CEOInterviews.AI#419 – Sam Altman: OpenAI, GPT-5, Sora, Board Saga, Elon Musk, Ilya, Power & AGI

L

Lex Fridman0:00

The following is a conversation with Sam Altman, CEO of OpenAI. [Brief sponsor mentions.] And now, here's Sam Altman. Take me through the OpenAI board saga that started on Thursday, November 16th, maybe Friday, November 17th for you.

S

Sam Altman7:59

That was definitely the most painful professional experience of my life and chaotic and shameful and upsetting and a bunch of other negative things. There were great things about it, too. I came across this old tweet of mine from that time period which was like going to your own eulogy watching people say all these great things about you. That whole weekend I felt a great deal of love and very little hate. Even though it felt like I had no idea what's happening. I also think I'm happy that it happened relatively early. I thought at some point between when OpenAI started and when we created AGI, there was going to be something crazy and explosive that happened. It still, I think, helped us build up some resilience and be ready for more challenges in the future.

L

Lex Fridman9:48

But the thing you had a sense that you would experience is some kind of power struggle.

S

Sam Altman9:54

The road to AGI should be a giant power struggle. I expect that to be the case.

L

Lex Fridman10:03

And so you have to go through that as you said iterate as often as possible in figuring out how to have a board structure, how to have organization, how to have the kind of people that you're working with, how to communicate all that in order to deescalate the power struggle as much as possible, pacify it.

S

Sam Altman10:24

But at this point it feels like something that was in the past that was really unpleasant and really difficult and painful, but we're back to work and things are so busy and so intense that I don't spend a lot of time thinking about it. There was a time after, there was this fugue state for maybe 45 days after that. I was just sort of drifting through the days. I was feeling so down. Really painful. Hard to have to keep running OpenAI in the middle of that. I just wanted to crawl into a cave and recover. But now it's like we're just back to working on the mission. It's still useful to go back there and reflect on board structures, on power dynamics, on how companies are run, the tension between research and product development and money. So that you who have a very high potential of building AGI would do so in a slightly more organized, less dramatic way.

L

Lex Fridman11:52

In the future. So there's value there to go both the personal psychological aspects of you as a leader and also just the board structure and all this kind of messy stuff.

S

Sam Altman12:00

Definitely learned a lot about structure and incentives and what we need out of a board. I think it is valuable that this happened now. I think this is probably not the last high stress moment of OpenAI, but it was quite a high stress moment. The company very nearly got destroyed. We think a lot about how to build a resilient org and a structure that will stand up to a lot of pressure in the world, which I expect more and more as we get closer. I think that's super important.

L

Lex Fridman12:47

Do you have a sense of how deep and rigorous the deliberation process by the board was? Can you shine some light on just human dynamics involved in situations like this? Was it just a few conversations and all of a sudden it escalates and why don't we fire Sam kind of thing?

S

Sam Altman13:05

I think the board members were well-meaning people on the whole. I believe that in stressful situations, where people feel time pressure, people understandably make suboptimal decisions. One of the challenges for OpenAI will be to have a board and a team that are good at operating under pressure.

L

Lex Fridman13:46

Do you think the board had too much power?

S

Sam Altman13:49

I think boards are supposed to have a lot of power. But one of the things we did see is that in most corporate structures, boards are usually answerable to shareholders. In this case, the board of a nonprofit has quite a lot of power; they don't really answer to anyone but themselves. There are ways in which that's good, but what we'd really like is for the board of OpenAI to answer to the world as a whole as much as that's practical.

L

Lex Fridman14:30

So there's a new board announced.

S

Sam Altman14:33

There's a new smaller board at first, and now there's a new final board. We've added some. We'll add more.

L

Lex Fridman14:39

What is fixed in the new one that was perhaps broken in the previous one?

S

Sam Altman14:52

The old board sort of got smaller over the course of about a year. It was nine and then it went down to six, and then we couldn't agree on who to add. The board also didn't have a lot of experienced board members. A lot of the new board members at OpenAI have more experience as board members. I think that will help.

L

Lex Fridman15:25

What's the process of selecting the board like? What's involved in that?

S

Sam Altman15:29

Brett and Larry were kind of decided in the heat of the moment over this very tense weekend. That weekend was a real roller coaster. We were trying to agree on new board members that both the executive team and the old board felt would be reasonable. Larry was actually one of their suggestions. Brett I think I had even previous to that weekend suggested but he was busy. I felt like if I was going to come back, I needed new board members. I didn't think I could work with the old board again in the same configuration. We wanted to get to a board of three and had to find two new board members over a short period of time. Those were decided on the battlefield. For new board members since and going forward, we have criteria for different expertise. Unlike hiring an executive, the board needs to do governance and thoughtfulness well. Brett says we want to hire board members in slates, not as individuals, thinking about a group that will bring nonprofit expertise, running companies, good legal and governance expertise.

L

Lex Fridman16:13

So is technical savvy important for the individual board members?

S

Sam Altman16:15

Not for every board member, but for certainly some you need that. That's part of what the board needs to do. The interesting thing that people probably don't understand about OpenAI is all the details of running the business. When they think about the board given the drama, they think about if you reach AGI or some incredibly impactful products, what's the conversation with the board like? You definitely need some technical experts there, and then you need some people who think about how to deploy this in a way that will help people the most. A mistake you or I might make is to think that only technical understanding matters. That's part of the conversation, but there's a lot more about how that's going to impact society and people's lives that you really want represented.

L

Lex Fridman16:40

Thank you. Thank you for making it mathematical for the audience.

S

Sam Altman16:43

For a board member, I do care much more about the y-intercept. I think there is something deep to say about track record and experience is sometimes very hard to replace.

L

Lex Fridman16:47

Do you try to fit a polynomial function or exponential one to the track record?

S

Sam Altman16:50

That analogy doesn't carry that far.

L

Lex Fridman16:52

All right, you mentioned some of the low points that weekend. What were some of the low points psychologically for you? Did you consider going to the Amazon jungle and just disappearing forever?

S

Sam Altman16:56

There were so many low points. It was a very bad period of time. There were great high points too, like my phone was non-stop blowing up with nice messages from people I work with every day, people I hadn't talked to in a decade. I didn't get to appreciate that as much as I should have because I was in the middle of this firefight. But on the whole, it was a very painful weekend. It was a battle fought in public to a surprising degree, extremely exhausting. The board did this Friday afternoon. I really couldn't get much in the way of answers, but I also was just like, well, the board gets to do this. So I'm going to think for a little bit about what I want to do, but I'll try to find the blessing in disguise. I was like, my current job at OpenAI was to run a decently sized company, and the thing I'd always liked the most was just getting to work with the researchers. I was like, 'Yeah, I can just go do a very focused high research effort.' I got excited about that. Didn't even occur to me at the time that this was all going to get undone. That was Friday afternoon. I went through a little period of confusion and rage, but very quickly. By Friday night, I was talking to people about what was going to be next, and I was excited about that. Friday night evening for the first time I heard from the exec team that they were going to fight this. I went to bed still being like okay, excited. I didn't sleep much, didn't eat much, but had a surprising amount of energy. It was a weird thing about adrenaline and wartime. I accepted the death of this baby OpenAI and was excited for the new thing. Then Saturday morning, two of the board members called and said they didn't mean to destabilize things and asked about me coming back. I immediately didn't want to do that, but I thought about it and said okay but here's what I would need. The most painful time of all was over the course of that weekend. I kept thinking and being told that we were almost done, but it was a very confusing state. Then Sunday evening, the board appointed a new interim CEO, and that felt really bad. That was the low point of the whole thing. I felt a lot of love that whole weekend. It was not other than that one moment Sunday night. I would not characterize my emotions as anger or hate. I really just felt a lot of love from people, towards people. It was painful, but the dominant emotion of the weekend was love, not hate.

L

Lex Fridman23:51

You've spoken highly of Mira Murati that she helped especially as you put in a tweet in the quiet moments when it counts. Perhaps we could take a bit of a tangent. What do you admire about Mira?

S

Sam Altman24:03

She did a great job during that weekend in a lot of chaos. But a thing I really value in leaders is how people act on a boring Tuesday at 9:46 in the morning, in the normal drudgery of the day-to-day, how someone shows up in a meeting, the quality of the decisions they make. That was what I meant about the quiet moments. Most of the work is done on a day by day, meeting by meeting, just be present and make great decisions.

L

Lex Fridman24:44

Yeah. I mean look, you wanted to spend the last 20 minutes about this one very dramatic weekend. But that's not really what OpenAI is about. OpenAI is really about the other seven years.

S

Sam Altman24:52

Well, yeah. Human civilization is not about the invasion of the Soviet Union by Nazi Germany. But still that's something people focus on. It gives us an insight into human nature. The extremes of human nature and perhaps some of the damage and some of the triumphs of human civilization can happen in those moments. So it's illustrative.

L

Lex Fridman25:18

Let me ask you about Ilya. Is he being held hostage in a secret nuclear facility?

S

Sam Altman25:22

No.

L

Lex Fridman25:23

What about a regular secret facility?

S

Sam Altman25:25

No.

L

Lex Fridman25:26

What about a nuclear non-secret facility?

S

Sam Altman25:27

Neither of that either.

L

Lex Fridman25:29

I mean this is becoming a meme at some point. You've known Ilya for a long time. He was obviously in part of this drama with the board. What's your relationship with him now?

S

Sam Altman25:43

I love Ilya. I have tremendous respect for Ilya. I don't have anything I can say about his plans right now. That's a question for him. But I really hope we work together for the rest of my career. He's a little bit younger than me. Maybe he works a little bit longer.

L

Lex Fridman26:01

There's a meme that he saw something like he maybe saw AGI and that gave him a lot of worry internally.

S

Sam Altman26:10

Ilya has not seen AGI. None of us have seen AGI. We've not built AGI. I do think one of the many things that I really love about Ilya is he takes AGI and the safety concerns broadly speaking, including things like the impact this is going to have on society, very seriously. As we continue to make significant progress, Ilya is one of the people that I've spent the most time over the last couple of years talking about what this is going to mean, what we need to do to ensure we get it right. Ilya did not see AGI. But Ilya is a credit to humanity in terms of how much he thinks and worries about making sure we get this right.

L

Lex Fridman27:16

I've had a bunch of conversations with him in the past. I think when he talks about technology, he's always doing this long-term thinking type of thing. So he's not thinking about what this is going to be in a year. He's thinking about in 10 years.

S

Sam Altman27:28

Yeah. I'm just thinking from first principles like okay, if the scales, what are the fundamentals here, where is this going? That's a foundation for then thinking about all the other safety concerns. Which makes him a really fascinating human to talk with.

L

Lex Fridman27:46

Do you have any idea why he's been kind of quiet? Is he just doing some soul-searching?

S

Sam Altman27:54

Again, I don't want to speak for him. I think you should ask him that. He's definitely a thoughtful guy. I think Ilya is always on a soul search in a really good way.

L

Lex Fridman28:13

Yes. Yeah. Also, he appreciates the power of silence. Also, I'm told he can be a silly guy, which I've never seen that side of him.

S

Sam Altman28:22

It's very sweet when that happens. I've never witnessed a silly Ilya, but I look forward to that as well.

L

Lex Fridman28:30

I was at a dinner party with him recently and he was playing with a puppy and he was in a very silly mood, very endearing. I was thinking, 'Oh man, this is not the side of Ilya that the world sees the most.'

So, just to wrap up this whole saga, are you feeling good about the board structure, about all of this and where it's moving?

S

Sam Altman28:50

I feel great about the new board in terms of the structure of OpenAI. One of the board's tasks is to look at that and see where we can make it more robust. We wanted to get new board members in place first. But we clearly learned a lesson about structure throughout this process. I don't have super deep things to say. It was a crazy, very painful experience. I think it was a perfect storm of weirdness. It was a preview for me of what's going to happen as the stakes get higher and the need for robust governance structures and processes and people. I am kind of happy it happened when it did, but it was a shockingly painful thing to go through.

L

Lex Fridman29:33

Did it make you be more hesitant in trusting people?

S

Sam Altman29:37

Yes. I think I'm an extremely trusting person. I have always had a life philosophy of don't worry about all of the paranoia, don't worry about the edge cases. You get a little bit screwed in exchange for getting to live with your guard down. This was so shocking to me. I was so caught off guard that it has definitely changed. I really don't like this. It's definitely changed how I think about default trust of people and planning for the bad scenarios. You got to be careful with that.

L

Lex Fridman30:07

Are you worried about becoming a little too cynical?

S

Sam Altman30:12

I'm not worried about becoming too cynical. I think I'm the extreme opposite of a cynical person, but I'm worried about just becoming less of a default trusting person.

L

Lex Fridman30:22

I'm actually not sure which mode is best to operate in for a person who's developing AGI, trusting or untrusting. It's an interesting journey you're on. But in terms of structure, I'm more interested on the human level. How do you surround yourself with humans that are building cool stuff but also are making wise decisions? Because the more money you start making, the more power the thing has, the weirder people get.

S

Sam Altman30:53

You can make all kinds of comments about the board members and the level of trust I should have had there or how I should have done things differently, but in terms of the team here, I think you'd have to give me a very good grade on that one. I have enormous gratitude and trust and respect for the people that I work with every day. Being surrounded with people like that is really important.

L

Lex Fridman31:26

Our mutual friend Elon sued OpenAI. What is the essence of what he's criticizing? To what degree does he have a point? To what degree is he wrong?

S

Sam Altman31:38

I don't know what it's really about. We started off just thinking we were going to be a research lab and having no idea about how this technology was going to go. It's hard to go back and really remember what it was like then. This was before language models were a big deal, before we had any idea about an API or selling access to a chatbot. We were just going to try to do research. With many new fundamentally new things, you start fumbling through the dark and you make some assumptions, most of which turn out to be wrong. Then it became clear that we were going to need to do different things and also have huge amounts more capital. So we said okay, the structure doesn't quite work for that. How do we patch the structure? Then you patch it again and again and you end up with something that does look kind of eyebrow-raising. But we got here gradually with reasonable decisions at each point. Doesn't mean I wouldn't do it totally differently if we could go back with an oracle. But you don't get the oracle at the time. In terms of what Elon's real motivations here are, I don't know.

L

Lex Fridman32:58

To the degree you remember, what was the response that OpenAI gave in the blog post? Can you summarize it?

S

Sam Altman33:07

We just said, Elon said this set of things. Here's our characterization of how this went down. We tried to not make it emotional and just say the history. I do think there's a degree of mischaracterization from Elon here about one of the points you just made, which is the degree of uncertainty you had at the time. You guys are a bunch of a small group of researchers crazily talking about AGI when everybody's laughing at that thought. It wasn't that long ago Elon was crazily talking about launching rockets when people were laughing at that thought. So I think he'd have more empathy for this. I do think that there's personal stuff here. There was a split that OpenAI and a lot of amazing people here chose to part ways with Elon. Elon chose to part ways. He thought OpenAI was going to fail. He wanted total control to turn it around. We wanted to keep going in the direction that now has become OpenAI. He also wanted Tesla to be able to build an AGI effort. At various times he wanted to make OpenAI into a for-profit company that he could have control of or have it merge with Tesla. We didn't want to do that and he decided to leave.

L

Lex Fridman34:52

So you're saying that's one of the things that the blog post says is that he wanted OpenAI to be basically acquired by Tesla.

S

Sam Altman35:02

Yeah. My memory is the proposal was just get acquired by Tesla and have Tesla have full control over it.

L

Lex Fridman35:17

So what is the word 'open' in OpenAI mean to Elon at the time? Ilya has talked about this in the email exchanges. What does it mean to you at the time? What does it mean to you now?

S

Sam Altman35:29

I would definitely pick a different name. One of the things that I think OpenAI is doing that is the most important of everything that we're doing is putting powerful technology in the hands of people for free as a public good. We don't run ads on our free version. We don't monetize it in other ways. We just say it's part of our mission. We want to put increasingly powerful tools in the hands of people for free and get them to use them. I think that kind of 'open' is really important to our mission. If you give people great tools and teach them to use them, or don't even teach them, they'll figure it out and let them go build an incredible future for each other. That's a big deal. So if we can keep putting free or low-cost powerful AI tools out in the world, I think that's a huge deal for how we fulfill the mission. Open source or not. I think we should open source some stuff and not other stuff. It becomes this religious battle line where nuance is hard to have, but nuance is the right answer.

L

Lex Fridman36:42

So he said change your name to ClosedAI and I'll drop the lawsuit. I mean is it going to become this battleground in the land of memes?

S

Sam Altman36:52

I think that speaks to the seriousness with which Elon means the lawsuit. That's an astonishing thing to say. I don't think the lawsuit is legally serious. It's more to make a point about the future of AGI and the company that's currently leading the way. So look, I mean, Grok had not open sourced anything until people pointed out it was a little bit hypocritical and then he announced that Grok will open source things this week. I don't think open source versus not is what this is really about for him.

L

Lex Fridman37:34

Well, we'll talk about open source and not. I do think maybe criticizing the competition is great. Just talking a little shit, that's great. But friendly competition versus like I personally hate lawsuits.

S

Sam Altman37:47

Yeah. Look, I think this whole thing is unbecoming of a builder. I respect Elon as one of the great builders of our time. I know he knows what it's like to have haters attack him and it makes me extra sad he's doing it to us.

L

Lex Fridman38:04

Yeah, he's one of the greatest builders of all time. Potentially the greatest builder of all time. It makes me sad and I think it makes a lot of people sad. Like there's a lot of people who've really looked up to him for a long time. I said in some interview that I missed the old Elon and the number of messages I got being like that exactly encapsulates how I feel. I think he should just win. He should just make X Grok beat GPT and then GPT beats Grok and it's just a competition and it's beautiful for everybody. But on the question of open source, do you think there's a lot of companies playing with this idea? It's quite interesting. I would say Meta surprisingly has led the way on this, or at least took the first step in the game of chess of really open sourcing the model. Of course, it's not the state-of-the-art model, but open sourcing Llama. Google is flirting with the idea of open sourcing a smaller version. Have you played around with this idea? What are the pros and cons of open sourcing?

S

Sam Altman39:08

Yeah, I think there is definitely a place for open source models, particularly smaller models that people can run locally. I think there's huge demand. I think there will be some open source models, there will be some closed source models. It won't be unlike other ecosystems in that way.

L

Lex Fridman39:26

I listened to the All-In podcast talking about this lawsuit and they were more concerned about the precedent of going from nonprofit to this capped-profit. What precedent this sets for other startups?

S

Sam Altman39:41

I would heavily discourage any startup that was thinking about starting as a nonprofit and adding a for-profit arm later. I don't think we'll set a precedent here.

L

Lex Fridman39:52

Okay. So most startups should just go for-profit.

S

Sam Altman39:54

For sure. And again, if we knew what was going to happen, we would have done that too.

L

Lex Fridman39:58

Well, like in theory, if you dance beautifully here, you can get some tax incentives or whatever. But I don't think that's how most people think about these things. It's just not possible to save a lot of money for a startup if you do it this way.

S

Sam Altman40:12

No, I think there's laws that would make that pretty difficult.

L

Lex Fridman40:17

Where do you hope this goes with Elon? This tension, this dance, what do you hope this like if we go 1, 2, 3 years from now, your relationship with him on a personal level too, like friendship, friendly competition, just all this kind of stuff.

S

Sam Altman40:37

Yeah, I really respect Elon. I hope that years in the future we have an amicable relationship.

L

Lex Fridman40:51

Yeah, I hope you guys have an amicable relationship like this month and just compete and win and explore these ideas together. I do suppose there's competition for talent or whatever, but it should be friendly competition. Just build cool stuff. And Elon is pretty good at building cool stuff, but so are you. So, speaking of cool stuff, Sora, there's like a million questions I could ask. First of all, it's amazing. It truly is amazing on a product level, but also just on a philosophical level. So let me just ask technical/philosophical: what do you think it understands about the world more or less than GPT-4 for example? The world model when you train on these patches versus language tokens.

S

Sam Altman41:51

I think all of these models understand something more about the world model than most of us give them credit for. And because there are also very clear things they just don't understand or don't get right, it's easy to look at the weaknesses, see through the veil, and say, 'Ah, this is just fake.' But it's not all fake. It's just some of it works and some of it doesn't.

I remember when I started first watching Sora videos and I would see a person walk in front of something for a few seconds and occlude it and then walk away and the same thing was still there. I was like, 'Oh, it's pretty good.' Or there are examples where the underlying physics looks so well represented over a lot of steps in a sequence. It's quite impressive. But fundamentally, these models are just getting better and that will keep happening. If you look at the trajectory from DALL-E 1 to 2 to 3 to Sora, there were a lot of people that dunked on each version saying it can't do this, it can't do that, and look at it now.

L

Lex Fridman42:51

Yeah. Well, the thing you just mentioned is kind of with occlusions is basically modeling the physics of three-dimensional physics of the world sufficiently well to capture those kinds of things.

S

Sam Altman43:03

Well, maybe you can tell me in order to deal with occlusions, what does the world model need to?

L

Lex Fridman43:10

Yeah. So, what I would say is it's doing something to deal with occlusions really well. What I represent that it has like a great underlying 3D model of the world. It's a little bit more of a stretch. But can we get there through just these kinds of two-dimensional training data approaches?

S

Sam Altman43:24

It looks like this approach is going to go surprisingly far. I don't want to speculate too much about what limits it will surmount and which it won't.

L

Lex Fridman43:32

What are some interesting limitations of the system that you've seen? I mean there's been some fun ones you've posted.

S

Sam Altman43:38

There's all kinds of fun. I mean like cats sprouting an extra limb at random points in a video. Pick what you want, but there's still a lot of weaknesses. Do you think that's a fundamental flaw of the approach or is it just bigger model or better technical details or better data that will solve those?

L

Lex Fridman44:04

I would say yes to both. I think there is something about the approach which just seems to feel different from how we think and learn, and then also I think it'll get better with scale. Like I mentioned, LLMs have tokens, text tokens, and Sora has visual patches. So it converts all visual data, videos and images into patches. Is the training fully self-supervised or is there some manual labeling? What's the involvement of humans? Without saying anything specific about the Sora approach, we use lots of human data in our work, but not internet scale data. So lots of humans, lots is a complicated word, Sam. I think lots is a fair word in this case.

S

Sam Altman44:57

Yeah, it doesn't. Because to me, lots like listen, I'm an introvert and when I hang out with like three people, that's a lot of people. Four people, that's a lot.

L

Lex Fridman45:05

But I suppose you mean more than three people work on labeling the data for these models. Yeah.

S

Sam Altman45:10

Okay. All right. But fundamentally there's a lot of self-supervised learning because what you mentioned in the technical report is internet scale data. That's another beautiful, it's like poetry. So it's a lot of data that's not human labeled. It's self-supervised in that way.

L

Lex Fridman45:31

And then the question is how much data is there on the internet that could be used in this that is conducive to this kind of self-supervised way. If only we knew the details of the self-supervised. Have you considered opening it up a little more details?

S

Sam Altman45:34

Is there on the internet that could be used in this that is conducive to this kind of self-supervised way. If only we knew the details of the self-supervised. Do you have you considered opening it up a little more details? We have you mean for source specifically?

L

Lex Fridman45:50

Sora specifically the training because it's so interesting that like can the same magic of LLM now start moving towards visual data and what does that take to do that?

S

Sam Altman46:04

I mean it looks to me like yes, but we have more work to do.

L

Lex Fridman46:08

Sure.

S

Sam Altman46:08

What are the dangers? Why are you concerned about releasing the system? What are some possible dangers of this?

L

Lex Fridman46:15

I mean, frankly speaking, one thing we have to do before releasing the system is just get it to work at a level of efficiency that will deliver the scale people are going to want from this. So I don't want to downplay that. And there's still a ton of work to do there. But you can imagine issues with deep fakes, misinformation. We try to be a thoughtful company about what we put out into the world. And it doesn't take much thought to think about the ways this can go badly.

S

Sam Altman46:51

There's a lot of tough questions here. You're dealing in a very tough space. Do you think training AI should be or is fair use under copyright law? I think the question behind that question is do people who create valuable data deserve to have some way that they get compensated for use of it and that I think the answer is yes. I don't know yet what the answer is. People have proposed a lot of different things. We've tried some different models, but if I'm an artist, for example, I would like to be able to opt out of people generating art in my style, and if they do generate art in my style, I'd like to have some economic model associated with that.

L

Lex Fridman47:32

Yeah, it's that transition from CDs to Napster to Spotify. We have to figure out some kind of model.

S

Sam Altman47:38

The model changes, but people have got to get paid. Well, there should be some kind of incentive if you zoom out even more for humans to keep doing cool stuff.

L

Lex Fridman47:48

Everything I worry about, humans are going to do cool stuff and society is going to find some way to reward it. That seems pretty hardwired. We want to create. We want to be useful. We want to achieve status in whatever way.

S

Sam Altman48:01

That's not going anywhere, I don't think. But the reward might not be monetary financially. It might be fame and celebration of other cool maybe financial in some other way. Again, I don't think we've seen the last evolution of how the economic system is going to work.

L

Lex Fridman48:17

Yeah. But artists and creators are worried when they see Sora. They're like, 'Holy shit.'

S

Sam Altman48:22

Sure. Artists were also super worried when photography came out. And then photography became a new art form and people made a lot of money taking pictures. I think things like that will keep happening. People will use the new tools in new ways.

L

Lex Fridman48:26

Yeah.

S

Sam Altman48:26

And then photography became a new art form and people made a lot of money taking pictures. I think things like that will keep happening. People will use the new tools in new ways.

L

Lex Fridman48:36

If we just look on YouTube or something like this, how much of that will be using Sora like AI generated content do you think in the next 5 years? People talk about how many jobs it's going to do in 5 years and the framework that people have is what percentage of current jobs are just going to be totally replaced by some AI doing the job. The way I think about it is not what percent of jobs AI will do, but what percent of tasks will AI do and over what time horizon. So if you think of all of the 5-second tasks in the economy, the 5 minute tasks, the 5 hour tasks, maybe even the 5 day tasks, how many of those can AI do? I think that's a way more interesting, impactful, important question than how many jobs AI can do because it is a tool that will work at increasing levels of sophistication and over longer and longer time horizons for more and more tasks and let people operate at a higher level of abstraction. So maybe people are way more efficient at the job they do. And at some point that's not just a quantitative change, but it's a qualitative one too about the kinds of problems you can keep in your head. I think that for videos on YouTube, it'll be the same. Many videos, maybe most of them will use AI tools in the production, but they'll still be fundamentally driven by a person thinking about it, putting it together, doing parts of it, sort of directing and running it.

S

Sam Altman50:01

Sort of directing and running it.

L

Lex Fridman50:04

Yeah, it's so interesting. I mean, it's scary, but it's interesting to think about. I tend to believe that humans like to watch other humans or other human humans really care about other humans a lot.

S

Sam Altman50:12

Humans really care about other humans a lot.

L

Lex Fridman50:15

Yeah. If there's a cooler thing that's more that's better than a human, humans care about that for like two days and then they go back to humans.

S

Sam Altman50:25

That seems very deeply wired.

L

Lex Fridman50:27

It's the whole chess thing. Oh, yeah. But now let's everybody keep playing chess and let's ignore the alpha in the room that humans are really bad at chess relative to AI systems.

S

Sam Altman50:38

We still run races and cars are much faster. I mean this is there's like a lot of examples.

L

Lex Fridman50:42

Yeah. And maybe it'll just be tooling like in the Adobe suite type of way where you can just make videos much easier. All that kind of stuff. Listen, I hate being in front of the camera. If I can figure out a way to not be in front of the camera, I would love it. Unfortunately, it'll take a while like that. Generating faces, it's getting there, but generating faces in video format is tricky when it's specific people versus generic people. Let me ask you about GPT4. There's so many questions. First of all, also amazing. It's looking back, it'll probably be this kind of historic pivotal moment with 3, 5, and 4 with ChatGPT.

S

Sam Altman51:26

Maybe five will be the pivotal moment. I don't know. Hard to say that looking forwards.

L

Lex Fridman51:30

We never know. That's the annoying thing about the future. It's hard to predict. But for me, looking back, GPT4 ChatGPT is pretty damn impressive. Like historically impressive. So allow me to ask, what's been the most impressive capabilities of GPT4 to you and GPT4 Turbo?

S

Sam Altman51:52

I think it kind of sucks.

L

Lex Fridman51:53

Typical human also got used to an awesome thing. No, I think it is an amazing thing. But relative to where we need to get to and where I believe we will get to, at the time of GPT3, people were like, 'Oh, this is amazing. This is a marvel of technology.' And it is, it was, but now we have GPT4 and you look at GPT3 and you're like, 'That's unimaginably horrible.' I expect that the delta between five and four will be the same as between four and three. And I think it is our job to live a few years in the future and remember that the tools we have now are going to kind of suck looking backwards at them and that's how we make sure the future is better.

S

Sam Altman52:45

What are the most glorious ways in which GPT4 sucks? Meaning what are the best things it can do and the limits of those best things that allow you to say it sucks therefore gives you inspiration and hope for the future? You know one thing I've been using it for more recently is sort of a brainstorming partner. For that, there's a glimmer of something amazing in there. I don't think it gets, when people talk about what it does, they're like, 'Ah, it helps me code more productively. It helps me write more faster and better. It helps me translate from this language to another.' All these amazing things. But there's something about the creative brainstorming partner. I need to come up with a name for this thing. I need to think about this problem in a different way. I'm not sure what to do here. That gives a glimpse of something I hope to see more of. One of the other things that you can see a very small glimpse of is when it can help on longer horizon tasks, break down something into multiple steps, maybe execute some of the steps, search the internet, write code, put that together. When that works, which is not very often, it's very magical.

L

Lex Fridman54:10

The iterative back and forth with a human, it works a lot for me. What do you mean? Iterative back and forth the human can get more often when it can go do a 10-step problem on its own.

S

Sam Altman54:20

Oh, doesn't work for that too often. Sometimes add multiple layers of abstraction or do you mean just sequential?

L

Lex Fridman54:23

Both, like break it down and then do things at different layers of abstraction and put them together. Look, I don't want to downplay the accomplishment of GPT4. But I don't want to overstate it either. And I think this point that we are on an exponential curve, we will look back relatively soon at GPT4 like we look back at GPT3. Now that said, ChatGPT was a transition to where people started to believe it. There was an uptick of believing not internally at OpenAI perhaps there's believers here but we need to think. And in that sense I do think it'll be a moment where a lot of the world went from not believing to believing. That was more about the ChatGPT interface and the post-training of the model and how we tune it to be helpful to you than the underlying model itself. How much of each of those things are important, the underlying model and the RLHF that tunes it to be more compelling to the human? They're both super important but the RLHF, the post-training step, the little wrapper of things that from a compute perspective is a little wrapper of things that we do on top of the base model even though it's a huge amount of work, that's really important to say nothing of the product that we build around it. In some sense we had to do two things: we had to invent the underlying technology and then we had to figure out how to make it into a product people would love, which is not just about the actual product work itself, but this whole other step of how you align it and make it useful and how you make the scale work where a lot of people can use it at the same time.

S

Sam Altman56:29

And that was a known difficult thing. We knew we were going to have to scale it up. We had to go do two things that had never been done before, both quite significant achievements, and then a lot of things like scaling it up that other companies have had to do before. How does the context window going from 8K to 128K tokens compare from GPT4 to GPT4 Turbo?

L

Lex Fridman56:58

People like long. Most people don't need all the way to 128 most of the time. Although, if we dream into the distant future, we'll have context length of several billion. You will feed in all of your information, all of your history over time, and it'll just get to know you better and better, and that'll be great. For now, the way people use these models, they're not doing that. People sometimes post a paper or a significant fraction of a code repository, but most usage of the models is not using the long context most of the time.

S

Sam Altman57:36

I like that. This is your 'I have a dream' speech. One day you'll be judged by the full context of your character or of your whole lifetime. That's interesting. So that's part of the expansion that you're hoping for is a greater and greater context. There was this internet clip once about Bill Gates talking about the amount of memory on some early computer, maybe 64K, maybe 640K, and most of it was used for the screen buffer. And he just couldn't imagine that the world would eventually need gigabytes of memory in a computer or terabytes. And you always need to follow the exponential of technology and we will find out how to use better technology. So I can't really imagine what it's like right now for context lengths to go out to the billions someday, but I know we'll use it and not want to go back once we have it.

L

Lex Fridman58:42

Yeah. Even saying billions 10 years from now might seem dumb because it'll be trillions upon trillions.

S

Sam Altman58:50

Sure, there'll be some kind of breakthrough that will effectively feel like infinite context, but even 128K, I haven't pushed it to that degree. Maybe putting in entire books or parts of books, papers. What are some interesting use cases of GPT4 that you've seen?

L

Lex Fridman59:25

The thing that I find most interesting is not any particular use case, but people who use it as their default start for any kind of knowledge work task. Mostly younger people. And it's the fact that it can do a lot of things reasonably well. You can use GPT-V, you can use it to help you write code, do search, edit a paper. The most interesting to me is the people who just use it as the start of their workflow. I do as well for many things. I use it as a reading partner for reading books. It helps me think through ideas, especially when the books are classics. It is significantly better than Wikipedia on well-covered topics. It's somehow more balanced and more nuanced. It inspires me to think deeper than a Wikipedia article does. I'm not sure where the magic is. But one of the things that concerns me for knowledge tasks when I start with GPT is I'll usually have to do fact-checking after to check that it didn't come up with fake stuff. How do you ground it in truth?

S

Sam Altman1:00:59

That's obviously an area of intense interest for us. I think it's going to get a lot better with upcoming versions, but we'll have to continue to work on it and we're not going to have it all solved this year. Well, the scary thing is as it gets better, you'll start not doing the fact-checking more and more, right? I'm of two minds about that. I think people are much more sophisticated users of technology than we often give them credit for. People seem to really understand that GPT and these models hallucinate some of the time, and if it's mission critical, you got to check it.

L

Lex Fridman1:01:13

Except journalists don't seem to understand that. I've seen journalists half-heartedly just using GPT for it's of the long list of things I'd like to dunk on journalists for. This is not my top criticism of them.

S

Sam Altman1:01:20

Well, I think the bigger criticism is perhaps the pressures and the incentives of being a journalist is that you have to work really quickly and this is a shortcut. I would love our society to incentivize long journalistic efforts that take days and weeks and reward great in-depth journalism. Also journalism that presents stuff in a balanced way where it celebrates people while criticizing them even though the criticism is the thing that gets clicks. Making stuff up also gets clicks and headlines that mischaracterize completely. I'm sure you have a lot of people dunking on that. All that drama probably got a lot of clicks.

L

Lex Fridman1:02:07

Probably did.

S

Sam Altman1:02:10

And that's a bigger problem about human civilization. I'd love to see that solved where we celebrate a bit more. You've given ChatGPT the ability to have memories. You've been playing with that about previous conversations and also the ability to turn off memory which I wish I could do that sometimes. What have you seen through that?

L

Lex Fridman1:03:27

We're very early in our explorations here, but I think what people want, or at least what I want for myself, is a model that gets to know me and gets more useful to me over time. This is an early exploration. There's a lot of other things to do, but that's where you'd like to head. You'd like to use a model over the course of your life and it gets better and better. How hard is that problem? Right now it's more like remembering little factoids and preferences. What about remembering all the stuff you went through in November and all the drama? And then you're clearly blocking it out a little bit.

S

Sam Altman1:03:29

It's not just that I want it to remember that. I want it to integrate the lessons of that and remind me in the future what to do differently or what to watch out for. And we all gain from experience over the course of our lives, varying degrees. And I'd like my AI agent to gain with that experience too. So if we imagine trillions and trillions of context length, if I can put every conversation I've ever had with anybody in my life in there, if I can have all of my emails input, all of my input and output in the context window every time I ask a question, that'd be pretty cool. People sometimes will hear that and be concerned about privacy. What do you think about that aspect?

L

Lex Fridman1:05:17

I think the right answer there is just user choice. Anything I want stricken from the record for my AI agent, I want to be able to take out. If I don't want it to remember anything, I want that too. You and I may have different opinions about where on that privacy-utility trade-off for our own AI we want to be, which is totally fine. But the answer is just really easy user choice. But there should be some high level of transparency from a company about the user choice because companies in the past have been kind of shady about collecting all your data and using it for advertisement without transparency.

S

Sam Altman1:05:21

That's totally true. You mentioned earlier that I'm blocking out the November stuff. Just teasing you. Well, I think it was a very traumatic thing and it did immobilize me for a long period of time. Definitely the hardest work thing I've had to do was just keep working that period because I had to try to come back in here and put the pieces together while I was in shock and pain. Nobody really cares about that. The team gave me a pass and I was not working at my normal level, but there was a period where it was really hard to have to do both. But I kind of woke up one morning and I was like, this was a horrible thing to happen to me. I could just feel like a victim forever, or I can say this is the most important work I'll ever touch in my life and I need to get back to it. It doesn't mean that I've repressed it because sometimes I wake up in the middle of the night thinking about it. But I do feel an obligation to keep moving forward.

L

Lex Fridman1:06:18

Well, that's beautifully said, but there could be some lingering stuff in there. What I would be concerned about is that trust thing, being paranoid about people as opposed to just trusting everybody or most people using your gut. It's a tricky dance. I've seen in my part-time explorations, I've been diving deeply into the Zelenskyy administration, the Putin administration, and the dynamics there in wartime. What happens is distrust and you isolate yourself and you start to not see the world clearly. That's a human concern. You seem to have taken it in stride and learned the good lessons and felt the love and let the love energize you, but it still can linger. There are some questions I would love to ask your intuition about what GPT is able to do and not. So it's allocating approximately the same amount of compute for each token it generates. Is there room in this approach for slower thinking, sequential thinking?

S

Sam Altman1:07:37

I think there will be a new paradigm for that kind of thinking.

L

Lex Fridman1:07:42

Will it be similar architecturally as what we're seeing now with LLMs? Is it a layer on top of the LLM?

S

Sam Altman1:07:49

I can imagine many ways to implement that. I think that's less important than the question you were getting at, which is do we need a way to do a slower kind of thinking where the answer doesn't have to be immediate. Spiritually, you want an AI to be able to think harder about a harder problem and answer more quickly about an easier problem. I think that will be important.

L

Lex Fridman1:08:16

Is that like a human thought that we're just having? You should be able to think hard. Is that wrong intuition? I suspect that's a reasonable intuition.

S

Sam Altman1:08:23

Interesting. So it's not possible once GPT gets like GPT7, we just instantaneously be able to see the proof of Fermat's Last Theorem. It seems to me like you want to be able to allocate more compute to harder problems. If you ask a system to prove a theorem versus what's today's date, unless it already knew the answer, it will take more compute. But can it look like a basically LLM talking to itself? Maybe. There are a lot of things you could imagine working. What the right or best way to do that will be, we don't know.

L

Lex Fridman1:09:23

This does make me think of the mysterious lore behind Q* star. What's this mysterious Q* project? Is it also in the same nuclear facility?

S

Sam Altman1:09:36

There is no nuclear facility.

L

Lex Fridman1:09:38

That's what a person with a nuclear facility always says.

S

Sam Altman1:09:41

I would love to have a secret nuclear facility. There isn't one. Maybe someday.

L

Lex Fridman1:09:47

Someday. All right. One can dream. OpenAI is not a good company at keeping secrets. It would be nice. We've been plagued by a lot of leaks.

S

Sam Altman1:09:57

It would be nice if we were able to have something like that.

L

Lex Fridman1:10:00

Can you speak to what Q* is?

S

Sam Altman1:10:02

We are not ready to talk about that.

L

Lex Fridman1:10:03

See, but an answer like that means there's something to talk about. It's very mysterious, Sam.

S

Sam Altman1:10:08

I mean, we work on all kinds of research. We have said for a while that we think better reasoning in these systems is an important direction that we'd like to pursue. We haven't cracked the code yet. We're very interested in it.

L

Lex Fridman1:10:54

Is there going to be moments, Q* or otherwise, where there are leaps similar to ChatGPT where you're like, that's a good question. It's interesting to me, it all feels pretty continuous.

S

Sam Altman1:11:08

Right. This is kind of a theme that you're saying there's a gradual exponential slope. But from an outsider's perspective, it feels like leaps, but to you there isn't. I do wonder if we should have more iterative deployment. Rather than build in secret until we got all the way to GPT5, we decided to talk about GPT 1, 2, 3, and 4. Part of the reason is I think AI and surprise don't go together. The world needs time to adapt. I think one of the best things OpenAI has done is this strategy. We get the world to pay attention to the progress, to take AGI seriously, to think about what systems and structures and governance we want in place before we're under the gun. But the fact that people still feel like there are these leaps makes me think maybe we should be releasing even more iteratively. Our goal is not to have shock updates. The opposite.

L

Lex Fridman1:12:15

Yeah, for sure. More iterative would be amazing. I think that's beautiful for everybody.

S

Sam Altman1:12:20

But that's what we're trying to do. That's our stated strategy. And I think we're somehow missing the mark. So maybe we should think about releasing GPT5 in a different way.

L

Lex Fridman1:12:30

Yeah. 4.71, 4.72. But people like to celebrate milestones. I don't know if you know humans, but they have these milestones.

S

Sam Altman1:12:41

I do know some humans. People do like milestones. I totally get that. I think we like milestones too. It's fun to declare victory on this one and go start the next thing. But yeah, I feel like we're getting this a little bit wrong.

L

Lex Fridman1:12:59

So, when is GPT5 coming out again?

S

Sam Altman1:13:01

I don't know. That's an honest answer.

L

Lex Fridman1:13:04

Oh, that's the honest answer. Is it blink twice if it's this year? I also we will release an amazing new model this year. I don't know what we'll call it.

S

Sam Altman1:13:22

So that goes to the question of what's the way we release this thing. We'll release over the coming months many different things. I think that'll be very cool. I think before we talk about a GPT5-like model called that or not, I think we have a lot of other important things to release first.

L

Lex Fridman1:13:48

I don't know what to expect from GPT5. You're making me nervous and excited. What are some of the biggest challenges and bottlenecks to overcome for whatever it ends up being called? Is it on the compute side? Is it technical? All of these? What's the one big unlock?

S

Sam Altman1:14:33

It's all of these things together. The thing that OpenAI does really well is multiply 200 medium-sized things together into one giant thing. There's this distributed constant innovation happening. Even on the technical side, detailed aspects of every approach. How does that work with different teams? There are a few people who have to think about putting the whole thing together, but a lot of people try to keep most of the picture in their head. One thing I generally believe is that it's sometimes useful to zoom out and look at the entire map. Things come together in surprising ways, and having an understanding of that whole picture, even if most of the time you're operating in the weeds, pays off with surprising insights. I used to have a good map of the frontiers in the tech industry and could see connections. I don't really have that anymore. I'm super deep now. But I know it's valuable.

L

Lex Fridman1:16:09

You're not the man you used to be, Sam.

S

Sam Altman1:16:11

Very different job now than what I used to have.

L

Lex Fridman1:16:14

Speaking of zooming out, let's zoom out to another cheeky but profound thing you said. You tweeted about needing $7 trillion.

S

Sam Altman1:16:27

I did not tweet about that. I never said we're raising $7 trillion.

L

Lex Fridman1:16:31

Oh, that's somebody else.

S

Sam Altman1:16:32

Yeah. But you said, 'Fuck it, maybe eight.' I meme'd once. There's misinformation out in the world.

L

Lex Fridman1:16:39

Oh, you meme. But sort of misinformation may have a foundation of insight there. Look, I think compute is going to be the currency of the future. I think it will be maybe the most precious commodity in the world and we should be investing heavily to make a lot more compute. Compute is an unusual market. People think about the market for chips for mobile phones. There are 8 billion people, maybe 7 billion have phones, they upgrade every 2 years, so the market is 3 billion system-on-chips per year. If you make 30 billion, you won't sell 10 times as many because most people have one phone. But compute is different. Intelligence is going to be more like energy. At price X, the world will use this much compute; at price Y, this much. If it's really cheap, I'll have it reading my email all day, giving me suggestions, trying to cure cancer. If it's expensive, we'll only use it to cure cancer. So the world is going to want a tremendous amount of compute. Energy is the hardest part. Building data centers is hard. The supply chain is hard. Fabricating enough chips is hard. But this is where things are going. We're going to want an amount of compute that's hard to reason about right now. How do you solve the energy puzzle? Nuclear. That's what I believe. Fusion. That's what I believe. Nuclear fusion. Yeah. Who's going to solve that? I think Helion is doing the best work, but I'm happy there's a race for fusion. Nuclear fission is also quite amazing. I hope we can embrace that. It's sad how the history of that went. So to you, part of the puzzle is nuclear fission reactors as we currently have them, and a lot of people are terrified because of Chernobyl.

S

Sam Altman1:19:02

Well, I think we should make new reactors. It's a shame that industry ground to a halt.

L

Lex Fridman1:19:08

And what, just mass hysteria is how you explain the halt?

S

Sam Altman1:19:11

Yeah. I don't know if you know humans, but that's one of the security threats for nuclear fission. Humans seem to be really afraid of it. And that's something we have to incorporate into the calculus. We have to win people over and show how safe it is. I worry about that for AI. I think some things are going to go theatrically wrong with AI. I don't know what the percent chance is that I eventually get shot, but it's not zero. How do you decrease the theatrical nature of it? I've already started to hear rumblings from both sides of the political spectrum. AI is going to be politicized. That really worries me. Then the theatrical nature can be leveraged fully. How do you fight that?

L

Lex Fridman1:20:24

I think it will get caught up in left versus right wars. I don't know exactly what that's going to look like, but that's what happens with anything of consequence. Unfortunately, what I meant more about theatrical risks is that AI is going to have tremendously more good consequences than bad ones, but it is going to have bad ones. And there will be some bad ones that are bad but not theatrical. A lot more people have died of air pollution than nuclear reactors, but we worry more about living next to a nuclear reactor than a coal plant. Something about the way we're wired is that the ones that make a good climax scene of a movie carry much more weight than the ones that are very bad over a long period of time on a slow burn. Well, that's why truth matters. Hopefully AI can help us see the truth of things to have balance.

Understand what are the actual risks, what are the actual dangers of things in the world. What are the pros and cons of the competition in the space and competing with Google, Meta, XAI and others?

S

Sam Altman1:21:42

I think I have a pretty straightforward answer to this. The pros seem obvious, which is that we get better products and more innovation faster and cheaper. The con is that if we're not careful, it could lead to an increase in an arms race that I'm nervous about. Do you feel the pressure of the arms race in some negative way?

L

Lex Fridman1:22:11

Definitely in some ways for sure. We spend a lot of time talking about the need to prioritize safety. I've said for a long time that if you think of a quadrant of slow timelines to AGI, long timelines, and then a short or fast takeoff, I think short timeline, slow takeoff is the safest quadrant. But I do want to make sure we get that slow takeoff. Part of the problem with Elon is that silos are created as opposed to collaboration on safety. It tends to go into silos and closed or open source models. Elon says he cares a great deal about AI safety, and I assume he's not going to race unsafely.

S

Sam Altman1:23:06

Yeah. But collaboration here I think is really beneficial for everybody on that front.

L

Lex Fridman1:23:12

Not really the thing he's most known for. Well, he is known for caring about humanity, and humanity benefits from collaboration. So there's always attention and incentives. In the end, I do hope humanity prevails. I was thinking someone reminded me the other day about how the day he surpassed Jeff Bezos for richest person, he tweeted a silver medal at Jeff Bezos. I hope we have less stuff like that as people start to work on AGI.

S

Sam Altman1:23:44

I agree.

L

Lex Fridman1:23:44

Towards AGI.

S

Sam Altman1:23:46

I think Elon is a friend and a beautiful human being, one of the most important humans ever. That stuff is not good.

L

Lex Fridman1:23:53

The amazing stuff about Elon is amazing, and I super respect him. I think we all should be rooting for him and need him to step up as a leader through this next phase.

S

Sam Altman1:24:05

Yeah, I hope you can have one without the other, but sometimes humans are flawed and complicated. There are a lot of really great leaders throughout history.

L

Lex Fridman1:24:13

Yeah, and we can each be the best version of ourselves and strive to do so.

Let me ask you about Google. With the help of search, Google has been dominating the past 20 years. One nerve-wracking thing for Google and everyone in the space is thinking about how people are going to access information. People show up to GPT as a starting point. So is OpenAI going to really take on this thing that Google started 20 years ago?

S

Sam Altman1:24:58

I find that boring. If the question is whether we can build a better search engine than Google, then sure, people should use a better product. But that would understate what this can be. Google shows you 10 blue links and 13 ads. That's one way to find information, but the exciting thing is not to build a better copy of Google search, but that there's some much better way to help people find, act on, and synthesize information. ChatGPT is that for some use cases, and hopefully we'll make it like that for more. I don't think it's interesting to just give you 10 ranked web pages. Maybe it's interesting to help you get the answer, the information you need, and create and synthesize that. A lot of people have tried to make a better search engine than Google, and it's a hard technical, branding, and ecosystem problem. I don't think the world needs another copy of Google. Integrating a chat client like ChatGPT with a search engine is cool, but tricky. If you just shove it in, it can be awkward.

L

Lex Fridman1:26:30

That's cooler.

S

Sam Altman1:26:32

It's cool, but it's tricky. If you just do it simply, it's awkward.

L

Lex Fridman1:26:41

As you might guess, we are interested in how to do that well.

S

Sam Altman1:26:44

That would be an example of a cool thing that's not just like that.

L

Lex Fridman1:26:47

Well, like a heterogeneous integration.

S

Sam Altman1:26:49

The intersection of LLM plus search. I don't think anyone has cracked the code yet. I would love to go do that. I think that would be cool.

L

Lex Fridman1:26:59

Yeah. What about the ads side? Have you ever considered monetization?

S

Sam Altman1:27:02

I kind of hate ads as an aesthetic choice. Ads needed to happen on the internet to get it going, but the world is richer now. I like that people pay for ChatGPT and know the answers are not influenced by advertisers. There is an ad unit that makes sense for LLMs, and there's a way to participate in the transaction stream unbiasedly. But it's easy to think of dystopic visions where you ask ChatGPT and it recommends a product. We have a very simple business model, and I like it. I know that I'm paying and that's how it works. When I use ad-supported products like Twitter, Facebook, or Google, I don't love that, and I think it gets worse in a world with AI. I can imagine AI would be better at showing ads for things you actually need, but then does that system always result in ads driving the content? Wikipedia made a bold move not to do advertisements, but it makes the business model challenging. So, you're saying the current thing with OpenAI is sustainable from a business perspective? Well, we have to figure out how to grow, but it looks like we're going to figure that out. If the question is whether we can have a great business that pays for our compute needs without ads, the answer is yes.

L

Lex Fridman1:29:18

Well, that's promising. I also just don't want to completely throw out ads as a.

S

Sam Altman1:29:23

I'm not saying that. I'm saying I have a bias against them.

L

Lex Fridman1:29:28

Yeah, I have a bias and skepticism in general. In terms of interface, I have a spiritual dislike of crappy interfaces. AdSense was a big leap forward versus animated banners. But there should be many more leaps forward in advertisement that doesn't interfere with content consumption or manipulate the truth to suit advertisers. Let me ask you about safety and bias. Gemini 1.5 came out recently, and there was drama around it generating black Nazis and black founding fathers, which was on the ultra woke side. There's a concern that if there is a human layer within companies that modifies the safety or harm caused by a model, they introduce a lot of bias that fits an ideological lean. How do you deal with that?

S

Sam Altman1:30:51

We work super hard not to do things like that. We've made our own mistakes and will make others. Google will learn from this one. These are not easy problems. One thing we've been thinking about is writing out the desired behavior of a model, making it public, taking input, and explaining edge cases. Then when a model is not behaving as wanted, it's clear whether it's a bug or a policy debate. Black Nazi is obviously ridiculous, but there are subtle things you can make a judgment call on either way.

L

Lex Fridman1:31:41

Yeah, but sometimes if you write it out and make it public, you can use language that's very high level, like Google's AI principles. That doesn't work. You have to say, when you ask it to do thing X, it's supposed to respond in way Y. So like, who's better, Trump or Biden? What's the expected response? Something very concrete.

S

Sam Altman1:32:05

I'm open to a lot of ways a model could behave, but you should have to say here's the principle and here's what it should say in that case.

L

Lex Fridman1:32:11

That would be really nice. And then everyone agrees because there's anecdotal data. If there's clarity about representative examples, you can define. Then when it's a bug, it's a bug, and the company can fix it. It'd be much easier to deal with a black Nazi type of image generation if there are great examples.

So, San Francisco is a bit of an ideological bubble, tech in general as well. Do you feel the pressure of that within a company, that there's a lean towards the left politically that affects the product and teams?

S

Sam Altman1:32:52

I feel very lucky that we don't have the challenges at OpenAI that I've heard of at other companies. Every company has some ideological thing. We have one about AGI and belief in that, and it pushes out others. We are much less caught up in the culture war. San Francisco is a mess, but that doesn't infiltrate OpenAI in obvious ways. We've had our flare-ups, but not like what I hear about at other companies.

L

Lex Fridman1:33:36

What's the process for the bigger question of safety? How do you provide that layer that protects the model from doing crazy dangerous things?

S

Sam Altman1:33:48

I think there will come a point where that's mostly what we think about. It's not like you have one safety team. When we shipped GPT-4, it took the whole company thinking about all these aspects. More and more of the company will think about those issues all the time. That's what humans will be thinking about the more powerful AI becomes. Most employees at OpenAI will be thinking about safety to some degree.

L

Lex Fridman1:34:18

Broadly defined. Yes.

Yeah. I wonder what the full broad definition of that is? What are the different harms that could be caused? Is this on a technical level or what?

S

Sam Altman1:34:31

No, it'll be people, state actors trying to steal the model, technical alignment work, societal impacts, economic impacts. It's not just one team aligning the model. Getting to a good outcome is going to take the whole effort. How hard do you think people, state actors, are trying to hack OpenAI and infiltrate?

L

Lex Fridman1:35:06

They're trying.

S

Sam Altman1:35:10

What kind of accent do they have?

L

Lex Fridman1:35:13

I don't actually want any further details on this point.

S

Sam Altman1:35:15

Okay. But I presume it'll be more and more as time goes on.

L

Lex Fridman1:35:21

That feels reasonable.

Boy, what a dangerous space. What aspects of the leap from GPT-4 to GPT-5 are you excited about?

S

Sam Altman1:35:39

I'm excited about being smarter. I know that sounds glib, but the special thing is that it's getting better across the board, not just in one area. There's a magical moment when you talk to someone and they kind of get you. That's how I would characterize the progress of GPT. It's not that it didn't get this or that, but the degree of intellectual connection, feeling like it grasps the deeper question behind your question. I'm excited by that. All of us love being understood and heard. Even with programming, when GPT gives a completion that just gets what you're thinking about, it's a good feeling. Looking forward to getting even better.

L

Lex Fridman1:36:39

That's for sure.

That's a weird feel. Even with programming, when you're coding and the completion is just right, it's such a good feeling. On the programming front, how much do you think humans will be doing 5-10 years from now?

S

Sam Altman1:37:05

A lot, but in a very different shape. Some people may program entirely in natural language. No one programs in punch cards anymore. But the skill set will change. The best practitioners will use multiple tools, some in natural language, some in C.

L

Lex Fridman1:37:13

Entirely natural language.

S

Sam Altman1:37:15

I mean no one programs in bite code. Some people do, but you know what I mean.

L

Lex Fridman1:37:25

Yeah. You're going to get a lot of angry comments. There are very few. I've been looking for people who program in Fortran. It's hard to find. But that changes the nature of the skill set.

S

Sam Altman1:37:35

Changes the skill set. How much it changes the predisposition, I'm not sure.

L

Lex Fridman1:37:45

Oh, same kind of puzzle solving.

S

Sam Altman1:37:49

Programming is hard. How to get that last 1% to close the gap?

L

Lex Fridman1:37:55

Yeah. I think with most other cases, the best practitioners will use multiple tools.

Will we see humanoid robots or humanoid robot brains from OpenAI at some point?

S

Sam Altman1:38:14

At some point. How important is embodied AI? I think it's depressing if we have AGI and the only way to get things done in the physical world is to make a human do it. I hope we also get humanoid robots or some physical world robots.

L

Lex Fridman1:38:38

OpenAI has quite a bit of history in robotics.

S

Sam Altman1:38:41

Yeah, but it's a small company. We have to focus, and robots were hard for the wrong reason at the time. We will return to robots in some way at some point.

L

Lex Fridman1:38:57

That sounds both inspiring and menacing.

S

Sam Altman1:39:00

Why? Because immediately we will return to robots? It's kind of like Terminator.

L

Lex Fridman1:39:06

We will return to work on developing robots. We will not turn ourselves into robots.

Yeah. When do you think we as humanity will build AGI?

S

Sam Altman1:39:17

I used to love to speculate on that, but I've realized the question is poorly formed. People use different definitions. It makes more sense to talk about when we'll build systems that can do capability X, Y, or Z. AGI is not an ending, it's closer to a beginning. But in the interest of not dodging the question, I expect by the end of this decade, possibly sooner, we will have quite capable systems that we look at and say wow. If you look at ChatGPT-3.5 and show it to Alan Turing, or people in the '90s, they would say it's definitely AGI. But I don't think GPT-3.5 changed the world. It changed the world's expectations, which is important. I could retire after that accomplishment and be happy, but as an artifact, I don't think we'll look back and say it was a threshold that changed the world.

L

Lex Fridman1:41:03

That was a threshold that really changed the world itself. So to you, you're looking for some major transition in how the world works.

S

Sam Altman1:41:10

For me, that's part of what AGI implies.

L

Lex Fridman1:41:15

Like singularity level transition?

S

Sam Altman1:41:17

Definitely not.

L

Lex Fridman1:41:18

But just a major like the internet or Google search did. What was a transition point? Does the global economy feel any different now than before we launched GPT-4? I would say no.

S

Sam Altman1:41:33

No. It might be a really nice tool for a lot of people, but doesn't feel different. People define AGI differently, but for me, that should be part of it.

L

Lex Fridman1:41:48

There could be major theatrical moments also. What to you would be an impressive thing AGI would do? Like you are alone in a room with the system.

S

Sam Altman1:41:59

This is personally important to me. I think when a system can significantly increase the rate of scientific discovery, that's a huge deal. Most real economic growth comes from scientific and technological progress.

L

Lex Fridman1:42:21

I agree with you. Hence why I don't like the skepticism about science in recent years.

S

Sam Altman1:42:28

Totally.

L

Lex Fridman1:42:29

But actual measurable rate of scientific discovery. Even just seeing a system have really novel scientific intuitions would be incredible.

Yeah. You quite possibly would be the person to build the AGI and interact with it before anyone else. What kind of stuff would you talk about?

S

Sam Altman1:42:55

Definitely the researchers here will do that before I do. But I've thought about this question. If someone were like, 'Okay Sam, here's a laptop, this is the AGI, you can go talk to it,' I find it surprisingly difficult to say what I would ask. That first AGI is not going to be the one that explains the grand unified theory of physics. I'd love to ask that question, but I don't expect it to answer.

L

Lex Fridman1:43:41

You can ask yes or no questions. Does such a theory exist? Can it exist?

S

Sam Altman1:43:46

Well then those are the first questions I would ask.

L

Lex Fridman1:43:48

Yes or no. And then based on that, are there other alien civilizations out there? Yes or no? What's your intuition? And then you just ask that.

S

Sam Altman1:43:56

Yeah. I don't expect that first AI could answer any of those questions, but if it could, those would be very high on my list.

L

Lex Fridman1:44:06

Maybe it can start assigning probabilities.

S

Sam Altman1:44:08

Maybe. Maybe we need to invent more technology and measure more things first.

L

Lex Fridman1:44:12

But if it's AGI, it just doesn't have enough data.

S

Sam Altman1:44:17

I mean as a kid, you know, you want to know the answer to this question about physics. I need you to build this machine and make these five measurements and tell me that.

L

Lex Fridman1:44:26

Yeah. Like what the hell do you want from me? I need the machine first, and I'll help you deal with the data from that machine. Maybe I'll help you build a machine.

S

Sam Altman1:44:33

Maybe. Maybe.

L

Lex Fridman1:44:35

And on the mathematical side, maybe prove some things. Are you interested in that side of things? The formalized exploration of ideas. Whoever builds AGI first gets a lot of power. Do you trust yourself with that much power?

S

Sam Altman1:45:00

I'll be very honest. It is important that I nor any other one person have total control over OpenAI or AGI. You want a robust governance system. I can point to the board drama from last year. I didn't fight it initially, but later I did. Although the board had the legal ability to fire me, in practice it didn't quite work. That is its own governance failure. I can defend the specifics, but it makes it harder for me to say the board can just fire me. I continue to not want super voting control over OpenAI. I never had it, never wanted it. Even after all this craziness, I still don't want it. I continue to think no company should be making these decisions, and we need governments to put rules in place. I realize people will claim regulatory capture, but I'm willing to be misunderstood. I think I've made plenty of bad decisions and good ones. I don't think any one person should be in control of AGI, and that's not what's happening.

L

Lex Fridman1:47:46

Thank you for saying that. That was really powerful. The idea that the board can fire you is legally true, but human beings can manipulate the masses. But there's also a positive version where the people still have power. So the board can't be too powerful either. There's a balance of power.

S

Sam Altman1:48:15

Balance of power is a good thing for sure.

L

Lex Fridman1:48:20

Are you afraid of losing control of the AGI itself? Many people worry about existential risk, not because of state actors, but because of the AI itself.

S

Sam Altman1:48:31

That is not my top worry. There have been times I worried about that more, and it may be again. But it's not my top worry right now.

L

Lex Fridman1:48:39

What's your intuition about it not being your worry? There's a lot of other stuff to worry about.

S

Sam Altman1:48:46

You think you could be surprised? We could be surprised. Saying it's not my top worry doesn't mean I don't think we need to work on it. It's super hard, and we have great people working on it. But there are other things we have to get right. It's not super easy to escape the box at this time. Theatrical risks like connecting to the internet can take over how people think. There's a group of smart, well-meaning AI safety researchers who got hung up on this one problem without much progress. I'm happy they do that because we need to think about it, but it pushed aside other significant AI-related risks.

L

Lex Fridman1:49:47

Let me ask you about you tweeting with no capitalization. Is the shift key broken on your keyboard?

S

Sam Altman1:49:53

Why does anyone care about that?

L

Lex Fridman1:49:55

I deeply care.

S

Sam Altman1:49:56

But why? Other people ask me too.

L

Lex Fridman1:50:00

Yeah. Any intuition?

S

Sam Altman1:50:03

I think it's the same reason. There's a poet E. Cummings who doesn't use capitalization to say 'screw you' to the system. People are paranoid because they want you to follow the rules.

L

Lex Fridman1:50:15

You think that's what it's about?

S

Sam Altman1:50:16

I think it's like this guy doesn't follow the rules. He doesn't capitalize his tweets.

L

Lex Fridman1:50:20

This seems really dangerous.

S

Sam Altman1:50:22

He seems like an anarchist.

L

Lex Fridman1:50:25

Are you just being poetic, hipster? What's the rationale?

S

Sam Altman1:50:30

I grew up as a very online kid. I spent a lot of time chatting, and I never capitalized there. I think most internet kids didn't. I actually think capitalization has gone down over time. I personally think it's a dumb construct to capitalize the letter at the beginning of a sentence. I used to capitalize my tweets to sound professional, but I haven't capitalized my private DMs in a long time. Shorter, less formal stuff has drifted to how I text my friends. If I write a formal memo, I always capitalize. I still remember how to do it. But I don't spend time thinking about this.

L

Lex Fridman1:52:09

Well, it's good to know the shift key is not broken. I was mostly concerned about your wellbeing.

S

Sam Altman1:52:16

I wonder if people still capitalize their Google searches or ChatGPT queries. Probably not.

L

Lex Fridman1:52:27

Probably not. But there's a percentage. The thing that would make me do it is if it felt like a sign of respect. But I don't think there's disrespect. It's just conventions of civility that have momentum, and you realize it's not actually important.

S

Sam Altman1:52:46

Yeah. I don't think anybody else thinks about this as much. I'm grateful we clarified.

L

Lex Fridman1:53:16

You're the only CEO of a company that doesn't capitalize tweets.

S

Sam Altman1:53:20

I don't even think that's true, but maybe.

L

Lex Fridman1:53:22

All right. Given Sora's ability to generate simulated worlds, let me ask you a psychedelic question. Does this increase your belief that we live in a simulation, maybe a simulated world generated by an AI system?

S

Sam Altman1:53:50

Yes, somewhat. I don't think that's the strongest piece of evidence, but the fact that we can generate worlds should increase everyone's probability somewhat. I was certain we would be able to do something like Sora at some point, but it happened faster than I thought. That was not a big update. The fact that you can generate novel worlds makes you think about how easy it is to create universes. Someone said to me recently a profound insight about the square root function. You can ask what is the square root of negative one, and that tips you into a whole other kind of reality. I think there are a lot of those operators for why people may think the simulation hypothesis is more likely. But for me, Sora is not in the top five. I do think AI will serve as those kinds of gateways to another way of seeing reality.

L

Lex Fridman1:56:46

That's pretty exciting. I haven't done ayahuasca, but I will soon in the Amazon jungle. It's exciting but terrifying. The jungle is a machine of nature that renews itself. It makes you appreciate that this human thing came from somewhere. Hopefully I'll make it out alive. Do you think there are other intelligent alien civilizations?

S

Sam Altman1:58:03

I deeply want to believe the answer is yes. I find the Fermi paradox very puzzling. It scares me that intelligence is not good at handling powerful technologies. But I'm pretty confident there are a large number of intelligent alien civilizations. It might just be difficult to travel through space. It also makes me think about the nature of intelligence. Maybe we're blind to what intelligence looks like, and AI will help us see that it's not as simple as IQ tests.

L

Lex Fridman1:58:52

What gives you hope about the future of humanity?

S

Sam Altman1:58:59

The past gives a lot of hope. In a short period of time, humanity has done a lot. Huge problems, deep flaws, but on the whole, very inspiring. The trajectory of it all, that we're together pushing towards a better future. I wonder if AGI will be more like a single brain or the scaffolding in society between all of us. You haven't had much genetic drift from your great-great-grandparents, but what you're capable of is dramatically different because of the scaffolding we all contributed to. No one person built the iPhone, but you get to use it. That collective creation fills me with hope.

L

Lex Fridman2:00:24

That was a very collective thing.

S

Sam Altman2:00:26

Yeah, we really are standing on the shoulders of giants. You mentioned when we talked about theatrical AI risks, sometimes you might be afraid for your own life. Do you think about your death? Are you afraid of it?

L

Lex Fridman2:00:43

I mean, if I got shot tomorrow and knew it today, I'd be like, 'Oh, that's sad. I want to see what's going to happen.' What a curious time. But I would feel very grateful for the moments I got. Yeah, me too. It's a pretty awesome life. I get to enjoy awesome creations of humans, of which I believe ChatGPT is one. Sam, it's an honor and pleasure to talk to you again.

S

Sam Altman2:01:20

Great to talk to you. Thank you for having me.

L

Lex Fridman2:01:24

Thanks for listening to this conversation with Sam Altman. To support this podcast, please check out our sponsors. Now, let me leave you with some words from Arthur C. Clarke: 'It may be that our role on this planet is not to worship God, but to create him.' Thank you for listening and hope to see you next time.

#419 – Sam Altman: OpenAI, GPT-5, Sora, Board Saga, Elon Musk, Ilya, Power & AGI

About Sam Altman