Sam Altman Speaks Out About What Happened at OpenAI | What Now? with Trevor Noah Podcast

1. 0:00 – 1:08

   OpenAI leadership whiplash: Altman fired, employee revolt, and rapid reinstatement

   1. TNTrevor Noah0:00

      Sam Altman is out as CEO of OpenAI

   2. SPSpeaker0:03

      A superstar CEO on one side, a disgruntled board on the other

   3. TNTrevor Noah0:07

      747 of 770 employees sent a scathing open letter to the board

   4. SPSpeaker0:12

      Five days after he was unexpectedly fired, Sam Altman is back Does this even count as a firing? This was a brutal... I guess I'm not really supposed to talk about this right now [upbeat music]

   5. TNTrevor Noah0:27

      This is What Now? with Trevor Noah [upbeat music] Hey, what's going on?

   6. SPSpeaker0:44

      Nice to meet you

   7. TNTrevor Noah0:45

      How you doing?

   8. SPSpeaker0:45

      Good. How are you?

   9. TNTrevor Noah0:45

      Absolute pleasure, man

   10. SPSpeaker0:46

       Mine too

   11. TNTrevor Noah0:49

       Thanks for taking the time

   12. SPSpeaker0:50

       Thank you

   13. TNTrevor Noah0:51

       [laughs] Uh, at what I feel like is a crazy time, right?

   14. SPSpeaker0:56

       Feels like the craziest time I have lived, yet lived through

   15. TNTrevor Noah0:59

       Yeah. I mean, you, you're at the, you're at the center of it all, so I wonder what that feels like. 'Cause I'm, I'm just an avid watcher of everything in this space-

   16. SPSpeaker1:08

       Yeah

2. 1:08 – 2:30

   Fame, scrutiny, and being the public face of “the future”

   1. TNTrevor Noah1:08

      ... and this world, you know? And I feel like you're somebody who's been affected by it all. I mean, I mean, just, just, just right now, we get the news Sam Altman was on, on the shortlist for Time Magazine Person of the Year, right?

   2. SPSpeaker1:21

      Thrilled not to get that

   3. TNTrevor Noah1:22

      Thrilled not to get it?

   4. SPSpeaker1:22

      Yeah, of course

   5. TNTrevor Noah1:23

      Wait, why?

   6. SPSpeaker1:24

      Um, I have had more attention this year than I would've liked to have my entire life, and that is a big one.

   7. TNTrevor Noah1:31

      Huh.

   8. SPSpeaker1:31

      And I'm happy for Taylor Swift.

   9. TNTrevor Noah1:32

      Okay, okay. Oh, so you don't want... You don't like the attention? You don't want the attention?

   10. SPSpeaker1:37

       No, it's been brutal. I mean, it's, like, fun in some ways, and it's, like, useful in some ways, but, uh, from, like, a personal life, like, quality-of-life trade-off-

   11. TNTrevor Noah1:46

       Yeah

   12. SPSpeaker1:47

       ... yeah, definitely not.

   13. TNTrevor Noah1:48

       Yeah.

   14. SPSpeaker1:48

       But, you know, this is it now. Like, this is what I signed up for

   15. TNTrevor Noah1:51

       Right. It's, it's, it's the, the infamy now.

   16. SPSpeaker1:53

       Yeah.

   17. TNTrevor Noah1:54

       Do people recognize you in the street?

   18. SPSpeaker1:55

       That's the kind of trade-off that's really bad. Yeah.

   19. TNTrevor Noah1:57

       Yeah

   20. SPSpeaker1:57

       You just feel like you never... I'm sure it happens to you, too, but I never get to be anonymous anymore.

   21. TNTrevor Noah2:00

       Yeah, but people don't ask me about the future.

   22. SPSpeaker2:02

       They don't ask you if you're gonna destroy the world. Yeah.

   23. TNTrevor Noah2:03

       Exactly.

   24. SPSpeaker2:04

       Absolutely.

   25. TNTrevor Noah2:04

       Exactly. There's a, there's a slight difference.

   26. SPSpeaker2:06

       [laughs]

   27. TNTrevor Noah2:06

       People might want a selfie from me. That's, that's, that's the extent of it

   28. SPSpeaker2:09

       A, a lot of selfies

   29. TNTrevor Noah2:10

       [laughs] Uh, well, congratulations. You are Time Magazine's CEO of the Year.

   30. SPSpeaker2:15

       Yeah.
3. 2:30 – 4:18

   Back in the CEO seat: emotional fallout and why he chose to return

   1. TNTrevor Noah2:30

      I don't know how it works. How does, how does it feel to be back in CEO?

   2. SPSpeaker2:35

      Um, I'm still, like, recompiling reality, to be honest.

   3. TNTrevor Noah2:41

      Yeah?

   4. SPSpeaker2:41

      Uh, I mean, it feels great in some sense because one of the things I learned through this whole thing is how much I love the company and the mission and the people.

   5. TNTrevor Noah2:51

      Right

   6. SPSpeaker2:51

      And, um, you know, I, I had a couple of experiences in that whole thing where I went through, like, all of the, like, the full range of human emotion it felt like-

   7. TNTrevor Noah3:02

      Mm-hmm

   8. SPSpeaker3:02

      ... in short periods of time. Um, but a very clarifying moment for me was, uh, the, like... So it all happened on, like, a Friday afternoon, a Friday at noon, and then the next S- the next morning, a Saturday morning, a couple of the board members called me and said, "You know, would, would you like to talk about coming back?" And I had really complicated feelings about that. Um, but it was very clarifying at the end of it to be like, "Yes, I do." Like, I really, I love this place and what we're doing, and I think this is, like, important to the world, and, like, the thing I care most about.

   9. TNTrevor Noah3:36

      It feels like in, in the world of tech, hiring and firing is something that everybody has to get used to. I mean, I know, I know back in the day, um, you know, you were at the, was the Y Combinator-

   10. SPSpeaker3:47

       Mm-hmm

   11. TNTrevor Noah3:47

       ... right? And you were fired from that position. And everyone has a story in... Wait, wait. What, what is-

   12. SPSpeaker3:52

       I don't wanna debate that. Uh, we can-

   13. TNTrevor Noah3:54

       No, no, no, no, no

   14. SPSpeaker3:54

       ... no, no, no. It's all right

   15. TNTrevor Noah3:55

       No, no, no. Don't. Tell me. Tell me. The- these are things. You know, you, you get the research and then you, you go from there.

   16. SPSpeaker4:00

       Oh, uh, I mean, I had, like, decided, like, a year earlier that I wanted to just come to OpenAI.

   17. TNTrevor Noah4:06

       Right.

   18. SPSpeaker4:06

       And it was, like, a complicated transition-

   19. TNTrevor Noah4:08

       Right

   20. SPSpeaker4:08

       ... to get here. But I, I, like, I, I had been working on both OpenAI and YC and, like, very much decided that I wanted to go do OpenAI. Um-

   21. TNTrevor Noah4:16

       Okay. Okay

   22. SPSpeaker4:17

       ... and I've never regretted that one.

   23. TNTrevor Noah4:18

       All right.

   24. SPSpeaker4:18

       That's-

4. 4:18 – 5:10

   Is it a firing if you come right back? Pain, fairness, and the human impact

   1. TNTrevor Noah4:18

      So then, so then you've never been fired, then. This is a tough place to be in as a person.

   2. SPSpeaker4:21

      Not-

   3. TNTrevor Noah4:22

      Does it, does this even count as a firing? Like, if you get fired and then you immediately hired back-

   4. SPSpeaker4:25

      Oh, no, what I was gonna say is not only... Like, this was a brutal... I guess I'm not really supposed to talk about this right now. Um, it was, this was a very painful thing-

   5. TNTrevor Noah4:33

      Well, I, I-

   6. SPSpeaker4:33

      ... and felt to me personally-

   7. TNTrevor Noah4:35

      Yeah

   8. SPSpeaker4:35

      ... just as a human, like, super unfair the way it was handled.

   9. TNTrevor Noah4:38

      Yeah. Yeah. I can imagine. You know, a lot of people will talk about, you know, getting, getting fired from their jobs. It became, it became a trend, I guess, during COVID especially. People would talk about getting an email or a mass video that would go out and then, you know, thousands of employees would, would, would be let go. You seldom think it would be possible for that to happen to a CEO of a company. And then, and then I think even more so, you don't think of it happening to a CEO who many people have, have termed, like, the Steve Jobs of this generation and, and, and the future. And I... You don't say that about yourself, by the way.

   10. SPSpeaker5:09

       Certainly not.

5. 5:10 – 8:22

   Prometheus, AGI, and why Altman thinks the short-term impact is overstated

   1. TNTrevor Noah5:10

      No, I think, I think a lot of people say that about you, you know, because... I mean, I was thinking about this and I was going, I, I, I think calling you the Steve Jobs of this generation is, is unfair. In my opinion, I think you're the Prometheus of this generation.

   2. SPSpeaker5:24

      Whoa.

   3. TNTrevor Noah5:24

      No, you really are. You really are. You... It, it seems like to me you have stolen fire from the gods-

   4. SPSpeaker5:31

      [laughs]

   5. TNTrevor Noah5:31

      ... and you are at the forefront of this, of this movement and this, and this time that we're now living through, where once AI was, was only the stuff of, of sci-fi and legend. You know, you, you are now the face at the forefront of what could change-

   6. SPSpeaker5:46

      Yeah

   7. TNTrevor Noah5:46

      ... like, civilization forever. Do you think it'll change everything-

   8. SPSpeaker5:49

      I do

   9. TNTrevor Noah5:49

      ... going forward?

   10. SPSpeaker5:49

       I do. I mean, I, I could totally be wrong about what I'm about to say, but my, my sense is we will build something that everybody, almost everybody agrees is AGI. The definition is hard, but we will build something that people will look at and say, "All right- You all did that. And then-

   11. TNTrevor Noah6:06

       That's artificial general intelligence.

   12. SPSpeaker6:08

       Yeah, yeah, yeah.

   13. TNTrevor Noah6:08

       Right?

   14. SPSpeaker6:09

       A hu- like, you know, a human-level or beyond human-level system.

   15. TNTrevor Noah6:11

       Well, let's, b- before you go into the details on that, like, what would you say is the biggest difference between what people think AI is and what artificial general intelligence is?

   16. SPSpeaker6:23

       We're getting close enough that the way people define it is important, and there are differences in it.

   17. TNTrevor Noah6:29

       Okay.

   18. SPSpeaker6:29

       So for some people, they mean a system that can, like, do some significant fraction of current human work.

   19. TNTrevor Noah6:35

       Yeah.

   20. SPSpeaker6:35

       Of course, we'll find new jobs, we'll find new things to do. Um, but for other people, uh, they mean something more like a system that can help discover new scientific knowledge.

   21. TNTrevor Noah6:43

       Okay.

   22. SPSpeaker6:44

       And those are obviously very different milestones, have very different impact on the world. Um, but the reason I don't like the term anymore, uh, even though I'm so stuck with it I can't stop myself from using it-

   23. TNTrevor Noah6:53

       You don't like which term?

   24. SPSpeaker6:54

       AGI.

   25. TNTrevor Noah6:55

       Okay.

   26. SPSpeaker6:56

       B- all, all that I think it really means to most people now is, like, really smart AI. But it's-

   27. TNTrevor Noah7:01

       [laughs]

   28. SPSpeaker7:02

       ... uh, it's become super fuzzy in what it is other than that bec-

   29. TNTrevor Noah7:04

       Right

   30. SPSpeaker7:04

       ... and I think largely just because we're getting closer. But the point I was gonna try and make was we're gonna make AGI, whatever you wanna call that.
6. 8:22 – 10:44

   Iterative deployment: why releasing gradually beats building in secret

   1. TNTrevor Noah8:22

      Hmm. Is that, is that something you think is good for us as humanity and society? Uh, is life supposed to just go on?

   2. SPSpeaker8:30

      I think-

   3. TNTrevor Noah8:31

      Or as, as, as, as one of the fathers of this product, one of the parents of this idea, do, do you wish that we all stopped and took a moment to, I guess, take stock of where we are?

   4. SPSpeaker8:44

      I think the resilience of humans individually and humanity as a whole is fantastic.

   5. TNTrevor Noah8:52

      Okay.

   6. SPSpeaker8:52

      And I'm very happy that we have this ability to absorb and adapt to new technology, to changes, and have it become, um, just like, you know, part of the world. The, it, it, it really is wonderful. Um, the, I think COVID was a recent example where we watched this. It-

   7. TNTrevor Noah9:14

      Yeah. Yeah. Yeah

   8. SPSpeaker9:14

      ... you know, like, the world kind of just adapted pretty quickly.

   9. TNTrevor Noah9:18

      Yeah.

   10. SPSpeaker9:18

       And then it felt pretty normal pretty quickly. Um, I mean, another example in a sort of non-serious way, but instructive, was when all of that UFO stuff came out.

   11. TNTrevor Noah9:27

       [laughs]

   12. SPSpeaker9:27

       This was a couple of years ago now.

   13. TNTrevor Noah9:29

       Yeah.

   14. SPSpeaker9:30

       A, a lot of my friends would say things like, "Hmm, maybe those are real UFOs or real aliens or whatever."

   15. TNTrevor Noah9:37

       Yeah. And, and yet everyone-

   16. SPSpeaker9:37

       People who were real skeptics.

   17. TNTrevor Noah9:38

       Yeah.

   18. SPSpeaker9:39

       And yet they just kinda, like, went to work and played with their kids the next day.

   19. TNTrevor Noah9:42

       Yeah, because I mean, what are you gonna do?

   20. SPSpeaker9:43

       What are you gonna do?

   21. TNTrevor Noah9:44

       What are you gonna do? If they're flying by, they're flying by. What are you gonna do?

   22. SPSpeaker9:47

       So, so do I wish that we had taken more, more of a time to take stock? Um, we are doing that as a world-

   23. TNTrevor Noah9:55

       Mm-hmm

   24. SPSpeaker9:55

       ... and I think that's great. I'm a huge believer that iterative deployment of these technologies are, are really important. We want-- We don't wanna go build AGI off in secret in a lab-

   25. TNTrevor Noah10:05

       Yeah

   26. SPSpeaker10:05

       ... have no one know it's coming-

   27. TNTrevor Noah10:06

       Uh-huh

   28. SPSpeaker10:06

       ... and then drop it on the world all at once and have people have to say, like, "Huh, here we are." Like-

   29. TNTrevor Noah10:12

       You think we have to get used to it gradually and, and sort of grow with it-

   30. SPSpeaker10:15

       Yeah
7. 10:44 – 12:03

   The ‘guest badge’ day: adrenaline, exhaustion, and interviewing for an entry-level role

   1. TNTrevor Noah10:44

      You know, it's interesting, you, today we, we walked into the, um, into the OpenAI buildings. It's like a little bit of a fortress, and it feels like-

   2. SPSpeaker10:51

      [laughs]

   3. TNTrevor Noah10:51

      ... feels like the, the home of the future, and I, I saw a post of yours where... D- did you come in as a guest today?

   4. SPSpeaker10:57

      Not anymore. I'm, I'm back now. I did-

   5. TNTrevor Noah10:59

      Okay. Okay

   6. SPSpeaker11:00

      ... one day during the middle of it.

   7. TNTrevor Noah11:00

      All right. I saw you, I saw you, I saw you had a post where you came in as a guest, and I was like, "Damn, that's, that's, that's a weird one." You know, it's like coming home, but then it's not home-

   8. SPSpeaker11:07

      Yeah, that was-

   9. TNTrevor Noah11:08

      ... but then it is home

   10. SPSpeaker11:09

       ... it, it felt like it should have been somehow a very strange moment, um, to, like, put on a guest badge here.

   11. TNTrevor Noah11:16

       Yeah.

   12. SPSpeaker11:17

       But I was like, everyone was, like, so tired, so exhausted, on so much adrenaline.

   13. TNTrevor Noah11:21

       Yeah.

   14. SPSpeaker11:21

       It really did not feel momentous in the way that I guess I could say I had hoped it would. It should've been, like, a funny, you know, moment to, like, reflect on and tell stories.

   15. TNTrevor Noah11:32

       Yeah.

   16. SPSpeaker11:32

       There, there were moments that day that were like that. Like, one of my proudest moments of that day is, uh, I, I, I was very tired, very distracted, um, and you know, we thought the board was gonna put me back as CEO that day, but in case they didn't, uh, I got interviewed for, uh, L3, which is, like, our lowest level software engineering job-

   17. TNTrevor Noah11:50

       [laughs]

   18. SPSpeaker11:50

       ... by one of our, like, best people, and, uh, you know, he gave me a yes. That was, like, a very proud moment.

   19. TNTrevor Noah11:55

       Okay. Okay.

   20. SPSpeaker11:56

       That, that sticks out in my memory that day.

   21. TNTrevor Noah11:57

       So you still got the, you still got, you still got the skills on the ground.

   22. SPSpeaker11:58

       But the, um, the, the, the, the badge was not as poignant as I would've hoped.

8. 12:03 – 15:46

   Why employees rallied: mission over personality and defining OpenAI’s purpose

   1. TNTrevor Noah12:03

      Right. I'd love to know what you think you've done right as CEO. To have the level of support that we've publicly seen from the people who work at OpenAI. You know, wh- when the story broke, and I, I won't ask you for the details 'cause I know you-

   2. SPSpeaker12:17

      Go ahead

   3. TNTrevor Noah12:17

      ... no, no, no, 'cause I know you can't comment-

   4. SPSpeaker12:18

      No, I might

   5. TNTrevor Noah12:18

      ... about the internal investigation stuff-

   6. SPSpeaker12:20

      Yeah, I won't

   7. TNTrevor Noah12:20

      ... but what, what, yeah, I mean that stuff. But what I mean is, you know, I, I know you can sort of speak about just the feelings and the, and, and, and what's been happening in the company as a whole. It's rare that we'll see a situation, um, unfold the way it did with OpenAI. You know, you, you, you have this company and this idea that for one minute doesn't exist for most people on the globe. The next minute, you release ChatGPT, this simple prompt, just a little chat box that changes the world. I think you go to one million users in, in the fastest time of any te- tech product, and then it-

   8. SPSpeaker12:52

      I think five days

   9. TNTrevor Noah12:52

      ... yeah, five days, and then it shoots to 100 million people.

   10. SPSpeaker12:55

       Yeah.

   11. TNTrevor Noah12:55

       And it, it very quickly, I know on an anecdotal level, for me, it went from nobody in the world knew what this thing was.

   12. SPSpeaker13:00

       Yeah.

   13. TNTrevor Noah13:01

       I was explaining it to people, trying to get them to understand it.

   14. SPSpeaker13:03

       Yeah.

   15. TNTrevor Noah13:03

       I'd have to show them, like, poetry-

   16. SPSpeaker13:04

       Yeah

   17. TNTrevor Noah13:04

       ... and simple things they would get. And then people are telling me about it, and now it just becomes this ubiquitous idea where people are trying to come to grips with what it is and what it means. But, but on the other side, you have this company that's trying to, in some way, shape, or form, harness and shape the future. And, and the people are behind you. You know, we see the story. Sam Altman is out, no longer CEO, and then the swirling everything. I mean-

   18. SPSpeaker13:30

       It was wild

   19. TNTrevor Noah13:30

       ... I, I don't know if you, I don't know if you saw some of the rumors. They were crazy. One of the craziest things I saw was they said, like s- [laughs] someone said, it was, it was, it was wild and funny, they said, "I have it on good, good, um, I have it from good sources that Sam was fired for trying to have sex with the AI." That's, that's what someone... [laughs]

   20. SPSpeaker13:47

       I, I mean, I don't even know how I'm supposed to react to that.

   21. TNTrevor Noah13:49

       [laughs] I saw that and I was like-

   22. SPSpeaker13:52

       I guess I, in the, given the moment, I, I, I should officially deny that, which did not happen.

   23. TNTrevor Noah13:56

       Yeah, and I, I don't think it could happen-

   24. SPSpeaker13:59

       Yeah

   25. TNTrevor Noah13:59

       ... because I don't think people understand, um, the combination of the two things. But, but what, what got me was how the salaciousness of the event seemed to bring OpenAI into a different spotlight and a different moment, and one of the big things was the support you had from your team. Like, people coming out and saying, "We're with Sam, no matter what happens. We're, we..." And, and that doesn't normally happen in companies. CEOs and its employees are generally in, in some way, shape, or form disconnected.

   26. SPSpeaker14:26

       Yeah.

   27. TNTrevor Noah14:27

       But it feels like this is more than just a team.

   28. SPSpeaker14:28

       Um, what I'm about to say is not false modesty at all. There's plenty of places I willingly take a lot of credit. Uh, I think this one, though, is not about me other than me as sort of like a, a figurehead representation, but it, like, I think one thing that we have done well here is a mission that people really believe in the importance of. And it was a, I think what happened there was, like, people realized that the mission and the organization and the team that we have all worked so hard on and made such progress to but have so much more to do-

   29. TNTrevor Noah15:09

       Mm-hmm

   30. SPSpeaker15:09

       ... like, that was, uh, under real threat. And I think that was the, that was what got the reaction. It was really not about me personally, although, you know, hopefully people like me and think I do a good job. It was about the, the shared loyalty we all feel, uh, and the sense of duty to completing the mission and wanting to maximize our chances at the ability to do that.
9. 15:46 – 20:19

   Capitalism, cost, and scaling: the economic reality of frontier AI

   1. TNTrevor Noah15:46

      Okay. That's, that's an interesting, um, you know, second line, and I would love to chat to you about that later, you know, getting into the safety of it all. When, when you look at OpenAI as a, as an organization, the very genesis of OpenAI was really strange. You know? And, and you'll correct-

   2. SPSpeaker16:00

      Totally

   3. TNTrevor Noah16:00

      ... you'll correct me at any point-

   4. SPSpeaker16:01

      Well-

   5. TNTrevor Noah16:01

      ... if I'm wrong, but, you know, it, it, it seems like it was started very much with safety in mind, you know? Where, where you brought this team of people together and you said, "We wanna start an organization, a company, a collective that is trying to create the most ethical AI possible-

   6. SPSpeaker16:18

      Yeah

   7. TNTrevor Noah16:18

      ... that'll benefit society." And you see that even in the, in, even in the, uh, I guess the profits, the, the, the way the company defines how its investors could receive the profit, et cetera. But even, even that changed at some point in Open, in OpenAI. D- do you think, do you think you can withstand the forces of capitalism? I mean, there's so much money in this. Do you think that you can, you can truly maintain a world where money doesn't define what you're trying to do and why you're trying to do it?

   8. SPSpeaker16:45

      It's, it has to be some factor. Like, just if you think about the costs of training these systems alone, we have to find some ways to play on the field of capitalism, for lack of a better phrase.

   9. TNTrevor Noah17:00

      Okay.

   10. SPSpeaker17:01

       But I don't think it will ever, uh, be our primary motivator. The-- And by the way, I like capitalism. I think it has huge flaws, but relative to any other system the world has tried, I think it is still the best thing we've come up with, but that doesn't mean we, we, we shouldn't strive to do better. Um, and I think we will find ways to, uh, spend the enormous, like, record-setting amounts of capital that we will need to be able to continue to advance the forefront of this technology.

   11. TNTrevor Noah17:35

       Mm-hmm.

   12. SPSpeaker17:36

       Um, that was, like, one of our learnings early on is just this stuff is way more expensive than we ever thought.

   13. TNTrevor Noah17:41

       Right.

   14. SPSpeaker17:41

       Like, we knew, we kind of knew, we had this idea of scaling systems, but we just didn't know how far it was gonna go.

   15. TNTrevor Noah17:47

       You've, you've always been a big fan of scaling. That's, that's something I-

   16. SPSpeaker17:49

       That's like a-

   17. TNTrevor Noah17:50

       ... I've read about you is like-

   18. SPSpeaker17:51

       Yeah

   19. TNTrevor Noah17:51

       ... you know, even when, when, when, um, one of your mentors and, and I think one of the people you invest with now in, um, Infusion-

   20. SPSpeaker17:57

       Yeah

   21. TNTrevor Noah17:58

       ... uh, Power, they said, "Whenever you bring an issue to Sam, the first thing he thinks about is how can we fix this, how can we solve it, and the second thing he says immediately is how do we scale the solutions?"

   22. SPSpeaker18:08

       Who said this?

   23. TNTrevor Noah18:08

       I, I don't remember.

   24. SPSpeaker18:09

       Oh. Interesting.

   25. TNTrevor Noah18:09

       I'm terrible with names.

   26. SPSpeaker18:10

       Interesting. Uh-

   27. TNTrevor Noah18:11

       But I know it was somebody you work with, um-

   28. SPSpeaker18:12

       Interesting

   29. TNTrevor Noah18:12

       ... you know?

   30. SPSpeaker18:13

       No, it is- It is right, but I haven't heard someone say that about me before.
10. 20:19 – 22:14

    Tools not a god-in-the-sky: how Altman’s view of AGI evolved

    1. SPSpeaker20:19

       You know, we had a very different approach for how we thought we were gonna build this before the language model stuff started to work.

    2. TNTrevor Noah20:24

       Mm-hmm.

    3. SPSpeaker20:24

       We also, I think, had a very different conception of what it was gonna be like to build an AGI, and we didn't understand this idea that it was gonna be, like, iterative tools that got better and better-

    4. TNTrevor Noah20:33

       Mm. Mm-hmm

    5. SPSpeaker20:33

       ... that you kinda just talk to like a human. Um, and so, and so our thinking was very confused about, well, when you build an AGI, what happens? And, and we sort of thought about it as there was this moment before AGI, and then this moment of AGI, and, you know, then you needed to, like, give that over to some other system and some other governance. I, I now think it can be—and I'm really happy about this, 'cause I think it's much easier to navigate—I think it can be a... I don't wanna say, like, just another tool-

    6. TNTrevor Noah21:03

       Mm-hmm

    7. SPSpeaker21:03

       ... because it's different in all these ways, but, but in some other sense, uh, we have made a new tool for humanity. We've added something to the tool chest.

    8. TNTrevor Noah21:10

       Right.

    9. SPSpeaker21:10

       People are going to use that to do all sorts of incredible things, but people remain the architects of the future, not one AGI in the sky. It's you can do things you couldn't do before. I can do things I couldn't do before.

    10. TNTrevor Noah21:23

        Right.

    11. SPSpeaker21:23

        We'll be able to do a lot more. And in that sense, I can imagine a world where part of the way we fulfill our mission is we just make really great tools that, that massively impact human ability and everything else, and I'm, I'm, I'm pretty excited about that. Like, I love that we offer free ChatGPT with no ads, um, 'cause I, I personally really think ads have been a problem for the internet.

    12. TNTrevor Noah21:51

        [laughs]

    13. SPSpeaker21:51

        Um, but we just, like, put this tool out there.

    14. TNTrevor Noah21:53

        That's the downside of capitalism, right?

    15. SPSpeaker21:55

        [laughs] Y- y- yeah, yeah, um, it, one of them. Uh, I think there's much bigger ones, personally. But we put this tool out there, and people get to use it for free, and we're not, like, trying to turn them into the product. We're not trying to make them use it more.

    16. TNTrevor Noah22:09

        Right.

    17. SPSpeaker22:09

        Uh, and I think that shows, like, a interesting path that we can do more on.

11. 22:14 – 26:24

    What happens now: board governance, stakeholder representation, and company stabilization

    1. TNTrevor Noah22:14

       So let's, so let's do this. You know, in our, in our time together in this conversation, th- there's so many things I would like us to get to, hopefully. You know, um, we won't be able to answer all questions, obviously, but there, there, there are a few ideas, you know, a few headings and a few spaces I wanted us to, to live within. I guess the first and most timely is what, what, what happens now for the future of the company? Where, where do you see it going? You know, one of the things I found particularly interesting was what the new board was, how the-

    2. SPSpeaker22:43

       Yeah

    3. TNTrevor Noah22:43

       ... new board was, was, was comprised, you know, for OpenAI. You know, where previously you, you had, like, women on the board, now you don't.

    4. SPSpeaker22:50

       Yeah.

    5. TNTrevor Noah22:51

       You know, where previously you had people who had, like, no financial incentive on the board, now you do. And I, I, I wonder if, if you worry that that guardrail that you were part of-

    6. SPSpeaker23:00

       Yeah

    7. TNTrevor Noah23:01

       ... implementing is now gone. You know? Do you, do you have a board that's now not focused on protecting people or, or, or, you know, or defining a safer future, as opposed to making money and getting this thing-

    8. SPSpeaker23:13

       Yeah

    9. TNTrevor Noah23:13

       ... to be as good or as-

    10. SPSpeaker23:13

        Yeah

    11. TNTrevor Noah23:14

        ... as, as, as big as it can be?

    12. SPSpeaker23:16

        Um, well, I think our current, our previous governance structure and board didn't work in some important sense, so I'm all for figuring out how to improve that, and I'll support the board in their work to do that. Um, obviously the board ne- board needs to grow and diversify, and that'll be something that I think happens quickly.

    13. TNTrevor Noah23:38

        Mm-hmm.

    14. SPSpeaker23:39

        Um, and voices of people who are going to, uh, advocate for people who are traditionally not advocated for and be really thoughtful about not only AI safety, but just the, the lessons we can take from the past about how to make these very complex systems that interact with society in all of these ways-

    15. TNTrevor Noah23:57

        Right

    16. SPSpeaker23:58

        ... um, as good as possible, which is both mitigating the bad and sharing the upside, um, that all needs to be represented. Uh, so I'm, I'm excited to have a second chance at getting all these things right-

    17. TNTrevor Noah24:11

        Mm-hmm

    18. SPSpeaker24:11

        ... and we clearly got them wrong before. Um, but, but yeah, like, diversifying the board, making sure that we represent all of the Major classes of stakeholders that need to be represented here, figuring out how we make this a more democratic thing, continuing to push for governments to make some decisions governing this technology, which I, I know is imperfect, but I think better than any other method of doing this that we can think of so far. Engaging with our user base more to like help let them help set the limits on how this works. That's all super important. Um, that'll be one major thing going forward is, is board and expanding the board and governance.

    19. TNTrevor Noah24:53

        Yeah.

    20. SPSpeaker24:53

        And again, really like I know our current board is small, but I think they're so committed to all the things you were just talking about. Then there's another big class of problems. Uh, if you asked me a week ago, I would've said stabilizing the company-

    21. TNTrevor Noah25:07

        Okay

    22. SPSpeaker25:08

        ... was my top thing, but uh, internally at least, I feel pretty good. We did not lose a single customer. We did not lose a single employee. We continued to grow.

    23. TNTrevor Noah25:15

        Oh, that's pretty amazing.

    24. SPSpeaker25:16

        We continued to ship new products. Uh, we- our key partnerships feel strengthened, not hampered by this. Uh, and things are on pace there. And the, the sort of research and product plan for the first half of next year I think feels better and more focused than ever. Um, but there's a lot of, clearly there's like a lot of external stabilization we still have to do. Um, and then, and then beyond that, like we're really confronting the possibility that we just like we have not been planning ambitiously enough for success. You know, we've had-

    25. TNTrevor Noah25:52

        Oh, interesting

    26. SPSpeaker25:52

        ... we've had like ChatGPT Plus, if you wanna subscribe to ChatGPT Plus right now, you have- you are not able to. You've not been able to.

    27. TNTrevor Noah25:58

        Because yeah, 'cause you-

    28. SPSpeaker25:58

        We, we just ran out of-

    29. TNTrevor Noah25:59

        You just have too many people.

    30. SPSpeaker26:00

        Yeah.
12. 26:24 – 33:54

    Inside the storm: Vegas firing call, message flood, and planning a future outside OpenAI

    1. TNTrevor Noah26:24

       I found myself constantly thinking about you as a person, you know, when, when, when the whole board saga was taking place. And, um, whenever, whenever there's a storm, I'm always interested in what's happening in the eye of the storm.

    2. SPSpeaker26:37

       Yeah.

    3. TNTrevor Noah26:38

       You know? And I, I wondered like where were you when this all, when this all broke? Like what, what were you doing? What, what was going on in your world on a personal level?

    4. SPSpeaker26:46

       I, I th- the reason I laughed is the thing people say about me is I'm like, I am good at sitting in the eye of the hurricane while it like turns around me-

    5. TNTrevor Noah26:53

       Oh, really?

    6. SPSpeaker26:53

       ... thoroughly and staying super calm. And this was, this time like turns out like not, uh... Th- this was the experience of like being in the eye of the storm and having it not be calm. Um, I was in Las Vegas at F1.

    7. TNTrevor Noah27:05

       Oh, okay.

    8. SPSpeaker27:06

       Yeah. Uh-

    9. TNTrevor Noah27:07

       You, you an F1 fan?

    10. SPSpeaker27:07

        I am, yeah.

    11. TNTrevor Noah27:08

        All right. Who's your team? Do you have one?

    12. SPSpeaker27:10

        Um, I honestly, I like... I mean, Verstappen is so good, it's hard to say.

    13. TNTrevor Noah27:15

        Uh-huh. Okay.

    14. SPSpeaker27:15

        But I feel like that's the answer everyone would say. I still think he's just like unbelievably-

    15. TNTrevor Noah27:18

        No, no, no. Actually, it depend, it depends on when they joined the sport. Like I was a, I was a Schumacher fan 'cause that's when I started watching.

    16. SPSpeaker27:23

        Schumacher obviously like-

    17. TNTrevor Noah27:23

        Well, I mean, obviously-

    18. SPSpeaker27:24

        ... started legendary

    19. TNTrevor Noah27:24

        ... Nigel Mansell then-

    20. SPSpeaker27:25

        Yeah

    21. TNTrevor Noah27:25

        ... like Ayrton Senna and you know what I mean? But yeah, okay. Okay.

    22. SPSpeaker27:28

        But he's, I like-

    23. TNTrevor Noah27:29

        No, Verstappen is, he's, he's precise with it. I see why.

    24. SPSpeaker27:32

        Uh, and just like, like I almost gets bored of like wa- boy watching him win so often.

    25. TNTrevor Noah27:36

        [laughs]

    26. SPSpeaker27:36

        But it's, it's, it's incredible. Um, so I was like so excited for that. Uh, that first night, I got in like late on a Thursday. That first night someone like, they forgot to weld down a manhole cover, so someone drove over it-

    27. TNTrevor Noah27:49

        Oh, wow

    28. SPSpeaker27:49

        ... on the first lap, blew up like one of Ferrari's engines-

    29. TNTrevor Noah27:51

        Oh, wow

    30. SPSpeaker27:51

        ... and stopped the, the practice. So I didn't get to watch it. I never got to watch any race that whole weekend. Uh, I was in my hotel room, took this call, I had no idea who it was gonna be, and got fired by the board, and it was just this like, it felt like a dream. I was like, I was confused. It was chaotic. It did not feel real. It was like-
13. 33:54 – 36:42

    Safety brakes and culture under pressure: deployment delays, non-releases, and crisis performance

    1. TNTrevor Noah33:54

       I c- I c- I can only, I can only imagine. When you, when you look towards the future of the company and, and your role in it, how do you no- how do you now find a balance between moving OpenAI forward, continuously propelling yourselves in the direction you believe, but then also, also, you know, do, do you, do you have, do you still have an emergency brake? Is there, is there some system within the company where you say, "If we feel like we're creating something that's gonna-"

    2. SPSpeaker34:23

       Totally

    3. TNTrevor Noah34:23

       "... adversely affect society, we, we will step in, we will stop this"? Do you have that ability and, and, and is it baked in?

    4. SPSpeaker34:29

       Yeah, o- o- of course. Like, we-- And we've had it in the past, like we've created systems that we've chosen not to deploy.

    5. TNTrevor Noah34:37

       Oh, interesting.

    6. SPSpeaker34:37

       Um, and I'm sure we will again in the future. Or we've created a system and just said, "Hey, we need much longer to make this safe before we can deploy it."

    7. TNTrevor Noah34:44

       Uh-huh.

    8. SPSpeaker34:45

       Like with GPT-4, um, it took us almost eight months after we finished training before we were ready to release it to do all of the alignment and safety testing that we wanted.

    9. TNTrevor Noah34:55

       Right. Right. I remember talking to some of the team about that, yeah.

    10. SPSpeaker34:57

        And that's not a board decision. That's just the people in here doing their jobs and being committed to the mission. So that will continue on, um, and I-- One of the things I'm really proud of about this team is, uh, the ability to operate well in chaos, crisis, uncertainty, stress. Uh, I give them, like, an A+ on that.

    11. TNTrevor Noah35:24

        Mm-hmm.

    12. SPSpeaker35:24

        They did such a good job. And as we get closer to more powerful, very powerful systems, I think that ability of the culture and the team we have built is maybe the most important element, you know. To, like, keep your head cool in a crisis and make good, thoughtful decisions. Um, I think the team here really proved that they can, they can do that, and that's super, that's super important. There were-- I saw this thing where someone was like, "You know, the thing we learned about OpenAI is that Sam can run the company without any job there," and I think that's totally wrong. I think that's not at all what happened. I think what happened is, or the right learning is the company can totally run without me. And-

    13. TNTrevor Noah36:09

        Oh. It's a, it's a culture that's-

    14. SPSpeaker36:11

        The team is ready. The culture is ready. Like, I think that's-- I'm just super proud of that. Really happy to be back and doing it, but, but like, I sleep better at night having watched the team manage through this, given the challenges ahead. And there will be bigger challenges than this that will come up, but I think in some subjective sense, I hope and believe this is the hardest one 'cause we were so unprepared, and now we kind of, like, realize the- ... the stakes and that we're not just, in some, in some, like, important sense, we're just not a regular company.

14. 36:42 – 45:04

    Where the product goes: base models, custom GPTs, and personal AI assistants

    1. TNTrevor Noah36:42

       Oh, yeah. Far from it. Far from it. Let- let's, let's talk a little bit about that. Um, ChatGPT, OpenAI, you know, whatever it may end up calling it- because, I mean, you've got DALL-E, you've got Whisper, you've got all these amazing products

    2. SPSpeaker36:56

       Yeah, if you have any name ideas-

    3. TNTrevor Noah36:57

       [laughs]

    4. SPSpeaker36:57

       ... name brand architecture ideas for us, I would love it.

    5. TNTrevor Noah37:00

       I feel like ChatGPT has, has done it. You know, I feel like it is now ubiqu- it is-

    6. SPSpeaker37:04

       Ah, yeah, it's a horrible name, but it may be too ubiquitous now to change

    7. TNTrevor Noah37:06

       But it's, it's... You can't change- you think you can change it at this point?

    8. SPSpeaker37:08

       Uh, I mean, could we drop it down to just, like, GPT or just Chat?

    9. TNTrevor Noah37:12

       I, I don't know.

    10. SPSpeaker37:13

        I don't know.

    11. TNTrevor Noah37:13

        Look, I don't know personally.

    12. SPSpeaker37:14

        Maybe not. Maybe not.

    13. TNTrevor Noah37:14

        Sometimes, sometimes I feel like a, a product or a name or an idea g- grows beyond the marketer's dream-

    14. SPSpeaker37:20

        Yeah

    15. TNTrevor Noah37:21

        ... space-

    16. SPSpeaker37:21

        Well-

    17. TNTrevor Noah37:21

        ... and then people just have it

    18. SPSpeaker37:22

        Yeah, no marketer ever would've picked ChatGPT-

    19. TNTrevor Noah37:24

        No

    20. SPSpeaker37:24

        ... as the name for this, but we may be stuck with it, and that might be all right.

    21. TNTrevor Noah37:27

        Yeah. And, and, and, and it's now, I mean, this, the, just the multimodal aspects of it, like, fascinate me. You know, I remember when I first saw DALL-E come out, uh, you know, and it was just an idea, and seeing how it worked and seeing this program that could create a picture-

    22. SPSpeaker37:42

        Yeah

    23. TNTrevor Noah37:42

        ... from nothing but noise and, and I was trying to explain it to people-

    24. SPSpeaker37:44

        Yeah

    25. TNTrevor Noah37:44

        ... and they were going, "But where did it get the picture from?" And I was like, "There was no picture. There was no source image."

    26. SPSpeaker37:48

        Right.

    27. TNTrevor Noah37:48

        And then they go, "That's not possible."

    28. SPSpeaker37:50

        Right.

    29. TNTrevor Noah37:50

        "It, it saw something." I was like, y- and it's so hard to explain some of this. Sometimes it's even hard to, to understand for myself.

    30. SPSpeaker37:56

        Yeah.
15. 45:04 – 56:04

    Defining AGI, interpretability, and hallucinations: creativity vs factuality

    1. TNTrevor Noah45:04

       Do you, [sighs] do you have an idea of when you will consider artificial general intelligence achieved? Like, like, how do we know-

    2. SPSpeaker45:13

       Me personally?

    3. TNTrevor Noah45:13

       Yeah, yeah

    4. SPSpeaker45:13

       Like when I'll feel like mission accomplished?

    5. TNTrevor Noah45:14

       Yeah, like what, what, what... 'Cause everyone talks about artificial general intelligence. But then I go, how do we know what that is?

    6. SPSpeaker45:21

       So this comes back to that point earlier where everyone's got a different definition.

    7. TNTrevor Noah45:24

       Yes.

    8. SPSpeaker45:24

       I'll tell you personally when I'll be thrilled.

    9. TNTrevor Noah45:26

       Uh-huh.

    10. SPSpeaker45:26

        When we have a system that can help discover novel physics-

    11. TNTrevor Noah45:29

        [clicks tongue] Okay

    12. SPSpeaker45:30

        ... I'll be very thrilled.

    13. TNTrevor Noah45:31

        But that feels like it's way beyond general intelligence. That [laughs] seems like you're... Do you know what I mean?

    14. SPSpeaker45:35

        It, it's beyond, I think, what most people would count as that.

    15. TNTrevor Noah45:38

        Like, like, maybe... Because this is what I think of sometimes is I go, how do we define that general intelligence? Are we defining it as brilliance in a certain field, or are we defining it as... Like, a child is artificially generally intelligent-

    16. SPSpeaker45:50

        That's for sure

    17. TNTrevor Noah45:50

        ... but you have to keep programming it, you know? They just, they come out, they don't speak, they don't know how to walk, they don't know how to... And you're constantly programming this, this, this, you know, AGI-

    18. SPSpeaker45:59

        Yeah

    19. TNTrevor Noah46:00

        ... to get to where it needs to go. So h- how will you... Like, if you get to the point where you have a four-year-old child version of AGI-

    20. SPSpeaker46:07

        If we have a system that can just, like, just figure it out, you know? Can just go autonomously with some help from its parents-

    21. TNTrevor Noah46:13

        Yeah

    22. SPSpeaker46:14

        ... uh, figure out the world in the way that a four-year-old kid does-

    23. TNTrevor Noah46:18

        Uh-huh

    24. SPSpeaker46:18

        ... uh, yeah, we can call that an AGI. If we can, if we can really address that truly generalizability to, to be confronted with a new problem and just figure it out.

    25. TNTrevor Noah46:30

        Mm-hmm.

    26. SPSpeaker46:30

        Not perfectly. A four-year-old doesn't always figure it out perfectly either.

    27. TNTrevor Noah46:33

        Right.

    28. SPSpeaker46:33

        But, you know, then we're clearly gonna get it.

    29. TNTrevor Noah46:36

        Are we able to get there if we don't fundamentally understand thinking and the mind and-

    30. SPSpeaker46:41

        It seems like it
16. 56:04 – 1:14:56

    Beyond human data, justice, democratized governance, and catastrophic-risk scenarios

    1. TNTrevor Noah56:04

       Yeah. I, I'm gonna lose you soon, so I, I wanna jump into a few, um, a few questions, um, that I think are, are really po- people will kill me if I don't ask these-

    2. SPSpeaker56:12

       All right

    3. TNTrevor Noah56:13

       ... of you. Um, okay. So one of the main ones, this is, this is from my side personally. We always talk about AI learning from the data, right? They, they're fed data sets, and we talk about this. They... That's why you need these mega computers that cost billions and billions of dollars, so that the computers can learn. How do we teach an AI to think better than the humans that have given it the data that is clearly flawed? So for instance, how do you, how does an AI learn-

    4. SPSpeaker56:41

       Yeah

    5. TNTrevor Noah56:41

       ... beyond the limited data that we've put out there? You know, when it comes to race, when it comes to economics, when it comes to the ideas, 'cause we're limited. How do we teach it to not be as limited as we are, if we're feeding it data that's limited?

    6. SPSpeaker56:52

       Uh, we don't know yet, but that's our big... that's, like, one of our biggest research thrusts in front of us, is-

    7. TNTrevor Noah56:58

       Okay

    8. SPSpeaker56:58

       ... is, like, how, how do we surpass human data?

    9. TNTrevor Noah57:02

       Mm-hmm.

    10. SPSpeaker57:03

        Um, and I, I hope that if we can do this again a year from now, I'll be able to tell you. But I don't know yet. We don't know yet.

    11. TNTrevor Noah57:09

        Okay.

    12. SPSpeaker57:09

        Uh, it's really important. However, a thing that I do believe is this is gonna be a force to combat injustice in the world in a super important way.

    13. TNTrevor Noah57:22

        Hmm.

    14. SPSpeaker57:22

        I think these systems will be... They won't have the same deep flaws that all humans do. Uh, they will be able to be made to be far less racist, far less sexist, far less biased. They'll be a force for economic justice in the world. I think, you know, if you make, if you make a great AI tutor or a great AI medical advisor available, um, that helps the poorest half of the world more than the richest half, even though it helps lift everybody up. So I don't have, like, an answer to the scientific question you asked, but I do at this point feel confident that these systems can be. Of c- of course, we have to do some hard societal work to make them in fact be, but can be, um, great for sort of increasing justice in the world.

    15. TNTrevor Noah58:08

        Okay. Maybe that, that leads then perfectly to a second question, which is, what are you doing? What is OpenAI d- doing? Are you, are you even considering doing anything to try and mitigate how much this new technology once again creates, you know, the haves and the have-nots?

    16. SPSpeaker58:25

        Yeah.

    17. TNTrevor Noah58:25

        Every new technology that's come out has been amazing for society as a whole-

    18. SPSpeaker58:30

        Yeah

    19. TNTrevor Noah58:30

        ... if we call it that, but you can't deny, it creates a moment in time where if you have it, you've got it all, and if you don't, you're out of the game.

    20. SPSpeaker58:37

        I think the... Our... We'll learn a lot more as we go, but currently, I think one really important thing we do is offer a truly free service, which means no ad supported, but just a free service to more than 100 million people who are using it every week.

    21. TNTrevor Noah58:53

        Mm-hmm.

    22. SPSpeaker58:54

        Um, and the fact that, it's not, it's not fair to say anyone, 'cause in some countries we, we have still blocked or we are still blocked, but trying to get closer and closer to everything we can do there, where anyone can access really high quality, easy to use, free AI, um, that is important to all of us personally. And I think that there's other things that we'd like to do with the technology, like if we can help cure diseases with AI and make those cures available to the world, that's clearly beneficial. But putting this tool in the hands of many, as many people as we can, and letting them use it to architect the future, um, that is super important. And, and I think we can push this much, much further.

    23. TNTrevor Noah59:38

        Okay. Two more questions.

    24. SPSpeaker59:40

        Can I add one more thing? That's-

    25. TNTrevor Noah59:41

        Oh, yeah, yeah. Yeah, go.

    26. SPSpeaker59:42

        Um-

    27. TNTrevor Noah59:42

        It's all your time. Go ahead, yeah.

    28. SPSpeaker59:43

        The other thing that I think is important to that is, who gets to make the decisions about what these systems say and not say, or do and not do? Like, who gets to set the limits?

    29. TNTrevor Noah59:52

        Yeah.

    30. SPSpeaker59:52

        And I... like, right now, it is basically the people who work at OpenAI deciding-

Episode duration: 1:14:56