[0:00:05] How is the baby? Great. It's a boy. It's a boy. You want to get more? Any more. Yeah. It's a lot of work, though. Yeah, I think four would be great. Yeah, so that's the ambition level. I think four would be great. Sam, before I start asking questions, I'd be interested, what is the question that you think most about at this moment? What is on top of your mind? So many things. One that has been top of mind is AI has progressed, I think, so much more than people are aware of. For most of the world, they're still just using ChatGPT.
[0:00:46] And yet we now have systems that are like as smart as our smartest humans and beating them in these like our hardest intellectual competitions. And so there's this huge overhang of what we have versus what the world is using and how that's going to play out, how long it takes the economy and society to catch up. That's a big one. What it's going to be like as these systems really do become self-improving.
[0:01:16] The speed of problem solving is always a little ahead of the speed of problem creation. But it looks quite good at the moment, doesn't it? It does. It does look quite good. So when we were sitting together here in Berlin last time, a couple of months ago, we spoke about Stargate Europe. So can you share, is there any progress on that project? Yeah. So we are very much inspired by you. We just announced OpenAI for Germany.
[0:01:45] And that we would build with SAP and our partners at Microsoft a sovereign cloud with our frontier models for Germany. This is the first of its kind partnership for us. We totally get why countries need both sovereignty and frontier models and the opportunity to deliver this and pioneer this here. Tremendously excited about seeing how this goes. And the idea of, let's say, a bigger project.
[0:02:14] Comparable to the Stargate project in the US, do you think that's still a realistic option in Europe? We'd be excited to do that. I think that'll be a question for the European people. You know, these data centers are huge investments, not just in capital, but energy, resources, focus. But we'd be very excited to support that. I think this is a great first step, and it will get people comfortable with this being a good idea, and we'd love to scale it from here. But you can't do it without nuclear energy, is that correct? So you can't do it...
[0:02:42] In Germany only, you would need, let's say, French or Czech nuclear energy? I mean, we could do it without nuclear energy. Again, I think that'd be a question for the German people and the German politicians, but it's possible. It's a lot of energy. Sam, you have a certain European patriotism. Where does that come from? Why do you care about Europe? It looks like a lost case for many Americans these days. Well, many reasons. I think a strong Europe is important to the world, just as I sort of try to understand.
[0:03:13] How the balance of power is likely to shift over time and where the alliances are likely to be that will help propagate the values that I hold so dear. I think a strong Europe is very important. I love Europe. I love coming to spend time here. The amount of culture and history and scientific advancement that has come out of Europe I don't think is an accident. So I do get sad when...
[0:03:38] Some of my American friends say, oh, it's a lost cause. It's a museum. There's no hope. It's certainly not what I think. I mean, in the bigger scheme of things, in the big, let's say, competition with China, it seems that America should be interested in having Europe as an ally. And the other way around. I think we can't do it alone. I strongly agree. Again, I'm realistic about some of the trade-offs and the challenges, but I think Europe is such an important
[0:04:13] Natural ally for the US. What does Europe get wrong at this moment? It's like way out of my expertise, but in the narrow lens that I have of looking at AI, I think we've had the regulatory conversation. I know it's ongoing and the EU will make more decisions there. And you have a critical approach. You said if they are regulating further, then it may be not a market for OpenAI.
[0:04:43] At some point, I mean, what we've said is we will try our best to be able to offer services here. And, you know, there are times where we can't launch a new model or not as fast. And our European users get very upset. And we explain like, this is, we're not doing this because we want to penalize European users. This is like, we have to comply. So we will certainly always try. But yeah, we could, you know, I can imagine for sure a set of regulations where we just wouldn't be able to comply. I don't think it'll happen. I hope it doesn't happen.
[0:05:12] And I kind of think what will happen is that different countries in the world will try different regulatory approaches. Some will work better, some won't. And, you know, it'll all like, people will learn what works and learn what doesn't and learn the, you know, risks and benefits of a more liberal and more conservative approach. And we're kind of so early in all of this that the world will adapt as we go.
[0:05:39] Ronald Reagan once famously said, the European understanding of the economy is, if a business moves, tax it. If it still moves, regulate it. If it's dead, subsidize it. Wow, I had not heard that. That's an evocative quote.
[0:05:57] I don't know. What do you think about the regulatory approach? Well, the whole world is deregulating Europe, is regulating up. That cannot work. I mean, it's self-damaging and we have to have smart regulation in certain areas. We need it across the Atlantic in America and in Europe. But this isolated way of Europe to be the world champion in regulation, that is leading to disaster and is really blocking innovation. And is that, I mean, my light observation on the outside is that Europe has said, okay, maybe we took this too far and we're going to pull back on that.
[0:06:27] Is that accurate? I think that's a realistic assessment. But what would be your concrete advice, let's say, to the government of the leading economy in Europe to Germany? What needs to happen in order to really stimulate AI progress and innovation in Europe? Well, first of all, I think AI progress in Europe is happening, particularly in Germany. The German usage of OpenAI products is quite amazing. It's our largest market in Europe.
[0:06:58] Virtually every young person, kind of 18 to 24, in Germany uses ChatGPT. I think it's now our fifth biggest market in the whole world. Crazy growth over the last year. And lots of startups and also bigger companies are building with our APIs. So I think it is happening. We hear a lot of complaints from some of those companies about, oh, we can't do this or we can't do that. So there's clearly things to address. My general advice,
[0:07:30] On AI is, you know, focus on the big risks, regulate the big risks. And that I think is important. There are the potential of real catastrophic risks with these frontier models, but then don't... Neglect the small risks. Let the small, like, be willing to accept some risk. But one of the main topics, if people are talking about AI, is always jobs. And what do you think? Is AI in the end destroying more jobs or is it creating more jobs?
[0:08:00] I think in the short term, it will destroy a lot of jobs. In the long term, like every other technological revolution, I assume we will figure out completely new things to do. I am willing to believe that human demand for more stuff, human desire to be useful to other people, to create value, to be part of the community, is pretty much limitless. And at every major technological revolution, very smart people have said,
[0:08:29] This is it. This is the end. This is like, you know, there's going to be no more jobs on the other side of the industrial revolution or whatever. And it has always been a failure of imagination. There are people who say it's different this time with AI because, you know, the special thing about humans is our intellectual capacity. And if it can do everything, what will be left for us? And my counter to that is we seem to really be very wired to care about other people. We don't worry that much about the world of machines. We're very focused on other people.
[0:09:00] And so I don't know exactly what the new jobs will be, but I bet they'll be very people-centric. And, you know, humans, human society, we have such main character energy. We don't really care that the machines are smarter than us. They already are. I totally agree. In the end, we just cannot imagine what's going to happen and which new forms of economy and growth and value creation are going to happen.
[0:09:23] When we had our dinner in Idaho a few weeks ago at the table, somebody asked you a fascinating question. Sam, imagine that I would destroy even your job. What would you do? And you said to our surprise, I would become a farmer. Why farmer? First of all, I think there will come a time when AI can be a much better CEO of OpenAI than me. And I will be nothing but enthusiastic the day that happens. You know, like there's plenty of other things for me to,
[0:09:54] So that doesn't scare me. That doesn't make me sad. That's just like, okay, I did this one thing. It's been automated. I wanted it to be automated. That's kind of what we're doing. There's still plenty of interesting roles for me at OpenAI. If I want to do that, that's great. But this has been a very stressful last 10 years for me. It's been a huge amount of work. I have a farm that I live on some of the time, and I really love it.
[0:10:26] And it sounds like a wonderful change of pace. And you really, I mean, you are not only enjoying the farm, you are really participating and working there or doing things in the nature. Is it like a kind of escape for you or what is it? It used to be. And then since the launch of ChatGPT, my life has gone so crazy that I do nothing but work and hang out with my family. And all of my hobbies have kind of have like gone by the wayside. But so I don't get to like...
[0:10:53] You know, drive tractors and pick stuff. But nature matters to you. Nature matters to me a lot. You're not a tree hugger? I'm not a tree hugger. No, I'm not a tree hugger, but I love time in nature and I love conservation. When I read the biography about you that was published recently, I learned a word that I didn't know before. Petrichor. Can you explain what that is?
[0:11:16] I know what that is. I don't know. I didn't read it. It's the smell after rain. Exactly. If rain drops on dry earth, the smell or the scent is called petrichor. And you said that, or people are saying that you said that that is your favorite scent. Is that true? I love that smell, yeah. Why can you describe it? What makes it...
[0:11:37] Special for you. What do you love about it? I mean, we all know that smell. It's hard for me to explain why I think like certain smells are better than other smells. It's like, I don't know. I just think that's like a particularly great. I'm actually, so in California, we have like a long dry season that's about to come to an end. And I was thinking the other day, I'm so excited for the first rains. But yeah, I love the smell that first day that happens. Okay, but there's no romantic memory? No, no, I just think it's a great smell.
[0:12:02] Are you actually really a believer of this simulation theory or you're just flirting with it? I'm confused. I think there is something strange going on. Like what an odd time for us to be alive, you know, in all of human history. What an odd time for us to be having this conversation right now. So I think there's something strange going on, but I don't know. I don't even in the simulation, like the simulation theory, people talk about that at many different levels.
[0:12:37] You know, how you want to think about that. I don't know. But one thing is sure, our consciousness and what we define as reality is only a small part of what is really going on. I mean, I don't know. I would say that our consciousness is the only thing we can be sure that's really going on. Like I kind of, I believe, I forget the name of that theory, but I believe that. And it's certainly the most important thing going on to any of us. But yeah, there's, it's an unusual time to be alive.
[0:13:06] Yuval Harari, the Israeli writer, wrote a fascinating book called Homo Deus, where he basically describes the next era of mankind where we become almost like gods. So eternal life, we don't deal with practical matters, work is basically done by others, unlimited wisdom and intelligence. So do you share that vision and when is it going to happen?
[0:13:36] There's a quote that people in Silicon Valley love to say. I'm going to get it wrong, but it's something like, we have the technology of gods, the moral framework of medieval times, and there's one more. But I think there is this sense that our technological capabilities are so outpacing our wisdom, our judgment, our time to develop what we want society to be.
[0:14:08] And it does feel unbalanced to me in a bad way. I don't know what to do about that because the technology, like, it's here. Like, this is no longer, you know, I've been talking about this for years. You know, this was the year that I think the AI got smarter than us. Like, that's a big, there's a big moment. And life goes on, but something fundamental has changed. I don't know how sci-fi it's going to get. No one does.
[0:14:39] You know, are we living forever in computers? Do we just cure a bunch of disease? Do we like whatever? I will say there's this new, not that new anymore. In the last few years, there's this show called Pantheon that explores some of these questions and it is the best. I don't watch much TV, but I made an exception and it is the best TV show I have seen in a very long time. So I think these questions are in the zeitgeist and sort of newly important and not theoretical. Certainly, I don't know how it's all going to go.
[0:15:07] Do you see any problem that AI is not going to solve? Yeah, I think we'll all still find ways to, you know, be miserable and manufacture a bunch of adversity. And like, I don't think that's going to get solved. Yuval Harari was making one wrong prediction. He said that there will be no more wars. Wars are outdated. We see the opposite these days. To which degree is AI changing warfare? And can it lead to a scenario where it gets completely out of control?
[0:15:41] I suspect AI will change warfare quite a bit. I think there's a saying about how the military always wants to fight the last war. They never understand enough about how technology changes the landscape. And so if the last war was fought with guns, they assume the next war will be fought with guns. And then, you know, we've seen technology change warfare hugely. I think already with AI, the fact that AI can now write software.
[0:16:17] The fact that AI is probably very close to being great at discovering cyber attacks has significantly changed the landscape of warfare or what it will look like over the coming decade. But I don't know exactly how it'll go. Would OpenAI work with the military sector and make a difference in winning wars? Or do you already? We certainly will in some ways. You know, I don't want us to become a weapons company, but that's always a...
[0:16:50] Easier thing to say than to adjudicate in every case. You know, like if you're a really great propagandist, are you a weapons company? Like a little bit if... And it's interesting how the approach has changed. I mean, a couple of years ago, the whole defense industry was not even invited for dinners. Now they are heroes that are saving democracy. Well, I think it is clearly a mistake not to support the defense industry. We're seeing right now in the world a reminder of why.
[0:17:20] Companies having strong national defense is important. And I certainly believe that. And I'm proud to work with them. Yeah. So we talked about godlike forms of existence and eternal life. Would you like to live forever? I don't think so. I can't really conceptualize what that would be like.
[0:17:49] Because, you know, by the time that would even be like a possible thing where we could like truly cure all disease and, you know, make copies of organs and stuff that forever seems like a long time. And I think the natural order of things and sort of continued forward progress like requires death and turnover and new people. And so I can't like, I can't totally wrap my heads around what that world would look like.
[0:18:15] Forever does seem like an awful long time. I mean, logically, we know that it would be probably boring if life is an unlimited quantity because we wouldn't enjoy things. On the other hand, on a more emotional basis, it's hard to accept that there will be an end. Totally. And also in that world where like living forever for a super long time is even possible, you know, maybe so much else has changed that it doesn't seem boring. I don't know.
[0:18:42] It's not something I've thought about very deeply. I think you once said that Oscar Wilde's Dorian Gray is a book that you particularly like, so the idea of eternal youth seems to be fascinating, or in any case, youth as such. I read that book in high school and thought it was a very interesting and beautifully written book, but yeah, not a particular favorite. How old do you want to get? I don't have a number. What I would like to do is stay healthy. What I don't...
[0:19:15] More than like, here's like the age I would like to live to, I would like the part of my life where I'm like very sick and failing to be short. So I'd like a long, healthy life and then a short end period. AI can help, absolutely.
[0:19:32] If it's about the creation of OpenAI and in more general, the, let's say, development of AI in Silicon Valley, there are a lot of people that seem today to be fierce competitors, but they used to work together at the beginning. I mean, Peter Thiel, Elon Musk, Demis Hassabis, Mustafa Suleiman, Paul Graham. So, Marc Andreessen, I don't know. Who?
[0:20:01] Who in that circle, in that group, in that gang, you think is next to you the most influential person in that context, played the biggest role? I don't know how to rank like the most, but the whole thing is like a pretty tightly webbed ecosystem. You know, there are people making chips and models and products.
[0:20:39] And obviously there's a lot of people trying to make competitive products and competitive models, but it reminds me of like what I've read about the early days of the internet or the early days of the semiconductor boom, where it's kind of this like collective that certainly no one person or one company drives. And there's a lot of people doing important work at every level of the stack. What would you say is your most important contribution? I mean, yeah, you have...
[0:21:07] Created OpenAI, now running it. But was there at the moment an epiphany where you realized something and then did something and we would say that has the biggest impact until today? Maybe making the big bet on that we were going to scale compute after we discovered scaling laws, but before it was obvious to the world. Everybody else, you know, like, I mean, at that time, DeepMind was still sane, which was our big competitor back then.
[0:21:36] That language models, they didn't think were important and not the right path. And they didn't think that even after we launched GPT-3, it took the moment until ChatGPT launched for the world to really say, okay, there's real utility here. Everybody else was in different directions at the time. So I think the conviction bet that we were going to bet the whole company on scaling up language models, which was much harder to do at the time than it sounded from a capital perspective and a research focus. I think that was a big contribution.
[0:22:05] The launch of ChatGPT, deciding to do that and really this belief that, because that also at the time was not obvious. There were a lot of people who thought AI should be developed in relative secret and that it wasn't going to be robust and safe enough to put out there and we didn't want people very involved. So maybe one of those, I don't know. It's like, I think my job is more like a lot of small decisions than like these one big moments. How is your relationship today with Elon?
[0:22:35] We don't talk much. We don't really talk at all actually. Are you suing you? A few times, like a bunch of times. I've lost track. Is it in the end also an ideological fight or is it just competitive? Where does it come from? I don't know. You should ask him for his side of that, but I don't think it's ideological because he basically has like, you know, XAI is very similar to what OpenAI does. Like if you look at their product, it's like very much a clone of ours.
[0:23:05] I mean, they do many things we don't do. So maybe he would say like, you know, we're too restrictive and that's the difference, but I think it's just competitive. But nevertheless, there is a big debate going on that these different large language models have a bias and one is more leaning to the right, the other one is more leaning to the left. Some people are saying chat GPT is too woke. Is there a point or would you say we are neutral?
[0:23:33] I mean, I hear at this point more people say we are not woke enough than too woke. That's a good sign if you're criticized from both sides. I was going to say, you are never, if our goal is for ChatGPT to represent the median views and then allow a lot of customization from there, you're going to get severe criticism from both sides if you're trying to be in the middle. And our goal is to sort of be at the weighted average.
[0:24:03] We spoke about the big and real problems and the small problems that we should neglect in 2015. You wrote in your blog, I quote, the development of superhuman machine intelligence is the biggest threat for the existence of mankind.
[0:24:30] As a consequence, you called for more regulation. Is that still your view or did you change your mind? That's still my view. And another conversation that we had, you said that the likelihood that... But I want to be clear that I think regulation of like...
[0:24:49] Existential risk to humanity is very different than regulation of like, you can use this for this banking application and not for this and this. I think we agree on that. Let's talk about the existential risk. And you quantified it when we spoke about it, that it is roughly around 2% that artificial superintelligence could develop consciousness. And once it's developed consciousness, it could destroy or extinct mankind. Could you elaborate a little? I don't think developing consciousness is the right.
[0:25:20] First of all, 2%. Don't take that as like a literal number I've calculated, but something that is non-zero. Big enough to take seriously, but, you know, not the dominant. But the statistical risk that an airplane crashes with, let's say, 300 people in it is 0.00001% for a relatively small accident. So 2% for the extinction of mankind is to be taken seriously.
[0:25:46] I think there are three, absolutely yes, hence that blog post, hence why I still believe this. I think there's like three main categories of ways that can go super wrong that we could talk about. Number one, there's no alignment failure at all. AI systems do exactly what we tell them to do, but somebody is able to build one and a country, we talked about warfare, a country that is using it to wage war is able to do, you know, sort of like a...
[0:26:16] An unimaginable amount of damage to the world. So that's one we can think about, right? Where AI is aligned, AI follows our instructions, but a human running a powerful country, fighting other powerful countries, decides to horribly misuse it and create a bioweapon or hack into someone's nuclear weapon systems or who knows what. That's not really like, that's not an aim.
[0:26:45] AI alignment failure in any sense of the word. It's someone built it and did exactly what they wanted them to do, but they had too much power relative to the rest of the world. This is why we think kind of like wide distribution of this is important because these capabilities are coming. And if one bad guy has them and a lot of good guys don't, category one. Category two is sort of the more classic sci-fi, you know, the sort of like AI, let's not even call it.
[0:27:15] Conscious, but the AI like develops some sense of agency and does not want to be stopped. And, you know, even, even if it's just like trying to accomplish a goal and there's no intentionality or consciousness, but it's like, I need to not be stopped by these humans in pursuit of this goal. That's that would be an alignment, a big alignment failure and category two, um, sort of the, the sci-fi thing. And then
[0:27:43] And the world spends a lot of time talking about those two. I think they're obviously important things we need to address. There's a third category that gets talked about relatively little compared to how important I think it could be. And this is one where neither of the obvious two failure cases happen, but the model in some abstract sense sort of accidentally takes over the world. And you can imagine
[0:28:16] A way this happens by thinking about what's happening with ChatGPT. So hundreds of millions of people talk to ChatGPT, soon billions. And people are relying on it for more and more important decisions in their lives and their work and whatever. And as these models get smarter, maybe right now you ask ChatGPT for advice about your job and sometimes it gives you good ideas and sometimes it doesn't. You certainly probably wouldn't let it do your job.
[0:28:46] But let's say it gets smarter and smarter and it gives you advice that you still understand, but now you think you should follow almost all the time. And then it gets smarter again. And now it gives you advice you don't understand. You're like, why is that the right thing to do? But it turns out to be right again and again because it can see around corners you can't. And now you either follow the advice that you don't understand or you are less competitive than your peers. And so the model has no ill will. It's really just trying to help you. You're trying to do good work.
[0:29:15] But like, you're like, if I want to be competitive, if I want to succeed in the world, if I want to live my best life, I kind of have to follow the advice of the model. And so does everybody else. So do the billions of other people talking about. So now we're all just doing what an AI system tells us to do. Or, you know, the ones of us that want to be most competitive are. And we are creating new training data that's going back in the model. It's continuing to learn about society. All of society, all the economy and the model keep
[0:29:48] Going in this sort of improvement loop. And who's really in control of that? It's like a kind of collective thing, but it does mean we're like now doing what the model wants us to do somehow for a very broad definition of once. I mean, you could say a artificial intelligence that is self-learning and has low limits has at a certain point to get rid of an inferior form of intelligence because it is just not healthy in the eyes of that super intelligence.
[0:30:17] That I don't agree with at all. I think that projects a lot of anthropomorphic or human-centric thinking into it. Like that assumes that the AI will be like us in some sense and our motivations and drives and all of that.
[0:30:40] Yeah, I mean, the counterargument would be that a superior form of intelligence, human intelligence, did not destroy ants, an inferior form of intelligence. But in any case, we should be prepared for the fact that there will be a form of intelligence that is basically superior and that may also give us direction and is going to guide us. A thing that my co-founder, Ilya Suskovor, used to say very early in the history of OpenAI, but I think was deeply wise,
[0:31:12] Was that what we'll get to kind of say like, okay, Mr. Superintelligence, here's your goal for the rest of time. And people would talk about what that goal should be. And Ilya would say that we should want all AGIs to treat us, treat humans as a loving parent with their child. And I think that's like a really beautiful.
[0:31:38] The friendly AI, the warm-hearted AI, the AI with empathy. It's a little bit different than just saying friendly. It's more than that. Yeah. But even if we have the best intentions not to fulfill the totally apocalyptic and dystopian visions of Eliezer Yudkowsky, who just wrote that book, where he basically is describing the total destruction of mankind as a necessity.
[0:32:07] Even if we don't let that happen because we don't program the machines like that, there may be other actors, bad actors, dictatorships, being it Islamist dictatorship, being it Chinese visions to conquer the world. How do you react to that threat, that it's not us, it's not the US economy, not the US society or the European society who does it?
[0:32:36] But other totalitarian societies are simply bad actors, criminals. Well, that was my category one of the ways this goes wrong, which is a bad actor. But is there any counter measure, any recipe to deal with it? Is there any governance that could avoid it? I think there are also sort of like three common approaches to this. Number one, and I think this is the LEA approach, you melt all the GPUs so that no one ever gets any technology again.
[0:33:06] Things like this in sci-fi, Dune is an example where people say this is the only way to prevent the rogue AI. Number two, you say that the answer to powerful AI is a lot more access to powerful AI. So yes, bad actors are going to get AI and they'll try to use it to attack. And the solution is lots more good AI in the hands of good guys using to defend. You know, we are not powerful enough as current unaided humans.
[0:33:36] To fight back against really bad AI, a super intelligence. But if all the good humans have super intelligence too, then hopefully the power balance of the world and more people are good than bad keeps going. And then the third, which I think we need to push for is a, some sort of global governance set of shared values. And we have done this before in other, when faced with other potentially catastrophic risks. The IEA is an example.
[0:34:07] You know, I think the world has done a good job so far with the threat of nuclear weapons. And I believe it's possible again. So you think with the balance of power of nuclear weapons, in a way, stabilize peace to a certain degree. Do you think that that could happen also with AI? Absolutely. I'd be surprised if it couldn't. It might not. But do I think it's possible? Yes.
[0:34:33] Sam, you are here in Berlin to receive the Axel Springer Award. Axel Springer is a publisher we live from and for journalism. Journalism is under attack, under attack by governments. The whole idea of freedom of expression seems to be here and there a pretty one-sided pleasure. And at the same time, it is also under attack by the development
[0:35:03] So what is your view on the future of journalism? Do we need journalism?
[0:35:34] My own company's work inadvertently. That would be a huge net loss. You mentioned freedom of expression. Freedom of expression is to me one of the most important and most difficult to uphold values. It's easy to commit to freedom of expression in the abstract sense. And then very quickly you realize people only mean it when it's the stuff that they like or that everyone thinks like they... It's painful. It is difficult. It truly is difficult. But to my mind,
[0:36:08] One of the clearest lessons of prosperous societies, not in the sense of economic prosperity, although that is a part of it, but societies where individuals thrive more and more over time and the quality of human life goes up and up. Freedom of expression seems just like an integral part. It seems like one of the best things we ever agreed upon. And watching that...
[0:36:37] Under continual attack in all these ways is very scary. Again, I get how difficult this value is to uphold. This is easier said than done. But I really hope that the Western world will turn its commitment back to this. On the question about answer machines, I will personally be very sad if ChatGPT somehow destroys journalism. I don't think it's likely, but that would be really upsetting to me. We need people to get...
[0:37:11] The information about what's happening in the world in real time and thoughtful analysis out into the ether. And certainly without that systems like ChatGPT can't do their job. Now, how much is that going to shift from a small number of professional journalists in the traditional sense to sort of something more like social media? That I don't know. How much is it going to shift from people really having a relationship?
[0:37:39] With a publication to individual journalists, that I don't know either. I certainly have shifted a lot of my consumption to caring about what I think very smart individuals say in their newsletters or whatever. But I think the need for the goings on of the world to be reported by people that we put a trust proxy into, that will keep going.
[0:38:07] I'm not so worried about the freedom of speech part. And I think it is up to us, the media companies, the journalists to defend it. And if we can't stand the heat, we should get into different jobs. It is the beauty of the open society model in autocracies. It doesn't exist and look to the quality of life of people. So in the end, I think there are self-correcting factors. I'm not so worried. And again, it's up to us and we are defending it day and night and we love it.
[0:38:35] I'm more worried about the other aspect because it is more fundamental. I mean, technically, the idea of copyright, and I'm not criticizing anybody, you, ChatGPT, and others, but simply the idea of copyright is broken. It ceased to exist in these days of large language models that are scraping data for their smart answers. And the offerings are so convincing. And at the same time, if for somebody who is basically
[0:39:05] Working to find out something that was not supposed to be found out. If there is no incentive to do so because there is no business model, there are no revenues coming back. It is immediately once it's published, it's used by others. Then we have a fundamental problem. Why am I also a little optimistic here? Because it is also in the mutual interest. If the system dries out, if you have no diversity of sources, no investigative journalists going around.
[0:39:30] Then there is nothing to scrape. There are not enough data, and the answers are getting more stupid. So in a way, it is in our mutual interest. Would you agree with that? Yeah, this is what I was trying to say. You need that constant reporting in the world, and that requires a lot of money to flow somehow to get people to find out those things. It may need a totally new economic model. It may only need a small tweak to the current models of how money flows.
[0:40:01] Like my personal hobby horse on this is it is fine. I have always wanted micropayments on the web to happen for valuable kinds of content. It has always been too difficult. I think in a world of AI agents who can finally maybe make it work, and that is the model I would love to see, you probably have a different idea. The world will try a lot of different things, but I hope there can be some much better way for this all to flow.
[0:40:29] Can I push on the principle a little bit?
[0:40:59] Some very valuable information. And that information is like directly reproduced within some constraints of what that means. It would be great to have like a new economic model where money can flow and you can imagine all sorts of things with agents. However, eventually that thing becomes historical fact. Or if we take another, maybe here's a different example.
[0:41:27] Let's leave news aside for a second. If I read a textbook on physics, I buy the textbook on physics, I read the textbook, I learn physics, I then shouldn't have to pay that writer every time I apply general relativity and do some math, nor should I have to pay Einstein probably for discovering it. So how do you differentiate between those two? Well, that is the challenge that the practical answer is very difficult and the incentive
[0:41:57] needs to remain intact for somebody to do something in order to create new IP, whether it's science, whether it's Einstein or whether it's a music song. If that incentive is not there, nobody is going to do it. And whether we do it in a micro payment form or whether we do it in a more kind of reach based form that you say this is.
[0:42:23] A quantity that we can measure and that leads to a certain revenue share or redistribution of money. I don't mind, but we need to find a principle that keeps that system intact. That is true for science and for any other form of IP creation. I wouldn't make a difference here. I think an interesting example is how the music industry in the US handles sampling. If you sample someone else's music in your song.
[0:42:47] And there's like some amount of time that it's longer than some amount of content. Then you negotiate and you pay them and it can go down the chain if it's multiple versions of that. And the industry kind of figured that out. However, if you just listen to people's music and you get inspired and you make different music, but you've learned something about that style, you don't pay for that.
[0:43:10] You as a musician are allowed to go listen to as much music as you want and learn about what makes a good song and a bad one and apply those concepts in brand new ways that are not reproducing it to generate your own stuff. The fact that you listen to music, you don't pay for that. There's a model like that in the example of music.
[0:43:30] I think we could apply maybe more broadly. I totally agree. And music is a good analogy because it always used to be a fluid thing. Everybody could copy it. Every DJ, every radio station could just play a song. So you needed that model. Whereas in the more kind of in less fluid worlds, it was handled differently. So, okay, let's hope that we will, in mutual interest of a healthy ecosystem, find a solution for that. Getting back before we conclude to OpenAI.
[0:43:56] and further developments. You started December 2015 as a nonprofit project, and now you are transforming into a for-profit company. The announcement is or the expectation is that by the end of this year, you will be a fully transformed for-profit company. Is that realistic? Are you going to deliver on that? And not really transforming, like the nonprofit will continue to exist. In fact, if we do a good job, it will be by far.
[0:44:26] The most well-resourced nonprofit in history. And I think it will continue not only to enforce the OpenAI mission and govern OpenAI, but do an amazing amount to deploy the resources using AI for benefit to the world. The thing I'm personally most excited about, although there's several things it can do, is using AI and using money.
[0:44:55] with AI to advance scientific understanding of the world and sharing those benefits. So if the nonprofit can help cure a bunch of diseases by applying AI to science and make those cures freely available, that'd be wonderful. We already do have a for-profit entity, but we're recapitalizing that.
[0:45:14] And so it'll have a more normal corporate structure. You raised a lot of money. Recently, it was a 300 billion round. Then a few days later, the announcement of a 500 billion round. Now, NVIDIA announced 100 billion investment, I think, on top of that. So if you look ahead, what is your vision for OpenAI, let's say, 10 years down the road? I mean, 10 years down the road is a long time. Like, you know, we're not even 10 years old yet.
[0:45:43] I wouldn't call it an endgame because there's never an endgame. Hopefully always development. But what is the boldest vision that you could imagine for that company? There's never an endgame. There is a smoothly rising exponential. But in five years, I really do hope we're in deep, not deep, we're somewhat into this world of AI discovering new science, where AI is being used to cure disease, to invent new materials.
[0:46:13] Fully understand physics. I am a true believer in the idea that all or almost all real sustainable human progress comes from discovery of new science and then how we apply that to the economy and society. And for the first time ever with GPT-5, we are finally beginning to deliver on that promise in small ways. But you see these reports online of scientists using AI to make new discoveries.
[0:46:44] And it's blowing their minds and I'm just so excited. So I hope over the next five years, we could really get into that phase. Sam, very last question. You only deal with things that really can change the world. Now, let's imagine that OpenAI has done it. What could be the next task for you? I want to go be that farmer. You know, the two things I care most about professionally are AI and energy.
[0:47:14] And if I can make a big contribution to those two, I think that's enough and then I should get to enjoy my life. Thanks very much.