[0:00:07] Welcome to Tel Aviv University. With us today, university supporters, management, faculty, students. We also have guests from Microsoft and other areas of the Israeli high-tech ecosystem. I'm Nadav Cohen, faculty at the School of Computer Science, and it is with great pleasure that I invite on stage Sam Altman and Ilya Sutskevel, CEO and Chief Scientist of OpenAI.
[0:00:41] Good to see you. Have a seat. So, thanks a lot for being here. I greatly appreciate it. Thank you for having us. I think we're going to start with a brief intro of yourselves. Ilya, please emphasize the Israeli roots. Oh, for me to introduce myself? Sure, yeah. I mean, hi everyone. So, indeed I...
[0:01:16] From the ages of 5 to 16, I lived in Jerusalem. I studied in the Open University from 2000 to 2002. After that, I moved to the University of Toronto, where I spent 10 years, and I got my bachelor, master's, and PhD degrees. And already during grad school, I was fortunate to contribute to important advances in deep learning. Then, with a few people, we started a company that was acquired by Google.
[0:02:00] And I worked there for some time. And then actually one day, I received a cold email from Sam saying, hey, let's hang out with some cool people. And I was very curious. I went. And that was the first original dinner with Elon Musk and Greg Brockman where we decided to start OpenAI. And we've been at it for quite a few years. So that's where we are right now. Thank you. Sam?
[0:02:28] I was very excited about AI as a little kid, like a big sci-fi nerd. Never really thought I'd get the chance to work on it, but then ended up at university. Studied it for a little while. It wasn't working at all. This was like 2004 timeline. Dropped out. Did startups, became a startup investor for a while. Really got excited about what was happening with AI after the advances that Ilya mentioned.
[0:02:56] Sent him that email and here we are. Okay. So to get things started, I wanted to ask you, what do you think is it about OpenAI that makes it a leader in generative AI? Especially when its competitors are often much larger, have more resources. So what do you think are the key advantages? Focus and conviction. We believe, I think we always...
[0:03:28] I'll add to Sam's. I can only add a small amount to Sam's answer, which is...
[0:03:58] You know, test one, two, three. Okay, to add to Sam's answer, I can only add 5% to Sam's answer, which is progress in AI is a game of faith. The more faith you have, the more progress you can make. And so if you have a very, very large amount of faith, you can make the most progress. And it sounds like I'm joking, but I'm not. You have to believe.
[0:04:27] You have to believe in the idea and to push on it. And the more you believe, the harder you can push. And that's what leads to the progress. Now, it's important that the thing you believe in is correct. But with that caveat, it's all about the belief. Thank you very much. So moving on to other topics. Progress in AI these days, and for a while now, is largely driven by industry. So I was wondering...
[0:04:58] What you feel should be the role of academic research in the field as it evolves? Yeah. No, it is true. Things have changed a lot. Academia used to be the place where the most cutting edge AI research has been happening. Now, not so much. For two reasons. The amount of compute and the engineering. Academia has less compute and generally does not have an engineering culture. And yet,
[0:05:35] Academia can make very dramatic and significant contributions to AI, just not to the most cutting-edge capabilities. The place that academia can contribute to, there are so many mysteries about the neural networks that we are training. We are producing these objects of miraculous and unimaginable complexity. What deep learning is, is a process of alchemy. We take the raw materials of data,
[0:06:04] Plus the energy source of compute, and we get this intelligence. But what is it? How does it work? What are its properties? How do we control it? How do we contain it? How do we understand it? How do we measure it? These are unknowns. Even the simple task of measurements. How good is our AI? We can't measure it. It wasn't a problem before because AI wasn't important. But now that AI is important, we are realizing we can't measure it. So this is just off the top of my head, some examples of problems.
[0:06:34] Which no one can solve. You don't need a giant compute cluster. You don't need a giant engineering team to ask these questions and to make progress on them. And if you do make progress, that will be a dramatic and a significant contribution that everyone will take note immediately. Thank you. So it sounds from your words, and actually relate to that, that there isn't exactly a balance between the progress in industry and in academia. We would like to see more.
[0:07:03] Contributions of those types. So I was wondering, is there anything you think that can be done to improve the situation, especially maybe from your position to somehow support? Yeah. So how can, like I would say two things. The first and the most important thing, I think, is the mindset shift. I think that, so I'm a little bit removed from academia these days, but I think there's a bit of a crisis of what are we doing?
[0:07:35] And one thing that creates a lot of confusion, I claim, is there's a lot of momentum around. A very large volume of papers is being written. But the important thing is to think about the most important problems. Just focus on them. The mindset shift on focus on the most important problems. What is it that we can't do? What is it that we don't know? We can't measure those systems. We can't understand them. Realize the problem. Once you understand the problem, you start moving towards it. That's where we can help.
[0:08:05] We have an academic access program where academic universities apply to get access to our most advanced models. They study them, they write papers. We've done it even with GPT-3, even before we've had our first product. Many universities have written papers studying their models, their properties, their biases. And I think there will be a few more ideas, by the way. I'd be happy to hear them. Yeah. So we should definitely discuss these things offline further.
[0:08:35] You know, I need to fit into the time that I have. You mentioned publishing. So it seems to me, as somebody in the field, that some believe, or at least it's a fair argument, that the level of scientific transparency is somewhat in decline with regards to research going on in industry. And while there are players that, or companies that really promote open source,
[0:09:05] Publishing their models, publishing their code. Others do so less, and then some say that includes also OpenAI. So I was wondering, what do you feel about, first of all, if you agree with this, and if so, what do you believe is the right strategy? Why is OpenAI's strategy the way it is? We do open source some models, and we'll open source more models over time. But I don't think it's the right strategy to open source everything. If the models of today are,
[0:09:34] Interesting, they have some usefulness, but they're quite primitive relative to the models we'll create. And I think most people would agree if you make a super powerful AGI that has wonderful upsides, but existential downsides, open source may not be the best answer for that. And so we're trying to figure out the balance. We will open source some things. We will, over time, as we understand models more, be able to open source more.
[0:10:04] And we have published a lot. I think a lot of the key ideas that other people are now using to build LLMs were published by OpenAI. And I think from the early GPT papers, scaling laws, some of the RLHF work. But it's a balance that we have to figure out as we go, and we have a lot of different tensions on us to successfully manage there.
[0:10:32] So are you considering models where you maybe publicize things to selected crowds, maybe not open source to the entire world, but to scientists, or is that something you're considering? When we finished training GPT-4, we spent a long time, almost eight months, working to understand it, to ensure the safety, to figure out how to align it, had external auditors, red teamers, and scientific...
[0:11:01] Community engagement. So we do that and we'll continue to do it. Okay. So I want to talk a little bit about the risks. I know it's a topic that's being discussed a lot before we get to the opportunity. So just a couple of minutes on that because I do think I agree it's important. So there are probably at least three classes of risks that one can imagine. One is economic dislocation, you know, jobs becoming redundant, things like that. Another one could be maybe
[0:11:28] A powerful weapon in the hands of few. One person, for example, a hacker could do probably something equivalent to thousands of hackers before if they are able to use these tools. And the last maybe, which is to some the most concerning, is a system that gets out of control. Even the people that triggered it to do something can't stop it. So I was wondering what you feel like is a likely scenario on each of these. OK.
[0:11:58] The likely scenario on each of the risks. Economic dislocation. Let's start with that. So you mentioned three. Economic dislocation, hacker, super intelligent, they have been out of control. Yeah, so economic dislocation indeed, like we already know that there are jobs that are being impacted or they're being affected. In other words, some chunks of the jobs can be done. You know, if you're a programmer, you don't write functions anymore. Copilot writes them for you.
[0:12:28] If you're an artist, though, it's a bit different, because a big chunk of the artist's economic activity has been taken by some of the image generators. I think that indeed it's going to be not a simple time with respect to jobs, and while new jobs will be created, it's going to be a long period of economic uncertainty. There is an argument to be made that even when we have full human-level AI, full AGI, people will still have economic activity to do. I don't know whether that's the case.
[0:12:59] But in either event, we will need to have something that will soften the blow to allow for a smoother transition either to the totally new professions that will exist, or even if not, then we want the government, the social systems will need to kick in. On the offense question, the hackers, yeah, that's a tricky one. Indeed, AI will be...
[0:13:28] And it could be used in powerful ways by bad actors. We will need to apply similar frameworks, similar to the one we apply with other very powerful and dangerous tools. Now, mind you, we are not talking about the AIs of today. We are talking about as time goes by and the capability keeps increasing, you know, and eventually it goes all the way to here, right? Right now we are here. Today, that's where we are. That's where we're going to get to.
[0:13:57] When you get to this point, then yeah, it's very powerful technology. It can be used for amazing applications. You can say, cure all disease. On the flip side, you can say, create a disease. Much more worse than anything that existed before. That'd be bad. So we will need to have structures in place that will control the use of the technology. It's powerful. You know, Sam has proposed a document where we said,
[0:14:27] The IAEA for AI to controlling very powerful technologies, but for AI specifically. That's the IAEA is the organization that controls nuclear power. To the last question, the super intelligent AI that's out of control. Yeah, that'd be pretty bad. Yeah, so it's like, it would be a big mistake to build a super intelligent AI that we don't know how to control. Can I add a few things? Yeah, of course.
[0:15:03] Bestow a bit of optimism. I have nothing to add to that last sentence. I strongly agree. On the economic point, I find it very difficult to reason about how this is going to go. I think there's so much surplus demand in the world right now, and these systems are so good at helping with tasks, but for the most part today, not current jobs, that I think in the short term the picture actually just looks pretty good. It's going to be a lot of...
[0:15:32] Dramatic productivity growth. And we're going to find out that if you can make programmers two times as productive, there's more than two times as much code that the world needs. So it's all good. In the longer term, I think these systems will do more and more complex buckets of stuff. And categories of jobs, some of them will go away. But some others will turn out to really need humans. And people really want humans in these roles in ways that are...
[0:16:02] Not very obvious. One example is that one of the first times the world saw AI was when Deep Blue beat Kasparov and everyone said chess is totally over. No one is ever gonna play chess again because it's not interesting. And that was just consensus that everybody agreed with. Chess has never been more popular than it is right now. Humans have gotten better at chess. The expectation has gone up. We can learn better with these tools.
[0:16:31] But people still really want to play and humans seem to still really care about what other humans do. You know, Dali can make great art, but people still care about the human behind the art that they want to buy and that sort of we all think is special and valuable. On the chess example, people watch humans play chess more than ever before too, but not very many people watch two AIs play each other. So I think there are just going to be all of these things that are difficult to predict. The human desire to differentiate,
[0:17:01] Create new things to sort of gain status. I think that's not going to go anywhere. But it will somehow look really different. And I would bet that the jobs of 100 years from now look almost nothing like the jobs of today. Many of them. Some things will turn out to be weirdly similar. But I do really agree with what Ilya was saying. That no matter what's going to happen, we're going to need some sort of different socioeconomic contract as automation reaches these.
[0:17:29] heretofore unimagined heights. Okay, thank you. So, another question on this topic. So, Sam, you recently signed a petition, right, calling for treating existential threat from AI with great seriousness. I'm not sure if I... Oh, you did too. Any of you signed it as well. So, I was wondering, kind of following this call, if there are any steps that you think we, mankind, and also maybe companies like OpenAI should take
[0:18:00] To address this problem. I really want to emphasize what we're talking about here is not the systems of today, not small startups training models, not open source, not the open source community. I think it would be a mistake to go put heavy regulation on the field right now or to try to slow down the incredible innovation. I hope we do get to talk about the benefits that's happening. But if we are
[0:18:29] I think what Ilya said about you really don't want to make a super intelligence that is not really well aligned. That seems inarguable. And I think the world should treat that not as a ha-ha, never going to come sci-fi risk, but something that we may have to confront in the next decade, which is not very long for the institutions of the world to adapt to something.
[0:18:55] One idea that we've contributed, and I hope that there's far better ones out there, is if we could get a global organization that at the very highest end, at the frontier of compute power and techniques, could have a framework to license models, to audit the safety of them, to propose tests that are required to be passed, that would help. That would be one way to treat this as a very serious risk. We do do the same thing for nuclear, for example. Okay.
[0:19:24] So let's indeed move on to talk about benefits a little bit. So this is kind of a scientific setting that we're in. So I was wondering, in terms of the role of AI in scientific discoveries, do you have any predictions or thoughts where we're going to be in a few years and maybe in the future beyond that? This is the thing that I am most personally excited about with AI. I think there's like tremendous, wonderful things that are going to happen all over the place.
[0:19:50] Huge economic benefits, huge healthcare benefits. But the fact that AI can help us do scientific discovery that we currently aren't capable of, we're going to get to understand the mysteries of the universe. And more than that, I really believe that scientific and technical progress is the only sustainable way that lives get better, that the world gets better. And if we can go unlock a gigantic amount of new science, new technological progress,
[0:20:19] Which I think we're already seeing the beginnings of with people using these tools to be more efficient. But if you imagine a world where you can say, AI, help me cure all disease, and it helps you cure all disease, this can be a dramatically better world. And I think we're not so far away from that. Okay. Another major problem alongside diseases is climate change. So I was wondering what your thoughts are in the potential role of AI there. Because I did see, Sam, that you did mention it as a...
[0:20:50] Potential area for contribution. I don't want to say this because climate change is so serious and so hard of a problem, but I think once we have a really powerful superintelligence, addressing climate change will not be particularly difficult for a system like that. We can even explain how. Here's how you solve climate change. You need a very large amount of...
[0:21:26] Of efficient carbon capture. You need the energy for the carbon capture, you need the technology to build it, and you need to build a lot of it. If you can accelerate the scientific progress, which is something that a powerful AI could do, we could get to a very advanced carbon capture much faster. We could get to a very cheap power much faster. We could get to cheaper manufacturing much faster. Now combine those three. Cheap power, cheap manufacturing, advanced carbon capture. Now you build lots of them.
[0:21:56] And now you sucked out all the excess CO2 from the atmosphere. And this plan today is a little bit difficult. If you have an AI which accelerates science and engineering very dramatically, it becomes very straightforward. And I think this illustrates how big we should dream. If you think about a system where you can say, tell me how to make a lot of clean energy cheaply. Tell me how to efficiently capture carbon.
[0:22:27] And then tell me how to build a factory to do this at planetary scale. If you can do that, you can do a lot of other things too. With one addition that not only you ask it to tell it, you ask it to do it. Okay, so I want to ask a couple of questions about OpenAI products. So first of all, in ChatGPT, I was wondering, so you released it. I heard you say that you didn't expect it to kind of spread like it did.
[0:22:58] So I was wondering, is there any application of ChatGPTs that you saw by others that really surprised you in terms of the value that it generated or the capabilities that it exposed? I'm going to go first. Yeah, I mean, I'm just, the thing which has given me, me personally, an endless amount of joy is when my parents told me that their friends use ChatGPT in their daily lives.
[0:23:29] So I would say that this was definitely very surprising and very enjoyable for me. It's hard to pick just like a couple of the favorite stories because it really is remarkable the creativity of the world and what people do when you give them powerful tools. Education has been amazing for us to watch. The number of people that write in saying like, this has changed my life because I can learn anything now or I learned this specific thing or...
[0:24:01] You know, I couldn't figure out how to do this and now I know. There's something that I find personally quite gratifying and wonderful to see about people learning in a new and better way and imagining what that's going to look like a few years from now. But if we can just unlock human potential at this rate, we didn't quite expect that to happen and it's been amazing. And then a fun story that I heard just yesterday, and I've heard like other versions of this in the past,
[0:24:32] was a guy that spends two hours every night with his kid collaborating to make up bedtime stories. And that it's just the kid's favorite thing and it's become this special moment every night that they do. Okay, thank you. And that's just one last small question before we move on to questions from the crowd. So in terms of what you can say, what is the most futuristic product that OpenAI is working on these days?
[0:25:03] Most futuristic product. System. Yeah. It's like we don't think about it in terms of products. We think about it in terms of can you improve the AI? Can you produce the next generation of the AI, of the model, of the neural network, which will be more reliable, better at reasoning, more controllable, better at the whole thing. So you do this, and then you get a whole new world of applications. A bit hard to predict, but we expect everything to become much better.
[0:25:33] And very significantly. I hope the world is never odd at us again. I hope that people had an update with ChatGPT. But from here on, it is one continuous smooth curve of progress. At every stage, we're confronting the risks successfully. It always feels like it's doing what you want and it's safe to use. But every year...
[0:25:57] Your expectations go up, and we deliver on them. And it feels like this gradual acceleration of technology, but in a way that very much is a tool that serves you. Okay, thank you. And now let's move on to some questions from the crowd. There's going to be a microphone. That was a quick draw. Yeah, but we will try moving forward. Have people raise their hands, and we're going to choose. So, wait.
[0:26:34] Both of you, the question is, could the open source LLM potentially match GPT-4's abilities without additional technical advances? Or is there a secret sauce in GPT-4 unknown to the world that sets it apart from other models? Or am I wasting my time installing stable Vicunia, 13 billion plus wizard? Am I wasting my time? Tell me. All right, so.
[0:27:17] To the open source versus non-open source models question, you don't want to think about it in binary black and white terms where there is a secret source that will never be rediscovered. What I will say, or whether GPT-4 will ever be reproduced by open source models, perhaps one day it will be. But when it will be, it will be a much more powerful model in the companies. So there will always be a gap.
[0:27:46] between the open source models and the private models. And this gap may even be increasing with time. The amount of effort and engineering and research that it takes to produce one such neural net keeps increasing. And so, even if there are open source models, they will be less and less produced by small groups of dedicated...
[0:28:14] Researchers and engineers. And it will only be the providence of a company. A big company. Hi. Can you tell us more about the base model before you lobotomized it, aligned it? The base model of GPT-4. What about it? How was it before you lobotomized it? We definitely realized that in the process of doing RLHF on the models, it loses important capability.
[0:28:47] We're studying how we can preserve as much of that as possible. The base model is not that easy to use, but what we'd like to get to is something that does follow instructions and gives users as much control and as much capability as possible and doesn't get us in legal trouble, although we've discovered a lot of stuff like refusals to help with that. So we totally hear the request for...
[0:29:18] More flexible models. And we're trying to figure out how to do that. And give users more customization over them. Okay. We have a question over there. First of all, thank you so much for this talk. It's truly invaluable. I'm really curious to know what in your eyes are the top sectors that can be impacted for the better by individuals and small companies. I didn't hear the question. Can you repeat the question, please? There is a lot of echo. Sorry.
[0:29:48] I'm really curious to know what in your eyes are the top sectors that can be impacted for the better by individuals and small companies. One of the reasons we're doing this trip around the world is to hear from people about what they want, what they'd like OpenAI to do, what their concerns are, how they're thinking about regulation, how they're thinking about how they want this to be integrated in society.
[0:30:15] But the other is to talk to people that are building on top of our API and understand what they're doing and what they want to do. And for me, the most fun part of this trip has been meeting developers and just being amazed at the creativity, the scale of the businesses being built. One, two, or three people that are building something that has now gotten to real scale and a product that people really love. And how that is happening in every industry.
[0:30:44] When we do these developer roundtables, almost never are two people working on the same kind of sector even. It's the diversity that is the coolest thing. I think any vertical you want to pick, AI is going to impact somehow. And this is probably the most magical period since the launch of the iPhone, at least for a technological tidal wave to go do incredible things.
[0:31:10] I think the most exciting part of this is it's not one or two sectors. It's just find some place that you're passionate about and go do it. Let's now, every person that asks a question, ask a question. Start with name and affiliation. Okay. Dror Globalman with Keshet Broadcast. First, thank you again for coming here. I appreciate this talk very much. And secondly, if you truly believe that AI imposes a danger to humankind, why keep developing it?
[0:31:39] Aren't you afraid for your own dear ones and family? And secondly, should regulation will be imposed upon you, upon OpenAI and other AI companies, will you obey or behave much like, say, Mark Zuckerberg, who tried to evade every regulation he finds? Thank you. I think it's a super fair and good question. And the most troublesome
[0:32:09] Part of our jobs is that we have to balance this incredible promise and this technology that I think humans really need, and we can talk about why in a second, with confronting these very serious risks. Why to build it? Number one, I do think that when we look back at the standard of living and what we tolerate for people today, it will look even worse.
[0:32:39] Then when we look back at how people lived 500 or 1,000 years ago and we'll say like, man, can you imagine that people lived in poverty? Can you imagine people suffered from disease? Can you imagine that everyone didn't have a phenomenal education and were able to live their lives however they wanted? It's going to look barbaric. I think everyone in the future is going to have better lives than the best people of today. And again, the upside there is tremendous. So I think there's like a moral duty to figure out how to do that.
[0:33:09] I also think this is unstoppable. This is the progress of technology. It won't work to stop it, and so we have to figure out how to manage the risk. We were formed as a company in large part because of this risk and the need to address it. We have an unusual structure. We have a capped profit. I believe that incentives are superpowers, and if you design the incentives right, you usually get the behavior you want.
[0:33:37] We're going to all do fine. We're not going to make any more or less money if we make the numbers go a little further up to the right. We don't have the incentive structure that a company like Facebook had. And I think there were very well-meaning people at Facebook. They were just in an incentive structure that had some challenges. So we tried to take AGI. We tried to, as Ilya always says, we tried to feel the AGI when we were setting up.
[0:34:00] Our company originally, and then we set up our capital profit structure. So how do we balance the need for the money for compute with what we care about is this mission. And one of the things we talked about is what's a structure that would let us warmly embrace regulation that would hurt us the most. And now that the time has come for that, we're out here advocating around the world for regulation that will impact us the most. So of course we'll comply with it. I think it's more easy to get...
[0:34:28] Good behavior out of people when they are staring existential risk in the face. And so I think all of the people at the leading edge here, these different companies, now feel this, and you will see a different collective response than you saw from the social media companies. I think all the skepticism, all the concern is fair. We wrestle with this every day, and there is not an easy soundbite answer. Hi, my name is Netanel Bezalel. I am a CEO at a small business.
[0:34:57] And I have to mention that we use GPT for a lot. It helps us a lot. And last year, I spoke with a VP in Microsoft, and she told me how they decided to listen to AI because all of the AI testing, the AI was right. And I'm just wondering, what is the gap between the AI you use, like...
[0:35:21] We have a lot of limitations with tokens and all of the things, and you don't have. But what is the gap between the power that you have to the power that we can use? The gap that there is between the models that you use and the models that we use is the question. Well, I mean, right now, GPT-4, you have access to GPT-4, and so do we. Indeed, we are working on the next future model.
[0:35:52] Maybe I'll describe the gap as follows. As we keep building AIs of increasingly greater capabilities, there will be a larger gap, a longer testing period, a longer period where we will read, team, understand the limitations of the models, understand all the way, you know, as many of the ways as possible in which it could be used in ways that we deem unacceptable, and then expand it gradually. So, for example, right now, GPT-4 has vision recognition abilities,
[0:36:23] Which you have not rolled out yet because the finishing touches weren't quite there. But soon we will. So maybe, so I think that would be an answer to your question. Hi. Probably not too far in the future. Hi, I'm David. I'm a data scientist at Classified. I'd just love to know what are your thoughts about the We Have No Modes document that was released lately. The leaked document.
[0:36:58] The thing that is special about OpenAI, and I think the thing that is so misunderstood by that document, aside from the fact that we have a gigantic number of users and people that have formed some sort of relationship with us and our products, is what OpenAI is special about is figuring out what comes next. It is easy to copy something once you know it can be done, and in that sense, sure. It is very hard to go figure out what to do next.
[0:37:27] The ideas, the big ideas, the medium-sized ideas, the small ideas, and the careful execution on them that it takes to get from here to superintelligence, that's what our mode is. So, sure, once we go do the next paradigm, everybody will get going trying to copy that too, but we'll already be working on the next one. Hey, Sam up here. Hello.
[0:37:49] We're up here. My name is Shayben Basat. I'm a YouTuber. I'm also a CEO of a new startup. I have a question regarding superintelligence and the Rokos Bazalisk dilemma. Can you kind of elaborate on how CGPT and OpenAI stands on that dilemma? So while Rokos Bazalisk is something that we are not very concerned about, we are definitely very concerned about superintelligence. And just for context, not everyone may...
[0:38:18] Not everyone in the audience may understand what we mean by superintelligence, right? What do we mean? One day, when it will be possible to build a computer, a computer cluster, a GPU farm, that is just smarter than any person, that can do science and engineering much, much faster than even a large team of really experienced scientists and engineers. And that is crazy. That is going to be unbelievably, extremely impactful.
[0:38:47] It could engineer the next version of the system, like AI building AI. It's just crazy. So our stance is that superintelligence is profound. It can be incredibly, unbelievably positive, but also very dangerous. And this danger needs to be approached with care. This is why we propose the IAEA approach to the very, very advanced cutting-edge system of the future, the superintelligence.
[0:39:16] And also, there is a lot of research that we'll need to do to contain the power of the superintelligence to align them so that their power and their capability will be used to our benefit, to the benefit of people. So that's our stance on superintelligence. It is the ultimate challenge of humanity, superintelligence. If you think about the evolutionary history of humanity, so four billion years ago, there was a single cell, some kind of a replicator.
[0:39:45] Then about a number of billions of years, you had various different single cellular organisms. Then about a billion years ago, you had multicellular life. Several hundred million years ago, you had maybe reptiles. Sixty million years ago, you had mammals. Ten million years ago, you had primates. One million years ago, you had the Homo sapiens. Then 10,000 years, we had the writing. Then we had the farming revolution. Then the industrial revolution, the technological revolution, and now finally the AGI, the super intelligence.
[0:40:14] It is the final, the ultimate challenge. It can create a life of unimaginable prosperity, which Sam alluded to. But it is also a great challenge. And it is a challenge that we need to face and overcome. Hello. What a time to live through. Ronnie Dorey from Kalkalist. Hi. This is a question for Sam Altman.
[0:40:43] I was wondering, what is your stand about data dignity in the context of AI? Yeah, we think it is very important that the people who contribute data to these systems or the people who in some other way help these systems, even if it's not training on their data, are rewarded, get the benefit from these systems. I think what these systems really want to be are reasoning engines.
[0:41:12] But they will be able to go off and access different data and they will also need people who help teach them how to reason correctly. And we are exploring a lot of ideas about how those people get aligned rewards with the success of the model. And also how if you're an artist and people are generating art in your style or inspired by you or whatever, you get economic benefit from that. So I think it's super important to figure out we're trying to come up with the right approach given both what
[0:41:41] I mean, if any.
[0:42:12] I think that in the near term, there are so many opportunities. There's a huge number of opportunities. I would say the near term is truly the golden age of AI. It's like you've got an uncharted territory of an incredible number of positive applications. And what I'll say, just go for it. Just do it. So, Seth, you worked with Israeli founders, startups, right, in the past? Yeah.
[0:42:46] The two things that I've observed that are particularly special about Israel, number one is the talent density. We're very focused on talent density, not just like absolute amount of talent. This is a smallish country that punches way above its weight and has lots of talented people that you can get clustered into areas. And then the second is just the sort of like relentlessness, drive, ambition level of Israeli entrepreneurs. Again, we had like
[0:43:15] Incredible success in all of the YC efforts we made. But those two things together, I think, ought to lead to incredible prosperity, both in terms of AI research and AI applications. Thank you. Yeah, we can take over there. Like in my ChachiBT history, it has basically replaced Wikipedia for me. I used to spend a lot of time learning stuff on Wikipedia.
[0:44:06] And I think the thing you would find if you looked through mine is how effective it can be at learning new knowledge in the Wikipedia style. So I don't do like the deep teach me everything about physics that I know some people do, but the like I heard about this thing and I want to learn about it as quickly as possible, you would find a surprisingly effective tool for that. Hi. Hi. Can you hear me? No, we need people with microphones. Hi.
[0:44:39] I'm here. Hi, Sam. Hi. We can't see. So my name is Arbel. I'm a volunteer here in Tel Aviv. I'm 18. And I wanted to ask you, what do you look for in a new employee in OpenAI? Thank you. Nice meeting you. Drive, taste, collaboration, intelligence, like the ability to contribute.
[0:45:10] Be a good team, like contribute to the output of the entire organization, which could mean come up with the next breakthrough or it could mean like really be a great engineer to help us build these systems or it could just mean like be really helpful to other people and contribute that way. Definitely a belief in super intelligence and a feeling the weight of the importance of getting this right in terms of getting the benefits but managing the risks. I don't know what else.
[0:45:40] Oh, sounds like a pretty comprehensive answer. All right. You should definitely apply. Hello. Hello. Up here again in the balcony to the right. We're just guessing. It's dark up there. My name is Alon. I'm the CEO of a company called Benny Goren. It's the leading mathematics. It's for teaching mathematics.
[0:46:18] We are the leading provider of mathematics textbooks and content for the last 40 years here in Israel. First of all, I wanted to ask, how do you plan to improve ChatGPT skills, Hebrew skills? And the second is related to something you were talking about. What is your vision of education and AI?
[0:46:48] And practically in schools for our kids. How we can improve and motivate them. And thank you very much. You want to take it? Yeah. Yeah. Education in math. Well, obviously, textbooks will be upgraded. You read a textbook, but the textbook doesn't answer your questions. So it will be possible with the aid of the kind of AIs that,
[0:47:27] We and others are building for you to have a conversation about the subject matter. So that makes for a much more efficient learning experience. It will apply to math, it will apply to everything else. Eventually we will be moving, or rather eventually, we already are moving to a world where every student has a dedicated private tutor. Not there yet. It's not quite good enough, but it will be. I'm actually not sure if it's a black swan event in employment.
[0:48:11] It may be a sort of more gradual and very predicted change that's going to happen where we now have these systems that are good at doing tasks but not hold jobs and they get better and do some jobs. But it's all sort of difficult to predict. The role of the government, I think, is going to be to provide some sort of new cushion. What the format of that will be, I think different governments will try different experiments and we'll see what works better and governments will copy the ones that work better.
[0:48:39] In almost every conversation I've had these last few weeks on the road, every government has been very thoughtful about this. They have different ideas about how best to solve it. But it is maybe the top of mind issue, at least the top three issue, that I think any world leader is thinking about with this. So I think people are on it. We have time for one or two questions. Hi, my name is Ben up here. Yeah, please. My name is Ben Shkalim.
[0:49:09] I'm studying computer science and I'm graduating soon. So what I wanted to know, why should I learn to still have a job in 10 to 15 years from now? Well, I think learning computer science is good no matter what. I almost never write code anymore, but it was one of the best things that I ever did in terms of learning how to think, learning how to address problems. So I think it's valuable for its own sake, even if you're the job of a computer programmer.
[0:49:39] Looks very different than it looks today. The main skill I think to learn is how to learn, how to learn fast, how to learn new things, how to give it a sense for what's coming, how to be adaptable, how to be resilient, taste, how to figure out what other people are gonna want, how you can be useful. So again, there's no question in our minds that jobs are gonna change, the nature of work is gonna change, but also I cannot imagine a world where people don't
[0:50:10] Do something with their time to create value for other people. And all of the benefits that come with that. And maybe in the future, the thing you and I care about is who has the cooler galaxy. But there's still going to be something. Hi. Up here. Last one. Hello. My name is Hensagiv. I'm co-founder of Deep Pathology. My question to both of you is, you guys are making history. How do you want history to remember you? I mean...
[0:50:47] In the best possible way. Hello. My name is Amir. I am 17 years old entrepreneur. And I want to know what is your tips for first time startups in these ages. I didn't quite hear the opinion for what? First time startup? I am 17 years old entrepreneur. This is truly, I think, the best time to start a startup that I have ever seen. This is, I think,
[0:51:30] Yeah, I think it's actually better than the iPhone. Maybe the only comparable thing is when the internet launched. If you are a first-time entrepreneur right now, you are the luckiest entrepreneur that has existed in a long time. You have an incredible, new, fast-moving technological wave, and those are when startups win. Those are when the incumbents screw up and get displaced. The ground is shaking right now. That's what you want as a startup.
[0:51:59] And things are possible that most people can't quite imagine. And the opportunity to build value with a new approach doesn't come along very often. And this is the big one. So like every entrepreneur is a summer child right now. And it's a super cool time. Hello? Doing what? Oh, well, I wouldn't, I totally disagree with that. Like that's, people are going to build.
[0:52:30] All of the stuff on top of us. Yes, if you're trying to come for ChatGPT, you've had a failure of imagination and probably you will not make something better than the pure version of ChatGPT. But the size of the universe of possibilities right now, just the companies that we've met with on this trip, is unbelievable. There's so much to go after that if you're somehow worried about us being the incumbent, I think you're really not thinking about the problem correctly.
[0:52:57] Sam, here. This is going to be the very, very last question. Here, here, here upstairs, here on the balcony. Go ahead. Sam, I'm Rinat. I'm a data scientist here in the university. And my question is about your future plans and the WorldCoin project, the Orb system. I'm, you know, I was, I'm an investor and I kind of like helped put the company together, but I'm not involved.
[0:53:25] Day-to-day at all. I think it's very exciting. I think experimenting with new ways to differentiate between, like to prove humanity in a privacy-preserving way and to think about things like global UBI and ways to fairly democratize access is a super great area to explore. But I'm not close enough to the company to meaningfully comment on the plans. Okay, so Sam Altman and Ilya Sutskevich, CEO and Chief Scientist of OpenAI.
[0:53:54] Thank you very much for being with us.