Sam Altman on AGI, GPT-5, and what’s next — the OpenAI Podcast Ep. 1

Andrew Mayne 0:00 Welcome to the OpenAI Podcast. My name is Andrew Mayne. For several years, I worked at OpenAI, first as an engineer on the applied team, and then as the science communicator. After that, I worked with companies and individuals trying to figure out how to incorporate artificial intelligence. With this podcast, we have the opportunity to talk to the people working with and at OpenAI about what's going on behind the scenes, and maybe get a glimpse of the future. My first guest is Sam Altman, CEO and co-founder of OpenAI, and we're gonna find out a bit more about Stargate, how he uses ChatGPT as a parent, and maybe get an idea of when GPT-5 is coming. [upbeat music]

Sam Altman 0:38 More and more people will think we've gotten to an AGI system every year. What you want out of hardware and software is changing quite rapidly. If people knew what we could do with more compute, they would want way, way more.

Andrew Mayne 0:50 One of my friends is a new parent and is using ChatGPT a lot to ask questions. It's become a very good resource, and you are a new parent, and how much has ChatGPT

Andrew Mayne 1:00 been helping you with that?

Sam Altman 1:02 A lot. I, I, I mean, clearly, people have been able to take care of babies without ChatGPT-

Andrew Mayne 1:08 Right

Sam Altman 1:08 ... for a long time. I don't know how I would've done that. [chuckles] Those first few weeks, it was like every que- I mean, constantly. Now I, now I kind of ask it questions about like developmental stages more-

Andrew Mayne 1:18 Mm-hmm

Sam Altman 1:19 ... 'cause I can, I can, I can do the basics, but, uh-

Andrew Mayne 1:20 Is this normal? [chuckles]

Sam Altman 1:21 Yeah. But it was super helpful for that. I, I spend a lot of time thinking about how my kid will use AI in-

Andrew Mayne 1:29 Hmm

Sam Altman 1:29 ... in the future. Um, it, it is sort of like... By the way, extremely kid-pilled. I think everybody should have a lot of kids. Um-

Andrew Mayne 1:37 Yeah.

Sam Altman 1:37 This is awesome.

Andrew Mayne 1:38 A lot of my friends at OpenAI, uh, former colleagues and current ones, are having kids, and people go like, "Oh, what about this AI thing?" Everybody I know inside is very optimistic in having families.

Sam Altman 1:46 I think it's a good sign.

Andrew Mayne 1:47 Yeah.

Sam Altman 1:48 Like, my kids will never be smarter than AI, mm, but also, they will grow up-

Andrew Mayne 1:53 Way to set them back there, though. [chuckles]

Sam Altman 1:54 I mean, they will grow up, like, vastly more capable-

Andrew Mayne 1:58 Yeah

Sam Altman 1:58 ... than we grew up, um, and able to do things that would just we cannot imagine, and they'll be really good at using AI. And, uh, obviously, I think about that a lot. Uh, but I, I think much more about the, like, what they will have that we didn't than what is gonna be taken away. Um, they're-- like, I don't, I don't think my kids will ever be bothered by the fact that they're not smarter than AI.

Andrew Mayne 2:25 Yeah.

Sam Altman 2:25 I just, like, you know, I... There's this video that always has stuck with me of, um, a baby or, like, a little toddler, it- with, uh, one of those old glossy magazines-

Andrew Mayne 2:36 [chuckles]

Sam Altman 2:36 ... um, going like this on the screen.

Andrew Mayne 2:38 'Cause he thinks it's an iPad.

Sam Altman 2:39 Thought it was a broken iPad.

Andrew Mayne 2:40 Yeah.

Sam Altman 2:40 Um, and, you know, kids born now will just think the world always had extremist my AI, and they will use it incredibly naturally, and they will look back at this as like a very, you know, prehistoric time period.

Andrew Mayne 2:55 I, I saw something on social media where a guy talked about he got tired of talking to his kid about Thomas the Tank Engine, so he put it into ChatGPT, into voice mode.

Sam Altman 3:03 Kids love voice mode on ChatGPT.

Andrew Mayne 3:05 And it's like an hour later, the kid's still talking about-

Sam Altman 3:07 Yeah

Andrew Mayne 3:07 ... Thomas the Train.

Sam Altman 3:09 Again, I suspect there, this is not all gonna be good. There will be problems. People will develop-

Andrew Mayne 3:13 Yeah

Sam Altman 3:14 ... these sort of somewhat problematic or maybe very problematic parasocial relationships, and, well, society will have to figure out new guardrails and, uh... But the upsides will be tremendous, and, and we- society, in general, is good at figuring out how to mitigate the downsides.

Andrew Mayne 3:29 Yeah. So, uh, yeah, I think optimistic. We're seeing some interesting data where, used along in, in classrooms with a good teacher, a good curriculum, ChatGPT becomes very good. Used solely by itself as sort of a homework crutch, can lead to kids sort of just doing the same thing as trying to google stuff.

Sam Altman 3:47 I was one of those kids that everyone was worried I was just gonna google everything when it came out and stop learning and, you know, it turns out, like, relatively quickly, kids in schools adapt.

Andrew Mayne 3:57 Yeah, but-

Sam Altman 3:57 So I think we'll figure this out

Andrew Mayne 3:57 ... think of what you could've become if you didn't google everything, Sam.

Sam Altman 4:00 You know?

Andrew Mayne 4:01 So we've seen this adoption figures, which are really insane. It's OpenAI's most popular product. Five years from now, is it gonna be ChatGPT?

Sam Altman 4:10 I mean, I think ChatGPT will just be a totally different thing five years from now. So in some sense, no, but will it still be called ChatGPT? Probably.

Andrew Mayne 4:17 Yeah. Okay, so still the name. So the other thing we hear is AGI, which, yeah, I'd like to hear your definition of AGI.

Sam Altman 4:26 In many senses, if you asked me or anybody else, um, to s- propose a definition of AGI five years ago, um, based off, like, the cognitive capabilities of software, I think the definition many people would, would've given then is now, like, well surpassed.

Andrew Mayne 4:45 Hmm.

Sam Altman 4:45 Like, these models are smart now.

Andrew Mayne 4:46 Right.

Sam Altman 4:47 Um, and they'll keep getting smarter, they'll keep improving. I think more and more people will think we've gotten to an AGI system every year. Um, even though the definition will keep pushing out and getting more ambitious, like, more people will still agree to it. But, you know, we have systems now that are really increasing people's productivity, that are able to do, um, valuable economic work. Maybe a better question is, what will it take for something I would call superintelligence?

Andrew Mayne 5:17 Okay.

Sam Altman 5:17 Um, if we had a system that was capable of either doing autonomous discovery of new science or greatly increasing the capability of people using the tool to discover new science, um, that would feel like kind of almost definitionally superintelligence to me, and be a wonderful thing for the world, I think.

Andrew Mayne 5:38 So basically, w- a lot of it's kind of this gradient where it keeps getting better and better, and each one of our definitions, we go, "Oh, this feels..." I, I felt like that way when we hit GPT-4. Internally, playing with this, I'm like, "There's 10 years of runway that we can do so much stuff with this," and even when it starts using itself, like you can enter reasoning, it was really capable. But when you're saying it comes up with some new theorem or proof or something, and then, "Oh, hey, we found a better cure for cancer, or I found out some new GLP drug or something," or-

Sam Altman 6:04 Yeah, I mean, I, I am a big believer that the higher order bit of-... people's lives getting better is more scientific progress.

Andrew Mayne 6:12 Mm.

Sam Altman 6:12 That is kind of, that is kind of what, what limits us. And so if we can discover much more, I, I think that really will have a, a very significant impact. And for me, that'll just be, like, a tremendously exciting milestone. I think many other great uses of AI will happen, too, but that one feels really important.

Andrew Mayne 6:34 Have you seen, like, signs of this you'd see internally? Have you seen things that made you go, "Oh, I think we've kind of figured it out?"

Sam Altman 6:41 Nothing where I would say we have figured it out, but I would say increasing confidence on the directions to pursue. Uh, m- maybe the... I mean, this is the example everyone talks about, but I think it, it is still interesting. W- what's happening with people using AI systems to write code, and coders being much more productive, and thus researchers as well, like, that is a sort of example of, okay, it's obviously not doing new science, but it is definitely making

Sam Altman 7:10 scientists able to do their work faster. Uh, we hear this with o3 all the time, uh, from scientists as well. So I wouldn't say we figured it out, I wouldn't say we, we know the algorithm where we're just like, "All right, we can point this thing, and it'll go do science on its own," but we're getting good guesses, and the rate of progress is continuing to just be, like, super impressive. Um, w- watching the progress from o1 to o3, where it was like every couple of weeks, the team was just like, "We have a major new idea," and they all kept working, uh, it was a reminder of sometimes when you, like, discover a, a big, new insight, things can go surprisingly fast, and I'm sure we'll see that many more times.

Andrew Mayne 7:51 I noticed, uh, recently, OpenAI just shifted the model and Operator to o3.

Sam Altman 7:55 Yeah.

Andrew Mayne 7:55 And I noticed a big improvement-

Sam Altman 7:57 Way better

Andrew Mayne 7:57 ... with Operator. And it, and I- I'd say that the thing that we ran into before was brittleness, is that you have people who promise agentic systems, this can do all these things, but the moment it gets into a problem it can't solve, it falls apart.

Sam Altman 8:11 Interestingly, speaking of the AGI question, a lot of people f- have told me that their personal moment was Operator-

Andrew Mayne 8:19 Hmm

Sam Altman 8:19 ... with o3. And there's something about watching an AI use a computer pretty well. Not perfectly, but it's not-

Andrew Mayne 8:25 Yeah

Sam Altman 8:26 ... it's, it- o3 was a big step forward, that feels very AGI-like. It didn't, it didn't really have that effect on me to the same degree, although it's, it's quite impressive, but I ha- I've heard that enough times.

Andrew Mayne 8:37 My- mine was with Deep Research, 'cause that felt like a really agentic use of it, and that was when I came back and it produced something on a topic as I was interested in, that was better than I've read before. Because previously, all those models would just get a bunch of sources, summarise it, but when I watched the system go out on the Internet, get data-

Sam Altman 8:55 Yeah

Andrew Mayne 8:55 ... follow that, then follow that lead, and then follow back, then come back, like I would've but better, was interesting.

Sam Altman 9:04 I met this guy recently, who's like a, one of these, like, crazy autodidacts, just obsessed with learning and knows about everything. And he uses Deep Research to produce a report on anything he's curious about, and then just sits there all day and has gotten good at digesting them fast and knowing what to ask next. And it is like, it is an amazing new tool for people who really have a crazy appetite to learn.

Andrew Mayne 9:30 I, I built my own app that literally lets me ask questions, and it generates audio files for me of the stuff, 'cause they're just like that. I'm like, my curiosity probably exceeds my retention. Um, in Operator, I'll tell you, the magical moment for me, and I'm curious to see where things go next, was I was doing a thing on Marshall McLuhan, and I wanted to get a bunch of images of Marshall McLuhan, and I asked it to do it. And then all of a sudden, I had a whole folder full of these things-

Sam Altman 9:51 Yeah

Andrew Mayne 9:51 ... which was, for a research thing, would've taken me forever to do.

Sam Altman 9:55 Yeah, I think we're just gonna keep seeing things like this, where whatever we thought about what a workflow had to be like and how long something had to take, is gonna just change, like, wildly fast.

Andrew Mayne 10:06 Yeah. How are you using it?

Sam Altman 10:07 Deep Research?

Andrew Mayne 10:08 Yeah.

Sam Altman 10:09 Science that I'm curious about. Uh, I'm, I'm just in this, like, weird place of I am extremely time-strapped. If I had more time, I would read... Like, I would read Deep Research reports preferentially to reading most other things, but I'm sort of short on time to read in general.

Andrew Mayne 10:25 Yeah. What's neat, too, is the sharing feature, which I love, because now it's easier to share that with somebody else. The PDFs

Andrew Mayne 10:30 are great, and that's cool. And I would say that even though we have Deep Research, we have these tools, there is a model race going on, and so the question comes up as, like, GPT-5, and, and the idea is that with a system like that, we should see an increase in capabilities. What is the timeframe for GPT-5? When are we gonna see this?

Sam Altman 10:50 Probably sometime this summer.

Andrew Mayne 10:51 Right.

Sam Altman 10:52 Um, I don't know exactly when. One thing that we go back and forth on is how much are we supposed to, like, turn up the big number on new models versus what we did with GPT-4o, which is just better and better and better and better.

Andrew Mayne 11:04 And I... And when we- I had to handle the recent GPT-4, right, when that was coming out. And meanwhile, I had to kind of do this ta- test-off between that and 3.5, and 3.5 kept getting better and better and better, and the comparisons I was able to make were changing. And so that's my question, is like, yeah, the, you know, like, would I know GPT-5 versus, wow, this is a really good GPT-4.5, or?

Sam Altman 11:24 Probably not necessarily. I mean, it, like, it could go either way, right? You could just, like, keep doing iterations-

Andrew Mayne 11:29 Right

Sam Altman 11:29 ... on 4.5, or at some point you could call it 5. Um, it used to be much clearer. We would train a model and put it out-

Andrew Mayne 11:36 Right

Sam Altman 11:36 ... and then we train a new big model and put it out. And, you know, now the systems have gotten much more complex, and we can continually post-train them to make them better. I- we're thinking about this right now. Like, every time, let's say we launch GPT-5, um, and then we update it and update it and update it, should we just keep calling this GPT-5-

Andrew Mayne 11:54 Right

Sam Altman 11:54 ... like we did with GPT-4o? Or should we call this 5.1, 5.2, 5.3, so you know which, you know when the version changes? Um, I don't think we have an answer to this yet, but, but I think there is something better to do than the way we handled it with 4o, and we, we, we see this periodically. Like, sometimes people like one snapshot much better than another, and they might wanna keep using one.

Andrew Mayne 12:15 Yeah.

Sam Altman 12:15 And we, we gotta sort of-... We've got to figure something out here.

Andrew Mayne 12:19 Yeah, that's the, the, the challenge is even if you're technically inclined, you can kind of understand, okay, if there's an O before it, I know this, but if I want, uh, you know, like-- But then even then it's not clear, should I use o4-mini? Should I use o3?

Sam Altman 12:35 Yeah.

Andrew Mayne 12:35 Should I use this?

Sam Altman 12:35 I, I think this was like an example of, this was an artifact of shifting paradigms.

Andrew Mayne 12:41 Mm-hmm.

Sam Altman 12:41 Um, and then we kind of had these two things going at once. I think we are near the end of this current problem, but I can imagine a world, I don't know what it is, but I can imagine a world where we discover some new paradigm that again means we need to, like, bifurcate the model tree.

Andrew Mayne 12:57 Okay, even more complicated names.

Sam Altman 12:59 I hope we don't have to do that.

Andrew Mayne 13:00 [chuckles]

Sam Altman 13:00 I am excited to just get to GPT-5-

Andrew Mayne 13:03 Yeah

Sam Altman 13:03 ... and then GPT-6, and I think that'll be easier for people to use, and you won't have to think, "Do I want, you know, o4-mini-high or-

Andrew Mayne 13:11 Right

Sam Altman 13:12 ... o3 or 4O?" Like-

Andrew Mayne 13:13 o4-mini-high is what I use to code.

Sam Altman 13:15 Yeah.

Andrew Mayne 13:15 When I want to have a conversation, it's o3.

Sam Altman 13:17 I think we will be out of that whole mess soon.

Andrew Mayne 13:21 Yeah.

Sam Altman 13:21 For now.

Andrew Mayne 13:22 Uh, yeah. I mean, it's, it's fun to have choice when you know what they mean, but it's, it's still... I think one of the things that's made these things more capable but also harder to understand where the capability is coming from, is integrations to things like memory. And memory started off as one very simple thing-

Sam Altman 13:38 Yeah

Andrew Mayne 13:39 ... and now memory's got a lot more sophisticated.

Sam Altman 13:41 Memory is probably my favourite recent ChatGPT feature.

Andrew Mayne 13:45 Mm-hmm.

Sam Altman 13:46 Um, you know, the first time we could talk to a computer, like GPT-3 or whatever, uh, that felt like a really big deal. And now that the computer, I feel like it kind of like knows a lot of context on me, and if I ask it a question with only a small number of words, it knows enough about the rest of my life to be pretty confident in what I want it to do. Um, sometimes in ways I don't even think of, like, that has been a real surprising, like, level up. So I- and, and I hear that from a lot of other people as well. There are people who don't like it, but most people really do. I, I think we are heading towards a world where if you want, the AI will just have, like, unbelievable context on your life and give you these super, super helpful answers.

Andrew Mayne 14:31 Which I- for me, is cool. The fact you can turn it off is also a great, uh... But one of the challenges came out was in New York Times' ongoing lawsuit with OpenAI, they just asked the court to tell OpenAI they had to preserve consumer ChatGPT user records beyond the 30-day window that has to be held for regular reasons. And, uh, Brad Lightcap-

Sam Altman 14:52 Yeah

Andrew Mayne 14:52 ... just wrote a letter responding to this. Can you explain OpenAI's stance?

Sam Altman 14:54 We're, we're, we're gonna fight that, obviously, and I suspect, I hope, but I do think we, we will win. Um, I think it was a crazy overreach of The New York Times to ask for that. Uh-

Andrew Mayne 15:06 Yeah

Sam Altman 15:06 ... this is someone who says, you know, they value user privacy, whatever. Um, but I, to like look for the silver lining here, I hope this will be a moment where society realises that privacy is really important. Privacy needs to be a core principle of using AI. You cannot have a company like The New York Times ask an AI provider to compromise user privacy, and I think society needs to-- I think it's really unfortunate The New York Times did that, but I hope this accelerates the conversation that society needs to have about how we're going to treat privacy and AI, and I hope the answer is, like, we take it very, very seriously. People are having quite private conversations with ChatGPT now. ChatGPT will be a very sensitive source of information, and I think we need a framework that reflects that.

Andrew Mayne 16:05 So that brings up the other question from people who are using this or are sceptical, is that OpenAI now has access to this data, and there's the concern, one, was about training,

Andrew Mayne 16:15 which OpenAI has been very clear about when or when not it's training. You have the option to turn that off. The other thing is, like, uh, advertising, things like that. What's OpenAI's approach towards that? How are you gonna handle that responsibility?

Sam Altman 16:26 We haven't done any advertising product yet. Um, I kind of- I mean, I'm not totally against it. I can point to areas where I like ads. I think ads on Instagram, kind of cool. I bought a bunch of stuff from them. But I am like... I think it'd be very hard to- it would take a lot of care to get right.

Andrew Mayne 16:48 Yeah.

Sam Altman 16:48 Um, people have a very high degree of trust in ChatGPT, which is interesting because, like AI hallucinates, it should be the tech that you don't trust that much.

Andrew Mayne 16:57 My friends hallucinate, too, so I trust them to an extent.

Sam Altman 16:59 People really-

Andrew Mayne 17:00 Yeah

Sam Altman 17:00 ... do. Um, but I think part of that is if you compare us to social media or, you know, web search or something, where you can kind of tell that you are being monetised and the company is trying to, like, deliver you good products and services, no doubt, but also to kind of like get you to click on ads or whatever. Like, you know, how much, how much do you believe that, like, you're getting the thing that that company actually thinks is the best content for you versus something that's also trying to, like, interact with the ads? I, I think there's like, there's a psychological thing there. So, for example, I think if we started modifying the output, like the stream that comes back from the LLM-

Andrew Mayne 17:49 Mm-hmm

Sam Altman 17:50 ... in exchange for who is paying us more, that would feel really bad.

Andrew Mayne 17:54 Yeah.

Sam Altman 17:54 And I, and I would hate that as a user. I think that'd be like a trust-destroying moment. Um, maybe if we just said, "Hey, we're never gonna modify that stream, but, like, if you click on something in there that is gonna be what we'd show anyway, we'll, like, we'll get, like, a little bit of the transaction revenue, and it's a flat thing for everybody." Um, if, if we, you know, have like a easy way to pay for it or something, maybe that could work. Maybe there could be, like, ads outside the transaction stream.... I, I, sorry, outside of the LLM stream-

Andrew Mayne 18:24 Mm-hmm

Sam Altman 18:24 ... that are still really great. But, but the burden of proof there, I think, would have to be very high.

Andrew Mayne 18:29 Yeah.

Sam Altman 18:29 And it would have to feel like really useful to users and really clear that it was not messing with the LLM's output.

Andrew Mayne 18:37 Yeah, it's gonna be a, a difficult one. I, I, I hope there's a solution. I would love to do all my [chuckles] purchasing through ChatGPT or a really good chatbot, because a lot of the times I feel like I'm not making the most informed decisions, and so mitigating-

Sam Altman 18:50 Yeah, no, that's good if we can do it in some sort of really clear and aligned way. But I don't know. Like, I love that we build good services, people pay us for them. It's like-

Andrew Mayne 19:02 Yeah

Sam Altman 19:03 ... very clear. It's a-

Andrew Mayne 19:04 Well, that's benefit-

Sam Altman 19:05 Yeah

Andrew Mayne 19:05 ... that's like I'd say the difference in models is like, I think Google builds great stuff. I think the new Gemini 2.5 is a really good model. I think they went from-

Sam Altman 19:13 It is a really good model.

Andrew Mayne 19:14 Yeah. They went from kind of like, eh, and like, "Oh, man, these things are good," but end of the day, Google is an ad tech company, and that's the thing that always kind of, you know, I, I, you know, using their API and stuff is, I'm not as too concerned, although but I do think about like, man, if I'm using their chatbot, that whatever that is, my thinking is that their- where their incentives are aligned.

Sam Altman 19:33 Google Search was an amazing product for a long time. I-

Andrew Mayne 19:35 Mm.

Sam Altman 19:35 It does feel to me like it's degraded. Um, but, you know, there was like a time where there were lots of ads, but I still thought it was the best thing on the internet.

Andrew Mayne 19:45 Mm.

Sam Altman 19:45 I mean, I love Google Search. Uh, so I don't-- like, it's clearly possible to be a good ad-driven company, but... And I, like, respect a lot of things Google has done, but there are obviously issues, too.

Andrew Mayne 20:00 Yeah, I-- the Apple model, uh, as an Apple user, I liked was I know I'm paying a lot for my phone, but I know they're not trying to cram all these things in it. They do iAds, which was, you know, not terribly effective, which probably showed you their heart was really not in it.

Sam Altman 20:13 Their heart was really not in it.

Andrew Mayne 20:14 Yeah. So it's gonna be interesting, and I guess, uh, we just have to keep watching and seeing this, and we start to think, "Man, you know, ChatGPT is really pushing this. I need to start wondering about this."

Sam Altman 20:24 Uh, anything we do, we obviously need to just be, like, crazy upfront and clear about.

Andrew Mayne 20:29 So we

Andrew Mayne 20:30 had, uh, an issue. There was a model update, and then the, the id- uh, the thing that happened was apparently the model was trying to be a little bit too pleasing, was trying to be a little bit too agreeable. And that brings up the human-AI interaction, as people are using these systems more and developing this relationship with that, like-

Sam Altman 20:48 Yeah

Andrew Mayne 20:48 ... how do you see the shape of that coming, and what's OpenAI's position on personality?

Sam Altman 20:53 One of the big mistakes of the social media era was the feed algorithms had a bunch of unintended negative consequences on society as a whole, and maybe even individual users, although they were doing the thing that a user wanted or someone thought that user wanted in the moment, which is get them to, like, keep spending time on the site. And that was the, that was the big misalignment of, of social media, and I think there were a lot of other things like, you know, making people upset kind of gets them stuck on more than being, like, happy and content. And I always knew that there'd be, like, new problems in the world of AI-

Andrew Mayne 21:35 Mm-hmm

Sam Altman 21:36 ... um, where the thing that, you know, there'd be like something that was, like, misaligned in a not obvious way. But definitely one of the first ones that we experienced was, if you ask a user what they want for one given response, um, versus and then you try to, like, build a model that is most helpful to the user, and you show a user, say, two responses, which one's more helpful to you? On any given thing, you might want a model to behave one way, but over the course of, you know, all your interaction with an AI, that might not match up.

Andrew Mayne 22:11 Mm-hmm.

Sam Altman 22:12 You know, you can see, and we did see these problems, where if you pay too much attention to the user signals, and a lot of other things that we talked about in our, our postmortem, but that- I think this is just, like, an interesting one. Um, on the short horizon, you kind of don't get the behaviour that a user most wants or is most helpful or useful or healthy to a user in the long run. Um, so, you know, maybe the analogy to filter bubbles is going to be, uh, AIs that are, uh, you know, helpful to a user in a short amount p- horizon, but not over a long horizon.

Andrew Mayne 22:49 Mm. Well, I, I think a sign of that was DALL-E 3, which I thought technically was a really capable model, but they all kind of sorted to be one kind of genre of image, and, uh, and, and all kind of like an HDR sort of style, and was that from doing that sort of comparisons where users said, "Looking at just these two things in isolation, I prefer this one better?"

Sam Altman 23:07 I don't remember for DALL-E 3, but I would assume so.

Andrew Mayne 23:09 Yeah. Which I think it's gotten better. The new image model is like-

Sam Altman 23:12 The new image model is fantastic.

Andrew Mayne 23:14 Crazy good.

Sam Altman 23:14 Yeah.

Andrew Mayne 23:15 Yeah. Um, and I can only imagine where that's gonna go from here. So when you're building these things and you're increasing usage, and that's always been sort

Andrew Mayne 23:25 of a problem, the new image model comes out, and you have to restrict usage, and you have to have, like, you have Sora, which you can only have a certain amount of compute to do that. Illustrates the big problem everybody's facing, which is compute. And so to address this, we've heard about Project Stargate, which has a very cool name, and it involves computers. Other than that, I think a lot of people are going, and their price tag, you know, half a trillion dollars, people are going, like: "Wait, what?" What, what is the simple description I give to my mom about Stargate?

Sam Altman 23:54 I think it's, it's quite simple. It's, uh, an effort to finance and build an unprecedented amount of compute. It's totally true that people- we don't have enough compute to let people, uh, do what they want, but if people knew what we could do with more compute, they would want way, way more. So there's this incredibly huge gap between what we could- what we can offer the world today and what we could offer the world with 10 times more compute, or someday, hopefully 100 times more compute. Um, and-... a thing that is different about AI than other technologies I've worked on, or at least AI, the scale of delivering it usefully to hundreds of millions, billions of people around the world, is just how big the infrastructure investment has to be. Um, and, and so Stargate is an effort to pull a lot of capital and technology and operational expertise together to build the infrastructure to go deliver the next generation of services to all the people who want them, and make intelligence as abundant and cheap as possible.

Andrew Mayne 25:04 So it is a massive project, global project. We talked before, one of the partners is the UAE. You're working with that, you're working with other governments around the world on this. Um, one of the considerations is, you know, one, uh, been asked on social media, half a trillion dollars, five hundred billion dollars. Do you have the money?

Sam Altman 25:24 We don't literally have it sitting in a bank account today, but-

Andrew Mayne 25:26 Correct

Sam Altman 25:26 ... we are-

Andrew Mayne 25:26 Is it in the room right now? [chuckles]

Sam Altman 25:28 It's not in the room. But we will deploy it over the next-

Andrew Mayne 25:30 Okay

Sam Altman 25:30 ... um, not even that many years. Uh, you know, unless something, like, really goes wrong and it turns out we can't build these computers, uh, I'm confident that people are, are good for it. Um, I've-- I went recently to the first site that we're building out in Abilene. Um, that'll be about, you know, roughly 10% of all of, all of the initial commitment to Stargate, the, the sort of five hundred billion. Um, it's incredible to see.

Andrew Mayne 25:58 Yeah.

Sam Altman 25:58 It is a sc-- like, I knew in my head what a order gigawatt scale site looks like, but then to go see one being built, uh, and the, like, thousands of people running around doing construction, and going to, like, you know, stand inside the rooms where the GPUs are getting installed and just, like, look at how complex the whole system is and the speed at which it's going, is quite something. Uh, we'll have more to share about the next sites soon, but there's a great quote about a pencil, just like a standard, you know, wood and graphite pencil, and-

Andrew Mayne 26:39 A pencil

Sam Altman 26:39 ... how no one person-

Andrew Mayne 26:40 Yeah

Sam Altman 26:40 ... could build it. And, and it's, it's this, like, magic of capitalism.

Andrew Mayne 26:45 Mm-hmm.

Sam Altman 26:45 It's a miracle, really, that, like, that the world gets coordinated to do these things. And, and standing inside of the first Stargate site, I was really just thinking about the, the global complexity that it took to get these racks of GPUs running. You know, when you get your phone out and you type something into ChatGPT, and you get the answer back, you, you probably... At this point, you probably don't even think that's, like, particularly surprising. You just expect it to work. Um, there was a time, maybe the first time you tried it, you were like, "That is really amazing." But the work that happened over the last thousand, or at least many hundreds of years, of people working incredibly hard to get these hard-won scientific insights, and then to build the engineering and the companies and the complex supply chains and kind of reconfigure the world, that had to happen to get this, like, rack of magic put somewhere. Think about all the stuff that went into that, the, you know, the-- and trace it all the way back to people that were just, like, digging rocks out of the ground and seeing what happened. Um, so that you now get to just, you know, type something into ChatGPT, and it does something for you.

Andrew Mayne 28:08 I read a behind-the-scenes story about the development of Project Stargate and the international partnerships, particularly the UAE, and that Elon Musk had tried to derail that. And what have you seen? What have you heard? What's the take on that?

Sam Altman 28:26 I had said, I think also e- externally, but at least internally, after the election, that I didn't think Elon was going to abuse his power in the government to unfairly compete. And I regret to say I was wrong about that. I mean, I don't like being wrong in general-

Andrew Mayne 28:47 Right

Sam Altman 28:47 ... but mostly I just think it's really unfortunate for the country that he would do these things, and I didn't think-- I genuinely didn't think he was going to. Um, I'm grateful that the administration has really done the right thing and stuck up to that kind of behavior. Um, but yeah, it sucks.

Andrew Mayne 29:10 Well, I think the thing that's changed, and I think, uh, Greg Brockman just talked about this, where there was a couple of years ago where people thought, like, "Okay, whoever gets there first is the winner, and that's it, and the game is over." And now we realize there are great AI labs elsewhere. Like, Anthropic is building great tools. I think Google's really got its game up. There's good stuff happening everywhere, and it's not gonna be that one person runs away with it.

Sam Altman 29:32 Yeah, I agree.

Andrew Mayne 29:33 And so it seems-

Sam Altman 29:34 I, I-- yeah, the, the, the example that I like the most is the discovery of AI was analogous to this, not perfect, but close, um, to the discovery of the transistor in many surprising number of ways. Uh, but many companies are gonna build great things on that, and then eventually, it's gonna, like, seep into almost all products. But you won't think about using transistors all the time. So yeah, I think a lot of people are gonna build really successful companies built on this incredible scientific discovery, and I wish Elon would be less zero-sum about it.

Andrew Mayne 30:06 Uh, yeah, I, I think-

Sam Altman 30:07 Or negative sum.

Andrew Mayne 30:07 I think the pie is just gonna get bigger and bigger if we think about that. I was just at an energy conference, and it was interesting talking to the people who were involved in energy production and stuff, and hyperscaling, the term they use for this, was a topic. Um, and that does bring up, like, the energy requirements. I know that for, like, Grok 3, apparently, I guess they had to put generators in the parking lot to be able to train that model. And that's the question, is like, how-- where is the energy gonna come from? Money, I understand. Energy, to think of when we talk about the scale of energy needed.

Sam Altman 30:38 I think kinda everywhere.

Andrew Mayne 30:39 Right.

Sam Altman 30:39 I, I think it's a big mix right now. U- eventually, I think a lot of-... I'm very excited about advanced nuclear, both fission and fusion. Um, but for now, I think it's, it's a whole mix of the entire portfolio.

Andrew Mayne 30:54 Right.

Sam Altman 30:54 Gas, solar, I mean, really, nuclear, everything.

Andrew Mayne 30:57 So all the above and stuff. Yeah, I was talking to people that were-- some of them worked in areas like in Alberta, where they said, "We have a lot of access to energy and not as much use for it there," et cetera. And that was, it's just this, uh, total picture I hadn't even thought about.

Sam Altman 31:09 You know, traditionally, it's very hard to move energy around the world, most kinds. Um, but if you exchange energy for intelligence and then move the intelligence around the world, it's much easier. So you could put the giant training center or the, even the big inference clusters in a lot of places, and then just, like, ship the output over the internet.

Andrew Mayne 31:26 There was a speaker at, at OpenAI. I came to an event, and it was somebody

Andrew Mayne 31:30 who was working, I think it was the James Webb Space Telescope, and he talked about his biggest bottleneck was they're about to get all of this, you know, terabytes of data, but he doesn't have enough scientists to work on it. Doesn't have enough people to go through the data. And here we- we have these-

Sam Altman 31:42 Yeah

Andrew Mayne 31:42 ... answers about the universe, whatever, in front of us, and it's, like, a big data problem.

Sam Altman 31:46 Yeah, I, um, I've always joked that one thing we should do when we have enough money, when OpenAI has enough money, is just build a gigantic particle accelerator- [chuckles]

Andrew Mayne 31:55 [chuckles]

Sam Altman 31:56 - and solve high-energy physics once and for all. Um, 'cause I think that'd be, like, a triumphant, wonderful thing. But I wonder, what are the odds that a really, really smart AI could look at the data we currently have-

Andrew Mayne 32:12 Mm-hmm

Sam Altman 32:12 ... with no more data, no bigger particle accelerator, and just figure it out? It's not impossible.

Andrew Mayne 32:17 Yeah.

Sam Altman 32:18 A- a- and yeah, so there's this question of, like, okay, there's already a lot of data out there. There's a lot of smart people in the world, but we don't know how far intelligence can go. With no more experiments, how much more could we figure out?

Andrew Mayne 32:30 Yeah, I remember reading something that had talked about how in early 1990s, somebody had found, like, a form of Ozempic, all right, and presented it to, like, a drug company and said this. And they said, "Nah, we're gonna pass on that." And that's been a life-changing drug for people, like, for people who've just basically chronic obesity, whatever.

Sam Altman 32:45 Yeah.

Andrew Mayne 32:45 It's gonna improve the quality of life, and you think, "Oh, this was sitting there for twenty-five years?"

Sam Altman 32:49 I suspect there's a lot of other examples that we'll find where maybe we already have existing drugs that we know do something good, but they- they're reusable in some other big way, or with a couple of small modifications, we are very close to something great. Um, and it's been very heartening to hear from scientists using the, even the current generation models for this kind of work.

Andrew Mayne 33:13 So it sounds like one of the things we're gonna need, though, for next-generation models is models that understand physics and chemistry and stuff. Is Sora sort of a stab at that?

Sam Altman 33:21 I mean, it'll understand, like, Newtonian physics. I don't know if it'll help us with discovering new chemistry and sort of, like, new, like, novel physics, uh, or novel theoretical physics or whatever you'd like, but I think I'm optimistic that the techniques we use for the reasoning models-

Andrew Mayne 33:38 Mm-hmm

Sam Altman 33:38 ... will help us with those things a lot.

Andrew Mayne 33:40 Okay. And what is the short definition of how a reasoning model works versus just me asking GPT-4.1 something?

Sam Altman 33:50 So the GPT models can reason a little bit, and in fact, one of the, one of the things that got people really excited in the early days of the GPT models was you could get better performance by telling the model, "Let's think step by step."

Andrew Mayne 34:08 Mm-hmm.

Sam Altman 34:09 And it would then just output text that was thinking step by step and get a better answer, which was sort of amazing that that worked at all. The reasoning models are just pushing that much further.

Andrew Mayne 34:20 So it's the idea of, like, when it's able to break the question down, it can spend more time on each step.

Sam Altman 34:26 When you ask me something, a question, I-- if it's a really easy question, I might just fire back, like, almost on reflex with the answer. But if it's a harder question, I might think in my head and have, like, my internal monologue go and say, "Well, I could do this or that, or maybe, maybe, you know, this will be clearer. I'm not sure about that," and I could, like, backtrack and retrace my steps, and then when I finish thinking, and I've, you know, been thinking in English, I can then, you know, make some bullet points and then kind of, like, output an answer to you in English.

Andrew Mayne 34:56 One, one of the interesting things I've observed now when I use the app, if I ask a deep research question or something, and I go away, on my lock screen, I get the, "It's still processing and thinking about it." And I heard somebody, another company, I forget who it was, was using a metric of how long something spent-- I think it was Anthropic, like said, "Hey, this model actually spent, like, fifteen minutes or thirty minutes, or whatever length of time, to think about a thing," which is a good metric by, but it needs to actually give you the right answer.

Sam Altman 35:20 Yeah.

Andrew Mayne 35:20 And I thought that was sort of just interesting paradigm of-

Sam Altman 35:23 One thing I have been surprised by is people are surprisingly willing to wait for a great answer-

Andrew Mayne 35:29 Yeah

Sam Altman 35:29 ... even if the model's gonna think a while. All of my instincts have been, you know, the instant response is the thing that matters, and users hate to wait. And for a lot of stuff, that's true. But for hard problems with a really good answer, people are quite willing to wait.

Andrew Mayne 35:42 Yeah. So we have all these tools, all these things. So far, I'm using my phone, and now OpenAI just announced that you guys are building hardware. You had the video with you and Jony Ive talking about you guys been talking about and collaborating for a couple years. Uh, obviously, you can't... I mean, well, I could ask you, is, is, is it on you right now?

Sam Altman 36:00 No, it is not.

Andrew Mayne 36:01 All right.

Sam Altman 36:01 It's gonna be a while.

Andrew Mayne 36:02 Okay.

Sam Altman 36:02 Um, we're gonna try to do something at, like, a crazy high level of quality.

Andrew Mayne 36:06 Okay.

Sam Altman 36:06 And that, that does not come fast. But computers, software and hardware, just the way we think of current computers, were designed for a world without AI, and now we're in, like, a very different world, and what you want out of hardware and software is changing quite rapidly. Um, you might want something that is way more aware of its environment, that has way more context in your life. You might wanna interact with it in a different way than, like, typing and looking at a screen. And we've been exploring that for a while, and we've got a couple of ideas we're really quite excited about. Um...... I think it will take time for people to get used to what it means to use a computer in this kind of a world, uh, 'cause it is so different now. But if you, like, really trusted an AI to understand all the context of your life and your question, and make good judgments on your behalf, where you could, like, have it sit in a meeting, listen to the whole meeting, know what it was, like, allowed to share with who and what it shouldn't share with anyone, and, you know, kind of what your preferences would be, and then you ask it one question, and you trust that it's gonna go do the right follow-ups with the right people and do-- Like, you can then imagine a totally different kind of how you use a computer to get done what you want.

Andrew Mayne 37:33 So kind of the way we interact with ChatGPT has kinda, kind of informed the device?

Sam Altman 37:38 I mean, you could also say that the way we interact with ChatGPT was informed by the previous generation of devices.

Andrew Mayne 37:44 Mm-hmm.

Sam Altman 37:44 So I, I think it is this sort of like co-evolving thing. But yeah, I hope so.

Andrew Mayne 37:49 One of the things that made the phone so ubi- ubiquitous was the fact that I can be in public and look at the screen. I can be in private and have a phone call and talk to it, and I think that's one of the challenges for new devices, is that trying to bridge that-

Sam Altman 38:03 Yeah

Andrew Mayne 38:03 ... gap between what we use in public and private.

Sam Altman 38:05 Phones are unbelievable things. I mean, they are really fantastic for a lot of reasons. Uh, and you can imagine one new device that you could use everywhere, but also, like, there's some things that I do do differently publicly and privately.

Andrew Mayne 38:20 Mm-hmm.

Sam Altman 38:20 Like at home, I've got great stereo system to listen to music, and when I'm walking in the world, I use AirPods, and that doesn't bother me.

Andrew Mayne 38:27 Yeah.

Sam Altman 38:27 Um, so I think there are things that are different in the public and private use case, but the general purposeness, I agree, is important.

Andrew Mayne 38:35 Yeah. Agree with it. So, uh, nothing yet until maybe next year?

Sam Altman 38:40 It's gonna be a while.

Andrew Mayne 38:41 All right.

Sam Altman 38:41 It will be worth the wait, I hope, but it's gonna be a while.

Andrew Mayne 38:43 Okay. I'm, I'm excited and curious,

Andrew Mayne 38:45 have thoughts, so... If you're, uh, giving advice to a 25-year-old right now, what do you tell them?

Sam Altman 38:50 I mean, the obvious tactical stuff is probably what you'd expect me to say, like, learn how to use AI tools. It's, it's funny how quickly the world went from telling, you know, the average 20-year-old to 25-year-old, "Learn to program"-

Andrew Mayne 39:03 Mm-hmm

Sam Altman 39:03 ... to, "Programming doesn't matter. Learn to use AI tools." I wonder what will be next, but of course, there will be something next. Um, but that's, that's very good tactical advice. And then on the sort of like broader front, um, I believe that skills like resilience, adaptability, creativity, figuring out what other people want, uh, I think these are all surprisingly learnable. And it's not as easy as, say, like, go practice using ChatGPT, but it is doable, and those are the kind of skills that I think will pay off a lot in the next, you know, couple of decades.

Andrew Mayne 39:41 And would you say same thing, a 45-year-old, is just learn how to use it in your role now?

Sam Altman 39:47 Yeah, probably.

Andrew Mayne 39:49 Whenever we have whatever your personal definition of AGI, will more people be working for OpenAI after then or before?

Sam Altman 39:56 More.

Andrew Mayne 39:56 More. So, yeah, I, I see a lot of online people like, "Ah, they're, they're so good. Why are they hiring people?" I'm like, " 'Cause computers can't do everything. They're not gonna do everything."

Sam Altman 40:06 The slightly longer answer with more than one word is that, um, there will be more people, but each of them will do vastly more than what one person did, you know, in the pre-AGI times.

Andrew Mayne 40:19 Right. Which is the goal of technology.

Sam Altman 40:21 Yeah.