Societal Misalignments Worse Than Killer Robots? What Sam Altman Said At World Govts Summit

Yeah, sometimes you read the news and really feel how audacious people can be. But I think if we're just going to come into the here and now and the topics that really matter for humanity, let's talk about large language models and AI systems and ChatGPT more specifically.

Look, we have a long way to go and a lot to prove. But I think if we can fulfill our mission, if we can even get close to it, the benefits to humanity of making intelligence broadly available, inexpensive, and sort of as a tool to let humanity build the future, I think is quite remarkable. I think abundant intelligence and closely related to that, abundant energy can unlock a future that is sort of difficult for me to even imagine how good it could be. And I think right now we don't realize how limited we are by the limits on intelligence and how expensive it is and how difficult it is. But if you imagine a world where everyone gets a great personal tutor, great personalized medical advice, we can use these tools to discover all sorts of new science, cure diseases, help the environment, discover new physics, who knows what else. I think that's pretty remarkable. And also just speaking personally, I think this is like the most exciting quest frontier I can imagine being on.

So the honest answer of course is we really don't know. You know, this is new science, we're discovering new things all the time. The rate of discovery is incredible, the rate of change is incredible, but it's sort of hard to know exactly how far we have to go. What I will say though is we hear all the time from scientists who say that our tools make them much more productive, and they don't have an easy way to quantify that, but they say it's substantial. We also don't know how much that difference, you know, if you could make every scientist on Earth twice as productive, what that would mean for the rate of scientific discovery. Because this is all so new, this is like, you know, a little bit more than a year old, but we'll find out.

So first of all, you touched on something that you sort of glossed over, but I want to spend a second on it because I think it's informative to what's happening, which was about education and cheating in schools. And when ChatGPT first came out, you know, the whole world had a moment of adjustment. But the first thing that happened is, at least in the United States in my experience, school districts were falling all over themselves to ban ChatGPT the fastest and declare, you know, this existential threat to education. And other people got concerned later, but it really started with education. Education was also the place it reversed the quickest. Teachers and school districts embraced it and said, you know, hey, this is part of the future, this is something that we all want, this is going to help our students learn better. And I really believe this will be the most, this is, and certainly will be one of the most transformative technologies for education we've had in recent times. And I think that leads into your question about governments, because there are all of these things that people were afraid of or maybe are still afraid of about large language models and AI. But governments who embrace it and say, let's find ways to help this deliver services for our citizens better, let's think about ways we can reform how the government does its job, how people fill out forms or do whatever they need to do, how the government helps provide education, healthcare as we discussed earlier, that seems to be working really well. And the governments that are saying, let's lean in, let's experiment with this, let's embrace it, let's, you know, make it available to our government, to our citizens, that I think is the best thing to do right now. And just say like, hey, we're all figuring this out together, we're all writing the rule book together, but let's lean in and try it.

Well, I think for that exact reason, as you said, we have not yet seen as much world-changing application as we'd like. Maybe we've seen some. There are a lot of people who use these services and get value out of them, but not as much as we'd like. And I think the reason is the current technology that we have is like, I mean, it's like that very first cell phone with the black and white screen that can only display those like numbers and, you know, it just didn't do much. But there was enough in there, you're like, hey, I can make a call, that's cool. And at the time that seems great. And then it took us, I don't know how long from that, but many decades from that to the iPhones we have today. And the thing we have today is incredible. And it took a massive amount of scaling in all these different ways to get there. But what we have now is like unimaginable at the time of those like first primitive cell phones. And I think that's what we have to push forward. We're at this barely useful cell phone, but people still like making phone calls, it turns out. And if you can make a better way for them to do it so they can go walk around the world while they do it, sure, that's great. But that's not what we want to deliver. We want to deliver the iPhone 16 or 15 or whatever the current one is.

That's like, you know, it took the world a while to do that last time around, so give us some time. But I will say I think in a few more years it'll be much better than it is now, and in a decade it should be pretty remarkable.

I was sort of laughing a little bit because this is going to sound like an annoying answer, but I think it is the important part. It's going to be smarter. There are all of these other things, you know, we can talk about, it'll be better at these kind of tasks, it'll be multimodal, it'll be faster, who knows what. But the thing that I think really matters is it's going to be smarter. And this is a bigger deal than it sounds, right? Because what makes these models so magical is that they're general. And so if it's a little bit better, if it's a little bit smarter, that means it's a little bit better at everything. And the thing that I think is most exciting is it's not like this model is going to get a little better at this task and not really better at these, or, you know, it's not that. It's because we're going to make the model smarter, it's going to be better at everything across the board.

I've heard of it, I haven't watched it.

Does that mean you have to take my job for a day?

You have fun with that. It's not as easy as it looks.

That'd be great. Anytime you want to switch, I will greatly look forward to that. What I would do... that was a curveball by the way. No, no, it's a really thoughtful question. I'm trying to give a thoughtful answer. I think what I would do is try to, and I know you've done some work in this direction and I really appreciate it, but I would try to find a way to create more of a regulatory sandbox where people could experiment with this technology and be able to figure out sort of like dream, imagine, whatever you want to call it, what the world could look like. And then I would try to see what makes sense and what doesn't and write the regulation around that. I think it's very hard. I think we have to try, and we're going to anyway, but I think it's very hard to get all of the regulatory ideas right in a vacuum. And if there was a sort of contained way that I could find a way to like give people the future and let them experiment with it, and then see what made sense, what went really wrong, what really right, and write the regulation around that, that seems like an interesting experiment.

Can I change my answer? I thought one... you can change your answer 100 times, go on. I think that's a good thing to do, but since I only have one day, a better thing to do... okay. One thing that I have been thinking about, so the world is going to try all of these different regulatory approaches. There will be your sandbox, I think it's awesome that you have that, other people do other things. But we are going to, and I think that's actually really good, but we are going to need, I believe, at some point some sort of global system. The example that I've given in the past is the IAEA, the International Atomic Energy Agency, for what happens with the most powerful of these systems, because they will have truly global impact. And what sort of auditing, what sort of safety measures do we want in place before you can deploy like a super intelligence or, you know, however you want to call an AGI. And I think for a bunch of reasons the UAE would be so well set up to be a leader in the discussions around that. I would like host a one-day conference with leaders from around the world to brainstorm about that.

Thank you. I'd love to come.

Yes, you hosted that.

Frankly speaking, I think we're still in the stage, and this is not necessarily bad, but we're still in the stage of a lot of discussion. So there's, you know, everybody in the world is having a conference, everybody's got an idea, a policy paper, and that's okay. I think we're still at a time where debate is needed and healthy. But at some point in the next few years, I think we have to move towards like an action plan with real buy-in around the world.

I honestly don't think that is for OpenAI to say. I mean, we have a lot of ideas, we've tried to contribute them to the discussion. I think a lot of other people have a lot of ideas. This is a big thing, like this is going to touch all of us. And I am very high conviction that we can manage our way through this, but it's going to take a great deal of collaboration. And it's going to take our leaders of the world coming together. And, you know, that's, I think, not for us to set.

We're giving this a lot of thought. We're trying to, you know, we want to have like an offering that makes sense for countries that want to have offer AI services. But in the meantime, I think what people are doing right now, which is just sort of use these APIs or run open source models, I think that'll make sense.

GPT-2 we did. I don't think three.

I think you should expect us to open source more things over time. But exactly what and when and how, we're trying to figure out. There are like great open source models, open source language models out in the world now. And, you know, I don't think what the world needs is like another similar model. So we'd like to do something that is helpful and new, and we're trying to figure out what that might be.

The keep up at night is easy. It's all of the sci-fi stuff. You know, I think sci-fi writers are a very smart bunch. And in the decades of sci-fi about AI, there have been unbelievably creative ways to imagine how this can go wrong. And I think most of them are like comical, but there's some things in there that are easy to imagine where things really go wrong. And I'm not that interested in like the killer robots walking down the street direction of things going wrong. I'm much more interested in the like very subtle societal misalignments where we just have these systems out in society and through no particular ill intention, things just go horribly wrong. But the thing that wakes me up in the morning with energy every day is what I actually believe is things are just going to go tremendously right. We got to work hard to mitigate all of the downside cases. They are, I think, very significant and real potentials to confront. But the reason that we think so hard about how to deploy this technology safely is the upside is remarkable. I think we can easily imagine a world in the not super distant future where everybody's got a better life than people have today. I think we can raise the standard of living so incredibly much if everybody has access to abundant amounts of really high quality intelligence and they can use those tools to create whatever they want to do. That's like pretty amazing. I think about it, but people are like, oh, that doesn't make any sense. I'm going to say it anyway. If you think about everybody on Earth getting the resources of a company of like hundreds of thousands of really competent people, and what that would do, you know, if you have like an AI programmer, AI lawyer, AI marketer, AI strategist, and not just one of those but many of each, and you get to sort of like decide how to use those to create whatever you want to create, we're all going to get a lot of great stuff. The creative power of humanity with tools like that should be remarkable. So that's I think what gets us all up every morning.

The first thing I would say is you are unbelievably lucky. You are coming of age at probably the best time in human history. You understand this technology. Young people are always the early adopters of technology, almost always, but certainly in this case. And you will be able to use these tools to do things that the people in the generation before you couldn't even imagine. You will have your entire career flooded with opportunity and the ability to do amazing new things. You'll be able to start companies that are phenomenal, more impactful and successful than people the generation before you could. You will live in this incredibly expansionary, opportunistic, just flooded with time of like massive, massive opportunity. And you can kind of go do whatever you want. I think the rule, like the ground under us all is shifting, the rules are changing, but the amount of value that will be created and the ability for an individual to express their creative vision and will, it's a great time.

Thank you soon. Thank you.