We started seven years ago because we felt like something really was happening in AI and we wanted to help steer in a positive direction. It’s honestly just really amazing to see how far this whole field come since then. And it’s really gratifying to hear from people like Raymond who using the technology we are building, and others, for so many wonderful things. We hear people who are excited, we hear from people who are concerned, we hear from people feel both those emotions at once. And honestly, that’s how we feel. all, it feels like we’re entering an historic period right now where as a world are going to define a technology that will be important for our society going forward. And I believe that we can this for good.
So today, I want to show you the current state that technology and some of the underlying design principles that we hold dear.
So first thing I’m going to show you is what it’s like build a tool for an AI rather than building for a human. So we have a new DALL-E model, which generates images, and we are it as an app for ChatGPT to use on your behalf. you can do things like ask, you know, suggest a nice post-TED meal and a picture of it.
(Laughter)
Now you get all the, sort of, ideation and creative back-and-forth and taking care the details for you that you get out of ChatGPT. And here we go, it’s not just the idea the meal, but a very, very detailed spread. So let’s see what we’re going get. But ChatGPT doesn’t just generate images in this case — sorry, it doesn’t generate text, it also an image. And that is something that really expands the power of what it can do on your in terms of carrying out your intent. And I’ll out, this is all a live demo. This is all by the AI as we speak. So I actually don’t know what we’re going to see. This looks wonderful.
(Applause)
I’m getting hungry looking at it.
Now we’ve extended ChatGPT with other tools too, example, memory. You can say “save this for later.” And the interesting thing about tools is they’re very inspectable. So you get this little up here that says “use the DALL-E app.” And the way, this is coming to you, all ChatGPT users, over upcoming months. And you look under the hood and see that what it actually did write a prompt just like a human could. And so you sort have this ability to inspect how the machine is using these tools, which allows us to provide feedback them.
Now it’s saved for later, and let me show you it’s like to use that information and to integrate with other applications too. You say, “Now make a shopping list for the tasty thing I suggesting earlier.” And make it a little tricky for the AI. “And tweet out for all the TED viewers out there.”
(Laughter)
So you do make this wonderful, wonderful meal, I definitely to know how it tastes.
But you can see that ChatGPT is selecting all these different without me having to tell it explicitly which ones use in any situation. And this, I think, shows a new way of thinking about the interface. Like, we are so used to thinking of, well, we have apps, we click between them, we copy/paste between them, and it’s a great experience within an app as long as you kind of know the menus and know the options. Yes, I would like you to. Yes, please. Always to be polite.
(Laughter)
And by having this unified interface on top of tools, the AI is able to sort of take away all those from you. So you don’t have to be the who spells out every single sort of little piece of what’s to happen.
And as I said, this is a demo, so sometimes the unexpected will happen to us. But let’s take a look at Instacart shopping list while we’re at it. And you can see we sent a of ingredients to Instacart. Here’s everything you need. And the thing that’s really interesting is that traditional UI is still very valuable, right? If you at this, you still can click through it and sort modify the actual quantities. And that’s something that I think that they’re not going away, traditional UIs. It’s just we have new, augmented way to build them. And now we a tweet that’s been drafted for our review, which is also a very important thing. can click “run,” and there we are, we’re the manager, we’re able inspect, we’re able to change the work of the if we want to. And so after this talk, you be able to access this yourself. And there we go. Cool. you, everyone.
(Applause)
So we’ll cut back to the slides. Now, the important thing how we build this, it’s not just about building these tools. It’s about teaching the AI how to them. Like, what do we even want it to do when we ask these high-level questions? And to do this, we use an old idea. If you go back Alan Turing’s 1950 paper on the Turing test, he says, you’ll never an answer to this. Instead, you can learn it. You could a machine, like a human child, and then teach it through feedback. Have a human who provides rewards and punishments as it tries things out does things that are either good or bad.
And this is exactly we train ChatGPT. It’s a two-step process. First, we produce what Turing would have a child machine through an unsupervised learning process. We just show it the whole world, whole internet and say, “Predict what comes next in you’ve never seen before.” And this process imbues it with sorts of wonderful skills. For example, if you’re shown a math problem, the only way to complete that math problem, to say what comes next, that green up there, is to actually solve the math problem.
But actually have to do a second step, too, which is to teach the what to do with those skills. And for this, we provide feedback. We have the try out multiple things, give us multiple suggestions, and then a human rates them, “This one’s better than that one.” And this reinforces not just specific thing that the AI said, but very importantly, the whole process that the used to produce that answer. And this allows it to generalize. It allows it teach, to sort of infer your intent and apply in scenarios that it hasn’t seen before, that it hasn’t feedback.
Now, sometimes the things we have to teach the AI are not you’d expect. For example, when we first showed GPT-4 Khan Academy, they said, “Wow, this is so great, We’re going be able to teach students wonderful things. Only one problem, it doesn’t double-check students’ math. there’s some bad math in there, it will happily pretend that one plus one equals three run with it.” So we had to collect some feedback data. Sal Khan himself was very kind and 20 hours of his own time to provide feedback to the machine alongside our team. over the course of a couple of months we able to teach the AI that, “Hey, you really should push back on humans in specific kind of scenario.” And we’ve actually made lots lots of improvements to the models this way. And when you push that thumbs down in ChatGPT, actually is kind of like sending up a bat to our team to say, “Here’s an area of weakness where should gather feedback.” And so when you do that, that’s one that we really listen to our users and make sure we’re something that’s more useful for everyone.
Now, providing high-quality feedback a hard thing. If you think about asking a kid to clean their room, all you’re doing is inspecting the floor, you don’t know if you’re just teaching to stuff all the toys in the closet. This a nice DALL-E-generated image, by the way. And the sort of reasoning applies to AI. As we move to tasks, we will have to scale our ability to provide high-quality feedback. for this, the AI itself is happy to help. It’s happy to us provide even better feedback and to scale our ability to supervise machine as time goes on. And let me show you what I mean.
For example, you ask GPT-4 a question like this, of how much time between these two foundational blogs on unsupervised learning and learning from human feedback. the model says two months passed. But is it true? Like, these models are 100-percent reliable, although they’re getting better every time we provide feedback. But we can actually use the AI to fact-check. it can actually check its own work. You can say, fact-check this for me.
Now, this case, I’ve actually given the AI a new tool. one is a browsing tool where the model can issue search and click into web pages. And it actually writes out its whole of thought as it does it. It says, I’m just going to search for this it actually does the search. It then it finds the publication date and the search results. It is issuing another search query. It’s going to click into blog post. And all of this you could do, but it’s a very task. It’s not a thing that humans really want do. It’s much more fun to be in the driver’s seat, to be in this manager’s position you can, if you want, triple-check the work. And out come so you can actually go and very easily verify any piece this whole chain of reasoning. And it actually turns out two months was wrong. months and one week, that was correct.
(Applause)
And we’ll cut back to the side. And so thing that’s interesting to me about this whole process is that it’s this many-step collaboration between human and an AI. Because a human, using this fact-checking tool is doing it order to produce data for another AI to become useful to a human. And I think this really shows the shape of that we should expect to be much more common in the future, where have humans and machines kind of very carefully and delicately designed in how they fit into problem and how we want to solve that problem. We sure that the humans are providing the management, the oversight, the feedback, and the are operating in a way that’s inspectable and trustworthy. And together we’re to actually create even more trustworthy machines. And I think that time, if we get this process right, we will able to solve impossible problems.
And to give you a sense of just how I’m talking, I think we’re going to be able to rethink almost aspect of how we interact with computers. For example, think about spreadsheets. They’ve been in some form since, we’ll say, 40 years ago VisiCalc. I don’t think they’ve really changed that much in that time. And here is specific spreadsheet of all the AI papers on the for the past 30 years. There’s about 167,000 of them. you can see there the data right here. But me show you the ChatGPT take on how to analyze a set like this.
So we can give ChatGPT access yet another tool, this one a Python interpreter, so it’s able run code, just like a data scientist would. And so you can just literally a file and ask questions about it. And very helpfully, you know, it knows the name of the file it’s like, “Oh, this is CSV,” comma-separated value file, “I’ll parse it for you.” The only here is the name of the file, the column like you saw and then the actual data. And that it’s able to infer what these columns actually mean. Like, semantic information wasn’t in there. It has to sort of, put together its world of knowing that, “Oh yeah, arXiv is a site that people submit and therefore that’s what these things are and that these integer values and so therefore it’s a number of authors in the paper,” all of that, that’s work for a human to do, the AI is happy to help with it.
Now I don’t know what I want to ask. So fortunately, you ask the machine, “Can you make some exploratory graphs?” And again, this is a super high-level instruction with lots of intent it. But I don’t even know what I want. And the kind of has to infer what I might be interested in. And so it comes up with good ideas, I think. So a histogram of the number of authors per paper, series of papers per year, word cloud of the paper titles. of that, I think, will be pretty interesting to see. And great thing is, it can actually do it. Here we go, a nice curve. You see that three is kind of the most common. It’s going then make this nice plot of the papers per year. Something crazy is happening in 2023, though. Looks like were on an exponential and it dropped off the cliff. could be going on there? By the way, all is Python code, you can inspect. And then we’ll see cloud. So you can see all these wonderful things that in these titles.
But I’m pretty unhappy about this 2023 thing. It makes this year look bad. Of course, the problem is that the year is over. So I’m going to push back on the machine. [Waitttt that’s not fair!!! 2023 isn’t over. percentage of papers in 2022 were even posted by April 13?] So April 13 was the cut-off date believe. Can you use that to make a fair projection? we’ll see, this is the kind of ambitious one.
(Laughter)
So you know, again, feel like there was more I wanted out of the machine here. I wanted it to notice this thing, maybe it’s a little bit of an overreach for to have sort of, inferred magically that this is I wanted. But I inject my intent, I provide additional piece of, you know, guidance. And under the hood, the AI is just writing again, so if you want to inspect what it’s doing, it’s very possible. And now, does the correct projection.
(Applause)
If you noticed, it even updates the title. I didn’t ask that, but it know what I want.
Now we’ll cut back the slide again. This slide shows a parable of how I think we … A vision how we may end up using this technology in the future. A brought his very sick dog to the vet, and veterinarian made a bad call to say, “Let’s just wait and see.” And the dog would not be today had he listened. In the meanwhile, he provided the blood test, like, full medical records, to GPT-4, which said, “I am not vet, you need to talk to a professional, here are some hypotheses.” He brought that information to a vet who used it to save the dog’s life. Now, systems, they’re not perfect. You cannot overly rely on them. But story, I think, shows that a human with a medical professional and ChatGPT as a brainstorming partner was able to achieve an outcome that not have happened otherwise. I think this is something should all reflect on, think about as we consider how integrate these systems into our world.
And one thing believe really deeply, is that getting AI right is going require participation from everyone. And that’s for deciding how we want it to in, that’s for setting the rules of the road, for what an will and won’t do. And if there’s one thing to take away from talk, it’s that this technology just looks different. Just from anything people had anticipated. And so we all have to become literate. that’s, honestly, one of the reasons we released ChatGPT.
Together, I believe that can achieve the OpenAI mission of ensuring that artificial general benefits all of humanity.
Thank you.
(Applause)
(Applause ends)
Chris Anderson: Greg. Wow. I mean … I suspect that within every out here there’s a feeling of reeling. Like, I that a very large number of people viewing this, look at that and you think, “Oh my goodness, pretty much every single thing about way I work, I need to rethink.” Like, there’s just new there. Am I right? Who thinks that they’re having to rethink the way that we things? Yeah, I mean, it’s amazing, but it’s also really scary. let’s talk, Greg, let’s talk.
I mean, I guess my first actually is just how the hell have you done this?
(Laughter)
OpenAI a few hundred employees. Google has thousands of employees working on artificial intelligence. Why is you who’s come up with this technology that shocked world?
Greg Brockman: I mean, the truth is, we’re all building shoulders of giants, right, there’s no question. If you look at the compute progress, the progress, the data progress, all of those are really industry-wide. But I think within OpenAI, we made a lot very deliberate choices from the early days. And the first one just to confront reality as it lays. And that we just thought hard about like: What is it going to take to make progress here? We tried a of things that didn’t work, so you only see things that did. And I think that the most important thing has been to get teams of who are very different from each other to work together harmoniously.
CA: we have the water, by the way, just brought here? I we’re going to need it, it’s a dry-mouth topic. But isn’t there something also just about the that you saw something in these language models that meant that if you continue to invest in and grow them, that something at some point might emerge?
GB: Yes. And I that, I mean, honestly, I think the story there is illustrative, right? I think that high level, deep learning, we always knew that was what we wanted to be, was deep learning lab, and exactly how to do it? I think in the early days, we didn’t know. We tried a lot things, and one person was working on training a model to predict the next in Amazon reviews, and he got a result where — this is a syntactic process, you expect, know, the model will predict where the commas go, where nouns and verbs are. But he actually got a state-of-the-art sentiment analysis classifier of it. This model could tell you if a was positive or negative. I mean, today we are like, come on, anyone can do that. But this was the first time that you saw emergence, this sort of semantics that emerged from this underlying syntactic process. there we knew, you’ve got to scale this thing, you’ve got to see where it goes.
CA: I think this helps explain the riddle that baffles everyone looking at this, because these things are described prediction machines. And yet, what we’re seeing out of feels … it just feels impossible that that could from a prediction machine. Just the stuff you showed us just now. the key idea of emergence is that when you get more a thing, suddenly different things emerge. It happens all the time, ant colonies, ants run around, when you bring enough of them together, you get ant colonies that show completely emergent, different behavior. Or a where a few houses together, it’s just houses together. But as you grow number of houses, things emerge, like suburbs and cultural and traffic jams. Give me one moment for you when you just something pop that just blew your mind that you just did see coming.
GB: Yeah, well, so you can try this in ChatGPT, you add 40-digit numbers —
CA: 40-digit?
GB: 40-digit numbers, model will do it, which means it’s really learned an internal circuit how to do it. And the really interesting thing actually, if you have it add like a 40-digit number plus 35-digit number, it’ll often get it wrong. And so can see that it’s really learning the process, but hasn’t fully generalized, right? It’s like you can’t memorize the 40-digit addition table, that’s more atoms than there in the universe. So it had to have learned something general, but it hasn’t really fully yet learned that, Oh, I can sort of generalize this to adding arbitrary of arbitrary lengths.
CA: So what’s happened here is that you’ve allowed it scale up and look at an incredible number of pieces of text. And it learning things that you didn’t know that it was to be capable of learning.
GB Well, yeah, and it’s more nuanced, too. So one science we’re starting to really get good at is predicting some of emergent capabilities. And to do that actually, one of the things I think is very in this field is sort of engineering quality. Like, we had to rebuild our stack. When you think about building a rocket, every tolerance has to be incredibly tiny. Same is in machine learning. You have to get every single piece of the stack engineered properly, and then you start doing these predictions. There are all these incredibly smooth curves. They tell you something deeply fundamental about intelligence. If you look our GPT-4 blog post, you can see all of these curves in there. And now we’re to be able to predict. So we were able to predict, example, the performance on coding problems. We basically look at some that are 10,000 times or 1,000 times smaller. And there’s something about this that is actually smooth scaling, even it’s still early days.
CA: So here is, one the big fears then, that arises from this. If it’s to what’s happening here, that as you scale up, things emerge that you maybe predict in some level of confidence, but it’s capable of surprising you. Why isn’t there just huge risk of something truly terrible emerging?
GB: Well, I all of these are questions of degree and scale and timing. And I one thing people miss, too, is sort of the integration with the is also this incredibly emergent, sort of, very powerful thing too. And so that’s one of the reasons we think it’s so important to deploy incrementally. And so I think that what we kind of see now, if you look at this talk, a lot what I focus on is providing really high-quality feedback. Today, the tasks we do, you can inspect them, right? It’s very easy to look at math problem and be like, no, no, no, machine, seven the correct answer. But even summarizing a book, like, that’s a hard to supervise. Like, how do you know if this book summary is any good? have to read the whole book. No one wants to that.
(Laughter) And so I think that the important will be that we take this step by step. And that we say, OK, as move on to book summaries, we have to supervise task properly. We have to build up a track record these machines that they’re able to actually carry out intent. And I think we’re going to have to produce even better, more efficient, more ways of scaling this, sort of like making the machine be aligned with you.
CA: So we’re to hear later in this session, there are critics say that, you know, there’s no real understanding inside, system is going to always — we’re never going to know it’s not generating errors, that it doesn’t have common sense and so forth. Is it belief, Greg, that it is true at any one moment, but that the of the scale and the human feedback that you talked about basically going to take it on that journey of actually to things like truth and wisdom and so forth, with a high of confidence. Can you be sure of that?
GB: Yeah, well, think that the OpenAI, I mean, the short answer yes, I believe that is where we’re headed. And I think that the OpenAI approach here always been just like, let reality hit you in face, right? It’s like this field is the field of broken promises, of all these experts saying X going to happen, Y is how it works. People have been neural nets aren’t going to work for 70 years. They haven’t been yet. They might be right maybe 70 years plus one something like that is what you need. But I think our approach has always been, you’ve got to push the limits of this technology to really see it action, because that tells you then, oh, here’s how we can on to a new paradigm. And we just haven’t exhausted the fruit here.
CA: I mean, it’s quite controversial stance you’ve taken, that the right way to this is to put it out there in public and then all this, you know, instead of just your team giving feedback, the world now giving feedback. But … If, you know, bad things going to emerge, it is out there. So, you know, original story that I heard on OpenAI when you were founded as a nonprofit, you were there as the great sort of check the big companies doing their unknown, possibly evil thing with AI. And you were to build models that sort of, you know, somehow them accountable and was capable of slowing the field down, need be. Or at least that’s kind of what heard. And yet, what’s happened, arguably, is the opposite. That your release GPT, especially ChatGPT, sent such shockwaves through the tech world that now Google Meta and so forth are all scrambling to catch up. And some of their criticisms have been, are forcing us to put this out here without guardrails or we die. You know, how do you, like, make case that what you have done is responsible here and reckless.
GB: Yeah, we think about these questions all the time. Like, seriously all the time. And don’t think we’re always going to get it right. But one thing I think has incredibly important, from the very beginning, when we were thinking how to build artificial general intelligence, actually have it benefit all of humanity, like, how are supposed to do that, right? And that default plan being, well, you build in secret, you get this super thing, and then you figure out the safety of it and then push “go,” and you hope you got it right. I don’t know to execute that plan. Maybe someone else does. But me, that was always terrifying, it didn’t feel right. And I think that this alternative approach is the only path that I see, which is that you do let reality hit you the face. And I think you do give people time to give input. You have, before these machines are perfect, before they are super powerful, that you actually have ability to see them in action. And we’ve seen from GPT-3, right? GPT-3, we really were afraid that the one thing people were going to do with it generate misinformation, try to tip elections. Instead, the number one was generating Viagra spam.
(Laughter)
CA: So Viagra spam bad, but there are things that are much worse. Here’s a thought experiment for you. Suppose you’re sitting a room, there’s a box on the table. You that in that box is something that, there’s a very strong it’s something absolutely glorious that’s going to give beautiful gifts to family and to everyone. But there’s actually also a one thing in the small print there that says: “Pandora.” And there’s a chance that this could unleash unimaginable evils on the world. Do you that box?
GB: Well, so, absolutely not. I think you don’t it that way. And honestly, like, I’ll tell you story that I haven’t actually told before, which is shortly after we started OpenAI, I remember I was in Puerto Rico for AI conference. I’m sitting in the hotel room just looking out over this wonderful water, all these having a good time. And you think about it for a moment, if you could choose for basically Pandora’s box to be five years away or 500 years away, which would you pick, right? On the hand you’re like, well, maybe for you personally, it’s better have it be five years away. But if it gets to 500 years away and people get more time to get it right, which do you pick? you know, I just really felt it in the moment. I was like, of course do the 500 years. My brother was in the military at the time and like, puts his life on the line in a much more real way any of us typing things in computers and developing technology at the time. And so, yeah, I’m really sold on the you’ve to approach this right. But I don’t think that’s quite playing the field as it lies. Like, if you look at the whole history computing, I really mean it when I say that this is industry-wide or even just almost like a human-development- of-technology-wide shift. And the more that you sort of, don’t together the pieces that are there, right, we’re still making faster computers, we’re improving the algorithms, all of these things, they are happening. And you don’t put them together, you get an overhang, which means that if someone does, or moment that someone does manage to connect to the circuit, then you suddenly have very powerful thing, no one’s had any time to adjust, who knows what of safety precautions you get. And so I think that one thing I take away is like, even think about development of other sort of technologies, think about weapons, people talk about being like a zero to one, sort of, change what humans could do. But I actually think that if you look at capability, it’s quite smooth over time. And so the history, I think, of every we’ve developed has been, you’ve got to do it incrementally and you’ve got figure out how to manage it for each moment you’re increasing it.
CA: So what I’m hearing is you … the model you want us to have that we have birthed this extraordinary child that may have superpowers that humanity to a whole new place. It is our responsibility to provide the guardrails for this child to collectively teach to be wise and not to tear us all down. Is basically the model?
GB: I think it’s true. And I think it’s important to say this may shift, right? We’ve got to take each step we encounter it. And I think it’s incredibly important today we all do get literate in this technology, figure how to provide the feedback, decide what we want it. And my hope is that that will continue be the best path, but it’s so good we’re honestly having this debate we wouldn’t otherwise if it weren’t out there.
CA: Greg Brockman, thank you so much for coming to and blowing our minds.
(Applause)