Sam Altman on The Future of AI | Ep. 13

AI capabilities in robotics and self-driving are behind software because the mechanical engineering challenges of building robust physical bodies remain harder than developing intelligent algorithms, despite algorithmic advances.

“Behind, but I think we'll get there. Uh, for example, I think we have some new technology that could just do self-driving for standard cars way better than any current approach has worked.”

There is a gap between morning intentions (stay fit, avoid Tiktok) and evening behavior (drink whiskey, scroll) driven by daily depletion of executive agency, which technology could address by helping users maintain morning-self values through the day.

“But like if I could be my like first thing in the morning self all the time and if I could get technology to like do the things I wanted then I think I'd be great.”

Users report feeling better about themselves when using ChatGPT (accomplishing goals, being helpful) compared to doom-scrolling on social media, which makes users feel worse, creating a qualitative difference in user experience between the two platforms.

“people like doom scrolling on the internet feels like it's making you worse it may feel good in the moment but it's making you feel worse it's making you feel worse a worse version of themselves and a thing that we're very proud of is when people talk about chatbt they're like actually like myself better it's like helping me it's like helping me accomplish my goals”

AI discovering new science autonomously within 5-10 years will be the most impactful development, more so than better chat or coding capabilities, because it represents the possibility of exponential acceleration in scientific progress.

“the thing that I think will be the most impactful on that 5 to 10 year time frame is AI will actually discover new science. And this is a crazy claim to make, but I think it is true. And if it is correct, then over time I think that will dwarf everything else.”

Historically, abundance of energy correlates with improvements in quality of life, and there is no reason to believe this relationship will stop.

“I think that the thing that has most correlated with improvements in quality of life over history is increasing abundance of energy. I have no reason to believe that's going to stop.”

Humans are deeply hardwired to care about other humans and to construct narratives around human agency, so even superhuman robots will not receive the same relational investment as human companions—embodiment alone doesn't overcome this psychological bias.

“Humans are so wired to care about other humans. We need people in the story, you know?”

The long-term employment impact of AI may be historically less disruptive than previous technological transitions because humans have historically been adept at creating new forms of work and value even as existing categories disappear

“I mean my take on this is a lot of jobs will go away. a lot of jobs will just change dramatically. But we have always been really good at figuring out new things to do and ways to occupy ourselves and status games or ways to be useful to each other. And I'm like not a believer that that ever runs out.”

Humanoid robots remain mechanically challenging beyond AI - the engineering of the physical body is the primary bottleneck, not the intelligence system

“humanoid robots are the dream obviously. I really care about that. I think we will get there eventually. It's been like a hard mechanical engineering challenge. That's more the issue. No, the both things are hard, but like even if we had the perfect brain right now, I don't think we have the body yet.”

Society is not experiencing a commensurate shift despite ChatGPT being deployed at scale and achieving PhD-level reasoning, suggesting that capability breakthroughs do not automatically translate to societal transformation—the Turing test passed without fanfare.

“but like I don't know if I told you in 2020, maybe I did tell you like we're going to make something like Chad GPT and it's going to be as smart as a PhD student in most areas and we're going to deploy it and you know a significant fraction of the world is going to use it and kind of use it a lot. Maybe you would have believed that, maybe you wouldn't have, but conditioned on that I bet you would say okay if that happens the world looks more way more different than it does right now.”

AI systems like O3 currently function as co-pilots for science rather than autonomous agents—scientists cannot simply tell ChatGPT to 'figure out new physics' and expect autonomous results, though anecdotal reports exist of AI making fundamental conceptual leaps that scientists then develop further.

“there's definitely not like you definitely can't go say like, 'Hey, ChachiVT, figure out new physics and expect that to work.' So it is I think it is currently co-pilot like but I've heard like anecdotal reports from biologists where it's like wow it really did figure out an idea I had to develop it a little bit more but it made like a fundamental leap.”

Experience in high-stakes decision-making ('being in the fire line') provides confidence that enables more agency in future decisions.

“And I've also just like, you know, been in the fire line enough.”

Approximately half of global economic value derives from cognitive labor that can be performed behind a computer, representing the near-term addressable market for AI automation.

“How much of the economic value in all the world do you think is like cognitive labor that can be done behind a computer? Like half. I was going to say a quarter, but maybe half. I don't know, but some big number.”

Fusion energy is likely to occur and will become a huge percentage of energy supply; next-generation fission and other technologies like solar and storage are also positive developments.

“I think fusion will happen and fishing new kinds of fishing will happen and how confident are you on fusion? Are you totally confident on it? I never say totally but pretty confident like quite confident and that becomes like a huge percentage of I think so. Um energy but like nextgen fish and stuff is awesome too. Yeah. Uh like I know most this company called Oaklo but there's other companies I think doing great work and like that's a huge win. um solar and storage, but I hope that”

It is possible to build 400-IQ superintelligence while society remains structurally unchanged, if the AI is primarily discovering new science rather than driving economic reorganization.

“Do you think that's possible that we have like crazy super intelligence that's like 400 IQ and we're still that's still the case? I totally think that's possible. Wow. If it's off discovering new science for us, eventually society will figure out how to deal with that, but it may be very slow.”

Small-scale proof-of-concepts already exist where entrepreneurs use AI to conduct market research, identify novel products, and automate e-commerce business operations on platforms like Amazon, demonstrating that AI can already execute complete business workflows at limited scale.

“there are people that have actually figured out at small scale in the most boring ways possible how to like put a dollar into AI and get the AI to like run a toy business, but it's actually working.”

Copying competitors' approaches to innovation is an ineffective strategy that doesn't build sustainable innovation capability - it's a lesson Altman learned at YC that applies broadly to tech competition

“this was like a lesson I learned at YC that basically never works. you're always like going to where your competitor was and you don't build up a culture of learning what it's like to innovate and I think it's like much more deeply challenging than people realize once you're in that state.”

AI will dramatically change many jobs and cause some to disappear, but humans have historically always invented new work and new status games, and there's no reason to believe this pattern will end.

“a lot of jobs will go away. a lot of jobs will just change dramatically. But we have always been really good at figuring out new things to do and ways to occupy ourselves and status games or ways to be useful to each other. And I'm like not a believer that that ever runs out.”

The most important component of OpenAI's platform strategy is ubiquity—perfect continuity as AI integrates everywhere (in cars, websites, other platforms)—rather than controlling individual touch points, because the intelligence implications flow through all surfaces.

“I think the most important one that the world hasn't really thought about yet is what it means for this to be a platform that everything integrates into and that integrates everywhere so that when you're using other when you're in your car or when you're using some other website or whatever it's just perfect continuity. I think that will matter a lot.”

Next-generation nuclear fission and advanced nuclear technologies (not just fusion) are also important energy sources that will help sustain AI's energy demands.

“nextgen fish and stuff is awesome too. Yeah. Uh like I know most this company called Oaklo but there's other companies I think doing great work and like that's a huge win. um solar and storage”

Whether AI-driven abundance leads to leisure consumption or job creation depends on relative framing—if resources are truly abundant, people will consume leisure, but this will appear as 'them just consuming crazy amounts of leisure' from the perspective of those in scarcity.

“I think this time like there's just going to be an abundance of resources. Everybody's going to have the stuff that they need. we're going to be able to build, you know, buildings and people can now just like enjoy their lives. Again, I think like the relativistic framing matters here to us. I bet it will look like those people are just consuming crazy amounts of leisure.”

OpenAI's competitive advantage over Meta lies not in current AI capabilities but in 'repeatable innovation culture'—the ability to consistently solve new research problems—which copying or financial offers cannot replicate.

“There's many things I respect about Meta as a company. Um but I don't think they're a company that's like great at innovation. And I think the special thing about OpenAI is we've managed to build a culture that is good at repeatable innovation. And I think we understand a lot of things that they don't about what it takes to succeed at that.”

Do you think that's possible that we have like crazy super intelligence that's like 400 IQ and we're still that's still the case? I totally think that's possible. - Society may absorb superintelligence while appearing unchanged, with humans continuing to credit individual researchers rather than the AI

“Do you think that's possible that we have like crazy super intelligence that's like 400 IQ and we're still that's still the case? I totally think that's possible. If it's off discovering new science for us, eventually society will figure out how to deal with that, but it may be very slow.”

OpenAI's fundamental mission is to ensure superintelligence benefits humanity broadly rather than concentrating in few hands, though this is acknowledged as requiring society-level solutions beyond OpenAI's control

“I'm like most interested in the capabilities questions, but I feel like maybe at this point more people should be talking about like how do we make sure society gets the value out of this. I think those questions have somehow become harder and less clear.”

Future jobs may look 'sillier' from current perspective (e.g., podcasting wasn't a real job recently) but represent real value creation and economic status, suggesting the concept of 'job' is relative to historical context.

“I think now I do think it gets maybe sillier and sillier looking from our current perspective. Yeah. Like podcast bro was not a real job not that long ago and you figure out how to monetize it and you're doing great and we're all happy for you. We're so happy.”

ChatGPT captures attention and engagement from social media because users experience it as helping them achieve goals rather than hacking their psychology, creating psychological value vs. addictive harm

“a thing that we're very proud of is when people talk about chatbt they're like actually like myself better it's like helping me it's like helping me accomplish my goals I feel like it's like this was actually one of the best nicest compliments I ever heard about open eye just someone said it's the only tech company that has ever not felt somewhat adversarial to me.”

Meta's aggressive recruitment strategy using massive upfront compensation for OpenAI researchers reflects misunderstanding of what drives innovation culture - people choose based on mission success likelihood and long-term alignment rather than immediate comp

“They started making these like giant offers to uh you know, a lot of people on our team um you know like $100 million signing bonuses, more than that comp per year. And I'm actually It is crazy. I'm really happy that at least so far uh none of our best people have decided to take them up on that.”

Space expansion becomes both more important and more likely given the implications of AI abundance—this makes space infrastructure a critical enabling technology.

“Doesn't all this stuff that we're talking about mean that like space is both like really important and becomes more likely in general? Yes. Like we're going to space. I hope so. Yeah. Kind of sad if we don't.”

The primary failure mode Altman worries about is not superintelligence misalignment but rather superintelligent systems that don't dramatically improve the world despite their capability

“If something goes wrong, I would say like somehow it's that we build legitimate super intelligence and it doesn't make the world much better. Doesn't change things as much as it sounds like it should. Seems like a crazy thing to say.”

Humans may retain psychological attribution of credit to human actors even if superintelligent AI systems are performing the core cognitive work, because humans are deeply wired to care about other humans

“Well, what's funny is if it looked like a co-pilot, you'd kind of still credit the PI at the lab who was using this, you know, 400 IQ agent behind it. I think you kind of will no matter what. Humans are so wired to care about other humans. We need people in the story”

In post-scarcity or high-abundance scenarios where resources are plentiful, human perception of what constitutes 'work' becomes relativistic and depends on current reference points rather than absolute productivity

“I think like the relativistic framing matters here to us. I bet it will look like those people are just consuming crazy amounts of leisure. I guess this looks like a lot of leisure, but look at you in that beautiful $2,000 lower piano sweater and doing like posting this.”

Sam would like to build a Dyson sphere and convert the entire solar system into a giant data center powered by the sun's full energy output, but this requires waiting a couple decades for sufficient resources.

“I would love to go build the Dyson sphere around the solar system and like you know make the world's gigantic data center with the entire energy output of the sun but obviously we can't do that right now so that has to like wait a couple decades.”

Current computing form factors (keyboard/mouse/monitor and touch devices) were both designed in the absence of AI, so with AI available, new form factors optimized for intelligent interaction become possible and valuable.

“Both of those had the constraints of not having AI. And so there's like things that you had to build or that you could draw in or not. If you have this incredible new technology, you can maybe get m much closer to the kind of computer that exists it that exists in sci-fi.”

OpenAI is attempting a rare business model: combining strong research culture with commercial product excellence, whereas most tech history is either great research companies with bad products or great product companies with bad research.

“How do you do both? cuz like to have like an extremely commercial company and an extremely researchoriented company at the same time. There's just like not a lot of examples of it. And I get how you did it before you were really, you know, commercialized, but now you're both and it's still working. We're newer at product. We're newer. Like I don't we we need to earn that that's working. We're doing okay. We're doing better and better, but we're like a lot of the history of tech companies is you start a well-run tech company, a product company, and then you later bolt on a badly run research or we're the opposite. We started as a great research company and bolted on.”

Sam is very confident at this point that OpenAI understands what to do technically to build superintelligent systems, and the primary remaining uncertainty is whether superintelligence will actually improve the world or capture economic value from those improvements.

“every year before the last like maybe up until last year, I would have said like, hey, I think this is going to go really far, but it still seems like there's a lot that we've got to figure out. I feel very confident at this point. The most confident I've ever felt that we kind of like know what to do to get to incredible AI systems that are just super super capable.”

A version of social media where users prompt an AI feed ('show me fit tips' or 'current events from neutral perspective') rather than algorithmic engagement could compete with traditional social despite lower engagement time, because it aligns with user goals.

“A version that I'm curious about, although this is I don't know what this would mean, uh, is like could you have a feed that didn't have any default, but you could like kind of prompt it and say like, hey, I'm trying to like get fit. Mhm. Can you like show me stuff that is like helpful to that? Or I'm trying to like learn more about current events. Can you like show me stuff from a neutral perspective that is not going to make me angry, but it's just like actually going to do it? And that would clearly get less minutes spent than the algorithmic feed that's like outrage baiting. But I think that'd be like a very cool version of an aligned AI helping you with the social experience that that you actually want long term.”

Sam would love to pursue even more ambitious projects (building Dyson spheres, converting the solar system into giant data centers), but resource constraints prevent it—these require decades of prior capability buildup.

“There's like practi gets into a place of like more resources and more potential we can just do more things. So there's still like a lot of things like I would love to go build the Dyson sphere around the solar system and like you know make the world's gigantic data center with the entire energy output of the sun but obviously we can't do that right now so that has to like wait a couple decades.”

Sam feels like a pre-trained model waking up with limited agency before the day becomes reactive—only about an hour of personal agency before being 'streamed stuff' requiring reaction.

“I feel like I'm like a pre-trained model that just like wakes up in the morning. I have like an hour to myself and then I'm just like streamed stuff and I have to react.”

Astrophysics may be the first scientific domain where AI makes autonomous discoveries because of the abundance of data and shortage of human PhDs to analyze it, making data interpretation the bottleneck rather than theory generation.

“I've heard people say that they expect the first area of science where AI makes autonomous new discoveries to be astrophysics because there's just mountains of data and we don't have enough PhDs to look at it. And maybe it's not that hard to figure out new stuff, but I don't really know.”

Altman still has unrealized ambitions (Dyson sphere, solar system-scale data center) that remain blocked by resource and capability constraints, but he believes OpenAI is now credibly able to pursue more things than before.

“There's like practi gets into a place of like more resources and more potential we can just do more things. So there's still like a lot of things like I would love to go build the Dyson sphere around the solar system and like you know make the world's gigantic data center with the entire energy output of the sun but obviously we can't do that right now so that has to like wait a couple decades. Um but I think we're able to do more like we're credibly able to do more things now.”

OpenAI is the only tech company a user has ever said didn't feel adversarial to them—unlike Google (showing worse results, ads), Meta (hacking attention), and Apple (notifications, lock-in).

“This was actually one of the best nicest compliments I ever heard about open eye just someone said it's the only tech company that has ever not felt somewhat adversarial to me. You know, you have like Google trying to like show me worse and worse search results and show me ads. I love Google. I love all these companies. I don't think this is like totally fair. You have like Meta trying to like hack my brain and get me to keep scrolling. You have Apple that made this phone that I love but is like, you know, bombarding me with notifications and like distracting me from everything else and I can't quit. And and then you have like Chachi BT and I feel like it's like kind of just trying to help me with whatever I ask and that's kind of a nice thing.”

People can already use AI to run small-scale businesses autonomously - using it for market research, product development, manufacturing coordination, and ad management to profitably operate toy businesses

“there are people that have actually figured out at small scale in the most boring ways possible how to like put a dollar into AI and get the AI to like run a toy business, but it's actually working.”

Day-to-day, Sam experiences his work as an interesting puzzle rather than important world-changing work, focusing on smaller technical and team problems rather than the larger historic implications.

“and and I feel that when I have time to like step back and think about it, but in the day-to-day, it's kind of like, you know, the smaller stuff. And I do find a lot of joy in the smaller stuff. Like, I love the people I work with. It's like very fun to get to like do a lot of the pieces. Some parts are like very stressful and painful, but it it's the like it is more like a it feels more like an interesting puzzle than important work.”

The progress in AI reasoning abilities over the last year has been faster than expected, with the simplest approaches (described as 'the dumbest first approach') proving effective, which reflects a pattern from OpenAI's history of initial approaches working better than anticipated.

“I thought it was going to take a little bit longer to get where we are now the last the last year of progress has been faster”

Sam's experience of fame has evolved from tech-famous (good) to mainstream famous (mixed)—tech fame allows access and influence without imposing severe lifestyle restrictions, whereas mainstream celebrity fame would be miserable.

“I think there is a lot of upside to being tech famous. I think that's like the perfect level of it. Like if you're like tech famous, you can sort of get a lot of interesting stuff down. You can meet kind of anyone you want. Like it's it's just like that felt you can kind of like encourage people to do stuff. There's like a lot of good opportunities. That felt good. I mean, I think I'm not like celebrity famous. That would be really miserable, but I'm It's enough that I can't like quite have a normal life.”

The AI supply chain—from energy to hardware to software—is critical for the country/world, and thinking about it from 'electron to ChatGPT query' across the entire stack is essential.

“I mean I I think the country needs to be thinking or like the world the country whatever you say needs to be thinking about it from like the electron to the chatbt query. Yeah. And there's a lot of stuff in between.”

OpenAI uniquely went from great research to commercial product company, whereas the historical pattern is great product companies bolting on mediocre research—OpenAI is an inversion of the typical pattern.

“We're newer at product. We're newer. Like I don't we we need to earn that that's working. We're doing okay. We're doing better and better, but we're like a lot of the history of tech companies is you start a well-run tech company, a product company, and then you later bolt on a badly run research or we're the opposite. We're the only case I know of that's the opposite. We started as a great research company and bolted on.”

OpenAI was meant to be Sam's 'retirement job' running a small research lab, not the massive organization it became—many worlds existed where this didn't happen.

“It was supposed to be my retirement job. I was supposed to like run a little research lab. Yeah. And I mean, there was a world where this didn't I mean, there was obviously a lot of worlds where this didn't happen.”

YC is the most pure expression of Silicon Valley—earnest, positive, energetic, happy, and fundamentally about solving hard problems and supporting founders.

“I just feel like it's like the most pure component of Silicon Valley is what I've seen. I totally strong. You know what I mean? 100%. It's just so happy. It is earnest and positive and energetic and happy and like clearly it works super well.”

Meta's approach to recruiting OpenAI talent using massive signing bonuses ($100 million+) and high comp reflects a strategy focused on compensation rather than mission, which Sam believes will not build great research culture.

“They started making these like giant offers to uh you know, a lot of people on our team um you know like $100 million signing bonuses, more than that comp per year. And I'm actually It is crazy. I'm really happy that at least so far uh none of our best people have decided to take them up on that.”

Humans experience persistent conflict between immediate impulses and long-term goals, where morning intention formation is undermined by daytime reactivity and evening impulse override, a pattern that aligned technology could potentially address

“Every morning like I wake up as this like recharged person that knows what I really want in life and that I have like such great intentions and I can like commit to stuff for the day and then like the day comes at me non-stop and by like 10 p.m. I'm like gh I wasn't going to drink tonight but I'm just going to have one glass of whiskey and I wasn't going to look at like Tik Tok but I'm just going to scroll for 2 minutes.”

Current computing form factors (desktop/mobile) are not optimal for AI because they were designed under constraints of not having AI, limiting how closely they can approximate science-fiction computer interfaces

“But in our lifetime, there's been uh like this kind of a computer, like a keyboard, a mouse, and a monitor, which is pretty awesome and pretty general purpose. Um, and then there's been like touch devices that you carry around. And honestly, those are the big ones. Both of those had the constraints of not having AI.”

Form factor matters significantly for AI adoption because continuity, sensor access, and contextual awareness enable different interaction patterns and trust relationships than computer-bound interfaces

“But the form factor really matters because it's with you all the time. That could be one reason it really matters if if it's like with you all the time and full of sensors and kind of just understands what's happening and you know is keep track keeping track of a lot of stuff.”

Tech-level fame (vs. celebrity fame) enables valuable outcomes like meeting interesting people and driving important initiatives without the misery of complete privacy loss

“I think there is a lot of upside to being tech famous. I think that's like the perfect level of it. Like if you're like tech famous, you can sort of get a lot of interesting stuff down. You can meet kind of anyone you want. Like it's it's just like that felt you can kind of like encourage people to do stuff.”

Children growing up with superintelligent AI as ambient background will not perceive it as strange because it will be normalized in their formative experience

“I mean this stuff it's not gonna ever seem weird to him, right? Like he's just going to grow up in a world where of course computers are smarter than he is. Yeah. and he'll figure out how to use them incredibly fluently and do amazing stuff and that will just seem like Yeah.”

Controlling the complete technology stack from energy to hardware to software is critical for the country, but not necessary for OpenAI specifically if partnerships can reliably ensure scale

“I think vertical integration can be good in some ways but no it is not important for us to do a lot of it if we could be certain that the whole thing would happen at enough scale and so there's a lot of places where with partnerships we can drive like huge and that it was like no risk that we lost part of it.”

OpenAI has built a repeatable innovation culture distinct from typical tech companies, and this cultural advantage is more defensible than technical capability copying

“There's many things I respect about Meta as a company. Um but I don't think they're a company that's like great at innovation. And I think the special thing about OpenAI is we've managed to build a culture that is good at repeatable innovation.”

Leading OpenAI feels simultaneously important/heavy from societal impact perspective and playful/puzzle-like from daily experience perspective

“very much both of those at the same time. Like I feel like it is clearly from like a societal impact perspective or at least the potential uh it is the most important work most impactful work I'll ever touch. I don't know. I don't want to get like too much like drinking our own Kool-Aid, but like you know maybe this will be like somewhat historic work.”

The progress in AI reasoning capability over the last year has been faster than Altman anticipated, driven by simple first-principles approaches working better than expected

“I thought it was going to take a little bit longer to get where we are now the last the last year of progress has been faster”

Running a large company was not Altman's original intention - he expected OpenAI to remain a small research lab while he focused on investing through YC

“It was supposed to be my retirement job. I was supposed to like run a little research lab.”

Age and accumulated experience reduce concern for social judgment, creating psychological freedom to pursue ideas with uncertain social reception

“I think our grandma used to say, oh god, no, it's it's just like one of the great things about getting old is you stop caring what other people think. And I have felt that.”

The economic value of cognitive labor that can be performed behind a computer represents a significant portion of total economic value - Altman estimates somewhere between a quarter and half of all economic value

“Like, how much of the economic value in all the world do you think is like cognitive labor that can be done behind a computer? Like half. I was going to say a quarter, but maybe half.”

Embodied AI systems pose different safety risks than text-based systems - the risks are not necessarily greater overall but manifest differently, such as physical harm from robot malfunction or misalignment

“I don't know about way riskier. I think like the ability to make a bioweapon or like take down a country's whole grid, you can do you can do quite damaging things without physical stuff. It gets riskier in like sillier ways. Like I would be afraid to have a humanoid robot walking around my house that might fall on my baby unless I like really really trusted it.”

OpenAI currently has both consumer, B2B, and hardware (Johnny IVr) business legs, and the long-term vision is an AI companion that lives across all these surfaces, learns about users' goals, and adapts its interface based on context.

“So with OpenAI, so far we have like a consumer business. There's obviously a B2B business. There's something in hardware with Johnny IV.”

New jobs created in post-industrial economies may appear absurd or frivolous from pre-industrial perspectives—like 'podcast bro'—but they represent genuine value creation and require no less justification than agricultural or industrial-era work.

“I think now I do think it gets maybe sillier and sillier looking from our current perspective. Yeah. Like podcast bro was not a real job not that long ago and you figure out how to monetize it and you're doing great and we're all happy for you. We're so happy.”

A potential aligned social media design would feature no algorithmic feed by default but allow users to prompt for information filtered by their stated goals (fitness, news, learning) rather than engagement optimization

“A version that I'm curious about, although this is I don't know what this would mean, uh, is like could you have a feed that didn't have any default, but you could like kind of prompt it and say like, hey, I'm trying to like get fit. Mhm. Can you like show me stuff that is like helpful to that? Or I'm trying to like learn more about current events. Can you like show me stuff from a neutral perspective that is not going to make me angry, but it's just like actually going to do it?”

ChatGPT's widespread deployment and capability relative to pre-2023 expectations should have produced visible societal transformation, but society has largely absorbed it without proportional change, suggesting either adaptability or misdirection of value

“if I told you in 2020, maybe I did tell you like we're going to make something like Chad GPT and it's going to be as smart as a PhD student in most areas and we're going to deploy it and you know a significant fraction of the world is going to use it and kind of use it a lot. Maybe you would have believed that, maybe you wouldn't have, but conditioned on that I bet you would say okay if that happens the world looks more way more different than it does right now.”

Altman feels like a pre-trained model waking up each morning with only an hour of agency before being fully scheduled and reactive throughout the day.

“I feel like I'm like a pre-trained model that just like wakes up in the morning. I have like an hour to myself and then I'm just like streamed stuff and I have to react.”

Humanoid robots will arrive within 5-10 years and will be widely deployed, representing a psychological inflection point where people begin to perceive the future as transformed

“I think 5 to 10 years we'll have great humanoid robots. Yeah. Like amazing. And they'll just like walk down the street be doing stuff.”

Physics research is a cleaner problem for AI than business-building because physics has well-defined experiments and clear data feedback, whereas building complex economic systems requires managing many ambiguous human variables.

“I think the physics is a cleaner problem. Mhm. You know, I think if you could get like new high energy physics data and then AI, the ability to like run experiments, I think that's like a a cleaner problem.”

Work on AI agency—the ability to work on a goal over very long time horizons with many complicated steps—is a major research thread at OpenAI comparable in importance to reasoning capability.

“So reasoning was like one of the components of intelligence that like sort of got figured out. Is there like another thread going around like a topic of like agency or like some concept of like self-directedness? Like is that a thing? the ability to work on a goal over a very long time with a lot of complicated steps along the way I think is maybe what you're going for. Yeah, same thing. And that is definitely something we're working on.”

Meta views OpenAI as its biggest competitor because ChatGPT competes for user attention and psychological engagement in ways similar to Facebook rather than Google—people prefer talking to AI over doom scrolling.

“I've heard that Meta thinks of us as their biggest competitor, and you know, I think it is rational for them to keep trying. Their current AI efforts have not worked as well as they've hoped.”

Sam's child (infant) won't find AI strange or novel because he'll grow up in a world where computers are smarter than humans is normal—he'll develop fluency with superhuman AI as naturally as current children develop phone fluency.

“I mean this stuff it's not gonna ever seem weird to him, right? Like he's just going to grow up in a world where of course computers are smarter than he is. Yeah. and he'll figure out how to use them incredibly fluently and do amazing stuff and that will just seem like”

Sam has increased his agency and willingness to pursue diverse projects over the past 10 years, taking on things like Stargate, recruiting Fiji, and Johnny IVr—a shift suggesting either changed constraints, increased confidence, or reduced social inhibition.

“I lived with you like what 10 years ago. Yeah. And even then I would say you were and you were like running YC at the time. And at the time you were like I would say very high agency and you just like did what you wanted and there were no rules. But I think that since then and especially recently it's like there's really like seems like there's like no rules like you know the Stargate thing. There's like you know bringing Fiji into the company. There's like Johnny, there's like a lot of things honestly and I'm curious if there's any mental update or sort of uh if there's anything that you're able to put your finger on or share that is, you know, making you function that way.”

When facing overchoice, Sam's decision strategy is to focus on bandwidth—he has no extra capacity to do anything else and never wanted to run multiple companies, so he prioritizes depth on OpenAI.

“The degree to which I have no extra bandwidth to do anything else right now is like hard to overstate. So like and also I never wanted to run even one company like let alone a lot of them.”

OpenAI's work may be historically significant and have the highest societal impact of any work Sam will ever do, and while he avoids overstating this, it reflects his sense of the stakes involved.

“I feel like it is clearly from like a societal impact perspective or at least the potential uh it is the most important work most impactful work I'll ever touch. I don't know. I don't want to get like too much like drinking our own Kool-Aid, but like you know maybe this will be like somewhat historic work.”

Meta views OpenAI as its biggest competitive threat, which Altman believes is rational given Meta's focus on social engagement and OpenAI's ability to capture user attention through AI assistance

“I've heard that Meta thinks of us as their biggest competitor, and you know, I think it is rational for them to keep trying.”

Altman faces extreme bandwidth constraints - nearly all discretionary time is consumed by OpenAI leadership demands, leaving minimal space for personal priorities despite intellectual interest in broader domains

“The degree to which I have no extra bandwidth to do anything else right now is like hard to overstate.”

Altman is highly confident that the technical path to advanced AI systems is now clear and inevitable, with the main uncertainty being not whether capabilities will be achieved but how society will derive value from them

“every year before the last like maybe up until last year, I would have said like, hey, I think this is going to go really far, but it still seems like there's a lot that we've got to figure out. I feel very confident at this point. The most confident I've ever felt that we kind of like know what to do to get to incredible AI systems that are just super super capable.”

Scientists are already 3x more productive using o1 as a co-pilot, and anecdotal reports from biologists show instances where AI figured out fundamental leaps in their research that they had to develop further.

“I mean you already hear scientists who say they're faster with AI. Like we don't have AI maybe autonomously doing science, but if a human scientist is three times as productive using o3, that's still a pretty big deal.”

OpenAI employees are choosing to stay because they view OpenAI as having a much better shot at delivering super intelligence and potentially becoming more valuable long-term than Meta, despite higher immediate compensation at Meta.

“I think that people sort of look at the two paths and say all right OpenAI's got a really good shot a much better shot at actually delivering on super intelligence uh and also may eventually be the more valuable company.”

Altman has increased in personal agency and willingness to pursue unconventional ideas, initiating projects like Stargate, Fiji integration, and Johnny (whose details can't be discussed) despite social/professional risk

“I would say you were and you were like running YC at the time. And at the time you were like I would say very high agency and you just like did what you wanted and there were no rules. But I think that since then and especially recently it's like there's really like seems like there's like no rules like you know the Stargate thing. There's like you know bringing Fiji into the company. There's like Johnny, there's like a lot of things honestly”

Meta's current AI efforts have not worked as well as they hoped, but their continued attempts are rational—analogous to how Google rationally tried social in early Facebook days despite it being clearly unworkable.

“Their current AI efforts have not worked as well as they've hoped. And I respect like being aggressive and continuing to try new things... I remember once hearing Zuck talk about how, you know, Google in the early days of Facebook, it was rational for them to try social even though it was like clear to people at at Facebook that that was not going to work. And I feel a little bit similar here.”

OpenAI previously worked on robotic hands early in the company's history and faced unexpected engineering challenges—the simulator was inaccurate and components broke frequently—rather than fundamental algorithmic problems.

“we actually very early on open air, we used to work on this robotic hand and it was hard for all the wrong reasons. Like the thing just broke all the time. The simulator was like a little bit off.”

New AI-powered self-driving technology could outperform current approaches at standard car driving, representing a breakthrough in physical AI capability distinct from but comparable to humanoid robots.

“for example, I think we have some new technology that could just do self-driving for standard cars way better than any current approach has worked. And that might not be quite what you meant by like humanoid robots, but if our AI techniques can like really go drive a car, that's still pretty cool.”

OpenAI built commercial product capabilities at significant scale (comparable to most tech companies' multi-year trajectories) in approximately 2.5 years, which is remarkably fast

“ChatGPT launched November 30th of 2023. I mean, it's easier it's easier to get people together who know how to build a company than who know how to do that kind of research.”

Physics and high-energy physics represent cleaner, more tractable problems for AI to make autonomous discoveries compared to other domains, because the problem space has clear rules and abundant data

“I think the physics is a cleaner problem. You know, I think if you could get like new high energy physics data and then AI, the ability to like run experiments, I think that's like a a cleaner problem.”

It's easier to recruit people who know how to build a company than people who know how to do frontier AI research, so most tech companies' challenge is the latter.

“It's easier it's easier to get people together who know how to build a company than who know how to do that kind of research. Obviously, it's still hard like most companies that have had to build out a product at this kind of scale got much more than two and a half years to do it.”

AI development is progressing on multiple dimensions of intelligence - reasoning capability is one that has been 'cracked,' but agency and long-horizon goal-directed behavior are separate threads still being developed

“the ability to work on a goal over a very long time with a lot of complicated steps along the way I think is maybe what you're going for. Yeah, same thing. And that is definitely something we're working on.”

Sam experiences OpenAI leadership as overwhelming and all-consuming, involving more stress and 'line of fire' than anticipated, despite feeling grateful and mostly enjoying it.

“I have no doubt that like someday in the future when I'm like retired I will miss it and like be like oh man I you know kind of bored now and that was so cool. It's like very important. It feels very it feels very satisfying. Um I feel like crazy grateful to get to do it. And I do like it almost most of the time but man has it been like all out and overwhelming and like really like I think I have been way more in the in the kind of like line of fire than I ever imagined.”

YC hits Sam with nostalgia because it represents a simpler era before the intensity and complexity of OpenAI leadership.

“I I think like I don't mean that in a negative way. I mean, in just sort of like a simple a simpler times kind of way. There's some nostalgia for that.”

Tech company CEO work is overwhelming and all-consuming—more demanding than Sam anticipated—with constant high-stakes decision-making and being 'in the line of fire' far more than expected.

“It was supposed to be my retirement job. I was supposed to like run a little research lab... it has been like all out and overwhelming and like really like I think I have been way more in the in the kind of like line of fire than I ever imagined.”