Sam Altman's Latest Interview: The Board, the Musk Lawsuit, GPT-5, Sora, AGI, $7 Trillion...
TechFlow Selected TechFlow Selected
Sam Altman's Latest Interview: The Board, the Musk Lawsuit, GPT-5, Sora, AGI, $7 Trillion...
32,000-word transcript of Sam Altman's latest interview.
Navigating Web3 tides with focused insightsCompiled by: Youxin
In the early hours today, Lex Fridman updated his second conversation with OpenAI CEO Sam Altman. They discussed topics including the OpenAI boardroom drama, Ilya Sutskever, Elon Musk's lawsuit, Sora, GPT-4, memory and privacy, Q*, GPT-5, a $7 trillion fundraising effort, Google Gemini, the transition to GPT-5, and AGI aliens.
Below is the full transcript of this dialogue, totaling 32,000 words:
The OpenAI Board Drama
Lex Fridman: Take me through the legendary OpenAI board saga that began on Thursday, November 16—perhaps you can also start from Friday the 17th.
Sam Altman: This was absolutely the most painful professional experience of my life—chaotic, shameful, disturbing, and many other negative things. But there were also amazing aspects to it. I wish I hadn’t been so adrenaline-fueled that I couldn’t stop and appreciate them at the time. But I found an old tweet of mine or one from that period. It felt like writing your own eulogy, watching people talk about all these great things about you and the incredible support from people I love and care about. That was truly wonderful, really wonderful. The entire weekend, except for one major exception, I felt a lot of love and very little hate, even though I had no idea what was happening here or what would happen next—which felt really bad. Sometimes I genuinely thought this would be one of the worst things to ever happen in AI safety history. Well, I’m also glad it happened relatively early. I think somewhere between OpenAI’s beginning and our creation of AGI, something crazy and explosive will occur—but probably even more such events will follow. I believe this incident still helped us build a certain resilience and better prepare for future challenges.
Lex Fridman: But did you have a sense that you’d go through some kind of power struggle?
Sam Altman: The path to AGI should be a massive power struggle. The world should… well, shouldn’t. I hope it turns out that way.
Lex Fridman: So you must go through this process, as you said, iterating as frequently as possible, figuring out how to build a board structure, how to build an organization, how to have the right people around you. We’re working hard on how to communicate all these issues so as to mitigate power struggles as much as possible.
Sam Altman: Yes.
Lex Fridman: To soothe it.
Sam Altman: But right now, it feels like a past thing—really unpleasant, really difficult, and painful—but we're back at work, and things are so busy and intense that I don’t know if I could spend much time thinking about it. Afterward, for a while—maybe a month later, perhaps 45 days—I was in this dissociative state, just drifting through those days. I was utterly pathetic. I was in a very low mood.
Lex Fridman: Just on a personal psychological level?
Sam Altman: Yes. It was really painful and hard to keep running OpenAI amid it all. I just wanted to crawl into a cave and recover. But now we seem to have just returned to the mission.
Lex Fridman: Hmm, reflecting back on the board structure, power dynamics, how the company operates, tensions between research and product development versus funding, and all these things remains useful, so you can achieve high potential in building AGI and do so in a slightly more organized, less dramatic way in the future. So there’s immense value here, both in terms of leadership psychology and board structure and all these messy elements.
Sam Altman: I did learn a lot about structure and incentives and what we need from our board. I think what happened now has value in a sense. I don’t think this might be OpenAI’s last high-pressure moment, but it was certainly quite a high-pressure one. My company almost got destroyed. We’ve thought a lot about everything else we need to get right for AGI, but thinking about how to build a resilient organization and how to create a structure capable of withstanding enormous global pressures becomes increasingly important as we draw closer, and I expect more of this, which I think is crucial.
Lex Fridman: Did you sense how deep and rigorous the board’s deliberation process was? Could you briefly describe the human dynamics in such a situation? Was it just a few casual remarks that suddenly escalated into “Why don’t we fire Sam”?
Sam Altman: I think board members overall were well-intentioned people, and I believe that when people feel time pressure or other stressors, they understand and make suboptimal decisions. One challenge OpenAI faces is that we need a board and team adept at operating under pressure.
Lex Fridman: Do you think the board had too much power?
Sam Altman: I think the board should have significant power, but one thing we clearly saw is that in most corporate structures, boards are typically accountable to shareholders. Sometimes people have super voting rights or similar mechanisms. In this case, I think in our structure one thing we might have needed to consider more is that unless otherwise specified, a nonprofit’s board holds considerable power. They aren’t really accountable to anyone except themselves. This is good in many ways, but what we truly want is for the OpenAI board to respond to the entire world, although that’s a practical challenge.
Lex Fridman: So a new board has been announced.
Sam Altman: Yes.
Lex Fridman: I guess there was initially a smaller interim board, and now there’s a new final one?
Sam Altman: Not yet final. We’ve added some. We’ll add more.
Lex Fridman: Added some. Okay. What has been fixed in the new version that might have been broken in the previous one?
Sam Altman: The old board had been shrinking over roughly a year. It started with nine, then reduced to six, and we couldn’t agree on adding anyone. I also think the board didn’t have many experienced board members, and many of OpenAI’s new board members simply have more board experience. I think that will help.
Lex Fridman: Some additions to the board have faced criticism. For example, I’ve heard many criticize Larry Summers’ inclusion. What was the selection process for board members? What does it involve?
Sam Altman: Brett and Larry made the decision during that tense weekend—it was a rollercoaster. There were many ups and downs. We tried to reach consensus on new board members, and both the executive team and former board members considered this reasonable. Larry was actually one of their suggestions, from the old board. Brett, I think I’d even suggested before that weekend, but he was busy and didn’t want to do it, then we really needed help [inaudible 00:09:22]. We discussed many others, but I felt that if I were returning, I needed new board members. I didn’t think I could work again with the old board in the same configuration, although afterward we decided I was grateful Adam would stay. We considered various configurations and decided we wanted a three-member board, needing to find two new members quickly.
So these were honest decisions—no... you do it on the battlefield. You don’t have time to design rigorous processes then. For subsequent board members and those we’ll add going forward, we have standards we believe are important for the board and different expertise we want the board to possess. Unlike hiring an executive where you need them for a specific role, the board needs to handle governance and thoughtful deliberation as a whole. One thing Brett said that I really liked is that we want to hire board members as a list rather than one at a time. Consider a group bringing nonprofit expertise, corporate operations expertise, strong legal and governance expertise—that’s what we’re trying to optimize for.
Lex Fridman: Is technical aptitude important for individual board members?
Sam Altman: Not necessary for every member, but certainly some should have it. It’s part of what the board needs to do.
Lex Fridman: It’s interesting that people may not understand OpenAI—I certainly didn’t—about all the operational details. When they think of a board, given the drama, they think of you. They wonder, if you reach AGI or some highly impactful product and build and deploy it, what would the board dialogue look like? They think, okay, what lineup would be appropriate in that scenario?
Sam Altman: I think you definitely need some technical experts. Then you need people asking, “How can we deploy this in a way that best helps people worldwide?” and those holding diverse perspectives. I think a mistake you or I might make is thinking only technical understanding matters, which is absolutely part of the conversation you want the board to have, but there’s more about how this affects society and people’s lives—you also want representation there.
Lex Fridman: Are you looking at people’s track records or just having conversations?
Sam Altman: Track record is important. Of course, you have many conversations, but for some roles I completely ignore track record and just look at slope, ignoring the Y-intercept.
Lex Fridman: Thanks. Appreciate the math lesson for the audience.
Sam Altman: As a board member, I actually care more about the Y-intercept. I think there’s something profound to say about track record, and experience is hard to replace.
Lex Fridman: Do you try to fit polynomial or exponential functions to track records?
Sam Altman: The analogy doesn’t go that deep.
Lex Fridman: Okay. You mentioned some lows that weekend. Psychologically, what were the lows for you? Did you ever consider going to the Amazon jungle, taking ayahuasca, and disappearing forever?
Sam Altman: It was a very rough period. There were also big highs. My phone kept getting messages from people I work with daily and haven’t spoken to in ten years. I wasn’t fully aware because I was in the middle of this firefight, but it was really nice. Overall, it was a very painful weekend. It was like an unexpectedly public fight, which was exhausting for me—more than I anticipated. I think fights are usually exhausting, but this one certainly was. The board did this on Friday afternoon. I couldn’t get many answers, but I also thought, okay, the board can do this, so I need to consider what I want to do, but I’ll try to find a silver lining here.
At the time, I thought, okay, my current job at OpenAI, or what it used to be, was running a fairly large company. What I always enjoyed most was working with researchers. At the time, I thought, yes, I could do a very focused AGI research job. I was excited about that. At the time, I didn’t even think all this might be reversed. This was Friday afternoon.
Lex Fridman: So you already accepted it.
Sam Altman: Very quickly. I went through a period of confusion and anger, but quickly, quickly. By Friday night, I was talking to people about what would happen next, and I was excited about it. I think it was Friday night when I first heard from the executive team here, saying, “Hey, we’re going to fight this.” Then I went to bed, but I still felt, okay, excited. Forward.
Lex Fridman: Could you sleep?
Sam Altman: Not much. One strange thing was that during those four and a half days, I slept little, ate little, but remained energetic. You learn strange things about adrenaline during wartime.
Lex Fridman: So you accepted the death of the OpenAI baby.
Sam Altman: I was excited about the new thing. I thought, “Well, this is crazy, but whatever.”
Lex Fridman: That’s a very good coping mechanism.
Sam Altman: On Saturday morning, two board members called saying, “Hey, we didn’t mean to destabilize. We don’t want to hold too much value here. Can we talk about you coming back?” I immediately didn’t want to do it, but I thought a bit, then I thought, hmm, I really care about the people here, partners, shareholders. I like this company. So I thought, “Okay, but this is what I need.” Then the most painful moment that weekend was constantly thinking and being told—not just me, but the whole team here—thinking, “When everyone is trying to break OpenAI apart, we’re trying to keep it stable.”
We kept being told, “We’re almost done. We’re almost done. We just need a bit more time.” It was such a chaotic state. Then Sunday night, every few hours, I expected we’d complete the task, figure out how to bring me back and restore things. The board then appointed a new interim CEO, and I thought, this feels really bad. That was the lowest point of the whole thing. Let me tell you one thing. It was painful, but I felt a lot of love throughout the weekend. Except for that moment Sunday night, I wouldn’t describe my emotions as anger or hatred, but I felt a lot of love from people toward others. It was painful, but the dominant emotion of the weekend was love, not hate.
Lex Fridman: You spoke highly of Mira Murati, as you mentioned in your tweets, especially helpful during important quiet moments. Maybe we can go off-topic a bit. What do you appreciate about Mira?
Sam Altman: Well, during that chaotic weekend, she did a great job, but people often see leaders in crisis moments, for better or worse. But one thing I truly value in leaders is how they perform during boring Tuesday mornings at 9:46 AM and in everyday mundane tasks. How someone behaves in meetings, the quality of their decisions. That’s what I call the quiet moments.
Lex Fridman: Meaning most work happens day-to-day in meetings. Just showing up and making key decisions.
Sam Altman: Yes. Look, you want to spend the last 20 minutes, I understand, on this very dramatic weekend, but that’s not really the purpose of OpenAI. OpenAI is truly about the other seven years.
Lex Fridman: Yeah, well. Human civilization isn’t Nazi Germany invading the Soviet Union, but that’s still what people focus on.
Sam Altman: It’s very easy to understand.
Lex Fridman: It gives insight into humanity, extremes of human nature, and perhaps some damage in human civilization’s triumphs might occur in those moments, so it’s illustrative. Let me ask you about Ilya. Is he being held hostage in a secret nuclear facility?
Ilya Sutskever
Lex Fridman: That’s become somewhat of a meme. You’ve known Ilya for a long time. Obviously, he was part of the board and such. What’s your relationship with him now?
Sam Altman: I love Ilya. I deeply respect Ilya. I have nothing to say about his plans right now. That’s his issue, but I truly hope we work together for the rest of my career. He’s a bit younger than me. Maybe he works a bit longer.
Lex Fridman: There’s a meme saying he saw something, like he might have seen AGI, which internally worried him greatly. What did Ilya see?
Sam Altman: Ilya hasn’t seen AGI. None of us have seen AGI. We haven’t built AGI yet. I do think one of the many reasons I deeply admire Ilya is that he takes AGI and safety extremely seriously, broadly speaking, including questions about its societal impact. As we continue making major progress, Ilya is one of the people I’ve spent the most time with in recent years discussing what this means, what we need to do to get it right, ensuring we successfully fulfill our mission. So Ilya hasn’t seen general artificial intelligence, but Ilya contributes to humanity by thinking and worrying about ensuring we do it right.
Lex Fridman: I’ve had many conversations with him before. I think when he talks about technology, he’s always doing this long-term thinking. So he’s not thinking about what happens a year from now. He’s thinking about 10 years from now, reasoning from first principles—okay, if scaled, what are the fundamentals here? Where is this going? So this forms the foundation for thinking about all other safety issues and such, making him a very interesting person. Do you know why he’s silent? Is he just introspecting?
Sam Altman: Again, I don’t want to speak for Ilya. I think you should ask him that question. He’s definitely a thoughtful person. I think Ilya always explores his soul in a very good way.
Lex Fridman: Yes. Also, he appreciates the power of silence. Also, I hear he might be a fool, but I’ve never seen that side of him.
Sam Altman: When that happens, it feels very sweet.
Lex Fridman: I’ve never seen a foolish Ilya, but I’m also looking forward to it.
Sam Altman: Recently, I attended a dinner with him; he was playing with a puppy, in a silly, cute mood. I thought, wow, this isn’t the side of Ilya the world sees most often.
Lex Fridman: To sum up the whole story, do you feel good about the board structure?
Sam Altman: Yes.
Lex Fridman: About all of this and where it’s headed?
Sam Altman: I feel good about the new board. Regarding OpenAI’s structure, one of the board’s tasks is to examine it and see where we can make it more robust. We wanted to get the new board members in place first, but we clearly learned lessons about structure throughout the process. I guess I don’t have anything very esoteric to say. It was a crazy, very painful experience. I think it was a perfect storm of weirdness. It was a rehearsal for me, as stakes grow higher and we need sound governance structures, processes, and people, for what will happen. I’m glad it happened, but experiencing it was shockingly painful.
Lex Fridman: Has this made you more hesitant about trusting others?
Sam Altman: Yes.
Lex Fridman: Just personally?
Sam Altman: Yes. I think I’m a very trusting person. I’ve always had a life philosophy of not worrying about all the paranoia. Don’t worry about edge cases. You’ll get messed up a little, but you live more freely. This shocked me. I was caught off guard, and it definitely changed; I really dislike this, but it absolutely changed my view on default trust and planning for bad scenarios.
Lex Fridman: You have to be careful about this. Are you worried about becoming a bit too cynical?
Sam Altman: I’m not worried about becoming overly cynical. I think I’m the extreme opposite of a cynic, but I worry about no longer being a default trustor.
Lex Fridman: I’m actually not sure which mode is best for developing AGI—trust or distrust. You’re on an interesting journey. But structurally, I’m more interested in the human aspect. How do you surround yourself with people creating cool things while making wise decisions? Because the more money you make, the greater this thing’s power, the weirder people get.
Sam Altman: I think you can make various comments about the level of trust I should receive or how I should do things differently. But regarding the team here, I think you should give me a very good grade. I have immense gratitude, trust, and respect for the people I work with daily, and I think being surrounded by such people is really important.
Elon Musk’s Lawsuit
Lex Fridman: Our mutual friend Elon sued OpenAI. What’s the essence of his criticism from your perspective? How valid is his viewpoint? How wrong is he?
Sam Altman: I don’t know exactly what it’s about. We initially just thought we’d be a research lab, not knowing how the technology would develop. Since that was only seven or eight years ago, it’s hard to go back and truly remember the context, but it was before language models became a big deal. Before we had any ideas about APIs or selling access to chatbots. Before we knew at all we’d productize. So we were like, “We just want to try doing research, but we don’t really know what we’ll use it for.” I think for many fundamentally new things, you start fumbling in the dark and make assumptions, most of which turn out wrong.
Then it became clear we needed to do different things and required substantial capital. So we said, “Okay, this structure doesn’t quite fit. How do we patch the structure?” Then you patch it again, patch it again, and end up with something that looks surprisingly odd, to say the least. But I think we gradually arrived here, making reasonable decisions at each point along the way. That doesn’t mean if we had an Oracle now, I wouldn’t do it completely differently, but you didn’t have the Oracle then. Anyway, I don’t know Elon’s real motivations.
Lex Fridman: To the extent you recall, what was OpenAI’s response in the blog post? Can you summarize it?
Sam Altman: Oh, we just said Elon said this series of things. This is our representation, or this isn’t our representation. Here’s a description of how events unfolded. We tried not to make it emotional, just stating, “This is the history.”
Lex Fridman: I do think there’s a degree of misrepresentation by Elon regarding the uncertainty you just described. You were a small group of researchers wildly talking about this idea while everyone mocked AGI.
Sam Altman: Not long ago, Elon was wildly talking about launching rockets while people mocked the idea, so I think he’d have more empathy for that.
Lex Fridman: I do think there’s a personal element here, that many great people at OpenAI chose to part ways with Elon, so there’s a personal—
Sam Altman: Elon chose to part ways.
Lex Fridman: Can you accurately describe it? Choosing to split up?
Sam Altman: He thought OpenAI would fail. He wanted complete control to turn it around. We wanted to keep moving forward in the direction of OpenAI. He also wanted Tesla to build AGI. He repeatedly wanted to make OpenAI a for-profit company he could control or merge with Tesla. We didn’t want to do that, and he decided to leave, which was fine.
Lex Fridman: So you’re saying, this is one thing the blog post mentioned, he wanted OpenAI essentially acquired by Tesla in the same way, or possibly similarly, or perhaps more dramatically than the Microsoft collaboration.
Sam Altman: My memory is that the proposal was like, acquired by Tesla and let Tesla have full control. I’m pretty sure that’s what it was.
Lex Fridman: So what did the term “open” in OpenAI mean for Elon at that time? Ilya talked about this in email exchanges and such. What did it mean to you at the time? What does it mean to you now?
Sam Altman: Speaking of using an Oracle, I’d pick a different name. I think the most important thing OpenAI is doing is providing powerful technology free to people as a public good. We don’t put ads on our website as a public good. Our free version doesn’t show ads. We won’t monetize it in other ways. We just say this is part of our mission. We want to provide increasingly powerful tools free to people and let them use them. I think this openness is very important for our mission. I think if you give people excellent tools and teach them to use them, or even don’t teach them, they figure it out, and let them build an incredible future for each other, that’s a big deal. So if we can keep rolling out powerful AI tools to the world for free or low cost, or free and low-cost, I think it’s hugely significant for how we accomplish our mission. Regardless of open source, yes, I think we should open-source some things and not others. It really became a hard religious line with little nuance, but I think nuance is the correct answer.
Lex Fridman: So he said, “Change your name to CloseAI, and I’ll drop the lawsuit.” I mean, will it become a meme battleground over the name?
Sam Altman: I think it shows how serious Elon is about the lawsuit, which I find surprising.
Lex Fridman: If I’m wrong, maybe correct me, but I think this lawsuit isn’t legally serious. More importantly, it highlights the future of AGI and who’s currently leading among companies.
Sam Altman: Look, I mean Grok hasn’t open-sourced anything until people pointed out it was a bit hypocritical, then he announced Grok will open-source this week. I don’t think open vs. closed source really matters to him.
Lex Fridman: Well, we’ll discuss open vs. closed source. I do think maybe criticizing competition is good. Just saying some nonsense would be great. But friendly competition differs from “I personally hate lawsuits” type competition.
Sam Altman: Look, I think the whole thing is inappropriate for builders. I respect Elon; he’s one of the greatest builders of our time. I know he knows what it feels like to be attacked by haters, which makes it sadder for me that he’s doing this.
Lex Fridman: Yes, he’s one of the greatest builders ever, maybe even the greatest ever builder.
Sam Altman: It makes me sad. I think it makes many people sad. Many people have long admired him. I’ve said in some interviews or other things that I miss the old Elon, and I received tons of messages saying, “That’s exactly how I feel.”
Lex Fridman: I think he should win. He should let X Grok beat GPT, then GPT beat Grok, just competition, beautiful for everyone. But on open source, do you think many companies are playing this idea? It’s interesting. I’d say Meta surprisingly leads here, or at least took the first step in the international chess game of truly open-sourcing models. Of course, it’s not the most advanced model, but open-sourcing Llama. Google is considering open-sourcing a smaller version. What are the pros and cons of open source? Have you thought about this idea?
Sam Altman: Yes, I think open-source models definitely have a place, especially small models people can run locally—I think demand is huge. I think there will be some open-source models and some closed-source models. It’s not different from other ecosystems in this regard.
Lex Fridman: I hear everyone on podcasts talking about this lawsuit and such. They’re more concerned about the precedent of shifting from nonprofit to capped-profit. What precedent does it set for other startups? Is that a thing—
Sam Altman: I strongly discourage any startup considering starting as a nonprofit and then adding a for-profit arm. I strongly discourage them from doing that. I don’t think we’ll set a precedent here.
Lex Fridman: Okay. So most startups should do—
Sam Altman: Of course.
Lex Fridman: Again—
Sam Altman: We’d do it ourselves if we knew what would happen.
Lex Fridman: Theoretically, if you jump nicely here, there might be tax benefits or something, but…
Sam Altman: I think most people don’t view these things that way.
Lex Fridman: It’s impossible to save a lot of money for startups if done this way.
Sam Altman: No, I think some laws make this quite difficult.
Lex Fridman: How do you want Elon to perform? This tension, this dance, what do you want this to be? If we look one or two years, three years from now, will your relationship with him also evolve personally, like friendship, friendly competition, etc.?
Sam Altman: Yes, I deeply respect Elon, and I hope we maintain a friendly relationship in the coming years.
Lex Fridman: Yes, I hope you maintain a friendly relationship this month, competing, winning, and exploring these ideas together. I do think there’s talent competition or other competition, but it should be friendly competition. Just building cool things. Elon is very good at creating cool things. So are you.
Sora
Sora, speaking of cool things. I could ask about a million questions. First, it’s amazing. It’s impressive on a product level, but philosophically too. So let me ask technically/philosophically, do you think it understands the world more or less than GPT-4? When you train with these patches and language tokens, the world model.
Sam Altman: I think all these models understand the world more than most of us imagine. And because they’re also very aware of what they don’t understand or get wrong, it’s easy to spot weaknesses, see through the veil, and say, “Ah, it’s all fake.” But it’s not all fake. Some things work, some don’t.
I remember when I first started watching Sora videos, I’d see a person walk in front of something for a few seconds, blocking it, then walk away, and the same thing would still be there. I thought, “Oh, that’s nice.” Or some examples where the underlying physics seemed well-represented across many steps in the sequence, like, “Oh, that’s really impressive.” But fundamentally, these models are just getting better, and this will continue. If you look at the trajectory from DALL·E 1 to 2 to 3 to Sora, every version had many people dunking on it saying it can’t do this, can’t do that, now look.
Lex Fridman: Well, you just mentioned occlusion is basically modeling a decent enough 3D physical world to capture such things.
Sam Altman: Well…
Lex Fridman: Or, maybe you can tell me, what does the world model need to do to handle occlusion?
Sam Altman: Yes. So I’d say it does a very good job handling occlusion. I said it has a great underlying 3D world model, which is a bit of a stretch.
Lex Fridman: But can this be achieved just with 2D training data methods?
Sam Altman: It seems this approach will yield surprising results. I don’t want to speculate too much on which limitations it will overcome and which it won’t, but…
Lex Fridman: What interesting limitations have you found in the system? I mean, you released some interesting content.
Sam Altman: All sorts of fun. I mean, cats sprouting extra limbs at random points in videos. Pick your favorite, but there are still many problems, many weaknesses.
Lex Fridman: Do you think this is a fundamental flaw of the method, or just that larger models or better technical details or better data, more data will fix the cat sprouting issue?
Sam Altman: I’d say yes to both. I think this approach seems different from how we think, learn, etc. And I think it gets better with scale.
Lex Fridman: Like I mentioned, LLMs have tokens, text tokens, while Sora has visual patches, so it converts all visual data—various visual data videos and images—into patches. Is training fully self-supervised, or is there some manual labeling? What’s the human involvement in all this?
Sam Altman: I mean, without revealing any specifics about Sora’s methods, we use a lot of human-generated data in our work.
Lex Fridman: But not internet-scale data? Humans so much. “A lot” is a complex word, Sam.
Sam Altman: I think “a lot” is fair in this context.
Lex Fridman: Because for me, “a lot”… look, I’m introverted, when I go out with three people, that’s a lot. Four people, that’s a lot. But I think you mean more than…
Sam Altman: Yes, more than three people are dedicated to labeling data for these models.
Lex Fridman: Okay. Right. But fundamentally, there’s a lot of self-supervised learning. Because you mentioned internet-scale data in the technical report. That’s another beauty… like poetry. So much data isn’t human-labeled. Is it self-supervised like that?
Sam Altman: Yes.
Lex Fridman: Then the question is, if we knew the details of self-supervision, how much data on the internet could be used in ways favorable to this self-supervision. Have you considered releasing more details?
Sam Altman: We have. Do you mean specifically about sources?
Lex Fridman: Specific sources. Because it’s so interesting, the same magic of LLMs could now start turning toward visual data? What’s needed to do that?
Sam Altman: I mean, yes, in my view, but we have more work to do.
Lex Fridman: Of course. What dangers? Why are you worried about releasing the system? What potential dangers exist?
Sam Altman: Honestly, I mean, one thing we must do before releasing the system is make it run efficiently enough to scale to what people want, so I won’t downplay that. There’s still a lot of work there. But you can imagine problems from deepfakes and misinformation. We strive to be a thoughtful company, deliberate about products we offer the world, and it doesn’t take much thought to see how this situation could go badly.
Lex Fridman: There are many tricky issues here; you’re dealing with a very tricky space. Do you think training AI should fall under fair use in copyright law?
Sam Altman: I think the question behind the question is whether creators of valuable data should be compensated in some way for its use, and I think the answer is yes. I don’t know the answer yet. People have proposed many different suggestions. We’ve tried some different models. But if I were an artist, for example, A, I’d want to opt out of having art created in my style. B, if they do create art in my style, I’d want some associated economic model.
Lex Fridman: Yes, it’s like the shift from CDs to Napster to Spotify. We need to figure out some model.
Sam Altman: The model changed, but people must be paid.
Lex Fridman: Well, if we zoom out further, for humans to keep doing cool things, there should be some incentive.
Sam Altman: Among all my concerns, humans will create cool things, and society will find ways to reward it. It looks pretty hardwired. We want to create, we want to be useful, we want status in any form. I don’t think that’ll change.
Lex Fridman: But the reward might not be monetary. It could be other cool fame and celebration.
Sam Altman: Perhaps financial aspects could be achieved in other ways. Again, I think we haven’t seen the final evolution of how economic systems operate.
Lex Fridman: Yes, but artists and creators are worried. When they see Sora, they say, “Wow.”
Sam Altman: Of course. When photography emerged, artists were very worried too, then photography became a new art form, and people made lots of money from photos. I think similar things will continue happening. People will use new tools in new ways.
Lex Fridman: If we just look at YouTube or similar, how much of its content do you think will be AI-generated using tools like Sora in the next five years?
Sam Altman: People talk about how much work AI will do in five years. The framework is often what percentage of current jobs will be entirely replaced by AI work? The way I think about it isn’t what percentage of jobs AI will do, but what percentage of tasks AI will do over time. So if you think about all five-second, five-minute, five-hour, even five-day tasks in the economy, how many can AI perform? I think this is a more interesting, impactful, and important question than how much work AI can do, because it’s a tool that can work at increasingly complex levels and complete more tasks over longer timeframes, allowing people to operate in more flexible, higher-level abstractions. So maybe people become more productive. At some point, it’s not just quantitative change but qualitative change in the types of problems you can remember. I think the same goes for YouTube videos. Many videos, perhaps most, will use AI tools in production, but fundamentally, they’re still driven by a person thinking, assembling, completing parts, guiding and running it.
Lex Fridman: Yes, that’s interesting. I mean, it’s scary, but interesting to think about. I tend to believe humans enjoy watching other humans or other human-like beings—
Join TechFlow official community to stay tuned Telegram:https://t.me/TechFlowDaily X (Twitter):https://x.com/TechFlowPost X (Twitter) EN:https://x.com/BlockFlow_News
