Lex Fridman PodcastIlya Sutskever: Deep Learning | Lex Fridman Podcast #94
EVERY SPOKEN WORD
150 min read · 30,014 words- 0:00 – 2:23
Introduction
- LFLex Fridman
The following is a conversation with Ilya Sutskever, co-founder and chief scientist of OpenAI, one of the most cited computer scientists in history with over 165,000 citations, and to me, one of the most brilliant and insightful minds ever in the field of deep learning. There are very few people in this world who I would rather talk to and brainstorm with about deep learning, intelligence, and life in general than Ilya, on and off the mic. This was an honor and a pleasure. This conversation was recorded before the outbreak of the pandemic. For everyone feeling the medical, psychological, and financial burden of this crisis, I'm sending love your way. Stay strong. We're in this together. We'll beat this thing. This is the Artificial Intelligence podcast. If you enjoy it, subscribe on YouTube, review it with five stars on Apple Podcast, support it on Patreon, or simply connect with me on Twitter @lexfridman, spelled F-R-I-D-M-A-N. As usual, I'll do a few minutes of ads now, and never any ads in the middle that can break the flow of the conversation. I hope that works for you and doesn't hurt the listening experience. This show is presented by Cash App, the number one finance app in the App Store. When you get it, use code LEXPODCAST. Cash App lets you send money to friends, buy Bitcoin, invest in the stock market with as little as $1. Since Cash App allows you to buy Bitcoin, let me mention that cryptocurrency in the context of the history of money is fascinating. I recommend A Cent of Money as a great book on this history. Both the book and audiobook are great. Debits and credits on ledgers started around 30,000 years ago, the US dollar, created over 200 years ago, and Bitcoin, the first decentralized cryptocurrency, released just over 10 years ago. So given that history, cryptocurrency is still very much in its early days of development, but it's still aiming to, and just might, redefine the nature of money. So again, if you get Cash App from the App Store or Google Play, and use the code LEXPODCAST, you get $10, and Cash App will also donate $10 to FIRST, an organization that is helping advance robotics and STEM education for young people around the world. And now, here's my conversation with Ilya Sutskever.
- 2:23 – 8:33
AlexNet paper and the ImageNet moment
- LFLex Fridman
You were one of the three authors, with Alex Krizhevsky, Geoff Hinton, of the famed AlexNet paper that is arguably, uh, the paper that marked the big catalytic moment that launched the deep learning revolution. At that time, take us back to that time, what was your intuition about neural networks, about the representational power of neural networks? And maybe you could mention, how did that evolve over the next few years up to today, over the 10 years?
- ISIlya Sutskever
Yeah, I can answer that question. At some point, in about 2010 or 2011, I connected two facts in my mind. Basically, the realization was this. At some point, we realized that we can train very large... I shouldn't say very, you know, they were tiny by today's standards, but large and deep neural networks end-to-end with backpropagation. At some point, different people obtained this result. I obtained this result. The first, th- the first moment in which I realized that deep neural networks are powerful was when James Martens invented the Hessian-free optimizer in 2010, and he trained a 10-layer neural network end-to-end without pre-training from scratch. And when that happened, I thought, "This is it." Because if you can train a big neural network, a big neural network can represent very complicated function. Because if you have a neural network with 10 layers, it's as though you allow the human brain to run for some number of milliseconds. Neuron firings are slow, and so in maybe 100 milliseconds, uh, your neurons only fire 10 times. So, it's also kind of like 10 layers. And in 100 milliseconds, you can perfectly recognize any object. So, I thought... So, I already had the idea then that we need to train a very big neural network on lots of supervised data, and then it must succeed because we can find the best neural network. And then there's also theory that if you have more data than parameters, you won't overfit. Today, we know that actually this theory is very incomplete and you won't overfit even if you have less data than parameters. But definitely, if you have more data than parameters, you won't overfit.
- LFLex Fridman
So, the fact that neural networks were heavily over-parameterized wasn't discouraging to you? So, y- y- you were thinking about the theory that the number of parameters, the fact there's a huge number of parameters is okay? It's gonna be okay?
- ISIlya Sutskever
I mean, there was some evidence before that it was okay-ish, but the theory was most, the theory was that if you had a big data set and a big neural net, it was going to work. The over-parameterization just didn't really, um, figure much as, as a problem. I thought, well, with images, you're just gonna add some data augmentation and it's gonna be okay.
- LFLex Fridman
So, where was any doubt coming from?
- ISIlya Sutskever
The, the main doubt was, can we train a big ne- will we have enough compute to train a big enough neural net?
- LFLex Fridman
With backpropagation.
- ISIlya Sutskever
Backpropagation, I thought was, would work. The thing which wasn't clear would, was whether there would be enough compute to get a very convincing result. And then at some point, Alex Krizhevsky wrote these insanely fast CUDA kernels for training convolutional neural nets, and that was, bam, let's do this, let's get ImageNet, and it's gonna be the greatest thing.
- LFLex Fridman
Was your intuition, most of your intuition from empirical results by you and by others? So, like, just actually demonstrating that a piece of program can train a 10-layer neural network? Or was there some pen and paper or marker and whiteboard thinking intuition? Like, 'cause you just connected a 10-layer large neural network to the brain. So, you just mentioned the brain. So, in your intuition about neural networks, does the human brain come into play as a intuition builder?
- ISIlya Sutskever
Definitely. I mean, you, you know, you gotta be precise with these analogies between neural, artificial neural networks and the brain. But...There is no question that the brain is a huge source of intuition and inspiration for deep learning researchers since all the way from Rosenblatt in the '60s. Like, if you look at- the wh- the whole idea of a neural network is directly inspired by the brain. You had people like McCollum and Pitts who were saying, "Hey, you got these n- n- neurons in the brain. And hey, we recently learned about the computer and automata. Can we use some ideas from the computer and automata to design some kind of computational object that's going to be simple, computational, and kind of like the brain?" And they invented the neuron. So, they were inspired by it back then. Then you had the convolutional neural network from Fukushima, and then later Yann LeCun, who said, "Hey, if you limit the receptive fields of a neural network, it's gonna be especially suitable for images," as it turned out to be true. So, there was- there was a very small number of examples where analogies to the brain were successful. And I thought, "Well, probably an artificial neuron is not that different from the brain if you squint hard enough, so let's just assume it is and r- and roll with it."
- LFLex Fridman
So, no- we're now at a time where deep learning is very successful. So, let us squint less, and say, let's, uh, open our eyes and say, w- what to you is an interesting difference between the human brain ... Now, I know you're probably not an expert, uh, neither a neuroscientist and neurobiologist, but loosely speaking, what's the difference between the human brain and artificial neural networks that's interesting to you for the next decade or two?
- ISIlya Sutskever
That's a good question to ask. What is in- what is an interesting difference between the neurons betwe- between the brain and our artificial neural networks? So, I feel like today, artificial neural networks ... So, we all agree that there are certain dimensions in which the human brain vastly outperforms our models. But I also think that there are some ways in which our artificial neural networks have a number of very important advantages over the brain. Look- looking at- at the advantages versus disadvantages is a good way to figure out what is the important difference. So, the brain uses spikes which may or may not be important.
- LFLex Fridman
Yeah, that's a really interesting question. Do- do you think it's important or not?
- ISIlya Sutskever
My-
- LFLex Fridman
That- that's one big architectural difference between-
- ISIlya Sutskever
Yeah.
- LFLex Fridman
... artificial neural networks and ...
- ISIlya Sutskever
It's hard to tell, but my prior is not very high, and I can te- I can say why. You know, there are people who are interested in spiking neural networks, and basically what they figured out is that they need to simulate the non-spiking neural networks in spikes.
- LFLex Fridman
Hmm.
- ISIlya Sutskever
And that's how they're gonna make- make them work. If you don't simulate the non-spiking neural networks in spikes, it's not going to work, because the question is, why should it work? And that connects to questions around back propagation and questions around
- 8:33 – 13:39
Cost functions
- ISIlya Sutskever
deep learning. You've got this giant neural network. Why should it work at all? Why should the learning rule work at all? It's not a self-evident question, especially if you, let's say, if you were just starting in the field and you read the very early papers. You can say, "Hey, people are saying, 'Let's build neural networks.'" That's a great idea, because the brain is a neural network, so it would be useful to build neural networks.
- LFLex Fridman
Yep.
- ISIlya Sutskever
Now, let's figure out how to train them. It should be possible to train them, probably, but how? And so the big idea is the cost function. That's the big idea. The cost function is a way of measuring the performance of the system according to some measure.
- LFLex Fridman
By the way, that is a big ... Well, l- actually, let me think. Is that- is that, uh, one, a difficult idea to arrive at? And how big of an idea is that, that there's a single cost function ... Let me ... Sorry, let me take a pause. Is supervised learning a difficult concept to come to?
- ISIlya Sutskever
I don't know. All concepts are very easy in retrospect.
- LFLex Fridman
Yeah, that's why it seems trivial now, but I ... 'Cause- 'cause the reason I ask that, and we'll talk about it, is- is there other things? Is there things that don't necessarily have a cost function, or maybe have many cost functions, or maybe have dynamic cost functions, or maybe a totally different kind of architectures?
- ISIlya Sutskever
Yeah.
- LFLex Fridman
'Cause we have to think like that in order to arrive at something new, right?
- ISIlya Sutskever
So, the only- so the good examples of things which don't have clear cost functions are GANs.
- LFLex Fridman
Right.
- ISIlya Sutskever
In a GAN, you have a game. So, instead of thinking of a cost function, where you wanna optimize- where you know that you have an algorithm gradient descent which will optimize the cost function, and then you can reason about the behavior of your system in terms of what it optimizes. With a GAN you say, "I have a game, and I'll reason abo- about the behavior of the system in terms of the equilibria of the game." But it's all about coming up with these mathematical objects that help us reason about the behavior of our system.
- LFLex Fridman
Right, that's really interesting. Yeah, so GAN is the only one. It's kind of a com- the cost function is emergent from the comparison.
- ISIlya Sutskever
It's- I don't- I don't know if it has a cost function. I don't know if it's meaningful to talk about the cost function of a GAN. It's kind of like the cost function of biological evolution or the cost function of the economy. It's- you can talk about regions to which it could be- go towards, but I don't think ... I don't think the cost function analogy is the most useful for GAN.
- LFLex Fridman
So, if evolution doesn't ... That's really interesting. So, if evolution doesn't really have a cost function, like a cost function based on its ... Something akin to our mathematical conception of a cost function, then do you think cost functions in deep learning are holding us back? Yeah, I ... So, y- you just kinda mentioned that cost function is a- is a nice first profound idea. Do you think that's a good idea? Do you think it's an idea we'll go past? So, self-play starts to touch on that a little bit, uh, in reinforcement learning, uh, systems.
- ISIlya Sutskever
That's right. Self-play, and also ideas around exploration where you're trying to take action that sur- that surprise a predictor. I'm a big fan of cost functions. I think cost functions are great and they serve us really well, and I think that whenever we can do things we c- with cost functions, we should. And, you know, maybe there is a chance that we will come up with some yet another profound way of looking at things that will involve cost functions in a less central way. But I don't know. I think cost functions are, I mean, I would not bet against cost functions.
- LFLex Fridman
Is there other things about the brain that pop into your mind that might be different and interesting for us to consider in designing artificial neural networks? So we talked about spiking a little bit.
- ISIlya Sutskever
I mean, one- one thing which may potentially be useful, I think people, neuroscientists have figured out something about the learning rule of the brain, or s- I'm talking about spike-timing-dependent plasticity, and it would be nice if some people were to study that in simulation.
- LFLex Fridman
Wait, sorry, spike-time-independent plasticity?
- ISIlya Sutskever
Yeah, that's right.
- LFLex Fridman
What's that?
- ISIlya Sutskever
STDP. It's- it's a particular learning rule that uses spike timing to figure out how to- to determine how to update the synapses. So it's kind of like, if a synapse fires into the neuron before the neuron fires, then it strengthen the synapse, and if the synapse fires into the neuron shortly after the neuron fired, then it weakens the synapse. Something along this line. I am 90% sure it's right, so if I said something wrong here, don't-
- LFLex Fridman
(laughs)
- ISIlya Sutskever
... don't get too angry.
- LFLex Fridman
But you sounded brilliant while saying it. But the timing, d- that's one thing that's missing. The- the temporal dynamics is not captured. I think that's like a fundamental property of the brain, is the timing of the s- of the signals.
- ISIlya Sutskever
Well, we have recurrent neural networks.
- LFLex Fridman
But, you- you think of that as the s- I mean, that's a very crude simplified, uh, what's that called? The- the- there's a clock, I guess, to, uh, recurrent neural networks.
- ISIlya Sutskever
Mm-hmm.
- LFLex Fridman
It's this- this seems like the brain is the general, the continuous version of that, the- the generalization, where all possible timings are possible, and then within those timings is contained some information.
- 13:39 – 16:19
Recurrent neural networks
- LFLex Fridman
You think recurrent neural networks, the recurrence in recurrent neural networks can capture the same kind of phenomena as the timing that seems to be important for the brain, i- i- in the- in the firing of neurons in the brain?
- ISIlya Sutskever
I- I mean, I think, I think recurrent neural netwo- r- recurrent neural networks are amazing, and they can do, I think they can do anything we- we'd want them to- we'd want a system to do. Right now, recurrent neural networks have been superseded by transformers, but maybe one day they'll make a comeback. Maybe they'll be back. We'll see.
- LFLex Fridman
(laughs) Let me, uh, on a small tangent say, do you think they'll be back? So, so much of, uh, the breakthroughs recently that we'll talk about on, uh, natural language processing and language modeling has been with transformers that don't emphasize recurrence. Do you think recurrence will make a comeback?
- ISIlya Sutskever
Well, some kind of recurrence I think very likely. Recurrent neural networks for proc- as they're typically thought of for processing sequences, I think it's also possible.
- LFLex Fridman
W- what is, to you, a recurrent neural network? In generally speaking, I guess, what is a recurrent neural network?
- ISIlya Sutskever
You have a neural network which maintains a high dimensional hidden state. And then when an observation arrives, it updates its high dimensional hidden state through its connections in some way.
- LFLex Fridman
So do you think, you know, that's what like expert systems did, right? Symbolic AI, uh, the knowledge-based, growing a knowledge base is- is maintaining a hidden state, which is its knowledge base and is growing it by sequential processing. Do you think of it more generally in that way? Or is it simply, is it the more constrained form th- of- of a hidden state with certain kind of gating units that we think of as today with LSTMs and that?
- ISIlya Sutskever
I mean, the hidden state is technically what you described there. The hidden state, it goes inside the LSTM or the RNN or something like this. But then what should be contained, you know, if you want to make the expert system, um, analogy, I'm not... I mean, you could say that the knowledge is stored in the- in the connections, and then the short-term processing is- is done in the- in the hidden state.
- LFLex Fridman
Yes. Could you say that?
- ISIlya Sutskever
Yes.
- LFLex Fridman
So, uh, sort of, do you think there's a future of building large-scale knowledge bases within the neural networks?
- ISIlya Sutskever
Definitely.
- LFLex Fridman
(laughs) So, we're gonna pause on that confidence 'cause I wanna explore that, but let me zoom back out and ask, um, back to the history of ImageNet,
- 16:19 – 19:57
Key ideas that led to success of deep learning
- LFLex Fridman
neural networks have been around for many decades, as you mentioned. What do you think were the key ideas that led to their success, that ImageNet moment and beyond the- the success in the past 10 years?
- ISIlya Sutskever
Okay, so the question is, to make sure I didn't miss anything, the key ideas that led to the success of deep learning over the past 10 years.
- LFLex Fridman
Exactly, even though the fundamental thing behind deep learning has been around for much longer.
- ISIlya Sutskever
So, the key idea about deep learning, or rather the s- the key fact about deep learning, before deep learning started to be successful, is that it was underestimated. People who worked in machine learning simply didn't think that n- neural networks could do much. People didn't believe that large neural networks could be trained. People thought that, well, there was lots of, there was a lot of debate going on in machine learning about what are the right methods and so on. And people were arguing because there were no, there- there were- there were no, there was no way to get hard facts. And by that I mean there were no benchmarks which were truly hard that if you s- do really well on them, then you can say, "Look, here is my system." That's when you switch from the, that's when this field becomes a little bit more of an engineering field. So in terms of deep learning, to answer the question directly, the ideas were all there. The thing that was missing was a lot of supervised data and a lot of compute. Once you have a lot of supervised data and a lot of compute, then there is a third thing which is needed as well, and that is conviction. Conviction that if you take the right stuff which already exists and apply and mix with a lot of data and a lot of compute, that it will in fact work.And so that was the missing piece. It was, you had the... You needed the, the data, you needed the compute which showed up in terms of GPUs, and you needed the conviction to realize that you need to mix them together.
- LFLex Fridman
So, that's really interesting. So, uh, I- I guess the presence of compute and the presence of supervised data allowed the empirical evidence to do the convincing of the majority of the computer science community. So, I guess there's a key moment with, uh, uh, Jitendra Malik and, uh, Alex, uh, Alyosha Efros who were very skeptical, right? And then there's a Geoffrey Hinton that was the opposite of skeptical, and there was a convincing moment, and I think ImageNet served as that moment.
- ISIlya Sutskever
That's right.
- LFLex Fridman
And that represented this kind of... or the big pillars of computer vision community kind of (laughs) th- the wizards got together, and then all of a sudden, there was this shift. And it's not enough for the ideas to all be there and the compute to be there. It's for it to convince the cynicism that existed that... That's interesting that people just didn't believe for a couple of decades.
- ISIlya Sutskever
Yeah. Well, but it's more than that. It's kind of... Wh- when- when- when put this way, it sounds like, "Well, you know, those silly people who didn't believe. What were they- what were they missing?" But in reality, things were confusing because neural networks really did not work on anything.
- LFLex Fridman
Right.
- ISIlya Sutskever
And they were not the best method on pretty much anything as well. And it was pretty rational to say, "Yeah, this stuff doesn't have any traction." And that's why you need to have these very hard tasks which are... which produce undeniable evidence, and that's how we make progress. And that's why the field is making progress today, because we have these hard benchmarks which represent true progress. And so... And this is why we are able to avoid endless debate.
- 19:57 – 29:35
What's harder to solve: language or vision?
- ISIlya Sutskever
- LFLex Fridman
So, incredibly, you've contributed some of the biggest recent ideas in AI in, in computer vision, language, natural language processing, reinforcement learning, sort of everything in between. Maybe not GANs. (laughs) Is there... (laughs) There may not be a topic you haven't touched, and of course the, the fundamental science of deep learning. What is the difference to you between vision, language, and as in reinforcement learning, action, as learning problems, and what are the commonalities? Do you see them as all interconnected? Are they fundamentally different domains that require different approaches?
- ISIlya Sutskever
Okay. That's a good question. Machine learning is a field with a lot of unity, a huge amount of unity. In fact-
- LFLex Fridman
What do you mean by unity? Like, overlap of ideas?
- ISIlya Sutskever
Overlap of ideas, overlap of principles. In fact, there is only one or two or three principles which are very, very simple, and then they, then they apply in almost the same way, in almost the same way to the different modalities, to the different problems. And that's why today, when someone writes a paper on improving optimization of deep learning and vision, it improves the different NLP applications, and it improves the different reinforcement learning applications. Reinforcement learn- so I would say that computer vision and NLP are very similar to each other. Today, they differ in that they have slightly different architectures. We use transformers in NLP, and we use convolutional neural networks in vision. But it's also possible that one day this will change, and everything will be unified with a single architecture. Because if you go back a few years ago in natural language processing, there were a hu- a huge number of architectures, for every different tiny problem had its own architecture. Today, there's just one transformer for all those different tasks. And if you go back in time even more, you had even more and more fragmentation. And every little problem in AI had its own little sub-specialization and sub... you know, a little set of collection of skills, people who would know how to engineer the features. Now, it's all been subsumed by deep learning. We have this unification. And so I expect vision to become unified with natural language as well, or rather, I shouldn't say "expect," I think it's possible. I don't want to be too sure because I think on the conventional neural net is very computationally efficient. RL is different. RL does require slightly different techniques, because you really do need to take action. You really do need to do something about exploration. Your variance is much higher. But I think there is a lot of unity even there. And I would expect, for example, that at some point, there will be some broader unification between RL and supervised learning, where somehow the RL will be making decisions to make the supervised learning go better. And it will be, I imagine, one big black box, and you just throw ev- you know, you shovel, shovel things into it, and it just figures out what to do with whatever you shovel at it.
- LFLex Fridman
I mean, reinforcement learning has some aspects of language and vision combined almost. There's elements of a long-term memory that you should be utilizing, and there's elements of a really rich sensory space. So it seems like the... it's- it's like the union of the two or something like that.
- ISIlya Sutskever
I'd- I'd say something slightly differently. I'd say that reinforcement learning is neither, but it naturally interfaces and integrates with the two of them.
- LFLex Fridman
Do you think action is fundamentally different? So, yeah, what is interesting about... what is unique about p- policy of learning to act?
- ISIlya Sutskever
Well, so one example, for instance, is that when you learn to act, you are fundamentally in a non-stationary world, because as your actions change, the things you see s- start changing. You, you experience the world in a different way, and this is not the case for the more traditional static problem where you have a dist- some distribution, and you just apply a model to that distribution.
- LFLex Fridman
You think it's a fundamentally different problem, or is it just a more difficult general- it's a generalization of the problem of understanding?
- ISIlya Sutskever
I mean, it's- it's- it's a question of definitions almost. There is a huge- I mean, although there's a huge amount of commonality, for sure. You take gradients, you try... You take gradients. We try to approximate gradients in both cases. In some ca- in the case of reinforcement learning, you have some tools to reduce the variance of the gradients. You do that.... there's lots of commonality, use the same neural net in both cases, you compute the gradient, you apply atom in both cases. So, I mean, there's lots in common for sure, but there are some small differences which are not completely insignificant. It's really just a matter of your po- of your point of view what frame of reference you, what, h- how much do you z- wanna zoom in or out as you look at these problems.
- LFLex Fridman
Which problem do you think is harder? So people like Noam Chomsky believe that language is fundamental to everything, so it underlies everything. Do you think language understanding is harder than visual scene understanding or vice versa?
- ISIlya Sutskever
I think that asking if a problem is hard is slightly wrong. I think the question is a little bit wrong and I wanna explain why. So what does it mean for a problem to be hard?
- LFLex Fridman
Okay. The non-interesting dumb answer to that is there's this s- s- there's a benchmark, and there's a human level of performance on that benchmark, and how is the effort required to reach the human level-
- ISIlya Sutskever
Okay.
- LFLex Fridman
... benchmark.
- ISIlya Sutskever
So from the perspective of how much until we get to human level...
- LFLex Fridman
On a very good benchmark.
- ISIlya Sutskever
Yeah, like some... I- I- I under- I understand what you mean by that. So what I was going, going to say that a lot of it depends on, you know, once you solve a problem, it stops being hard, and that's-
- LFLex Fridman
Right. (laughs)
- ISIlya Sutskever
... all, all, that's always true, and so-
- LFLex Fridman
That's a...
- ISIlya Sutskever
But if something is hard or not depends on what our tools can do today. So you know, you say today, true human level language understanding and visual perception are hard in the sense that there is no way of solving the problem completely in the next three months.
- LFLex Fridman
Right.
- ISIlya Sutskever
So I agree with that statement. Beyond that, I'm just, I'd be, my s- my guess would be as good as yours. I don't know.
- LFLex Fridman
Oh, okay, so you don't have a fundamental intuition about how hard language understanding is?
- ISIlya Sutskever
Well, I think... I- I- I, now I changed my mind. I'd say language is probably going to be harder... I mean, it depends on how you define it. Like, if you mean absolute top-notch 100% language understanding, I'll go with language.
- LFLex Fridman
And so, uh, th-
- ISIlya Sutskever
But then if I show you a piece of paper with letters on it, is that... You, you see what I mean? And so, you have a vision system, you say it's the best f- human level vision system. I show you... I open a book and I show you letters. Will it understand how these letters form into word and sentences and meaning? Is this part of the vision problem? Where does vision end and language begin?
- LFLex Fridman
Yeah, so Chomsky would say it starts at language, so vision is just a little example of the kind of, uh, structuring we... And, you know, fundamental hierarchy of ideas that's already represented in our brain somehow, that's represented through language. But where does vision stop and language begin? That's a really interesting, uh, question. (sighs) I- so one possibility is that it's impossible to achieve really deep understanding in either images or language without basically using the same kind of system. So you're going to get the other for free. (laughs)
- 29:35 – 36:04
We're massively underestimating deep learning
- ISIlya Sutskever
- LFLex Fridman
So, forgive the romanticized question, but looking back, to you, what is the most beautiful or surprising idea in deep learning or AI in general you've come across?
- ISIlya Sutskever
So I think the most beautiful thing about deep learning is that it actually works.
- LFLex Fridman
(laughs)
- ISIlya Sutskever
And I mean it, because you got these ideas, you got the little neural network, you got the back propagation algorithm, and then you got some theories as to, you know, this is kinda like the brain, so maybe if you make it large, if you make the neural network large and you train it on a lot of data, then it will do the same function that the brain does. And it turns out to be true. That's crazy.
- LFLex Fridman
(laughs)
- ISIlya Sutskever
And now we just train these neural networks and you make them larger and they keep getting better.... and I find it unbelievable. I find it unbelievable that this whole AI stuff with neural networks works.
- LFLex Fridman
Have you built up an intuition of why? Are there little bits and pieces of intuitions, of insights of why this whole thing works?
- ISIlya Sutskever
I mean, some, definitely. While we know that optimization... We- we- we now have good re- you know, we've take, we- we've had lots of empirical, you know, huge amounts of empirical reasons to believe that optimization should work on all, most problems we care about.
- LFLex Fridman
Do you have insights of what... So, you just said empirical evidence. Is most of your (sighs) s- sort of empirical evidence kind of convinces you? It's like evolution is empirical. It shows you that, look, this evolutionary process seems to be a good way to design, uh, organisms that survive in their environment. But it doesn't really get you to the insights of how the whole thing works.
- ISIlya Sutskever
Well, I think it's- it's... A good analogy is physics. You know how you say, "Hey, let's do some physics calculation and come up with some new physics theory and make some prediction," but then you gotta run the experiment. You know, you gotta run the experiment. It's important. So, it's a bit the- the same here, except that maybe som- sometimes the- the experiment came before the theory. But it still is the case. You know, you have some data, and you come up with some prediction. You say, "Yeah, let's make a big neural network. Let's train it. And it's going to work much better than anything before it. And it will, in fact, continue to get better as we make it larger." And it turns out to be true. That's- that's amazing when a theory is validated like this, you know? It's not a mathematical theory. It's more of a biological theory almost. So, I think there are not-terrible analogies between deep learning and biology. I would say it's like the geometric mean of biology and physics.
- LFLex Fridman
(laughs)
- ISIlya Sutskever
That's deep learning.
- LFLex Fridman
The geometric mean of biology and physics (laughs) . I think I'm gonna need a few hours to wrap my head around that. Uh, 'cause just to find the geometric... Just to find, uh, the set of what biology represents.
- ISIlya Sutskever
Well, biology, p- in- in biology things are really complicated and theories-
- LFLex Fridman
Yeah.
- ISIlya Sutskever
... are really, really... It's really hard to have good predictive theory. And if... In physics, the theories are too good. F- in th- in theory, in physics, people make these super precise theories which make these amazing predictions. And in machine learning, we're kind of in between.
- LFLex Fridman
Kind of in between. But it'd be nice if machine learning somehow helped us discover the unification of the two as opposed to sort of the in between. But you're right. That's... You're- you're kind of trying to juggle both. So, do you think there's still beautiful and mysterious properties in neural networks that are yet to be discovered?
- ISIlya Sutskever
Definitely. I think that we are still massively underestimating deep learning.
- LFLex Fridman
What do you think it will look like? Like, what...
- ISIlya Sutskever
If I knew, I would have done it, you know? (laughs) So, uh... But if you look at all the progress from the past 10 years, I would say most of it, I would say there have been a few cases where some... Where things that felt like really new ideas showed up. But by and large, it was every year we thought, "Okay. Deep learning goes this far. Nope, it actually goes further."
- LFLex Fridman
(laughs)
- ISIlya Sutskever
And then the next year, "Okay. Now, it... You... Now, this is- this is peak deep learning. We are really done. Nope, goes further." It just keeps going further each year. So, that means that we keep underestimating, we keep not understanding it. Has surprising properties all the time.
- LFLex Fridman
Do you think it's getting harder and harder?
- ISIlya Sutskever
To make progress?
- LFLex Fridman
Yeah, to make progress.
- ISIlya Sutskever
It depends on what you mean. I think the field will continue to make very robust progress for quite a while. I think for individual researchers, especially people who are doing re- um, research, it can be harder, because there is a very large number of researchers right now. I think that if you have a lot of compute, then you can make a lot of very interesting discoveries. But then you have to deal with the challenge of managing a huge computes- a huge clus- a huge compute cluster to run your experiments. So, it's a little bit harder.
- LFLex Fridman
So, I'm asking all these questions that nobody knows the answer to. But you're one of the smartest people I know, so I'm gonna keep asking. The... So, let's imagine all the breakthroughs that happen in the next 30 years in deep learning. Do you think most of those breakthroughs can be done by one person with one computer? Sort of in the space of breakthroughs, do you think compute will be, uh, compute and large efforts will be necessary?
- ISIlya Sutskever
I mean, I can't be sure. When you say one computer, you mean how- how large?
- LFLex Fridman
(laughs) Ah, you're- uh, you're clever. I mean, one com- one GPU.
- 36:04 – 41:20
Deep double descent
- ISIlya Sutskever
- LFLex Fridman
Can you maybe, sort of o- on the topic of the- the science of deep learning, talk about one of the recent papers that you released?
- ISIlya Sutskever
Sure.
- LFLex Fridman
The deep double descent?
- ISIlya Sutskever
Mm-hmm.
- LFLex Fridman
Where bigger models and more data hurt. Uh, I think it's a really interesting paper. Can you- can you describe the main idea and...
- ISIlya Sutskever
Yeah, definitely. So, what happened is that some...Over, over the years, some small number of researchers noticed that it is kind of weird that when you make the neural network larger, it works better. And it seems to go in contradiction with statistical ideas. And then some people made an analysis showing that actually you got this double descent bump. And what we've done was to show that double descent occurs for all, for pretty much all practical deep learning systems. And that it will be also-
- LFLex Fridman
So can you step back? Um, what's the X axis and the Y axis of a double descent plot?
- ISIlya Sutskever
Okay, great. So you can, you can look, you can do things like, you can take a neural network and you can start increasing its size slowly while keeping your data set fixed. So if you increase the size of the neural network slowly, and if you don't do early stopping, that's a pretty important detail, then when the neural network is really small, you make it larger, you get a very rapid increase in performance. Then you continue to make it larger, and at some point performance will get worse. And it gets, and- and it gets the worst exactly at the point at which it achieves zero training error, precisely zero training loss. And then as you make it larger, it starts to get better again. And it's kind of counterintuitive because you'd expect deep learning phenomena to be monotonic. And it's hard to be sure what it means, but it also occurs in, in the case of linear classifiers. And the intuition basically boils down to the following. When you, when you have a lot, when you have a large data set and a small model, then small, tiny, random... So, basically what is overfitting? Overfitting is when your model is somehow very sensitive to the small random unimportant stuff in your data set.
- LFLex Fridman
In the training data.
- ISIlya Sutskever
In the training data set, precisely. So if you have a small model and you have a big data set, and there may be some random thing, you know, some training cases are randomly in the data set and others may not be there. But the small model, but the small model is kind of insensitive to this randomness because it's the sa- You, there is pretty much no uncertainty about the model when the data set is large.
- LFLex Fridman
So, okay, so at the very basic level, to me, it is the most surprising thing that neural networks don't overfit every time very quickly, uh, (laughs) before ever being able to learn anything. The huge number of parameters.
- ISIlya Sutskever
So here is... So there is one way... Okay, so maybe the... So let me try to give the explanation, maybe that will be... That will work. So you got a huge neural network. Let's suppose you've got a... You are, you have a huge neural network, you have a huge number of parameters. And now let's pretend everything is linear, which is not, but let's just pretend. Then there is this big subspace where your neural network achieves zero error.
- LFLex Fridman
Mm-hmm.
- ISIlya Sutskever
And SGT is going to find approximately the point-
- LFLex Fridman
Stochastic gradient, yeah.
- ISIlya Sutskever
That's right, yeah, approximately the point with the smallest norm in that subspace.
- LFLex Fridman
Okay.
- ISIlya Sutskever
And that can also be proven to be insensitive to the small randomness in the data when the dimensionality is high. But when the dimensionality of the data is equal to the dimensionality of the model, then there is a one-to-one correspondence between all the data sets and the models. So small changes in the data set actually lead to large changes in the model, and that's why performance gets worse. So this is the best explanation more or less.
- LFLex Fridman
So then it would be good for the model to have more parameters, sort of, uh, to be bigger than the data?
- ISIlya Sutskever
That's right. But only if you don't early stop. If you introduce early stop in your regularization, you can make the double descent bump almost completely disappear.
- LFLex Fridman
What is early stop?
- ISIlya Sutskever
Early stopping is when you train your model and you monitor your test, your validation performance. And then if at some point validation performance starts to get worse, you say, "Okay, let's stop training. We are good. We are good. We are good enough."
- LFLex Fridman
So the, the magic happens after, after that moment. So you don't want to do the early stopping?
- ISIlya Sutskever
Well, if you don't do the early stopping, you get these very, you get the very pronounced double descent.
- LFLex Fridman
Do you have any intuition why this happens?
- ISIlya Sutskever
Double descent? Oh, sorry, early stopping?
- LFLex Fridman
No, the double descent. So the-
- ISIlya Sutskever
Oh, yeah, so I try... Let's see. The intuition is basically, is this, that when the data set has as many degrees of freedom as the model, then there is a one-to-one correspondence between them. And so small changes to the data set lead to noticeable changes in the model. So your model is very sensitive to all the randomness. It is unable to discard it. Whereas it turns out that when you have a lot more data than parameters or a lot more parameters than data, the resulting solution will be insensitive to small changes in the data set.
- LFLex Fridman
Oh, so it's able to, that's nicely put, discard the small changes, the- the randomesses?
- 41:20 – 42:42
Backpropagation
- ISIlya Sutskever
- LFLex Fridman
Geoff Hinton suggested we need to throw backpropagation. We already kind of talked about this a little bit, but he suggested we need to throw away backpropagation and start over. I mean, of course some of that is a little bit, um, wit and humor. But what do you think? What could be an alternative method of training neural networks?
- ISIlya Sutskever
Well, the thing that he said precisely is that, to the extent that you can't find backpropagation in the brain, it's worth seeing if we can learn something from how the brain learns. But backpropagation is very useful and we should keep using it.
- LFLex Fridman
Oh, you're saying that once we discover the mechanism of learning in the brain or any aspects of that mechanism, we should also try to implement that in neural networks?
- ISIlya Sutskever
If it turns out that we can't find backpropagation in the brain.
- LFLex Fridman
If we can't find backpropagation in the brain. Well, so I guess your answer to that is backpropagation is pretty damn useful. So why are we complaining?
- ISIlya Sutskever
I mean, I- I personally am a big fan of backpropagation. I think it's a great algorithm because it solves an extremely fundamental problem, which is finding a neural circuit subject to some constraints.... and I don't see that problem going away. So, that's why I- I really, I think it's pretty unlikely that we'll have anything which is going to be dramatically different. It could happen, but I wouldn't bet on it right now.
- 42:42 – 50:35
Can neural networks be made to reason?
- ISIlya Sutskever
- LFLex Fridman
So, let me ask a sort of big picture question. Do you think can- do you think neural networks can be made to reason?
- ISIlya Sutskever
Why not?
- LFLex Fridman
(laughs)
- ISIlya Sutskever
Well, if you look for example at AlphaGo or AlphaZero, the neural network of AlphaZero plays Go, which aga- which we all agree is a game that requires reasoning, better than 99.9% of all humans. Just the neural network, without the search, just the neural network itself. Doesn't that give us an existence proof that neural networks can reason?
- LFLex Fridman
To push back and disagree a little bit, we all agree that Go is reasoning. Uh, I think I- I agree. I don't think that's a trivial... So, obviously, reasoning, like intelligence, is, um, is a loose, gray area term, a little bit. Maybe you disagree with that. But yes, I think it has some of the same elements of reasoning. Reasoning is almost, like, akin to search, right? There's a sequential element of stepwise consideration of possibilities, and sort of building on top of those possibilities in a sequential manner until you arrive at some insight. So, yeah, I guess playing Go is kind of like that. And when you have a single neural network doing that without search, it's kind of like that. So, there's an existent proof in a particular constrained environment that a- a process akin to what many people call reasoning exists, but more general kind of reasoning. So, off the board.
- ISIlya Sutskever
There is one other existence proof.
- LFLex Fridman
Oh, boy.
- ISIlya Sutskever
(laughs)
- LFLex Fridman
Which one? Us humans?
- ISIlya Sutskever
Yes.
- LFLex Fridman
Okay. All right. So, do you think the architecture that will allow neural networks to reason will look similar to the neural network architectures we have today?
- ISIlya Sutskever
I think it will. I think... Well, I don't want to make too overly definitive statements. I think it's definitely possible that the neural networks that will produce the reasoning breakthroughs of the future will be very similar to the architectures that exist today. Maybe a little bit more recurrent, maybe a little bit deeper. But, like, these- these neural nets are so insanely powerful. Why wouldn't they be able to learn to reason? Humans can reason. So, why can't neural networks?
- LFLex Fridman
So, do you think the kind of stuff we've seen neural networks do is a kind of just weak reasoning? So, it's not a fundamentally different process? Again, this is stuff we don't- nobody knows the answer to.
- ISIlya Sutskever
So, when it comes to our neural networks, I would- the thing which I would say is that neural networks are capable of reasoning, but if you train a neural network on a task which doesn't require reasoning, it's not going to reason. This is a well-known effect where the neural network will solve exactly the- it will solve the problem that you pose in front of it in the easiest way possible.
- LFLex Fridman
Right. That takes us to the- to o- o- one of the brilliant sort of ways you've described neural networks, which is, uh, you've referred to neural networks as the search for small circuits, and maybe general intelligence as the search for small programs, which I found as a metaphor very compelling. Can you elaborate on that difference?
- ISIlya Sutskever
Yeah. So, the thing which I said precisely was that if you can find the shortest program that outputs the data in your- at your disposal, then you will be able to use it to make the best prediction possible.
- LFLex Fridman
Mm-hmm.
- ISIlya Sutskever
And that's a theoretical statement which can be proven mathematically. Now, you can also prove mathematically that it is- that finding the shortest program regenerates some data is not com- is not a computable operation. No, uh, finite amount of compute can do this. So then, with- with neural networks, neural networks are the next best thing that actually works in practice. We are not able to find the best- the shortest program which generates our data, but we are able to find, you know, a small, but now- now that statement should be amended, even a large circuit which fits our data in some way.
- LFLex Fridman
Well, I think what you meant by the small circuit is the smallest needed circuit.
- ISIlya Sutskever
Well, I- th- the thing- the thing which I would change now, back- back then, I really have- I haven't fully internalized the over-parametr- the over-parameterized results. The re- the things we know about over-parameterized neural nets, now, I would phrase it as a large circuit that can- wh- whose weights contain a small amount of information, which I think is what's going on. If you imagine the training process of a neural network as you slowly transmit entropy from the dataset to the parameters, then somehow, the amount of information in the weights ends up being not very large, which would explain why they generalize so well.
- LFLex Fridman
So that's- the- the large circuit might be one that's helpful for the regulariz- for the generalization?
- ISIlya Sutskever
Yeah, something like this.
- LFLex Fridman
But do you see their- do you see it important to be able to try to learn something like programs?
- ISIlya Sutskever
I mean, if we can, definitely. I think it's kind of- the answer is kind of yes, if we can do it. We should do things that we can do it.
- LFLex Fridman
(laughs)
- ISIlya Sutskever
It's- it's- the reason we are pushing on deep learning, the fundamental reason, the cau- the- the- the- the root cause is that we are able to train them. So, in other words, training comes first. We've got our pillar, which is the training pillar, and now we are trying to contort our neural networks around the training pillar. We gotta stay trainable. This is an invo- this is an invariant we cannot violate. And so...
- LFLex Fridman
... being trainable means starting from scratch, knowing nothing, you can actually pretty quickly converge towards knowing a lot.
- ISIlya Sutskever
Or even slowly. But it means that given the resources at your disposal, you can train the neural net and get it to achieve useful performance.
- LFLex Fridman
Yeah, that's a pillar we can't move away from.
- 50:35 – 56:37
Long-term memory
- LFLex Fridman
What about long-term memory? Can neural networks have long-term memory or something like knowledge bases? So, being able to aggregate important information over long periods of time that would then serve as useful sort of representations of state that, uh, you can make decisions by, so have a long-term context based on which you're making the decision.
- ISIlya Sutskever
So, in some sense, the parameters already do that. The parameters are an aggregation of the da- of the neural- of the entirety of the neural net's experience, and so they count as the long- as long-form, long-term knowledge. And people have trained various neural nets to act as knowledge bases and, you know, investigated th- invest- people have investigated language models as knowledge bases, so... there is work, there is work there.
- LFLex Fridman
Yeah, but in some sense. Do you think in every sense? Do you think there's, um... it's- it's all just a- a matter of coming up with a better mechanism of forgetting the useless stuff and remembering the useful stuff? 'Cause right now, I mean, there's not been mechanisms that do remember really long-term information.
- ISIlya Sutskever
What do you mean by that precisely?
- LFLex Fridman
Precisely... I like, I like the word precisely. So... (sighs) I'm thinking of the kind of compression of information the knowledge bases represent, sort of creating a... Now, I apologize for my sort of human-centric thinking about what knowledge is, 'cause n- neural networks aren't in- interpretable necessarily with the kind of knowledge they have discovered. But a good example for me is knowledge bases, being able to build up over time something like the knowledge that Wikipedia represents. It's a really compressed, structured, f- uh, knowledge base. O- obviously not the actual Wikipedia or the language, but like a semantic web, the dream that a semantic web represented. So, it's a really nice compressed knowledge base, or something akin to that in a non-interpretable sense as, um, neural networks would have.
- ISIlya Sutskever
Well, the neural networks would be non-interpretable if you look at their weights, but their outputs should be very interpretable.
- LFLex Fridman
Okay, so yeah, how do you, how do you make very smart neural networks, like language models, interpretable?
- ISIlya Sutskever
Well, you ask them to generate some text and the text will generally be interpretable.
- LFLex Fridman
Do you find that the epitome of interpretability, like can you do better? Like c- can you a- 'cause you can't... Okay, I would like to know what does it know and what doesn't it know? I would like the neural network to come up with examples where it- it's completely dumb and examples where it's completely brilliant. And the only way I know how to do that now is to generate a lot of examples and use my human judgment. But it would be nice if a neural network had some awar- self-awareness about it. (laughs)
- ISIlya Sutskever
Yeah, 100%. I'm- I'm a big believer in self-awareness and I think that...
- LFLex Fridman
(laughs)
- ISIlya Sutskever
... I think, I think new- neural net self-awareness will allow for things like the capabilities like the ones you described, like for them to know what they know and what they don't know and for them to know where to invest to increase their skills most optimally. And to your question of interpretability, there are actually two answers to that question. One answer is, you know, we have the neural net, so we can analyze the neurons and we can try to understand what the different neurons and different layers mean and you can actually do that and OpenAI has done some work on that. But there is a different answer which is that, I would say this is the human-centric answer, where you say, you know, you look at a human being, you can't read... you know, h- h- how do you know what a human being is thinking? You ask them, you say, "Hey, what do you think about this? What do you think about that?" And you get some answers.
- LFLex Fridman
The answers you get are sticky in the sense you already have a mental model. You already have an... uh, yeah, a mental model of that human being. You already have an understanding of like a- a big conception of what it... of that human being, how they think, what they know, how they see the world, and then everything you ask, you're s- adding on to that.... and, and that stickiness seems to be... That's one of the really interesting qualities of the, the human being, is that information is sticky. You don't- you seem to remember the useful stuff, aggregate it well, and forget most of the information that's not useful. The- that process. But, uh, that's also pretty similar to the processing neural networks do. It's just that neural networks are much crappier at, at this time. It's not- it doesn't seem to be fundamentally that different. But just to stick on reasoning for a little longer. (laughs) You said, "Why not? Why can't I reason?" What- what's a good impressive feat, benchmark, to you, of reasoning, that you would be impressed by if neural networks were able to do? Is that something you already have in mind?
- ISIlya Sutskever
Well, I think writing, writing really good code. I think proving really hard theorems. Solving open-ended problems with out-of-the-box solutions.
- LFLex Fridman
And, uh, sort of theorem type mathematical problems.
- ISIlya Sutskever
Yeah, I think tho- tho- those ones are a very natural example as well. You know, if you can prove an unproven theorem, then it's hard to argue you don't reason. And so by the way, and this comes back to the point about the hard results, you know? If you got a hard, if you have... Be- machine learning, deep learning as a field, is very fortunate, because we have the ability to sometimes produce these unambiguous results, and when they happen, well, the debate changes, the conversation changes. It's a convers- you- we, we have the ability to produce conversation-changing results.
- LFLex Fridman
Conversation. And then, of course, just like you said, people kind of take that for granted and say, "That wasn't actually a hard problem."
- ISIlya Sutskever
Well, I mean, at some point, we'll probably run out of hard problems.
- LFLex Fridman
Yeah. That whole mortality thing is kinda, kind of a sticky problem that we haven't s- quite figured out. Maybe we'll solve that one.
- 56:37 – 1:00:35
Language models
- LFLex Fridman
I think one of the fascinating things in, in your entire body of work, but also the work at OpenAI recently, one of the conversation-changers has been in the world of language models. Can you briefly kinda try to describe the recent history of using neural networks in the domain of language and text?
- ISIlya Sutskever
Well, there's been lots of history. I think, I think the Elman network was, was a s- was a s- was a small, tiny recurrent neural network applied to language back in the '80s. So, the history is really, you know, fairly long, at least. And the thing that started, th- the thing that changed the trajectory of neural networks and language is the thing that changed the trajectory of all deep learning, and that's data and compute. So, suddenly, you move from small language models which learn a little bit. And with language models in particular, you can... There's a very clear explanation for why they need to be large to be good, because they're trying to predict the next word.
- LFLex Fridman
Mm-hmm.
- ISIlya Sutskever
So we don't, when you don't know anything, you'll notice very, very broad strokes, surface-level patterns, like sometimes there are characters, and there is space between those characters. You'll notice this pattern, and you'll notice that sometimes there is a comma and then the next character is a capital letter. You'll notice that pattern. Eventually, you may start to notice that there are certain words that occur often. You may notice that spellings are a thing. You may notice syntax. And when you get really good at all these, you start to notice the semantics. You start to notice the facts. But for that to happen, the language model needs to be larger.
- LFLex Fridman
So that's... Let's linger on that, 'cause that's where you and Noam Chomsky disagree. (laughs) So you think we're actually taking, uh, incremental steps, a sort of larger network, larger compute will be able to get to the semantics, be able to understand language without what, uh, Noam likes to sort of think of as a fundamental understandings of the structure of language, like, uh, imposing your theory of language onto the learning mechanism? So you're saying the learning, you can learn from raw data the mechanism that underlies language?
- ISIlya Sutskever
Well, I thi- I think it's pretty likely, but I also want to say that I don't really k- know precisely what is, what, uh, Chomsky means b- when he talks about im- you, you said something about imposing your structure on language. I'm not 100% sure what he means, but empirically, it seems that when you inspect those larger language models, they exhibit signs of understanding the semantics, whereas the smaller language models do not. We've seen that a few years ago when we did work on the sentiment neuron. We trained a small, you know, smallish LSTM to predict the next character in Amazon reviews, and we noticed that when you increase the size of the LSTM from 500 s- LSTM cells to 4,000 LSTM cells, then one of the neurons starts to represent the sentiment of the article, oh, sorry, of the review. Now, why is that? S- sentiment is a pretty semantic attribute. It's not a syntactic attribute.
- LFLex Fridman
And for people who might not know, I don't know if that's a standard term, but sentiment is whether it's a positive or a negative review.
- ISIlya Sutskever
That's right, like this, is the person happy with something, or is the person unhappy with something? And so here we had very clear evidence that a small neural net does not capture sentiment, while a large neural net does. And why is that? Well, our theory is that at some point you run out of syntax to model, so you start, gotta focus on something else.
- LFLex Fridman
And with size, you quickly run out of syntax to model, and then you really start to focus on the semantics, is, would be the idea.
- ISIlya Sutskever
That's right. And so I don't wa- I don't want to imply that our models have complete semantic understanding, because that's not true, but they definitely are showing signs of semantic understanding, partial semantic understanding. But the smaller models do not show that, those signs.
- LFLex Fridman
Can you
- 1:00:35 – 1:07:14
GPT-2
- LFLex Fridman
take a step back and say what is GPT-2, which is one of the big language models that was the conversation-changer in the past couple of years?
- ISIlya Sutskever
Yeah, so it's, so, so GPT-2 is a transformer-... with one and a half billion parameters that was trained on a, on about 40 billion tokens of text, which were obtained from web pages that were linked to, from Reddit articles with more than three up votes.
- LFLex Fridman
And what's a transformer?
- ISIlya Sutskever
A transformer is the most important advance in neural network architectures in recent history.
- LFLex Fridman
What is attention maybe too? 'Cause I think that's an interesting idea, not necessarily sort of technically speaking, but the idea of attention versus maybe what recr- recurrent neural networks represent.
- ISIlya Sutskever
Yeah. So the thing is, the transformer is a combination of multiple ideas simultaneously of which attention is one.
- LFLex Fridman
Do you think attention is the key?
- ISIlya Sutskever
No. It's a key, but it's not the key. The transformer is successful because it is the simultaneous combination of multiple ideas. And if you were to remove either idea, it would be much less successful. So, the transformer uses a lot of attention, but attention existed for a few years, so that can't be the main innovation. The transformer is designed in such a way that it runs really fast on the GPU, and that makes a huge amount of difference. This is one thing. The second thing is a transformer is not recurrent, and that is really important too because it is more shallow and therefore much easier to optimize. So, in other words, it uses attention, it is, it is a, a really great fit to the GPU, and it is not recurrent, so therefore less deep and easier to optimize. And the combination of those factors make it successful. So, now it makes, it makes great use of your GPU. It allows you to achieve better results for the same amount of compute, and that's why it's successful.
- LFLex Fridman
Were you surprised how well transformers worked and GPT-2 worked? So, you worked on language... You've had a lot of great ideas before transformers came about in language, so you got to see the whole set of revolutions before and after. Were you surprised?
- ISIlya Sutskever
Yeah, a little.
- LFLex Fridman
A little? (laughs)
- ISIlya Sutskever
Yeah. I mean, it's hard, it's hard to remember because you adapt really quickly, but it definitely was surprising. It definitely was. In fact, I'll... You know what? I'll, I'll retract my statement. It was, it was pretty amazing. It was just amazing to see it generate this text of this... And, you know, you gotta keep in mind that we've seen... At that time, we've seen all this progress in GANs, in improving... You know, the samples produced by GANs were just amazing.
- LFLex Fridman
Mm-hmm.
- ISIlya Sutskever
You have these realistic faces. But text hasn't really moved that much. And suddenly, we moved from, you know, whatever GANs were in 2015 to the best, most amazing GANs in one step.
- LFLex Fridman
Right.
- ISIlya Sutskever
And that was really stunning. Even though theory predicted, yeah, you train a big language model, of course you should get this. But then to see it with your own eyes, it's something else.
- LFLex Fridman
And yet, we adapt really quickly, and now there's, uh, sort of... Some cognitive scientists write articles saying that GPT-2 models don't truly understand language. So, we adapt quickly to how amazing the fact that they're able to model the language so well is. So, what do you think is the bar?
- ISIlya Sutskever
For what?
- LFLex Fridman
For impressing us that it...
- ISIlya Sutskever
I don't know.
- LFLex Fridman
Do you think that bar will continuously be moved?
- ISIlya Sutskever
Definitely. I s- I s- I think when you start to see really dramatic economic impact, that's when. I think that's in some sense-
- LFLex Fridman
I see.
- ISIlya Sutskever
... the next barrier, because right now, if you think about the work in AI, it's really confusing. It's really hard to know what to make of all these advances. It's kind of like, okay, you got an advance, and now you can do, uh, more things. And you got another improvement, and you got another cool demo. At some point, I think people who are outside of AI, they can no longer distinguish this progress anymore.
- LFLex Fridman
So, we were talking offline about translating Russian to English and how there's a lot of brilliant work in Russian that the, the rest of the world doesn't know about. That's true for Chinese. That's true for a lot of, uh, for, for a lot of scientists and just artistic work in general. Do you think translation is the place where we're going to see sort of economic big impact?
- ISIlya Sutskever
I, I don't know. I, I think, I think there is a huge number of applic- I mean, so first of all, I would wanna s- I wanna point out that translation already today-
- LFLex Fridman
That's true.
- ISIlya Sutskever
... is huge. I think billions of people interact with, uh, big chunks of the internet primarily through translation. So, translation is already huge, and it's hu- hugely, hugely positive too. I think self-driving is going to be tr- hugely impactful. And that's... You know, it's, it's unknown exactly when it happens, but again, I would, I would not bet against deep learning, so I-
- LFLex Fridman
So, that's deep learning in general. But you, you th- you think-
- ISIlya Sutskever
Deep learning for self-driving.
- 1:07:14 – 1:08:52
Active learning
- ISIlya Sutskever
- LFLex Fridman
Yeah. T- listen, I love active learning. So, let me, (laughs) let me ask, does the selection of data, can you just elaborate that a little bit more? Do you think the selection of data is, um... Like, I- I have this kind of sense that the optimization of how you select data, so the active learning process, is going to be a place for a lot of breakthroughs, even in the near future. Because there hasn't been many breakthroughs there that are public. I feel like there might be private breakthroughs that companies keep to themselves, 'cause it's a fundamental problem that has to be solved if you wanna solve self-driving, if you wanna solve a particular task. But d- do you, wha- what do you think about this space in general?
- ISIlya Sutskever
Yeah. So, I think that for something like active learning, or in fact for any kind of capability, like active learning, the thing that it really needs is a problem. It needs a problem that requires it. It's very hard to do research about a capability if you don't have a task, because then what's going to happen is that you will come up with an artificial task, get good results, but not really convince anyone.
- LFLex Fridman
Right. Like, we're- we're now past the stage where getting a result on MNIST, some clever formulation of MNIST, will- will convince people.
- ISIlya Sutskever
That's right. In fact, you could quite easily come up with a simple active learning scheme on MNIST and get a 10X speedup. But then, so what? And I think that with active learning there needs, there need- active learning will naturally arise as there are, as problems that require it pop up. That's how I would, that's my- my take on it.
- 1:08:52 – 1:13:41
Staged release of AI systems
- ISIlya Sutskever
- LFLex Fridman
There's another interesting thing that OpenAI has brought up with GPT-2, which is when you create a- a powerful artificial intelligence system, and it was unclear what kind of detrimental, once you release GPT-2, what kind of detrimental effect it'll have. Because if you have an- a model that can generate pretty realistic text, you can start to imagine that, you know, on the, it would be used by bots in some- some way that we can't even imagine. So like, there's this nervousness about what it's possible to do. So you- you did a really kind of brave and I think profound thing, which is start a conversation about this. Like, how do we release powerful artificial intelligence models to the public, if we do at all? How do we privately discuss with other, even competitors, about how we manage the use of the systems and so on? So from that, this whole experience, you released a report on it, but in general, are there any insights that you've gathered from just thinking about this, about how you release models like this?
- ISIlya Sutskever
I mean, I think that my take on this is that the field of AI has been in a state of childhood, and now it's exiting that state, and it's entering a state of maturity. What that means is that AI is very successful and also very impactful. And its impact is not only large, but it's also growing. And so for that reason, it seems wise to start thinking about the impact of our systems before releasing them, maybe a little bit too soon, rather than a little bit too late. And with the case of GPT-2, like I mentioned earlier, the results really were stunning. And it seemed plausible, it didn't seem certain, it seemed plausible that something like GPT-2 could easily be used to reduce the cost of disinformation. And so there was a question of, what's the best way to release it? And a staged release seemed logical. A small model was released, and there was time to see the... Many people used these models in lots of cool ways. There have been lots of really cool applications. There haven't been any negative applications that we know of. And so eventually it was released. But also other people replicated similar models.
- LFLex Fridman
That's an interesting question though, that we know of. So in your view, staged release is, um, at least part of the answer to the question of how do we... Uh, how, what do we do once we create a system like this?
- ISIlya Sutskever
It's part of the answer, yes.
- LFLex Fridman
I- is there any other insights? Like, say you don't want to release the model at all because it's useful to you for whatever the business is.
- ISIlya Sutskever
Well, there are plen- plen- plen- plenty of people don't release models already.
- LFLex Fridman
Right. Of course. But is there some moral, ethical responsibility when you have a very powerful model to sort of communicate? Like, just as you said, when you had GPT-2, it was unclear how much it could be used for misinformation. It's an open question. And getting an answer to that might require that you talk to other really smart people that are outside of, uh, out- outside of your particular group. It, have you, please tell me there is some optimistic pathway for people across the world to collaborate on these kinds of cases. Or is it still really difficult from- from one company to talk to another company?
- ISIlya Sutskever
So it's definitely possible. It's definitely possible to discuss these kind of models with colleagues elsewhere, and to get the, get their take on what's, on what to do.
- LFLex Fridman
How hard is it though?
- ISIlya Sutskever
I mean...
- LFLex Fridman
Do you see that happening?
- ISIlya Sutskever
I think that's- that's a place where it's important to gradually build trust between companies. Because ultimately, all the AI developers are building technology which is be- going to be increasingly more powerful. And so it's... The way to think about it is that ultimately we're all in it together.
- LFLex Fridman
Yeah, it's, uh, I tend to believe in the, uh, th- the better angels of our nature. But I do hope that, um, that when you build a really powerful AI system in a particular domain, that you also think about the potential negative consequences of, um... Yeah. It's an interesting and scary possibility that there will be a race for AI dev- AI development that would push people to close that development, and not share ideas with others.
- ISIlya Sutskever
I don't love this. I've been in a ca- a pure academic for 10 years. I really like sharing ideas, and it's fun. It's exciting.
- 1:13:41 – 1:25:00
How to build AGI?
- LFLex Fridman
What do you think it takes to... Let's talk about AGI a little bit. Wh- what do you think it takes to build a system of human level intelligence? We talked about reasoning. We talked about long-term memory. But, in general, what does it take, you think?
- ISIlya Sutskever
Well, I can't be sure. But I think that deep learning plus maybe another small idea.
- LFLex Fridman
Do you think self-play will be involved? Sort of like you've spoken about the powerful mechanism of self-play, where systems learn by sort of, uh, exploring the world in a competitive setting against other entities that are similarly skilled as them, and so incrementally improve in this way. Do you think self-play will be a component of building an AGI system?
- ISIlya Sutskever
Yeah. So, what I would say, to build AGI, I think is going to be deep learning plus some ideas. And I think self-play will be one of those ideas. I think that that is a very... Self-play has this amazing property that it can surprise us in truly novel ways. For example, like, we... I mean, pretty much every self-play system, both our Dota bot. I don't know if em- eh- OpenAI had a release about multi-agent, where you had two little agents who were playing hide and seek.
- LFLex Fridman
Mm-hmm.
- ISIlya Sutskever
And, of course, also AlphaZero. They were all produced surprising behaviors. They all produced behaviors that we didn't expect. They are creative solutions to problems. And that seems like an important part of AGI that our systems don't exhibit routinely right now. And so, that's why I like this area, I like this direction, because of its ability to surprise us.
- LFLex Fridman
To surprise us.
- ISIlya Sutskever
And an AG-
- LFLex Fridman
And what-
- ISIlya Sutskever
AGI system would surprise us fundamentally.
- LFLex Fridman
Yes. But... And to be precise, not just a r- not just a random surprise, but to find a surprising solution to a problem that's also useful.
- ISIlya Sutskever
Right.
- LFLex Fridman
Now, a lot of the self-play mechanisms have been used in the game context or at least in the simulation context. W- how much, how much do you... How far along the path to AGI do you think will be done in simulation? How much faith, promise, do you have in simulation versus having to have a system that operates in the real world, wheth- whether it's the real world of digital real world data or real world like actual physical world with robotics?
- ISIlya Sutskever
I don't think it's an either/or. I think simulation is a tool, and it helps. It has certain strengths and certain weaknesses, and we should use it.
- LFLex Fridman
Yeah, but-
- ISIlya Sutskever
Okay.
- LFLex Fridman
... uh, I understand that. That's, um, that's true. But one of the criticisms of self-play, one of the criticisms in reinforcement learning is one of the, the... Its current power, its current results, while amazing, have been demonstrated in a simulated environments or very constrained physical environments. Do you think it's possible to escape them, escape the simulated environments and be able to learn in non-simulated environments? Or do you think it's possible to also just simulate in the photorealistic and physics realistic way the real world in a way that we can solve real problems with self-play in simulation?
- ISIlya Sutskever
So, I think that sim- transfer from simulation to the real world is definitely possible and has al- and has been exhibited many times in, by many different groups. It's been especially successful in vision. Also, OpenAI in the summer has demonstrated a robot hand which was trained entirely in simulation in a certain way that allowed for sim-to-real transfer to occur.
- LFLex Fridman
Is this, uh, for the Rubik's Cube?
- ISIlya Sutskever
Yeah, that's right. And so that-
- LFLex Fridman
I wasn't aware that was trained in simulation.
- ISIlya Sutskever
It was trained in simulation entirely.
- LFLex Fridman
R- really? So, it wasn't in the phys-... The, the hand wasn't trained?
- ISIlya Sutskever
No. 100% of the training was done in simulation. And the policy that was learned in simulation was trained to be very adaptive. So adaptive that when you transfer it, it could very quickly adapt to the physical, to the physical world.
- LFLex Fridman
So, the kind of perturbations with the giraffe or whatever the heck it was, those weren't... Were those part of the sim- simulation?
- ISIlya Sutskever
Well, the simulation was generally... So, the s- the simulation was trained to be robust to many different things, but not the kind of perturbations we've had in the video. So-
- LFLex Fridman
I see.
- ISIlya Sutskever
... it's never been trained with a glove. It's never been trained with a, uh, uh, p- a, a stuffed giraffe.
- LFLex Fridman
So, in theory, these are novel perturbations?
- ISIlya Sutskever
Correct. It's not in theory, in practice.
Episode duration: 1:37:27
Install uListen for AI-powered chat & search across the full episode — Get Full Transcript
Transcript of episode 13CZPWmke6A
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome