing is writing code for me I'm a software engineer at XYZ start-up and the next is like I'm a new parent and like I don't know how to take care of a baby but like I ask this thing 80 questions a day and it kind of like helps me understand how to like navigate life as a new mom. Um, and like the same tool can power each one of those experiences and when you have something that's like that fundamentally, uh, diverse, um, and I think that, uh, kind of fund- you know, fundamentally, um, uh, accessible like it- it's just bound to have a really important impact, uh, in-... like, in adoption and, you know, in how people use it. And I think... I mean, that obviously translates to a business impact. But our focus is, is, is just continuing to push on that front. Um, the, the B2B business is obviously different, different, different f-... kind of, um, cadence to that business. Um, there's, there's more of an adoption cycle, uh, in the enterprise. We've had, um, amazing success on the developer side. So we've, we've always been a company that has really prided itself, I think, on just... we kind of build for who we know.

    6. HSHarry Stebbings31:02

       Yeah.

    7. BLBrad Lightcap31:02

       Um, and so we've, we've tried to build the best developer platform in the world for AI. Um, enterprise is, is, is a new focus for us. Um, and so, you know, that, that will have more of a, uh... there'll be more of a process to building for the enterprise. But, um, it's, it's one that we're excited to take on. And, uh, and so, uh, a lot more to come.

    8. HSHarry Stebbings31:19

       Can I ask, on talent,

11. 31:21 – 32:18

    Talent Attraction

    1. HSHarry Stebbings31:22

       is it bad if talent wants to join because OpenAI is the hottest company, it's the fastest-growing company?

    2. BLBrad Lightcap31:27

       Probably.

    3. HSHarry Stebbings31:28

       So everyone has to join for the mission. 'Cause I'm always like, "Does it actually..." Like, we always say-

    4. BLBrad Lightcap31:32

       No.

    5. HSHarry Stebbings31:32

       ... mission, mission, mission.

    6. BLBrad Lightcap31:33

       Uh, I mean, I think it's bad just 'cause it makes us, like, harder to filter. Uh, it makes it harder for us to filter. I... But, but yeah, like, I do kinda want people to think that they're doing something that's really important. Um, I watched what has happened to other tech companies when they just become the place you wanna work because it's a good resume item. And y- you can, like, filter against that to varying degrees. Um, and as you said, it doesn't literally need to be total- 100% true in 100% of cases. But I think companies that lose their mission orientation, um, and get taken over by mercenaries, usually come to regret that.

    7. HSHarry Stebbings32:17

       It's interesting.

12. 32:18 – 33:46

    Learning from Exceptional Founders

    1. HSHarry Stebbings32:18

       You've invested in some of the best founders. Are there any that stand out as ones that you've learned from, that you've invested in and have shaped how you think about building?

    2. BLBrad Lightcap32:26

       I have been extremely fortunate to work and, like, be along for a small part of the ride, I think with, like, many of the best founders of my generation. And, uh, and I'm also happy that, that they have been willing to, like, spend so much time now helping me. Uh, and-

    3. HSHarry Stebbings32:46

       Can I push you? Are there one or two that stand out? And has there been a lesson or two from them?

    4. BLBrad Lightcap32:51

       Chesky has been incredibly hands-on and helpful to me over the last year and a half, uh, and is really good at a lot of things that, uh, I'm not good at and have had to, like, come up to speed quickly on. Um, how to think about how we talk about our products, um, how to think about how to build great products. Uh, he is really a special person. Uh, the Collison brothers are incredible, and like every time I talk to them, I, I'm like, "Hmm, that is a new deep insight that I just never would have thought of." It's like a totally non-linear thing. I invested in a lot of companies for a long time, so I have like a long list of incredible founders and... to, that, that are l-... have been like... I'm very grateful to, but have been, like, very willing to, uh, really kind of, like, help out in different areas. And I think in the same way that I tried to, like, learn a little bit each from a lot of different investors, trying to learn a little bit each from a lot of different founders has been, uh,

13. 33:46 – 37:47

    AI Go-to-Market Strategies for Enterprise Adoption

    1. BLBrad Lightcap33:46

       a great strategy.

    2. HSHarry Stebbings33:47

       Can I go back to usage? You mentioned the kind of divergence in usage from kind of consumers every day, maybe parents, maybe scientific researchers. You've also built an incredible go-to-market with some of the largest enterprises in the world. What have been some of the biggest lessons on enterprise adoption and how large enterprises are thinking about it, approaching it, adopting it, that you think are noteworthy?

    3. BLBrad Lightcap34:09

       I think the biggest one is, enterprises have a very natural desire, I think, to want to throw the technology into a business process with the pure intent of driving a very quantifiable ROI. I know what none of those words mean.

    4. HSHarry Stebbings34:23

       But it sounds great. I mean... (laughs)

    5. BLBrad Lightcap34:25

       This is my joke that I can't do. I couldn't do Brett's job.

    6. HSHarry Stebbings34:27

       There's three strategic levers... (laughs)

    7. BLBrad Lightcap34:29

       I manage my supply chain, and it costs me X per year, and I wanna take AI and throw it at a specific process in supply chain management and cut 20% of my spend out of this specific area that I spend money on, that type of thing. And that's great. Um, we are here and happy to help you think through that problem. I think people, though, criminally underrate how important it is actually and how much, like, return you really get on just giving people access to the technology, and that there's this kind of... because you, you can't quite quantify exactly how it works. But, like, someone that used to spend two days doing something that now spends two minutes doing something and is freed up to do, like, 85 other things in their daily life, that doesn't really show up in how you would think about ROI as an enterprise. But imagine doing that now 10,000 times over, 100,000 times over.

    8. HSHarry Stebbings35:16

       How do you explain that to enterprises? 'Cause you're right, it's not like a budget line where you're like, "Oh, we got rid of X."

    9. BLBrad Lightcap35:21

       Yeah.

    10. HSHarry Stebbings35:21

        It's difficult to show that supply-of-time shift.

    11. BLBrad Lightcap35:24

        Yeah, I mean, uh, part of it is just having time to show it. Um, uh, ChatGPT is, is a, is... as a business product, it's still so new. We released Enterprise back basically in s-... you know, late August, September of last year.

    12. HSHarry Stebbings35:35

        Yeah.

    13. BLBrad Lightcap35:35

        And, and Teams is a, a... so, our self-serve product-

    14. HSHarry Stebbings35:37

        Yeah.

    15. BLBrad Lightcap35:37

        ... uh, we released earlier this year. So the time in market's been virtually zero, and enterprise adoption cycles are slower. But, um... so I think part of it will just come with time and part of it just comes with expectations of, uh, your workforce will want these tools. And also, like, you're gonna start to hire people who, uh, will have come from a world where they could only ever use these tools, and they can use it as much as they'd like, um, and they will expect to be able to use them in the workplace. Uh, and so I think that, like, over time, we will start to see that shift. Um, but right now, I think that's... there's this kind of weird miscalibration of, um, of where people think they should be deploying AI that's gonna have high impact with where I would say they should be deploying AI, where there will be high impact.

    16. HSHarry Stebbings36:17

        What questions do you think the biggest companies don't ask that they should ask?

    17. BLBrad Lightcap36:22

        Questions the biggest companies don't ask that they should ask?

    18. HSHarry Stebbings36:25

        Yeah. About how to use AI, about how to integrate it, about concerns that they should think through.

    19. BLBrad Lightcap36:30

        A lot of companies think it's static, so a lot of companies think GPT-4 is the best the model will ever get. That's understandable. Every technology they've ever had to adopt has been st-... relatively static. If you think about, like, what the iPhone looked like, you know, what mobile looked like in 2009 versus today-

    20. SASam Altman36:44

        ... it kind of is the same thing. Like the form factors change a little bit, they're faster, they're like higher resolution, but like the technology's pretty much the same. Application development's pretty much the same. Same thing with cloud. And so here, they've been handed this new technology and I think their expectation is like, "Well, this is it." Um, and I think they don't ask enough about really how steep that rate of change is and like what, how to think about like what the next wave of the technology will be and the n- and the wave after that, um, and how to think through implementing how to-

    21. HSHarry Stebbings37:11

        Do you think they're set up for that rate of change? Like, you know, we're obviously in London now. Uh, European corporates are not that fast moving. Um, when you change as fast as you are changing, it's almost very difficult because they get used to their workflows and processes, and then you change and you update and it's like, "Oh, fuck. Well, they're all gone. They're out the window." Do you see what I mean?

    22. SASam Altman37:31

        Sure.

    23. HSHarry Stebbings37:31

        It's almost hard.

    24. SASam Altman37:31

        Yeah. No, it's, it is hard. Um, and that's what makes our job hard, right? Is I think companies have a desire to want to move that fast, but there's this kind of, um... When you're operating at 100,000 person or 200,000 person scale, it can be really, really hard.

    25. HSHarry Stebbings37:44

        Yeah.

    26. SASam Altman37:44

        Um, and so I think that'll be the- the big question over the next

14. 37:47 – 39:15

    Challenges in Blending Product & Sales Cultures

    1. SASam Altman37:47

       few years for us.

    2. HSHarry Stebbings37:48

       Uh, Sam, you mentioned the research and th- and culture and the importance to retain that. When you bring in a go-to-market function and sales leaders and wholesale teams, it's very difficult to blend kind of product and sales functions or cultures so efficiently. How do you think about the challenges that one faces?

    3. SASam Altman38:07

       I think this is where Brad and I have a, a great partnership in that we have different opinions about maybe how to balance any particular decision, and we're, I think, very good at deferring to the other based off of where it has like more context or feels like it will have a more important impact. But we have really deep agreement I think in a way that many people in Brad's role wouldn't about the critical, um, focus of making sure that, that we let research drive product and product drive sales. Now, that doesn't exclusively mean that, of course. There's got to be feedback the other direction. And one of the reasons that we love having users now is this is like the most important reward signal you can get for if the model's good or not. It's like, how useful is it really to people? Like that, that's what matters. But we also know that the best thing we can do to sell more product is to make the product better. And the best thing we can do to make the b- product better is to have a better, to have better research. And there's like zero disagreement between us ever on that. And that is really important.

15. 39:15 – 43:19

    Evolution of Growth Mindset Post-OpenAI

    1. SASam Altman39:15

    2. HSHarry Stebbings39:15

       It's funny, you mentioned the users. I was chatting to Alex Schultz from Meta before, and he said, "Ah, a- ask Sam about, um, growth and ask him how his mindset has been changed on growth post OpenAI because it is such a, a different story."

    3. SASam Altman39:29

       I think there i- like Alex Schultz is a legitimate growth genius. He'll be there, and then he'll talk about this retention curve and the 30D here and the that and the, this acronym. And I mean, he really understands the dials of things. I think you usually don't learn that much from failure. You learn more from success. Um, but I think you also don't learn that much from like extreme, break-all-the-rules, unrepeatable success either. And what we had with ChatGPT, I would be hesitant to say I've learned anything at all about growth. Like have a once-in-a-generation technological revolution that's not really like actionable advice. So if I wanted to learn about growth, which I do, I'm now very interested in it, uh, uh, you know, Alex probably can't advise me on it at this point, but that's who I would normally ask.

    4. HSHarry Stebbings40:13

       Why do you not learn from failure? See, I always disagreed.

    5. SASam Altman40:16

       You learn something from failure for sure. Um, you learn some things to exclude. But at least in my own experience, having failed at many, many things and succeeded at some, uh, I have learned much more from the successes.

    6. HSHarry Stebbings40:31

       What's been your biggest learning from a success?

    7. SASam Altman40:33

       I mean, so many. Like what to look for when hiring people. Um, what, you know, I've now like... I don't hire externally that often. I'm like a big beli- or for like my direct reports, I'm like a big believer in try to like promote into that when you can. But certainly, what to look for when promoting someone. What to look for in a founder. Uh, I would say like, yeah, I can like point to my extremely long track record of failed investments and say, "Ah, I made this mistake here. I made this mistake there. I made, you know, this one over there."

    8. HSHarry Stebbings41:02

       Josh Kushner asked that one. He said, "Ask him what he looks for in founders 'cause the track is so strong."

    9. SASam Altman41:07

       Well, all of the obvious things and then some of... I think some of the things that I look for, uh, more than other people are founders that are going after, uh, something that seems big if it works. Um, I think that is way more important than people realize to like the really outlier returns. So I'm, you know, happy to like lose nine times out of ten and like really succeed on the tenth company rather than kind of like do okay seven times out of ten. Um, I think founders that are like very good at generating lots of new ideas. Um, founders that have like a very fast iteration cycle. Obviously, like, you know, smart and determined and all of those things matter. Oh, great communication skills, uh, are something that I really look for.

    10. HSHarry Stebbings41:58

        Do you? Okay, but I've fucked up so many... I mean, I've missed so many great companies. But I've fucked up because you get an engineering-led CEO, and respectfully, especially at Seed or Series A where I tend to invest, they're not so honed, and so they don't have that communication.

    11. SASam Altman42:12

        Yeah. Polished, I don't worry about, but like as that great CEO used to... Like, like I don't mean communication like can someone sit in an interview and be like super charismatic and, you know, like very, you know, hit the talking points and like... No, clearly not me either.

    12. HSHarry Stebbings42:32

        (laughs)

    13. SASam Altman42:32

        Um, but, but I do think a lot of the job is communications-driven. Like you have to be able to like explain to the company what we're going to do and why, and you have to be able to like-... hire people and get them to want to work with you. And you have to be able to, like, sell things to customers and get people to, like, try your product. At some point, you may have to, like, talk to wider audiences. So I- I don't mean it, like, literally as, you know, can the person give a polished interview 'cause I may make it my whole life without being able to do that. We'll see. Um, but in the day-to-day, you know, able to clearly explain what you're doing, why people should care about it, what you'd like them to do to help you, uh, that's super important.

16. 43:19 – 46:59

    Strategies for Hiring: Experience vs. Hunger

    1. HSHarry Stebbings43:19

       Final one before we do a quick-fire. I do have to ask, on the people that you hire at OpenAI, one thing that's quite striking is, they're a little bit older actually or it certainly s- appears that way. How do you feel about hiring for experience versus hiring people who would be new to a job, but may have that hustle and hunger? And am I wrong to say that you hire for experience and that little bit older?

    2. SASam Altman43:42

       Um-

    3. HSHarry Stebbings43:42

       What do you think?

    4. BLBrad Lightcap43:44

       I think, at least in my orgs where I set hiring policy and whatnot, I ... there's a difference between kind of what the composition of your hires are and kind of what the composition of responsibility is on the team. And I- I try and keep it, uh, keep this kind of, um, th- this team where, like, great ideas can, are like kind of always elevated. Um, and I ... By and large actually, I would say, like, the really, really good ideas come from unexpected places on the team, not from, like, the most experienced end of the- the team always. And so, that's kind of my advice is, like, find a way to make sure that there's, um, there's this very, very flat kind of, like, very, very even playing field when it comes to how you kind of, like, look to the team for perspective, for decision-making, for- for judgment, um, and for creativity. You do need experienced hires, I think, in that they- they bring a little bit of, like, a little bit more perspective obviously. Um, but I tend to think that, like, really the- the, like, company-changing ideas actually, by and large, come from places that are not- not those hires.

    5. HSHarry Stebbings44:47

       Do you agree?

    6. SASam Altman44:48

       I think there's, like, some roles where experience really matters and some where it either doesn't matter or is a slight negative, or could be a big negative. Um, I think, like, our leadership team is probably more like 30s and 40s than the 20s and 30s you would see at other startups. And I think our technical people skew, like, slightly older. Um, I don't have numbers but, you know, maybe I would guess that, like, the average age of the technical team is, like, early 30s instead of the average being, like, late 20s at some other tech companies. I think part of that is just the sort of, like, path to becoming a great researcher. There's huge exceptions in both sides. Um, and I don't want to say I don't care about experience on the whole, but I think there's, like, amazing people with tons of experience, there's amazing people with, like, almost no experience at all. I think whatever we're doing seems to be working. But it's not like ... I don't think about it as a, like, do we want more or less experience? I think it's very much like, who is the per- like, is this the person?

    7. BLBrad Lightcap45:57

       D- I'll- I'll add one thing, which is, there's a lot of areas explicitly where people coming in with experience, I think ... What we do is so categorically different. Like, it's, it is an entirely new category. The way that people kind of engage with, consume, use, talk about, put, you know, put your verb in there, uh, this- this technology is different. So, the playbooks for how you actually, like, bring it to the world are really different. There- there aren't playbooks for a lot of these things. And so, like, the approach you take to solving problems doesn't nesc- ... You don't necessarily benefit in all ways, at least in my world, from people who-

    8. SASam Altman46:31

       (laughs) I think

    9. BLBrad Lightcap46:31

       ... who have done it for 20 years before.

    10. HSHarry Stebbings46:32

        Yeah, one of the joys of new industries is it levels the playing field.

    11. SASam Altman46:35

        It does.

    12. HSHarry Stebbings46:35

        I think you saw this in crypto in particular, where suddenly 19-year-olds were just as impactful (laughs) as a 45-year-old-

    13. SASam Altman46:40

        Yeah.

    14. HSHarry Stebbings46:40

        ... because it doesn't matter.

    15. SASam Altman46:41

        Yeah.

    16. BLBrad Lightcap46:41

        I think- I think in general, if you could, like, sample someone at OpenAI, you know, look at the role they're doing and the level of responsibility they have and the impact they have and say, you know, would I have expected this person to be more experienced or less experienced given that? You would say, "On the whole, I would've expected/maybe even hoped that this person was more experienced."

17. 46:59 – 53:06

    Quick-Fire Round

    1. BLBrad Lightcap46:59

    2. HSHarry Stebbings46:59

       Are you ready for a quick-fire?

    3. SASam Altman47:00

       Sure.

    4. HSHarry Stebbings47:01

       Okay. So it's 60 seconds or less. Uh-

    5. BLBrad Lightcap47:03

       Per question.

    6. HSHarry Stebbings47:04

       Let's start. Sam, what's the single biggest challenge to OpenAI over the next 12 months and then five years? 30 seconds each.

    7. SASam Altman47:10

       Doing the best research and the best productization of- of, like, the best innovation on that stuff, uh, over the next 12 months. And was it five years for the second thing?

    8. HSHarry Stebbings47:20

       Yeah.

    9. SASam Altman47:21

       Sufficient, like, supply chain and compute.

    10. HSHarry Stebbings47:23

        Brad, what have you changed your mind on most over the last 12 months?

    11. BLBrad Lightcap47:27

        I would say actually it really is that, um, the r- I think the rate of adoption in the enterprise is actually gonna be way faster than people realize.

    12. HSHarry Stebbings47:32

        Huh.

    13. BLBrad Lightcap47:33

        Um, I think we will buck convention on that. Uh, p- enterprises have an, uh, a- a reputation as being slow adopters of technology. I think that will not be true here.

    14. HSHarry Stebbings47:41

        Does that differ by geography?

    15. BLBrad Lightcap47:42

        No.

    16. HSHarry Stebbings47:43

        Hmm. Do we have loads of experimental budgets?

    17. BLBrad Lightcap47:47

        Do we have loads of experimental budgets? Well, we have real budgets.

    18. HSHarry Stebbings47:50

        (laughs)

    19. BLBrad Lightcap47:50

        Um, and that'll help.

    20. HSHarry Stebbings47:51

        Yep. Uh, Sam, what are you most concerned about in the world today?

    21. SASam Altman47:56

        The- the whole thing just feels like way more uns- ... The whole situation of the world, the geopolitical thing, the sort of socioeconomic stuff, politics, it feels more unstable to me than it has felt since I've been paying attention.

    22. HSHarry Stebbings48:10

        Mm.

    23. SASam Altman48:10

        And there's no, like, one thing I would say, uh, that I- I- I- I couldn't with confidence tell you, like, here is the- the crux of it or here is the root cause. But the- the general macroinstability feels high.

    24. HSHarry Stebbings48:24

        Brad, what's been the most unexpected thing in the scaling of OpenAI for you?

    25. BLBrad Lightcap48:29

        I think it's how consistently s- the scaling of models has worked. Um, it still breaks my brain. Uh, like, I don't ... Maybe I've been ... If we've ... I've watched the same trend line for six years now, but I still find it incredible that, uh, you can make these models bigger and they get predictably better. Um, and that is a tremendous gift.

    26. HSHarry Stebbings48:49

        Brad, what do you know now that you wish you'd known when you started?... at OpenAI?

    27. SASam Altman48:53

        I wish I'd appreciated the order in which the technology was actually going to get, uh, have impact. Um, it, it caught us somewhat by surprise how important the technology is and is going to be in creative industries, for example, um, relative to, uh, more knowledge-based industries or relative to even more industrial industries. Um, we were doing robotics really early on. And so I was preparing for a world where we were working with robotics companies, uh, and building robots and working with gaming companies and building agents. And we've gone completely the other way.

    28. HSHarry Stebbings49:24

        Sam, what do you not do much of that you'd like to do more of? I guess time is not particularly your friend these days.

    29. SASam Altman49:29

        I don't really read anymore. Um, I used to read a lot. That's a sort of sad change.

    30. HSHarry Stebbings49:34

        Would you like to make more room for it?

Episode duration: 53:06