Stanford CS153 Frontier Systems | Scale, AGI, and the Future of Everything

1. 0:09 – 1:29

   Returning to Stanford: from CS183 to the AI-startup era

   1. SPSpeaker0:09

      Please join me in welcoming Sam Altman. [audience applauding] This class was designed as an inspiration from a, you know, from a set of different experiences, uh, while I was a student here. One of them was Terry Winograd's, uh, intro seminar, CS47N, Computers and the Open Society. Uh, but a second one that was a pretty formative experience, uh, for me and a lot of my friends and peers on campus at the time in 2014 was, uh, CS183, How to Start a Startup by Sam. Um, and so it's really cool to have you back. Uh, what's it like? How, how's it feeling for you to be back?

   2. SASam Altman0:48

      I was thinking as I was walking in, if I had just a little more time, I would do, uh, an update to that class because I think everything about starting a startup has changed so much, and I have not seen anyone do a good version of how you're supposed to make a startup now. Uh, so I, I had that, like, just walking in here, I had that, like, ah, it'd be fun to do it again.

   3. SPSpeaker1:06

      So, uh, timeline-wise, you know, you, you taught that in '14. I think OpenAI was founded in 2015. Is that right?

   4. SASam Altman1:13

      '16, basically.

   5. SPSpeaker1:14

      '16. Okay. So, so then you went, you know, it was like you were-- it, it, it felt to me from the, from an observer perspective that you had, like, come up with your working theory for how to do it right, and then you went and tried to implement it. Is that, is that a fair assessment, or is that not the case?

2. 1:29 – 2:25

   OpenAI as an ‘upside-down’ startup: research lab first, product later

   1. SASam Altman1:29

      O-OpenAI was, like, the strangest startup of the last maybe couple of decades in Silicon Valley. Uh, 'cause it started as a research lab. It was, it was really not a company at all.

   2. SPSpeaker1:39

      Right.

   3. SASam Altman1:40

      Um, and that-- the kind of normal course of, of startups is that you start a product company, and then it, like, grows for a while, and then growth slows down, and then you start a research lab, and you, like, bolt that on, and you try to figure out the next thing to do. And we were the opposite of that. We were a research lab first that later had to bolt on a startup.

   4. SPSpeaker1:59

      Right.

   5. SASam Altman2:00

      And, uh, I don't really recommend that. It's kind of an unusual thing. But that, that's not quite what I meant. What I meant is, like, we still followed the pre-AI rules of a startup because-

   6. SPSpeaker2:11

      Mm-hmm

   7. SASam Altman2:11

      ... we were trying to make AI. We didn't have it yet. But now, like, watching what the best startups do is so different than how startups worked even a couple of years ago, um, that I think someone, I'm probably not gonna do it, someone should do that class again.

3. 2:25 – 3:11

   The biggest startup update: token spend as a substitute for huge engineering teams

   1. SPSpeaker2:25

      And what would be the biggest updates you'd, you'd make based on new data?

   2. SASam Altman2:29

      Um, you-- with, like, an affordable amount of spend on tokens, you can do what a hundred-person incredibly great engineering team would do as a startup, and that was just totally impossible. That was, like, not in the set of options for a startup, and now it is. So, so I think what you can take on, uh, the level of ambition you can have, the speed at which you can move, the amount of stuff you can do at once, uh, is just totally different.

   3. SPSpeaker3:00

      And, um, does that change the shape of the problems you feel like you'd assign at the end of the class for people to attack, you know, at the end of that quarter if you were teaching it again?

4. 3:11 – 4:59

   Why you can’t assign great startup ideas: hunt for non-obvious, newly-possible markets

   1. SASam Altman3:11

      I don't think assigning problems to attack ever works because if you, like-- If I can think of a problem, if I can think of, like, a really great startup idea, uh, if it's, like, obvious enough to me, uh, then it's probably obvious to a lot of people. When we started OpenAI, we were, we were, like, the, uh, you know, one of maybe generously speaking, four AGI efforts in the world.

   2. SPSpeaker3:32

      Right.

   3. SASam Altman3:32

      And you wanna find something like that, and I'm sure that there exists something today that just wasn't possible at all pre, like, automated coding era, uh, that is totally not obvious that will be, you know, a multi-trillion-dollar market soon, uh, and that only four companies are working on right now. But I don't know what that is. It's much more likely you all know what that is than I know what that is. I just, you know, my brain has, like, taken over by OpenAI. Um, but, you know, the kind of idea someone can assign you to work on is probably not what you want.

   4. SPSpeaker4:02

      Yep. Um, okay. So that, that's fair. Um, but I, I think it'd be helpful since this is a systems class to maybe, uh, reason about a particular problem that you had to reason through so that they can then apply the shape of the techniques used to break down from a systems perspective that problem to solutions to their own problem.

   5. SASam Altman4:21

      Yeah.

   6. SPSpeaker4:21

      Um, and a, and a concept that, uh, you had started to tease in the class, you know, back in 2014 and then, uh, clearly you've talked about publicly over the years is, um, scale, right? Scale is its own beast. It's, it's, it's, you know, quantity is its own quality. You know, what-what, uh, scale as a concept has been something it seems like you've, um, empirically investigated in all kinds of ways over the last 10 years. So, um, could you help un- help us first unpack, like, what you mean by scale now 10 years later? How would you deconstruct that as a systems design, uh, attribute to apply, whether it's a, as, as a tool? Um, can, can we start there?

5. 4:59 – 8:00

   Scale as a systems principle: emergent properties and underestimated returns

   1. SASam Altman4:59

      Yes. Uh, so I don't know why the following observation is true. I offer no theory that I find satisfying to explain it, and that makes me a little bit nervous to suggest you follow it, but I'm going to anyway because empirically it does seem to be true, which is all of the most interesting things I have observed in my career in watching other, uh, things happen, all of the most interesting ones, uh, have had something to do with emergent properties that scale or scale continuing to provide returns far beyond what the consensus thinks will work. And this obviously happens with, like, scaling laws for AI models, um, but this happens with, uh, you know, getting more smart people together to think about one problem. This ha- in a, in a research setting, um, this happens with, uh, companies and the sort of economy of scale you can get all the, in all these different ways. I really learned this at Y Combinator when, uh, it became clear to me that everybody was saying, "Oh, Y Combinator's gotten too big. It should shrink. We should fund less companies per batch." You know, the best times of Y Combinator were when it was, like, 10 companies per batch. And a lot of, like, very smart people were saying this. And, and it was, like, tempting 'cause it would've been, like, much less work, and the theory was that, you know, the best companies are always kinda obvious, and then you fund the rest, and it's not as helpful. Um, but a huge part of the magic of what made YC work were, uh, was the sort of the network effects inside of the batch, and that was an emergent property at scale that just hadn't been discovered before. No one had tried to fund startups at scale in the same way, and, and thus no one had ever happened upon this observation of when you do that, um, there's, there's something important that happens that just didn't exist at all at the one-tenth or one-one-hundredth of the scale. There's a bunch of other examples like this. Uh, I-- And I'll skip them in the interest of time, but I, I would say, again, I offer no explanation for why, but empirically speaking, when you find a time that you can push on-- you can push something to a scale people have not tried before, and it's already working in some interesting way at the smaller scale, more often than not, that seems to be a good idea. And it also seems to be something that most people don't do enough.

   2. SPSpeaker7:29

      Mm.

   3. SASam Altman7:30

      And I don't of- offer an explanation for this either, but, like, in, you know, when we were like, "We're really gonna scale AI models," um, all of the, like, geniuses in the field, most of them were, "Oh, this isn't really working. You know, that's ne- that's barely a scientific result. It's not interesting that it gets better at scale. You've already shown that. Why keep scaling it?" So I mentioned the YC example. Um, I've seen a lot of startup founders where they're like, "Well, you know, there might be something interesting that would happen if I scaled this up, but I, I'm a little worried about it for non-specific

6. 8:00 – 10:27

   Why scaling is hard: what breaks (technical, capital, culture) and how to decompose it

   1. SASam Altman8:00

      reasons." And again, looking back at, like, a huge data set of people that have scaled their companies in all these different ways, there's almost always interesting stuff there. So I think directionally that's, like, a interesting thing to push on and, and severely underexplored. Um, on the systems design part of that, uh, I think one reason people don't do it as much is stuff breaks, uh, at an accelerating rate and in an unpredictable way as you scale it. And if you are gonna really scale something, um, it's always, like, a little bit broken. There are always, like, very smart people who say why you shouldn't do this. You know, "Don't get too ambitious. Don't get too big. Let's try this smaller." And so breaking that down is a systems problem. I'll use the thing of when we were, like, scaling up AI models. There was, "Technically, can we do this at all? This seems crazy." Like, no one had ever thought about trying to do a run across 10,000 or 100,000 GPUs, and that was gonna require stacks of engineering talent. Um, there was the capital requirements and what it was going to take to do this, and, like, how is there ever gonna be a business? How can you think about taking this risk? Uh, there was the sort of, like, cultural stuff of researchers saying, "Well, if we're gonna get all this compute, why do we put it all into this one project where we're not gonna learn something? Why not divide it up among all these, all these projects?" And this also happens in kind of every area I've looked at, or almost every area for scale. And breaking it down into the sort of each difficult area or each reason not to do it and trying to address them one at a time-

   2. SPSpeaker9:35

      Yeah

   3. SASam Altman9:35

      ... that's been really important.

   4. SPSpeaker9:36

      Um, I- I'm gonna push on that a little bit because there's very few people who've been able to sort of s- repeatedly scale new products and systems the way, uh, the OpenAI team has over the years. But it seems like one of the issues is there, there are all these prior conditioning, uh, sort of mental models and expectations humans have, and you said things break. And one of the things it seems often breaks that's har- the hardest to refactor is, is human, the human side of the, the system's design, right? Wherever there's human implementers or there's, uh, human participants in that. And so what have you learned about humans at scale, like organizing humans at scale to participate in a system that may not be, uh, like just a, a redo of some past system that they, they get naively on at, you know, a priority on first blush?

7. 10:27 – 12:43

   Humans at scale: aligning organizations around clear goals and exponential thinking

   1. SASam Altman10:27

      Um, I think, like, clear, a clear goal, a clear plan to get there, uh, and, like, a clear answer to the way that you're gonna get there and kind of how you're gonna make decisions along the way, that's, that's very important. So, um, y- you know, if we go back to the example of when we decided to scale up models, there were a lot of people who were like, "Ah, this isn't really gonna work. It's gonna have these problems. It's also not, you know, we need a more diversified portfolio." But once we say, "No, we're gonna make a bet on scaling deep learning," like, that's our thing. If we're wrong, we'll fail, but we're gonna do that. Here's why we're gonna do that. Here's what we believe about what the state of the world could be like if we get there. Uh, that's very powerful. And then for whatever reason, um, we did not evolve to be good at thinking about exponentials. People have a hard time imagining that scaling laws are gonna continue exponentially, that revenue will grow exponentially, that an organization can take on exponential complexity. And in my experience, it takes a lot of time to really reason through first principles with people about why, why that can happen.

   2. SPSpeaker11:39

      Can we take two examples, uh, to walk through that? The first being ChatGPT and the second being Codex. You know, both of these have transformed... Can, can everyone hear? I'm gonna try to project it. Yeah? Okay. Um, so let, let me put in a frame, and you can challenge both the assumption, and then we can hopefully reason through example of what happened. In the case of ChatGPT, you know, for a long time in scaling of models, a big mental block that- Seem to be prevalent in the space is wha-what are these things gonna be useful for? This is, you know, it's a research, uh, sort of solution, uh, solution chasing a problem, research first approach. It's not a product. Um, and then, you know, ChatGPT came out and it proved to the world that con- you know, ch- that chat experience was a killer app for general models, um, at scale in, for consumers. And then a couple of years later, you know, s- it's clear that coding has been the killer enterprise app. So wh- what, how would you compare and contrast the systems you guys used to discover those use cases, ship them, scale them, monetize them? Any, any salient learnings from those two systems?

8. 12:43 – 16:41

   ChatGPT’s path: GPT-3 API, user behavior signals, viral breakout, and emergency scaling

   1. SASam Altman12:43

      Yes. Um, so we had made GPT-3, and we needed to make money 'cause we wanted to go scale up to, you know, a billion and multi-billion dollar computers, and we had GPT-3, and it was kind of interesting. It was a cool demo, but we couldn't figure out a product to build around it. And we had been thinking, thinking. We just couldn't do it. We had tried a few things. They, they hadn't worked. Um, and so we knew the models were gonna get better, but we also wanted to, like, start a revenue engine sooner. And we said, "Well, since we can't figure out what product to build, we're just gonna put this into an API, and we're gonna hope that somebody else can figure out what product to build." And so we launched in, like, I don't know, somewhere in the summer of 2020, the GPT-3 API. And initially, it kinda got no traction at all, and then about a month later, randomly as far as we can tell, it went viral on Twitter. On the same day, uh, a few different developers kinda found, got it to do something cool, posted it, other people started trying. And, and then, like, a lot of people started trying the API. Um, but it was shockingly bad. If you go back and use GPT-3 or 3.5, um, you will be astonished at how bad the models were then, uh, relative to the amount of excitement they generated at the time. Uh, so people tried all of these things, and really, the only business that people got to work in a significant way with GPT-3 was copywriting. Um, and that was, like, not that great and not that exciting, and we were kind of like, you know, "Ah, it's just gonna have to wait for a better model." But although n- that was the only business that was working, developers had figured out how to, like, put in a prompt and get, and be able to chat with it. And we saw this a lot. Like, more people were using... They couldn't get the API to work for their business, but they were using their API key to just chat. And we said, "Well, we can build a good chatbot. People clearly want that." And we had a new model. We actually had GPT-4 done, but we had a new model we were ready to release in between called 3.5, and we had figured out a new kind of post-training where we could get the models to do, like, a good job with instruction following so it can make it easier to chat with. And we said, "Well, you know, the API is not working great." Maybe it was like a 10 or a $20 million run rate kinda business. But there is this thing that people love, uh, and under the YC principle of see what your users love and do that, we said, "Well, we'll build a chatbot around it." And we put that out, and we still didn't think it was gonna do that well. Uh, there was-- It was really meant as, like, a research demo, uh, to convince other people that they should build chat-like products and pay us for the API. But that went, like, crazy viral. And another thing I had learned from YC is when something really starts growing and it's not very good, you have, like, a guaranteed hit on your hands. And so we had, like, five days where the traffic would shoot up, fall off, and everybody would be like, "Well, that was just a hype cycle." But then the next day it would get to a higher peak, fall off again later in the day. People would say, "That's a hy-hype cycle." By the fourth or fifth day, I was like, "I know how this works. I know what's gonna happen." Like, we have the potential here-

   2. SPSpeaker15:49

      Hmm

   3. SASam Altman15:50

      ... at a killer product. Um, and we knew we could make it much better. We knew we could, we knew we had GPT-4. We knew we could keep scaling. Um, but by that fifth day, we got everybody together and said, "This is an emergency. This is a good kind of emergency, but we have to build a company and a product all at once."

   4. SPSpeaker16:11

      Hmm.

   5. SASam Altman16:11

      Uh, we then had, like, two months of crazy scaling. Uh, and then we said, you know, "We have to figure out a business model later. For now, we're just gonna charge people so that we don't, like, run out our compute bills, but that's obviously not the long-term answer." That also turned out just to work. Um, and that was the story of ChatGPT. And then there was so much utility that people just had not gotten over the activation energy to find that that has worked really well. Um, and then Codex... Actually, the plan before ChatGPT was that

9. 16:41 – 17:31

   Codex and the ‘actuators’ thesis: code for computers, robots for the physical world

   1. SASam Altman16:41

      we were gonna go all in on code.

   2. SPSpeaker16:43

      Hmm.

   3. SASam Altman16:44

      Um, we knew these models could write code. Uh, we knew that they could be really-- A-and we knew that that would be, like, a valuable area, but then we had this incredibly exciting thing happen. Um, but our kinda internal belief at the time was that coding was how these models would control things on computers, and robots were how these models would control things in the physical world. And if you made a smart enough model that had sort of the actuators of writing code and robot, and driving a robot, you could then kind of actually get this intelligence to do stuff for you in the world.

   4. SPSpeaker17:16

      Hmm.

   5. SASam Altman17:16

      So, uh, then it took us a while to get there. And then I think Codex got really good by early this year, but with 5.5 is when we saw this real inflection point where people are now, like, doing just incredible things with it.

10. 17:31 – 18:20

    The modern capability pipeline—and why it may be rewritten

    1. SPSpeaker17:31

       And, um, you know, that, the, w- early in the class, we've talked about how the capabilities pipeline, uh, is starting to look, it's starting to become somewhat more legibly standard across different research groups. You've got, you know, pre-training, mid-training, post-training, then you've got the RL and supervised feedback loop. Is th- Do you think that's roughly, like, the shape of the pipeline that allowed Codex to, you know, go through a capability jump, and that will basically stay stable now and consistent, or are we gonna go through a major rewrite of that pipeline?

    2. SASam Altman17:59

       I think that is definitely the current pipeline. I expect we will go through a major rewrite. I don't know when it'll happen or exactly how. Um, but It is a little odd to me that it's so happens as a pipeline and doesn't quite feel like the optimal solution.

    3. SPSpeaker18:17

       Um, what would be an optimal solution in your head?

11. 18:20 – 19:57

    AI as research intern to autonomous researcher: compute-backed milestones

    1. SASam Altman18:20

       I think that's a research problem for the AIs to figure out. Um, I think we're at a point where-- And we've set this goal that by September of this year, we will use 500,000 A100 equivalent GPUs, like a lot of computing power, as an AI research intern, and by March of 2028, that we will have a full end-to-end very talented researcher, like figuring out complete new architectures. Um, so I think we are gonna get like, with the current pipeline, the current architectures, I think we're gonna get over the line of when AIs can do incredible, incredible work.

    2. SPSpeaker18:52

       Um, you know, o-one of the things that you, you just described there i-is you, you-- We, we've been talking a lot in the class about systems, frameworks, and analogies to make s-concepts from one domain legible to other people who may not have all the context in another, and that sometimes because of the translation problem, you know, reasoning by analogy is not helpful because then errors compound.

    3. SASam Altman19:16

       Yeah.

    4. SPSpeaker19:17

       Um, right there you said, you know, our goal is to try to use it as an AI intern, which obviously is a very useful metaphor within the context of, you know, Silicon Valley, a cl- a class that understands how these pipelines work and so on. And then as, as you scale actually that metaphor globally, people who might not have all that context go start analogizing these models in ways that they shouldn't be. Like, how should we think about the limits of, of that, of, of-- What, what are the limits to scale of, um... What, what are the product analogies, the research analogies you find most useful within the valley, and which one of the-- What have you found about, found about the limits of those analogies scaling, and now how do you navigate between those two problems?

12. 19:57 – 22:50

    Explaining AI to the world: limits of analogies and the ‘intelligence utility’ frame

    1. SASam Altman19:57

       I, I've been very interested in studying how... Like, I think what is happening is we are, we are in the process of creating a new utility. This doesn't happen very often. You know, electricity is a utility, internet's a utility. There-- Water, I guess. There's not a lot of these. Uh, and so there are not a lot of examples that we can study for good metaphors or learnings about how to explain this to the world. Um, but I was recently looking at what happened when electricity became a utility, and it's a good analogy for many reasons. It's imperfect, of course, too. But the electricity companies, at least the ones I could find information about, they didn't talk about selling electricity because no one knew what that was or why they wanted it. It sounds, like, very scary. It's this thing that's, like, gonna come into your house, and it can kill you in this, like, gruesome way, and y-you know, it feels sort of, like, very different than the world before. Uh, and maybe they tried to sell electricity or market electricity at first. I don't know. But in any case, that didn't work. And then what they started marketing, selling to people, was light at night. You know-

    2. SPSpeaker20:59

       Mm

    3. SASam Altman20:59

       ... we are gonna-- What you are getting from us is not electricity. It's light at night. By the way, you can use the same thing that lets you get light for all these other things. But people are like, "Well, why would I want that?" And they're like, "Well, you know, it'll wash your clothes for you someday." And, "No, no, it won't. I can't-- That's too far of a jump for me."

    4. SPSpeaker21:15

       Right.

    5. SASam Altman21:15

       Um, so I don't know what our analogy for this should be. Um, but I suspect that even if, even if we're totally right and intelligence is gonna become this new utility that every company, every customer, uh, every government just n-needs access to and is gonna use in all sorts of incredible ways, and you will have, like, a OpenAI token subscription that you will plug into everything and use to access everything, and you have running for you all the time and doing this amazing stuff, I kind of don't think, at least right now, the right way for us to analogize that is we're selling intelligence, 'cause people are just like, somehow not resonating.

    6. SPSpeaker21:54

       Mm.

    7. SASam Altman21:54

       I don't know what our equivalent of we're selling you light at night is going to be. But I think if we're gonna become a new utility, we need to find a way to explain to the world what it means to have this, like, intelligence pipe that you can just do whatever you'd like with.

    8. SPSpeaker22:09

       It-- So, um, o-one question that has emerged, uh, an emergent property of this class is of, of having a diversity of different speakers is that the utility analogy has come up several times but in reference to different things. So Jensen likened util- like compute to a utility, um, and why there should be access and so on, and talked about how Stanford should pool budget and so on and, and, and procure that as a utility for everybody on campus. Whereas you just likened the intelligence part to a util- I-i-- Are both of these things true? Is one of them true? One, is one more likely to be true? How should people reason about compute as a utility versus tokens as a utility? Uh, and, and by compute, I mean here chips versus tokens. Does, does that make sense?

13. 22:50 – 25:53

    Compute vs tokens: what users will buy, plus the ‘one-person frontier lab’ advice (inference)

    1. SASam Altman22:50

       I think as a consumer, as, like, a business or an individual, um, you will think in something closer to tokens or probably even one level up from tokens. I don't think you'll care very much about, you know, where the hardware is, what particular chip it is, what's powering it. I think that stuff will be abstracted out. And what you will care about is when you're interacting with the system, um, can you use it a lot? Is it cheap? Is it doing a good job? Um, so right now, it's like tokens. It may get-- As we move into a world where we all just have, like, this constant agent running for us, being useful to us all of the time, um, you may think about it as even one level up. But yeah, my, my guess is, is you-- When you, like, pay for your cell phone bill, you're like, "All right. I'm buying access to airtime and some number of gigabytes, and, you know, it's gonna do all these things, and I'll use all these apps and whatever else." But, like, what you think about paying for the kind of internet utility in this case is just, like, access to the whole system and the particular hardware at the base station and how it connects to the internet. You don't think about that as much.

    2. SPSpeaker23:58

       Um, I know I, I could nerd out about utility infrastructure for a long time, but I wanna make sure we switch a little bit to being relevant for the students. Usually, we have, uh, questions, but we're not hearing those today. Uh, unless you're comfortable.

    3. SASam Altman24:09

       Happy to.

    4. SPSpeaker24:09

       Oh, okay. Great. How about that? Improv. Okay. Uh- So one final question to start getting the creative juices flowing is, um, the final project for this class, according to private 183, is the one-person Frontier Lab. So everybody here is working on projects where they're simulating being an individual, uh, as a lab with access to all the right tools. They've got hundreds of thousands of dollars of credits from Cloudflare. I think we've got some OpenAI tokens maybe, but there's a bunch of compute at their disposal. Um, what would you, if you were in the class, what would you be working on for your one-person Frontier Lab project?

    5. SASam Altman24:44

       First of all, I think that's an awesome project. Um, I think this is top of mind because, uh, you-- we, we were just, like, talking about utility frame- frameworks. I think there's a lot of very smart people working on, uh, great training ideas, and we're gonna have incredible models. No matter what you all do, we're gonna have incredible models, I promise you, uh, like, pretty quickly. But I, I, I think we have not invested enough in being able to deliver at scale huge amounts of cheap intelligence. So maybe I would go work on, like, the inference part of the stack-

    6. SPSpeaker25:20

       Hmm

    7. SASam Altman25:20

       ... and how are we going to get this incredible intelligence to be cheap and abundant. Uh, I think that's under-invested in, and, and I think all of the Frontier Labs are going to have to become inference companies to a significant degree.

    8. SPSpeaker25:33

       Um, okay. It might be too late to pivot your projects, but better late than never.

    9. SASam Altman25:39

       Work on whatever you wanna work on. [laughs]

    10. SPSpeaker25:42

        Uh, okay, let's start taking questions, and I'm gonna moderate and try to be not, you know, please try to be productive and not spicy, et cetera. Remember, it's a CS class, but up to you, Sam, if you wanna answer.

    11. SASam Altman25:51

        Real spicy is fine.

14. 25:53 – 29:03

    Q&A: LLM ‘dead end’ debate, identity traps, and why scaling keeps surprising

    1. SPSpeaker25:53

       Oh, we've got questions. Oh, perfect. All right, first one. There are questions about your views on Yann LeCun's view that LLMs are a dead end.

    2. SASam Altman26:01

       Um, first of all, in terms of achieving human-level intelligence, these models have already far surpassed human intelligence in some ways, and then they're wildly worse than others. Like, for example, they seem much worse than people are at very long horizon, kind of high judgment signal and tasks. Um, on the other hand, yesterday, we had one of our models, uh, discover or disprove a conjecture, one of the Erdos problems that had-- smart people had worked on for a long time. And a lot of people, a lot of smart scientists, I don't know if LeCun was one of them or not, had even quite recently said something like that was not going to happen. Uh, and then, like, the model just did it, and, you know, now you have all these mathematicians saying, like, "Is math over? What does this mean for our field?" So clearly, LLMs are capable of figuring out new knowledge, and clearly, they are capable of doing some things that, some intelligence tasks that humans just can't do. Um, they are going to scale much further, so how much better and what distribution of the tasks they can do better than humans, we'll find out, but I suspect it's a lot. And the, you know, in terms of this, like, lack of a belief in the exponential we were talking about earlier, um, I think the field was honestly held back by a generation of scientists who just were way too certain on what wouldn't, what, what scaling was not going to produce. And then some people just looked at the graphs and said, "Well, it looks like it's continuing beautifully. Let's keep going." Um, I think world models are clearly important, and to- we'll need that for things like robotics. Uh, but betting against LLM scaling at this point, uh, feels quite misguided to me.

    3. SPSpeaker27:52

       Uh, uh, does it get annoying to be the I-told-you-so guy?

    4. SASam Altman27:55

       No. I mean, there are these, like, Twitter trolls that, you know, for years have just been like, "It's not gonna work. It's not gonna work. This is so dumb." Like, you know, "This is a fraud. This company's gonna fail. This research approach is gonna fail." And I used to get more bothered by them, but I don't even, like, feel the I-told-you-so at this point. It's like, you were just wrong.

    5. SPSpeaker28:13

       Like, she's Nirvana.

    6. SASam Altman28:14

       You're still going on about it. Like, the data is-

    7. SPSpeaker28:18

       Right

    8. SASam Altman28:19

       ... quite strong on our side. And I don't think it'd be that fun to say I told you so, and also the fact that you're, like, still saying we're wrong doesn't really bother me.

    9. SPSpeaker28:27

       I think there's that-

    10. SASam Altman28:28

        Kind of move on

    11. SPSpeaker28:29

        ... there's that saying that, like, insanity is doing the same thing over and o- over again when presented with data that is not working, and if they keep repeating that, in a sense, it's, it's, it's a form of insanity, I think.

    12. SASam Altman28:39

        I, I think there's something that happens, which is if you make your identity about a particular thing is going to work or not work, and you associate yourself with that belief, and then the science or the empirical results disprove you, and you're, like, too hung up on your identity, you can't let it go, you can't see the truth.

    13. SPSpeaker28:59

        Yeah.

    14. SASam Altman29:00

        And I think this is, like, a important reminder in both directions.

15. 29:03 – 32:17

    Education in a post-ChatGPT world: slow adaptation and risk of critical-thinking atrophy

    1. SPSpeaker29:03

       Yeah. How do you see education?

    2. SASam Altman29:07

       Um, it clearly has to super adapt, and I am worried. I, I thought by now it would have. Um, the, the-- I think if we continue to teach and evaluate students as if we were in a pre-AGI world, um, it's not gonna work, and it is gonna lead to, like, atrophy of learning how to think or whatever. And I thought that was gonna be obvious enough that I wasn't that worried. You know, when ChatGPT launched, I was like, yeah, we're gonna have one year of, like, students, like, cheating and not learning that much, and then the educational system is just gonna redesign itself, and there's-- and we're gonna teach people so much better. You know, people are going to really get projects where they have to, they have to use AI to be able to do it, but they still have to, like, stretch their brain more and think more and figure out new things to do. And honestly, I struggle to point to any significant systemic change that I've seen in the education system at large in the three and a half years since ChatGPT launched, and I-- that was a prediction error for me. I thought, I thought that would have happened. So I have no doubt that we can, uh, like we have done with every other technological leap before, redesign how education works so that you still have to learn how to think. And there will be some things, like I- I, I am a person who thinks by writing, and I write a lot of stuff that I never show anyone else, but it's still important to me to figure something out, and so I'm grateful that I, I learned to write. People say the same thing about programming. Um, so there will be some things that we teach people to do that machines can do better just because it's helpful to teach them the meta skill of thinking and learning, and that makes sense. But there are a lot of other things where we should just totally teach-- totally change how we teach or how we learn or how we evaluate. And if we don't do that, I think there will be, like, significant atrophy in people's critical thinking skills.

    3. SPSpeaker30:58

       Uh, question is, what was your favorite class, and what, what do you wish you had taken while, when you were at Stanford?

    4. SASam Altman31:03

       Does Stanford still do intro sem? I did, like, all the-- I did, like, three intro sem a quarter my freshman year. Like, and I loved all of them. Uh, they were all super different. Uh, I-- But looking back, the fact that I was able to get such a broad exposure to stuff and ha-have, like, a, a very shallow understanding of lots of different fields was an incredible thing. If it had not been for that, I just would've taken, like, CS and physics classes, which still would've been great. But, um, I, I think more about the stuff, the classes I took that were, like, totally random and unrelated to what I do now, but in some important way gave me a perspective than I d-- I, I think I would've, like, learned to program no matter what. Uh, so I-- And I didn't think that at the time. I was, like, kind of like, you know, you know, this is, this stuff is all cool, but it's mostly gonna be about, like, learning CS. Um, I only did two years of school. Uh, so there was a lot of stuff I wanted to take that I didn't get to. Um, but that's kinda the surprising thing.

    5. SPSpeaker32:08

       My question is, what is your spiciest take of all?

16. 32:17 – 38:45

    Spicy forecast: ten-year forks—democratization, wealth distribution, and compute allocation

    1. SASam Altman32:17

       I, I think with more time to think, uh, I could come up with a much spicier one. But, um, I think AI is just gonna keep going, and I think this is considered-- I don't, I don't think this is, like, widely believed yet. And I think if this were widely believed, there would be, like, significantly more reverberations than are happening through society right now. And maybe I don't have the spicier take. Actually, maybe this is the high order bit, that if AI progress continues on the exponential that it's on for another-- It's been three and a half years since ChatGPT. If even for another three and a half years on that same trajectory, the world, the potential, the way that society, what society is capable of are just completely different.

    2. SPSpeaker33:06

       Well, le-let me try to prompt you with more thinking tokens on that one. Um, you, you have-- If we treated you as a model, like as a frontier model, and you have some inherent capabilities, and we're gonna, we're gonna try to elicit capabilities that people don't know about for the next few minutes. Um, one of them is that you've been post-trained now on-- You, you've been continuously RL'd on OpenAI, as well as the external feedback loop of the world on what doesn't work and, uh, what works and doesn't work. So now, if we're gonna treat you as a prediction engine for a sec, the prompt is what are the three most likely forks of the universe you see over the next ten years? And what is your, what is your probability assessment on each of those? Does that make sense?

    3. SASam Altman33:47

       One that feels very important is, uh, like, how much is this technology going to be very widely democratized versus how much is it going to sit in a few companies? I, I think a world-- There are all of these reasons why you could imagine the default is that this gets concentrated to a few companies, and they become, like, you know, a significant fraction of the wealth on Earth. That would obviously be terrible, and we work super hard to push against that. But I think that's gonna require, like, the will of the world to, to really avoid, um, because there is a sort of attractor state there. And I think part of the reason that we need to push to this kinda utility model of the world is that, A, it's quite unstable and quite bad and will feel quite unfair if a few companies have all of this. But, B, I think it is a real alignment failure and a very fragile world. Uh, and the best way to get to a world we want that represents, like, everybody winning and everybody's values being represented, everybody having agency, is to just put, push this technology out into the world. Um, but there will be a very strong argument against that around sort of safety and stability. And I think that will be a big fork, and it's very important, and I encourage all of you in your careers to push hard that this is a technology. It can bring us an incredible sci-fi future. Life can be unbelievably much better. We are going to incur some risk to get there, but the risk of keeping this concentrated in a handful of companies, even though we would be one of, like, these companies, is not something we should tolerate. So I think that will be a big fork. Uh, in terms of probability, I think it's-- The world should have such an interest in it happening this way that I think it's, like, eighty percent we end up on the democratic path. But there will be a very strong safety message and, you know, there will be a lot of power-seeking people who, who wanna concentrate the power.

    4. SPSpeaker35:36

       And o-o-one of the problems with forecasting this or, or that you have and we all have as humans is once you make that forecast, then you have agency to affect the forecasts, right, and the fork.

    5. SASam Altman35:49

       Well, I mean, we're clear on what we're gonna use our agency for. Like, this is what we believe in. We think that, uh, you know, we're gonna do everything we can to push it in this direction. We just, we see the forces in the other direction. Maybe a related fork, uh, there's a lot of talk about, like, future economic models and are we gonna do universal basic income? Are we gonna have everybody gets to, like, own a slice of every company? Uh, like, are we gonna-- Is it capitalism with no change? Is it, like, full-on communism? There's, like, a lot of talk about this. One thing that I think is not talked about much is how, specifically how we distribute compute.

    6. SPSpeaker36:25

       Mm.

    7. SASam Altman36:26

       So maybe a lot of the economy can work In a way that it's going to work. And I've actually, I've become much less of a even short-term jobs doomer. I've always been optimistic we find new things to do, but this may not even be as disrupted as I originally thought in the short term. Um, but we are seeing compute shortages now. I can imagine them getting much worse, and I can imagine compute being, like, the most important utility that people need. Uh, so if the price of compute from a supply and demand perspective gets way out of whack, then I think there will be a very interesting fork about what it means to equitably distribute compute.

    8. SPSpeaker37:02

       So you said two very interesting things there, which you said on the economic side, we might have need universal basic income. Everybody owns a piece of shares. You know, one of the speakers in this class is, um, Nicolai Tangen, who runs the Norwegian Sovereign Wealth Fund.

    9. SASam Altman37:17

       He's awesome.

    10. SPSpeaker37:18

        He's awesome. You know, the Norwegian Sovereign Wealth Fund owns 1.5% of all publicly traded companies on the planet. They also have effectively universal ba- basic income. You could argue there's flavors of this already today because, you know, the largest employer now in the United States is the government, and you could argue, like, large sections of that are a, a way for the government to redistribute income from taxpayers. So are these solutions that actually need to be novel or just re-implemented for this era? How do you think about the novelty of those solutions where we often, you know, in Silicon Valley ma- have this tendency to be like, reinvent, you know, old things from first principles? And so sh- should we just look to existing systems and tweak them?

    11. SASam Altman37:54

        I, I don't think that these things require deeply new ideas. Although I will say, um, I am much more excited about people having some sort of ownership stake than a fixed monthly cash dividend.

    12. SPSpeaker38:09

        Right.

    13. SASam Altman38:09

        Um, and I, I funded, like, a big universal basic income study a while ago. I've also watched what happens when people, like, invest in startups, and I know which model I think, like, hits human psychology better. So what I would love to see is as leverage in the world shifts from labor to capital, which I think is gonna keep happening, that we find a way to have something like a citizen's wealth fund in the country or in the world eventually, where you, like, you basically own a slice of capitalism.

    14. SPSpeaker38:43

        Right.

    15. SASam Altman38:43

        Own a slice of these companies.

17. 38:45 – 41:04

    The compute shortage: pricing, demand uncapped, and why ‘shortage’ may be permanent

    1. SPSpeaker38:45

       And then on the second fork there on compute bottlenecks, you said, uh, at some point when compute prices get out of whack. Between January and this year, my, my current understanding is based on data we've seen that H100 prices and Blackwell prices, the spreads between long-term reservations and spot is like 5X.

    2. SASam Altman39:04

       I don't know if it's that high anymore. I think it got a little better, but yeah.

    3. SPSpeaker39:07

       It's high. Or if you can even find H100s-

    4. SASam Altman39:09

       Yeah

    5. SPSpeaker39:09

       ... 'cause they're pretty much all gone-

    6. SASam Altman39:11

       Yeah

    7. SPSpeaker39:11

       ... for this year. Does that sound right?

    8. SASam Altman39:13

       No argument. There's a gigantic compute short- shortage, yeah.

    9. SPSpeaker39:16

       So that, that's a good example of an, of a systems problem right now that's live. Uh, at least to some folks, it feels like COVID, you know, for the compute era, like all the toilet paper's gone.

    10. SASam Altman39:27

        Yeah.

    11. SPSpeaker39:28

        Wh- why are people not freaking out about this?

    12. SASam Altman39:31

        Well, I think people assume we will make big inference gains on the hardware we have. Uh, I also think there is a tsunami of hardware coming.

    13. SPSpeaker39:38

        Mm.

    14. SASam Altman39:39

        But maybe the, the demand tsunami is even bigger, and pe- I think people should be freaking out somewhat.

    15. SPSpeaker39:43

        And, and would you say it's fair-- Like, how, how long are we gonna exist in a compute shortage, at least, you know, based on current data you have?

    16. SASam Altman39:54

        I think like other... You, you can't talk really about, like, worldwide demand for electricity without talking about the price. Like, it's-- There's an extremely different demand a- about how much energy people need to use in the world if the price comes down by a factor of ten or goes up by a factor of ten. And I think AI is like that too.

    17. SPSpeaker40:14

        Mm.

    18. SASam Altman40:14

        Uh, the-- If we can make models sufficiently smart and at a sufficiently low cost, I think demand is, like, kinda uncapped. And so in some sense, as long as we can s- continue to make progress on this, there will be a shortage forever, and things will be bid among, a- above what the price we think, we think the price should be, even though people are getting better, smarter, more whatever intelligence. Just because you can use, like... If we make really great personal agents, then you can have ten of them running and working for you all the time, or a hundred and, you know, you'll want the hundred, I think.

    19. SPSpeaker40:53

        It's a lot of inference, a lot of compute. Awesome. With that, I'm gonna give you the swag for the class, which is [audience applauding] [laughs] Thank you for coming.

    20. SASam Altman41:03

        Thank you. Thank you all.

Episode duration: 41:09