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