# How To Build The Future: Sam Altman Markdown copy of https://realalpha.app/v/xXCBz_8hM9w/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy. Sam Altman says AI gives startups an unusual edge: move faster than incumbents, but still build a durable business. - Show: Lightcone - Hosts: [Garry Tan](https://realalpha.app/people/garry-tan/) - Guest: [Sam Altman](https://realalpha.app/builders/sam-altman/) is CEO of OpenAI. - Published: Nov 8, 2024 - Length: 47 min ## The best minutes - [13:42](https://realalpha.app/v/xXCBz_8hM9w/?t=822) [OpenAI bet on deep learning and scale before the field agreed](https://realalpha.app/s/openai-bet-on-deep-learning-and-scale-before-the-field/) (6 min) - [26:13](https://realalpha.app/v/xXCBz_8hM9w/?t=1573) [GPT-4 demand showed when models became commercially useful](https://realalpha.app/s/gpt-4-demand-showed-when-models-became-commercially-useful/) (3 min) - [40:33](https://realalpha.app/v/xXCBz_8hM9w/?t=2433) [Current models may enable innovation sooner than expected](https://realalpha.app/s/current-models-may-enable-innovation-sooner-than-expected/) (4 min) ## Also worth watching - [22:28](https://realalpha.app/v/xXCBz_8hM9w/?t=1348) [Conviction should give way when the evidence arrives](https://realalpha.app/s/conviction-should-give-way-when-the-evidence-arrives/) (3 min) - [44:12](https://realalpha.app/v/xXCBz_8hM9w/?t=2652) [AI gives startups an edge when they move faster than incumbents](https://realalpha.app/s/ai-gives-startups-an-edge-when-they-move-faster-than/) (2 min) - [31:10](https://realalpha.app/v/xXCBz_8hM9w/?t=1870) [A failed startup can still compress years of learning](https://realalpha.app/s/a-failed-startup-can-still-compress-years-of-learning/) (3 min) - [7:00](https://realalpha.app/v/xXCBz_8hM9w/?t=420) [Founder peers can raise your ambition more than school](https://realalpha.app/s/founder-peers-can-raise-your-ambition-more-than-school/) (4 min) - [34:04](https://realalpha.app/v/xXCBz_8hM9w/?t=2044) [A prior exit helped Altman finance OpenAI's first steps](https://realalpha.app/s/a-prior-exit-helped-altman-finance-openais-first-steps/) (2 min) Skip the opening ASI and abundance discussion, the YC Research backstory, Loopt's initial product story, and OpenAI's leadership changes. ## Full analysis: Sam Altman says GPT-3.5 drew buyers before GPT-4 GPU requests - Altman said GPT-3 built no great businesses, with a possible copywriting exception. - GPT-3.5 drew buyers, then GPT-4 brought GPU requests. - That supports customer pull, not lasting economics, so founders should test repeat use and business value. Its 6 sections, what it means for you and what to watch are on [the page](https://realalpha.app/v/xXCBz_8hM9w/#analysis), free with an account: - [19:24](https://realalpha.app/v/xXCBz_8hM9w/?t=1164) Altman says one focused bet helped OpenAI compete - [23:04](https://realalpha.app/v/xXCBz_8hM9w/?t=1384) Altman says conviction should expire when evidence arrives - [27:47](https://realalpha.app/v/xXCBz_8hM9w/?t=1667) Altman says customer demand separated demos from products - [41:06](https://realalpha.app/v/xXCBz_8hM9w/?t=2466) Altman thinks current models may accelerate innovation - [44:43](https://realalpha.app/v/xXCBz_8hM9w/?t=2683) Altman says startups can exploit speed against larger firms - [45:25](https://realalpha.app/v/xXCBz_8hM9w/?t=2725) Altman says faster execution still needs durable value ## Summary Sam Altman is CEO of OpenAI. He started Loopt at 19, later worked at YC, and says he wrote early checks for OpenAI. Altman argues that rapid AI progress makes this an unusually good time to start a company, because focused startups can move faster than larger organizations. He traces OpenAI’s early conviction in deep learning and scaling, while emphasizing that conviction must give way to evidence when results contradict assumptions. He says current models are enabling more ambitious work, but founders still need to build a real business and an enduring advantage. He also describes OpenAI’s recent organizational changes and says clearer research, infrastructure, and product priorities now help the company move faster. ### Minute by minute - [0:56](https://realalpha.app/v/xXCBz_8hM9w/?t=56) **AI gives startups a timely opening.** Altman calls this the best time yet to start a technology company, since major shifts give upstarts an edge over larger companies that move slowly. He says ASI could be thousands of days away if recent progress compounds, while acknowledging that progress could hit a wall. He links abundant intelligence and energy to advances in both ideas and physical work. - [6:22](https://realalpha.app/v/xXCBz_8hM9w/?t=382) **Peers expand what founders attempt.** Altman recalls joining YC as a Stanford sophomore after Paul Graham suggested he wait a year. He says the ability to simply try things resonated with him, and that YC’s peer group mattered even more than encouragement from an individual: founders should seek ambitious peers early. - [10:48](https://realalpha.app/v/xXCBz_8hM9w/?t=648) **OpenAI began with a bold research agenda.** Altman says YC Research appealed to him as a way to fund smart people and try research projects, though AI was not an obvious winner. He and the early team, including Greg Brockman and Ilya, made pursuing AGI part of OpenAI’s pitch, which attracted young researchers despite criticism. Early plans included unsupervised learning and reinforcement learning. - [17:08](https://realalpha.app/v/xXCBz_8hM9w/?t=1028) **A focused bet beat spreading resources.** Altman says the early belief that deep learning would improve with scale was controversial, and that the team initially had a hunch before later results made the gains more predictable. With fewer resources than DeepMind and others, OpenAI chose to concentrate on what seemed to work instead of spreading its effort across many bets. - [21:40](https://realalpha.app/v/xXCBz_8hM9w/?t=1300) **Conviction must respond to evidence.** Altman says founders often have to act before they have data, but should change course once results show they are wrong. OpenAI made many incorrect assumptions and explored areas such as robots and game-playing agents before GPT-3. He credits Alec Radford’s work on a sentiment-related neuron in generated Amazon reviews with pointing toward the GPT series. - [26:13](https://realalpha.app/v/xXCBz_8hM9w/?t=1573) **Customers validated GPT-4.** Garry Tan describes how Case Text founder Jake Heller used GPT-4 in legal workflows after GPT-3.5 hallucinated too much, and says Heller sold the company for $650 million. Altman says GPT-3 had few major businesses beyond possible copywriting uses, while GPT-3.5 drew more buyers. Demand for GPUs after GPT-4 access helped confirm its commercial potential. - [28:53](https://realalpha.app/v/xXCBz_8hM9w/?t=1733) **Loopt taught Altman about platform shifts.** Altman says he started Loopt at 19 because mobile phones were becoming computers people could carry. He says Loopt never found product-market fit or escape velocity, but the experience taught him a great deal about startups and how messy platform shifts can be. - [33:11](https://realalpha.app/v/xXCBz_8hM9w/?t=1991) **Startups can move ahead of incumbents.** Altman says many established companies are still underestimating AI, creating an opening for startups. He and Garry Tan point to mobile as a past platform shift that incumbents struggled to adapt to. Altman says it helped that he could write early checks for OpenAI, though he also says he might have chosen a better project than Loopt. - [36:48](https://realalpha.app/v/xXCBz_8hM9w/?t=2208) **Clearer priorities can speed OpenAI.** Altman says OpenAI compressed the changes of a much larger tech-company lifecycle into a short period, bringing painful transitions and management changes. He hopes for a calmer stretch but expects further change. He says the company now has clearer research and infrastructure paths, a clearer product direction, and a stronger basis for aligning teams. - [39:56](https://realalpha.app/v/xXCBz_8hM9w/?t=2396) **Current models can support innovation.** Garry Tan describes a YC hackathon project from Camfer that used O1 to iteratively improve an airfoil until it produced competitive lift. Altman says the demo and others changed his view of how difficult it would be to reach the next level of AI-enabled innovation. He outlines a progression from chatbots and reasoners to agents, innovators, and systems operating at the scale of an organization. - [44:12](https://realalpha.app/v/xXCBz_8hM9w/?t=2652) **Build with AI, then build a business.** Altman advises founders to build with AI because models are not near saturation and startups can react faster than companies tied to quarterly planning. He warns that using AI is not itself a moat: founders still need a better product, a competitive edge, and enduring value. Garry Tan adds that demos are easy; building a business remains the harder goal. ## Transcript Lightcone, hosted by Garry Tan, with Sam Altman. Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there. [0:00](https://realalpha.app/v/xXCBz_8hM9w/?t=0) We said from the very beginning we were going to go after AGI at a time when in the field you weren't allowed to say that because that just seemed impossibly crazy. I remember a rash of criticism for you guys at that moment. We really wanted to push on that and we were far less resourced than DeepMind and others and so we said okay they're going to try a lot of things and we've just got to pick one and really concentrate and that's how we can win here. most of the world still does not understand the value of like a fairly extreme level of conviction on one bed. That's why I'm so excited for startups right now. It is because the world is still sleeping on all of this to such an astonishing degree. [0:43](https://realalpha.app/v/xXCBz_8hM9w/?t=43) We have a real treat for you today. Sam Altman, thanks for joining us. Thanks, Garry. This is actually a reboot of your series, How to Build the Future. And so welcome back to the series that you started. That was like eight years ago. I was trying to think about that, something like that. That's wild. I'm glad it's being rebooted. That's right. Let's talk about your newest essay on the age of intelligence. You know, is this the best time ever to be starting a technology company? Let's at least say it's the best time yet. Hopefully there'll be even better times in the future. I sort of think with each successive major technological revolution, you've been able to do more than you could before. And I would expect the companies to be more amazing and impactful and everything else. So yeah, I think it's the best time yet. Big companies have the edge when things are like moving slowly and not that dynamic. And then when something like this or mobile or the internet or semiconductor revolution happens, or probably like back in the days of the industrial revolution, that was when upstarts have their edge. So yeah, this is like, and it's been a while since we've had one of these. So this is pretty exciting. In the essay, you actually say a really big thing, which is ASI, super [1:53](https://realalpha.app/v/xXCBz_8hM9w/?t=113) intelligence, is actually thousands of days away. Maybe. I mean, that's our hope, our guess, whatever. But that's a very wild statement. Yeah. Tell us about it. I mean, that's big. That is really big. I can see a path where the work we are doing just keeps compounding and the rate of progress we've made over the last three years continues for the next three or six or nine or whatever. You know, nine years would be like 3,500 days or whatever. If we can keep this rate of improvement or even increase it, like that system will be quite capable of doing a lot of things. I think already even a system like a one is capable of doing like quite a lot of things It was from just like a raw cognitive IQ on a closed end, well-defined task in a certain area. [2:45](https://realalpha.app/v/xXCBz_8hM9w/?t=165) I'm like, oh, one is like a very smart thing. And I think we're nowhere near the limit of progress. I mean, that was an architecture shift that sort of unlocked a lot. And what I'm sort of hearing is that these things are going to compound. We could hit some like unexpected wall or we could be missing something. But it looks to us like there's a lot of compounding in front of us still to happen. I mean, this essay is probably the most techno-optimist of almost anything I've seen out there. Some of the things we get to look forward to, fixing the climate, establishing a space colony, the discovery of all of physics, near limitless intelligence, and abundant energy. [3:24](https://realalpha.app/v/xXCBz_8hM9w/?t=204) I do think all of those things, and probably a lot more we can't even imagine, are maybe not that far away. And one of, and I think it's like tremendously exciting that we can talk about this even semi-seriously now. One of the things that I always have loved most about YC is it encourages slightly implausible degrees of techno-optimism and just a belief that like, ah, you can figure this out. And, you know, in a world that I think is like sort of consistently telling people, this is not going to work, you can't do this thing, you can't do that. I think the kind of early PG spirit of just encouraging founders to like think. [3:59](https://realalpha.app/v/xXCBz_8hM9w/?t=239) a little bit bigger is like it is a special thing in the world the abundant energy thing seems like a pretty big deal you know there's sort of path a and path b you know if we do achieve abundant energy seems like this is a real unlock almost any work not just you know knowledge work but actually like real physical work yeah could be unlocked with robotics and with language and intelligence on tap like there's a real age of abundance i think these are like the key to in the two key inputs to everything else that we want there's a lot of other stuff of course that matters but the unlock that would happen if we could just get truly abundant intelligence truly abundant energy what we'd be able to make happen in the world like both like come up with better ideas more quickly and then also like make them happen in in the physical world like to say nothing of it'd be nice to be able to run lots of AI and that takes energy too. I think that would be a huge unlock. And the fact that it's, I'm not sure it to be whether, like whether to be surprised that it's all happening at the same time, or if this is just like the natural effect of an increasing rate of technological progress, but it's certainly a very exciting time to be alive and great time [5:17](https://realalpha.app/v/xXCBz_8hM9w/?t=317) to do a startup. Well, so we sort of walked through this age of abundance, you know, maybe robots can actually manufacture, do anything, almost all physical labor can then result in material progress, not just for the most wealthy, but for everyone. You know, what happens if we don't unleash unlimited energy, if, you know, there's some physical law that prevents us from exactly that? Solar plus storage is on a good enough trajectory that even if we don't get a big nuclear breakthrough we would be like okay-ish but for sure it seems that driving the cost of energy down the abundance of it up has like a very direct impact on quality of life and eventually we'll solve every problem in physics so we're going to figure this out it's just a question of when and we deserve it uh there's you know someday we'll be talking not about fusion or whatever, but about the Dyson sphere. And that'll be awesome too. [6:17](https://realalpha.app/v/xXCBz_8hM9w/?t=377) Yeah. This is a point in time, whatever feels like abundant energy to us will feel like not nearly enough to our great grandchildren. And there's a big universe out there with a lot of matter. Yeah. I wanted to switch gears a little bit to sort of your earlier, you were mentioning Paul Graham, who brought us all together, really created Y Combinator. He likes to tell the story of how, you know, how you got into YC was actually you were a Stanford freshman. And he said, you know what, this is the very first YC batch in 2005. And he said, you know what, you're a freshman and YC will still be here next time. You should just wait. And you said, I'm a sophomore and I'm coming. And you're widely known in our community as, you know, one of the most formidable people. [7:03](https://realalpha.app/v/xXCBz_8hM9w/?t=423) where do you think that came from that one story i think i i would happy i'd be happy if that like drifted off well now it's it's purely immortalized here here it is my memory of that is that like i needed to reschedule an interview one day or something um and pg tried to like say like i'll just do it next year or whatever and then i think i said some nicer version of i'm a sophomore and I'm coming. But yeah, you know, these things get slightly apocryphal. It's funny, I don't, and I say this with no false modesty, I don't like identify as a formidable person at all. In fact, I think there's a lot of ways in which I'm really not. I do have a little bit of a just like, I don't see why things have to be the way they are. And so I'm just going to like do this thing that from first principles seems like fine. [8:00](https://realalpha.app/v/xXCBz_8hM9w/?t=480) always felt a little bit weird about that. And then I remember one of the things I thought was so great about YC and still that I care so much about YC about is it was like a collection of the weird people who are just like, I'm just going to do my thing. The part of this that does resonate as a like accurate self-identity thing is I do think you can just do stuff or try stuff a surprising amount of the time. And I think more of that is a good thing. And then I think one of the things that both of us found at YC was a bunch of people who all believed that you could just do stuff. For a long time, when I was trying to like figure out what made YC so special, I thought that it was like, okay, you have this like very amazing person telling you, you can do stuff I believe in you. And as a young founder, that felt so special and inspiring. And of course it is. But the thing I didn't understand until much later was it was the peer group of other people doing that. [8:59](https://realalpha.app/v/xXCBz_8hM9w/?t=539) **Sam Altman:** And one of the biggest pieces of advice I would give to young people now is finding that peer group as early as you can was so important to me. And I didn't realize it was something [9:11](https://realalpha.app/v/xXCBz_8hM9w/?t=551) that mattered. I kind of thought, ah, like I have, you know, I'll figure it out on my own. But man, being around like inspiring peers, so, so valuable. What's funny is both of us did spend time at Stanford. I actually did graduate, which is, I probably shouldn't have done that, but I did. Stanford's great. You pursued the path of, you know, far greater return by dropping out. But, you know, that was a community that purportedly had a lot of these characteristics, but I was still beyond surprised at how much more potent it was with a room full of founders. It was, I was just going to say the same thing. Actually, I liked Stanford a lot. [9:48](https://realalpha.app/v/xXCBz_8hM9w/?t=588) Yeah. But I did not feel surrounded by people that made me like want to be better and more ambitious and whatever else. And to the degree I did, the thing you were competing with your peers on was like who was going to get the internship at which investment bank, which I'm embarrassed to say I fell on that trap. This is like how powerful peer groups are. It's a very easy decision to not go back to school after like seeing what the like YC vibe was like. Yeah. There's a powerful quote by Carl Jung that I really love. It's, you know, the world will come and ask you who you are. And if you don't know, it will tell you. It sounds like being very intentional about who you want to be and who you want to be around as early as possible is very important. [10:35](https://realalpha.app/v/xXCBz_8hM9w/?t=635) Yeah, this was definitely one of my takeaways, at least for myself, is no one is immune to peer pressure. And so all you can do is like pick good peers. Yeah. Obviously, you went on to create Looped, sell that, go to Green Dot, and then we ended up getting to work together at YC. Talk to me about the early days of YC research. One of the really cool things that you brought to YC was this experimentation. I remember you coming back to partner rooms and talking about some of the rooms that you were getting to sit in with the Larry and Sergeys of the world. and that, you know, AI was sort of at the tip of everyone's tongue because it felt so close, and yet it was, you know, that was 10 years ago. [11:18](https://realalpha.app/v/xXCBz_8hM9w/?t=678) The thing I always thought would be the coolest retirement job was to get to, like, run a research lab. And it was not specific to AI at that time. When we started talking about YC Research, well, not only was it going to, it did end up funding, like, a bunch of different efforts. and I wish I could tell the story of like if I was obvious that AI was gonna work and be the thing but like we tried a lot of bad things too. Around that time I read a few books on like the history of Xerox Park and Bell Labs and stuff and I think there were a lot of people like it was in the era of Silicon Valley at the time that we need to like have good research labs again and I just thought it would be so cool to do. And it was sort of similar to what YC does in that you're going to allocate capital to smart people and sometimes it's going to work and sometimes it's not going to. And I just wanted to try it. AI for sure was having a mini moment. This was like kind of late 2014, 2015, early 2016 was like this super intelligence discussion, like the book superintelligence was happening. [12:26](https://realalpha.app/v/xXCBz_8hM9w/?t=746) Bostrom, yep. Yeah. The DeepMind had had a few impressive results, but a little bit of a different direction. I had been an AI nerd forever, so I was like, oh, it'd be so cool to try to do something, but it's very hard to say. Was ImageNet out yet? ImageNet was out. Yeah. Yeah. For a while at that point. So you could tell if it was a hot dog or not. You could sometimes. Yeah, that was getting there, yeah. you know how did you identify the initial people you wanted involved in you know yc research and open ai i mean greg greg brockman was early in retrospect it feels like this movie montage and there were like all of these like you know at the beginning of like the bank eyes movie when you're like driving around to find the people and whatever and they're like you son of a bitch i'm in right right like ilia i like heard he was really smart and then i watched some video of his. And he's extremely smart, like true, true, genuine genius and visionary. But also he has this incredible presence. And so I watched this video of his on YouTube or something. I was like, I got to meet that guy. And I emailed him and he didn't respond. So I just like went to some conference he was speaking at and we met up. And then after that, we started talking a bunch. [13:33](https://realalpha.app/v/xXCBz_8hM9w/?t=813) And then like Greg, I had known a little bit from the early Stripe days. What was that conversation like, though? It's like, I really like what your ideas about AI and I want to start a lab. Yes. And one of the things that worked really well in retrospect [13:49](https://realalpha.app/v/xXCBz_8hM9w/?t=829) **Sam Altman:** was we said from the very beginning we were going to go after AGI at a time when in the field you weren't allowed to say that [13:57](https://realalpha.app/v/xXCBz_8hM9w/?t=837) because that just seemed impossibly crazy and, you know, borderline irresponsible to talk about. So that got his attention immediately. It got all of the good young people's attention and the derision, whatever that word is, of the mediocre old people. and I felt like somehow that was like a really good sign and really powerful and we were like this ragtag group of people I mean I was the oldest by a decent amount I was like I guess I was 30 then and so you had like these people who were like those are these irresponsible young kids who don't know anything by anything and they're like saying these ridiculous things and the people who that was really appealing to I guess are the same kind of people who would have said like it's a you know I'm a sophomore and I'm coming or whatever and they were like let's just do this thing let's take a run at it and so we kind of went around and met people one by one and then in different configurations of groups and it kind of came together over the course of in fits and starts but over the course of like nine months and then it started i mean and then it started started happening and one of my favorite like memories of all of open ai was ilia had some reason that google or something that we couldn't start in we announced in [15:09](https://realalpha.app/v/xXCBz_8hM9w/?t=909) December of 2015 but we couldn't start until January of 2016 so like January 3rd something like that of 2016 or like very early in the month people come back from the holidays and we go to Greg's apartment maybe there's 10 of us something like that and we sit around and it felt like we had done this monumental thing to get it started and everyone's like so what do we do now and what a great moment it reminded me of when startup founders work really hard to like raise around and they think like oh i accomplished this we did it we did it and then you sit down and say like fuck now i gotta like figure out what we're gonna do it's not time for popping champagne that was actually the starting gun and now we gotta run yeah and you have no idea how hard the race is going to be it took us a long time to figure out what we're gonna do um but one of the things that I'm... [16:00](https://realalpha.app/v/xXCBz_8hM9w/?t=960) Really amazingly impressed by Ilya in particular, but really all of the early people about, is although it took a lot of twists and turns to get here, the big picture of the original ideas was just so incredibly right. And so they were like up on like one of those flip charts or white boys, I don't remember which, in Greg's apartment. And then we went off and, you know, did some other things that worked or didn't work or whatever, and some of them did, and eventually now we have this like system. And it feels very crazy and very improbable looking backwards that we went from there to here with so many detours on the way, but got where we were pointing. [16:41](https://realalpha.app/v/xXCBz_8hM9w/?t=1001) Was deep learning even on that flip chart initially? Yeah. I mean, more specifically than that, like do a big unsupervised model and then solve RL was on that flip chart. One of the flip charts from a very, this is before Greg's apartment, but from a very early offsite. I think this is right. I believe there were three goals for the effort at the time. It was like, figure out how to do unsupervised learning, solve RL, and never get more than 120 people. Missed on the third one. But the predictive direction of the first two is pretty good. So deep learning, then the second big one sounded like scaling. Like the idea that you could scale. That was another heretical idea that people actually found even offensive. [17:26](https://realalpha.app/v/xXCBz_8hM9w/?t=1046) You know, I remember a rash of criticism for you guys at that moment. When we started, yeah, the core beliefs were deep learning works and it gets better with scale. And I think those were both somewhat heretical beliefs. At the time, we didn't know how predictably better it got with scale. That didn't come for a few years later. it was a hunch first and then you got the data to show how predictable it was but people already knew that if you made these neural networks bigger they got better yeah um like that was we were sure of that yeah um before we started and what took the like word that keeps coming to mind is like religious level of belief was that that wasn't going to stop everybody had some reason of oh it's not really learning it's not really reasoning it can't really do this it's you know it's like a parlor trick and these were like the eminent leaders of the field and more than just saying you're wrong they were like you're wrong and this is like a bad thing to believe or a bad thing to say it was that there's gotta you know this is like you're gonna perpetuate an AI winter you're gonna do this you're gonna do that and we were just like looking at these results and saying they keep getting better then we got the scaling results [18:40](https://realalpha.app/v/xXCBz_8hM9w/?t=1120) it just kind of breaks my intuition even now and at some point you have to just look at the scaling laws and say we're going to keep doing this and this is what we think it'll do and it also it was starting to feel at that time like something about deep learning was just this emergent phenomenon that was really important and even if we didn't understand all of the details in practice yet which obviously we didn't and still don't that there was something really fundamental going on it was the pgism for this is we had like discovered a new square in the periodic table yeah and so it we just we really wanted to push on that and we were far less resourced than deep mind and others and so we said okay they're gonna try a lot of things and we've just got to pick one and really concentrate and that's how we can we can win here which is totally the right startup takeaway and so we said well we don't know we don't know we do know this one thing works so we're going to really concentrate on that and I think some of the other efforts were trying to outsmart themselves in too many ways and we just said we'll just we'll do the thing in front of us and keep pushing on it. Scale is this thing that I've always been interested in [19:54](https://realalpha.app/v/xXCBz_8hM9w/?t=1194) at kind of just the emergent properties of scale for everything for startups. For deep learning models, for a lot of other things, I think it's a very underappreciated property and thing to go after. And I think it's, you know, when in doubt, if you have something that seems like it's getting better with scale, I think you should scale it up. I think people want things to be, you know, less is more, but actually... More is more. More is more. We believed in that. We wanted to push on it. I think one thing that is not maybe that well understood about OpenAI is we had just this... Even when we were like pretty unknown, we had a crazy talented team of researchers. [20:32](https://realalpha.app/v/xXCBz_8hM9w/?t=1232) You know, if you have like the smartest people in the world, you can push on something really hard. Yeah, and they're motivated. And or you created sort of one of the sole places in the world where they could do that. Like one of the stories I heard is just even getting access to compute resources. even today is this crazy thing and embedded in some of the criticism from maybe the elders of the industry at the moment was sort of that, you know, you're going to waste a lot of resources and somehow that's going to result in an AI winter. Like people won't give resources anymore. It's funny. People were never sure if we were going to waste resources or if we were doing something kind of vaguely immoral by putting in too much resources and you were supposed to spread it across lots of bets rather than like conviction on one. Most of the world still does not understand the value of like a fairly extreme level of conviction on one bet. And so we said, okay, we have this evidence. We believe in this thing. We're gonna, at a time when like the normal thing was we're gonna spread against this bet and that bet and that bet. You're a definite optimist. [21:35](https://realalpha.app/v/xXCBz_8hM9w/?t=1295) You're a definite optimist. And I think across like many of the successful YC startups, you see a version of that again and again. Yeah, that sounds right. When the world gives you sort of push back and the pushback doesn't make sense to you, you should do it anyway. Totally. One of the many things that I'm very grateful about getting exposure to from the world of startups is how many times you see that again and again and again. And before, I think before YC, I really had this deep belief that somewhere in the world, there were adults in charge, adults in the room, and they knew what was going on. And someone had pushing back on you, they probably knew what was going on. And the degree to which I now understand that, you know, to pick up the earlier phrase, you can just do stuff. You can just try stuff. [22:23](https://realalpha.app/v/xXCBz_8hM9w/?t=1343) No one has all the answers. There are no like adults in the room that are going to magically tell you exactly what to do. And you just kind of have to like iterate quickly and find your way. That was like a big unlock in life for me to understand. There is a difference between being high conviction just for the sake of it and if you're wrong you don't adapt and you don't try to be like truth-seeking it still is really not that effective the thing that we tried to do was really just believe whatever the results told us and really kind of try to go do the thing in front [22:58](https://realalpha.app/v/xXCBz_8hM9w/?t=1378) **Sam Altman:** of us and there were a lot of things that we were high conviction and wrong on but as soon as we realized we were wrong we tried to like fully embrace it conviction is great until the moment [23:08](https://realalpha.app/v/xXCBz_8hM9w/?t=1388) you have data one way or the other and there are a lot of people hold on it past the moment of data so it's iterative it's not just they're wrong and i'm right you have to go show your work but there is a long moment where you have to be willing to operate without data and at that point you do have to just sort of run on conviction yeah it sounds like there's a focusing aspect there too like you had to make a choice and that choice had better you know you didn't have infinite choices and so you know the prioritization itself was an exercise that made it much more likely for you to succeed i wish i could go tell you like oh we knew exactly what was going to happen and it was you know we had this idea for language models from the beginning and you know we kind of went right to this but obviously the story of open ai is that we did a lot of things that helped us develop some scientific understanding, but... [24:00](https://realalpha.app/v/xXCBz_8hM9w/?t=1440) We're not on the short path. If we knew then what we know now, we could have speed run this whole thing to like an incredible degree. It doesn't work that way. Like you don't get to be right at every guess. And so we started off with a lot of assumptions, both about the direction of technology, but also what kind of company we were going to be and how we were going to be structured and how AGI was going to go and all of these things. And we have been like humbled and badly wrong many, many, many times. and one of our strengths is the ability to get punched in the face and get back up and keep going this happens for scientific bets for uh you know being willing to be wrong about a bunch of other things we thought about how the world was going to work and what the sort of shape of the product was going to be again we had no idea or i at least had no idea maybe alec radford did i had no idea that language models were going to be the thing um you know we started working on robots and agents playing video games and all these other things then a few years later gpt3 happened that was not so obvious at the time yeah it sounded like there was a a key insight around positive or negative sentiment around ngp1 even before gpt1 oh before he i think the paper was [25:14](https://realalpha.app/v/xXCBz_8hM9w/?t=1514) called the unsupervised sentiment neuron and i think alec did it alone by the way alec is this unbelievable outlier of a human. And so he did this incredible work, which was just looking at, he noticed there was one neuron that was flipping positive or negative sentiment as it was doing these generative Amazon reviews, I think. Other researchers might have hyped it up more or made a bigger deal out of it or whatever, but it was Alex, so it took people a while to, I think, fully internalize what a big deal it was. And he then did GPT-1 and somebody else scaled it up into GPT-2. But it was off of this insight that there was something amazing happening where, and at the time, unsupervised learning was just not really working. So he noticed this one really interesting property, which is there was a neuron that was flipping positive or negative with sentiment. And yeah, that led to the GPT series. I guess one of the things that Jake Heller from Case Text, I think of him as maybe, I mean, not surprisingly, a YC alum who got access to both 3, 3.5, and 4. And he described getting 4 as sort of the big moment revelation. Because 3.5 would still do yeah i mean it would hallucinate more than he could use in a legal setting [26:40](https://realalpha.app/v/xXCBz_8hM9w/?t=1600) and then with four it reached the point where if he chopped the prompts down small enough into workflow he could get it to do exactly what he wanted and he built you know huge test cases around it and then sold that company for 650 million dollars so it's uh you know i think of him as like one of the first to commercialize gpt4 in a relatively grand fashion i remember that conversation with him yeah with one gpt4 like that was one of the few moments in that thing where i was like okay we have to be really great on our hands um when we first started trying to like sell gpt3 to founders they would be like it's cool it's doing something amazing it's an [27:22](https://realalpha.app/v/xXCBz_8hM9w/?t=1642) **Sam Altman:** incredible demo but with the possible exception of copywriting no great businesses were built on gpt3 and then 3.5 came along and people startups like yc startups in particular started to do [27:37](https://realalpha.app/v/xXCBz_8hM9w/?t=1657) interesting like it no longer felt like we were pushing a boulder uphill it's like people actually wanted to buy the thing we were selling totally and then four we kind of like got the like just how many GPUs can you give me? Moment, like very quickly after giving people access. So we felt like, okay, we got something like really good on our hands. So you knew actually from your users then? Totally. Like when the... model dropped itself and you got your hands on it. It was like, well, this, this is better. We were totally impressed then too. We had all of these like tests that we did on it that were very, it looked great. And it could just do these things that we were all super impressed by. Also, like when we were all just playing around with it and like getting samples back, I was like, wow, it's like, it can do this now. And there were, it can rhyme and it can like tell a funny joke, slightly funny joke and it can like you know do this and that and so it felt really great but you know you never really know if you have a hit product on your hands until you like put it in customers hands yeah you're always too impressed with your own work yeah and and so we were all excited about it we're like oh this is really quite good but until like the test happens it's [28:46](https://realalpha.app/v/xXCBz_8hM9w/?t=1726) like the real test yeah yeah the real test is users yeah so there's some anxiety until that until that moment happens. Yeah. I wanted to switch gears a little bit. So before you created, obviously, one of the craziest AI labs ever to be created, you started at 19 at YC with a company called Looped, which was basically Find My Friends geolocation, you know, probably, what, 15 years before Apple ended up making it. Too early in any case, yeah. Yeah. What drew you to that particular idea i was like interested in mobile phones and i wanted to do something that got to like use mobile phone this was when like mobile was just starting it was like you know still three years or two years before the iphone but it was clear that carrying around computers in our pockets was somehow a very big deal i mean that's hard to believe now that there was a moment when phones were actually literally you just they were just a phone they were an actual phone yeah yeah i mean i try not to use it as an actual phone ever really i still remember the first phone i got that had internet on it and it was this horrible like text-based mostly text-based browser it was really slow you could like you know do like you could so painfully and so slowly [30:03](https://realalpha.app/v/xXCBz_8hM9w/?t=1803) check your email um but i was like a i don't know in high school sometime in high school when i got a phone that could do that versus like just text and call and i was like hooked right then yeah I was like, this is not a phone. This is like a computer we can carry. And we're stuck with a dial pad for this accident of history. But this is going to be awesome. And I mean, now you have billions of people who they don't have a computer. Like to us growing up, you know, that actually was your first computer. This is a replica or like another copy of my first computer, which is a Mac LC2. Yeah. So this is what a computer was to us growing up. [30:41](https://realalpha.app/v/xXCBz_8hM9w/?t=1841) and the idea that you would carry this little black mirror like kind of we've come a long way unconscionable back then yeah so you know even then you like technology and what was going to come was sort of in your brain yeah i was like a real i mean i still am a real tech nerd yeah but i always that was what i spent my friday nights thinking about and then uh one of the harder parts of it was we didn't have the app store the iphone didn't exist uh you ended up being a big part of that launch, I think. A small part, but yes, we dig it to be a little part of it. It was a great experience for me to have been through because I kind of like understood what it is like to go through a platform shift and how messy the beginning is and how much like little things you do can shape the direction it all goes. [31:27](https://realalpha.app/v/xXCBz_8hM9w/?t=1887) I was definitely on the other side of it then. Like I was watching somebody else create the platform shift, but it was a super valuable experience to get to go through and sort of just see how it happens and how quickly things change and how you adapt through it. What was that experience like? You ended up selling that company. It was probably the first time you were managing people and, you know, doing enterprise sales. All of these things were useful lessons from that first experience. I mean, it obviously it was not a successful company. It was and so it's a very painful thing to go through but the rate of experience and education was incredible another thing that pg said or quoted somebody else saying but always stuck with me is your 20s are always an apprenticeship but you don't know for what and then you do your real work later and i did learn quite a lot and i'm very grateful for it it was like a difficult experience and we never found product market fit really and we also never like really found a way [32:26](https://realalpha.app/v/xXCBz_8hM9w/?t=1946) **Sam Altman:** to get to escape velocity, which is just always hard to do. There is nothing that I have ever heard of that has a higher rate of generalized learning than doing a startup. So it was great [32:37](https://realalpha.app/v/xXCBz_8hM9w/?t=1957) in that sense. You know, when you're 19 and 20, like riding the wave of some other platform shift, this shift from, you know, dumb cell phones to smartphones and mobile. And, you know, here we are many years later and your next act was actually you know i mean i guess two acts later literally spawning one of the major platforms we all get old yeah but that's really what's happening you know uh 18 20 year olds are deciding that they could get their degree but they're gonna miss the wave like because all of this stuff that's great everything's happening right now i am proud what is do you have an intuitive sense like speaking to even a lot of the you know really great billion-dollar company founders, some of them are just not that aware of what's happening. [33:26](https://realalpha.app/v/xXCBz_8hM9w/?t=2006) They're CTOs. It is astonishing to me. It's wild, right? That's why I'm so excited for startups right now, is because the world is still sleeping on all of this to such an astonishing degree. And then you have the YC founders being like, no, no, I'm going to do this amazing thing and do it very quickly. Yeah. It reminds me of when Facebook almost missed mobile, because they were making web software and they were really good at it. Yeah. And like they, I mean, they had to buy Instagram. Like Snapchat. And WhatsApp. Yeah. And WhatsApp. So it's interesting. The platform shift is always built by the people who are young with no prior knowledge. [34:07](https://realalpha.app/v/xXCBz_8hM9w/?t=2047) I think it's great. So there's this other aspect that's interesting in that I think you and Elon and Bezos and a bunch of people out there like they sort of start their journey as founders you know really you know whether it's looped or zip2 or you know really in maybe pure software like it's just a different thing that they start and then later they you know sort of get to level up you know is there a path that you recommend at this point if people are thinking you know i want to work on the craziest hard tech thing first should they just run towards that to the extent they can or Or is there value in, you know, sort of solving the money problem first, being able to invest your own money, like, very deeply into the next thing? [34:53](https://realalpha.app/v/xXCBz_8hM9w/?t=2093) It's a really interesting question. [34:55](https://realalpha.app/v/xXCBz_8hM9w/?t=2095) **Sam Altman:** It was definitely helpful that I could just, like, write the early checks for OpenAI. [35:00](https://realalpha.app/v/xXCBz_8hM9w/?t=2100) And I think it would have been hard to get somebody else to do that at the very beginning. And then Elon did it a lot at a much higher scale, which I'm very grateful for. And then other people did after that. and and there's other things that i've invested in that i'm really happy to have been able to support and i don't i think it would have been hard to get other people to to do it um so that's great for sure and i did like we were talking about earlier learn these extremely valuable lessons but i also feel like i kind of like was wasting my time for lack of a better phrase working on looped i don't i definitely don't regret it it's like all part of the tapestry of life and i learned a ton and whatever else what would you have done differently or what would you tell yourself from like now to in a time cap in like time travel capsule that would show up on your desk at stanford when you were 19. well it's hard because ai was always the thing i most wanted to do and ai just like i went to school to study ai but at the time i was working in the ai lab the one thing. [36:00](https://realalpha.app/v/xXCBz_8hM9w/?t=2160) They told you it was definitely don't work on neural networks. We tried that. It doesn't work. That's a long time ago. I think I could have picked a much better thing to work on than Loopt. I don't know exactly what it would have been, but it all works out. It's fine. Yeah. There's this long history of people building more technology to help improve other people's lives. And I actually think about this a lot. Like I think about the people that made that computer and I don't know them. You know, there are many of them probably long retired, but I am so grateful to them. Yeah. And some people worked super hard to make this thing at the limits of technology. I got a copy of that on my eighth birthday and it totally changed my life. And the lives of a lot of other people too. They worked super hard. They never like got a thank you from me but I feel it to them very deeply. And it's really nice to get to like add our brick to that long road of progress. Yeah. It's been a great year for OpenAI, not without some drama. Always. Yeah. We're good at that. [37:03](https://realalpha.app/v/xXCBz_8hM9w/?t=2223) What did you learn from, you know, sort of the ouster last fall? And how do you feel about some of the, you know, departures? I mean, teams do evolve, but how are you doing, man? Tired, but good. Yeah. It's, we've kind of like speed run, like medium size or even kind of like pretty big sized tech company arc that would normally take like a decade and two years like chat gpt is less than two years old yeah and and there's like a lot of painful stuff that comes with that um and there are you know any company as it scales goes through management teams at some rate uh and you have to sort of the people who are really good at the zero to one phase are not necessarily people that are good at the one to ten or the ten to the hundred phase we've also kind of like changed what we're going to be made plenty of mistakes along the way done a few things really right and that comes with a lot of change and I think the goal of the company the immersion AGI or whatever however you want to think about it is like just keep making the best decisions we can at every stage but it does lead to a lot of change I hope that we are heading towards a period now of more calm, but I'm sure there will be other periods in the future where [38:25](https://realalpha.app/v/xXCBz_8hM9w/?t=2305) things are very dynamic again. So I guess, how does OpenAI actually work right now? You know, I mean, the quality and like the pace that you're pushing right now, I think is like beyond world class compared to a lot of the other, you know, really established software players like who came before this is the first time ever where i felt like we actually know what to do like i think from here to building an agi will still take a huge amount of work there are some known unknowns but i think we basically know what to go what to go do and it'll take a while it'll be hard but that's tremendously exciting i also think on the product side there's more to figure out but roughly we know what to shoot at and what we want to optimize for that's a really exciting time and when you have that clarity i think you can go pretty fast yeah if you're willing to say we're going to do these few things we're going to try to do them very well and our research path is fairly clear our infrastructure path is fairly clear our product path is getting clearer you can orient around that super well we for a long time did not have that we were a true research lab and even when you know that it's hard to act with the conviction on it [39:45](https://realalpha.app/v/xXCBz_8hM9w/?t=2385) because there's so many other good things you'd like to do but the degree to which you can get everybody aligned and pointed at the same thing is a significant determinant in how fast you can move I mean sounds like we went from level one to level Well, too, very recently, and that was really powerful. And then we actually just had our O1 hackathon at YC. Yeah, that was so impressive. That was super fun. And then weirdly, one of the people who won, I think they came in third, was Camfer. And so CAD CAM startup, you know, did YC recently last year or two. And they were able to, during the hackathon, build something that would iteratively improve an airfoil from something that wouldn't fly to literally something that had a competitive amount of lift. And I mean, that sort of sounds like level four, which is, you know, the innovator stage. It's very funny you say that. [40:43](https://realalpha.app/v/xXCBz_8hM9w/?t=2443) I had been telling people for a while, I thought that the level two to level three jump was going to happen. But then the level three to level four jump was, level two to level three was going to happen quickly. And then the level three to level four jump was somehow going to be much harder [40:57](https://realalpha.app/v/xXCBz_8hM9w/?t=2457) **Sam Altman:** and require some medium-sized or larger new ideas and that demo and a few others have convinced me that you can get a huge amount of innovation just by using these current models in really creative ways well yeah i mean it's uh what's interesting is basically camfer already built [41:19](https://realalpha.app/v/xXCBz_8hM9w/?t=2479) sort of the underlying software for CAD CAM. And then, you know, language is sort of the interface to the large language model, which then can use the software-like tool use. And then if you combine that with the idea of CodeGen, that's kind of a scary, crazy idea, right? Like not only can the large language model code, but it can create tools for itself and then compose those tools similar to, you know, Chain of Thoughts with O1. Yeah, I think things are going to go a lot faster than people are appreciating right now. Yeah. Well, it's an exciting time to be alive, honestly. You know, you mentioned earlier that thing about discover all of physics. [42:03](https://realalpha.app/v/xXCBz_8hM9w/?t=2523) I always wanted to be a physicist. I wasn't smart enough to be a good one. Had to, like, contribute in this other way. But the fact that somebody else, I really believe, is now going to go solve all the physics with this stuff, like, I'm so excited to be alive for that. let's get to level four so happy for whoever that person is yeah do you want to talk about level three four and five briefly yeah so we realized that agi had become this like badly overloaded word and people meant all kinds of different things and we tried to just say okay here's our best guess roughly of the order of things you have these level one systems which are these chatbots there'd be level two that would come which would be these these reasoners we think we got there earlier this year with the 01 release. Three is agents' ability to go off and do these longer-term tasks, you know, maybe, like, multiple interactions with an environment, asking people for help when they need it, working together, all of that. And I think we're going to get there faster than people expect. Four is innovators, like, that's, like, a scientist, and you know that's ability to go explore like a not well understood phenomena over like a long [43:13](https://realalpha.app/v/xXCBz_8hM9w/?t=2593) period of time and understand what's just kind of go just figure it out and then and then level five this is the sort of slightly amorphous like do that but at the scale of the whole company or you know a whole organization or whatever that's going to be a pretty powerful thing yeah and it It feels kind of fractal, right? Like even the things you had to do to get to two sort of rhyme with level five and that you have multiple agents that then self-correct that work together. I mean, that kind of sounds like an organization to me, just at like a very micro level. Do you think that we'll have, I mean, you famously talked about it. [43:48](https://realalpha.app/v/xXCBz_8hM9w/?t=2628) I think Jake talks about it. It's like you will have companies that make, you know, billions of dollars per year and have like less than 100 employees, maybe 50, maybe 20 employees. maybe one. It does seem like that. I don't know what to make of that other than it's a great time to be a startup founder. Yeah. But it does feel like that's happening to me. Yeah. You know, it's like one person plus 10,000 GPUs. Could happen. Pretty powerful. Sam, what advice do you have for people watching who, you know, either are about to start or just started their startup? Bet on this tech trend. Like, bet on this trend. [44:28](https://realalpha.app/v/xXCBz_8hM9w/?t=2668) It's, this is, we are not near the saturation point. The models are going to get so much better so quickly. What you can do as a startup founder with this versus what you could do without it is so wildly different. And the big companies, even the medium-sized companies, even the startups that are a few years old, they're already on quarterly planning cycles. And Google is on a year or decade planning cycle. I don't know how they even do it anymore. But your advantage with speed and focus [44:54](https://realalpha.app/v/xXCBz_8hM9w/?t=2694) **Sam Altman:** and conviction and the ability to react to how fast the technology is moving, that is the number one edge of a startup kind of ever, but especially right now. So I would definitely like build [45:07](https://realalpha.app/v/xXCBz_8hM9w/?t=2707) something with AI and I would definitely like take advantage of the ability to see a new thing and build something that day rather than like put it into a quarterly planning cycle. I guess the other thing I would say is it is easy when there's a new technology platform to say, well, because I'm doing some of the AI, the laws of business don't apply to me. I have this magic technology and so I don't have to build a moat or a competitive edge or a better product. It's because I'm doing AI and you're not. So that's all I need. And that's obviously not true. But what you can get are these short-term explosions of growth by embracing a new technology more quicker than somebody else. And remembering not to fall for that and that you still have to build something of enduring value. I think that's a good thing to keep in mind too. Yeah. Everyone can build an absolutely incredible demo right now, but... Everyone can build an incredible demo. [46:04](https://realalpha.app/v/xXCBz_8hM9w/?t=2764) But building a business, man, that's the brass ring. The rules still apply. You can do it faster than ever before and better than ever before, but you still have to build a business. What are you excited about in 2025 what's to come AGI yeah excited for that uh what am I excited for um or any kid I'm more excited for that then congratulations I've ever been incredible yeah probably that that's but that's gonna be really that's the thing I've like most excited for ever in life yeah it changes your life completely so I cannot wait well here's to building that better world for you know our kids and really hopefully the whole world there's a lot of fun Thanks for hanging out, Sam. [46:43](https://realalpha.app/v/xXCBz_8hM9w/?t=2803) Thank you.