YouSaid · the spoken record

Ilya Sutskever

lines on the record: 106
first: 2020-05-08
most recent: 2020-05-08
sittings or episodes: 1
sources: podcast

Every line below is reproduced as it was said and linked to the record it came from. Nothing here is summarised or generated. Directory · Search · Corrections

“Thanks for listening to this conversation with Ilias Etsr, and thank you to our presenting sponsor, Cash App. Please consider supporting the podcast by downloading Cash App and using the code Lex Podcast. If you enjoy this podcast, subscribe on YouTube, review it with five stars and Apple podcasts, support it on Patreon, or simply connect with me on Twitter at Lex Friedman. Now let me leave you with some words from Alan Turing on machine learning. Instead of trying to produce a program to simulate the adult mind, why not rather try to produce one which simulates the child's? If this were then subjected to an appropriate course of education, one would obtain the adult brain. Thank you for listening and hope to see you next time.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Being humble in the face of the uncertainty seems to be also part of this whole happiness thing. Well, I don't think there's a better way to end it than meaning of life and discussions of happiness, so Ilya, thank you so much. You've given me a few incredible ideas. You've given the world many incredible ideas. I really appreciate it. And thanks for talking today.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“So, your academic accomplishments, all the papers, you're one of the most cited people in the world, all the breakthroughs I mentioned in computer vision and language and so on. What is the source of happiness and pride for you?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Let me ask two silly questions about life. One do you have regrets moments that if you went back you would do differently, and two are there moments that you are especially proud of that made you truly happy.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“But they're so dynamic. There's got to be some underlying sort of Freud thinks there's like sexual stuff. There's people who think it's the fear of fear of death. And there's also the desire for knowledge and, you know, all these kinds of things, procreation, sort of all the evolutionary arguments. It seems to be there might be some kind of fundamental objective function from which everything else emerges. But it seems like that's very difficult to meet.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“It's funny because action does require an objective function. It's definitely there in some form, but it's difficult to make it explicit and maybe impossible to make it explicit, I guess, is what you're getting at. And that's an interesting. The fact of an RL environment.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“So on that topic of the objective functions of human existence, what do you think is the objective function that simplicity in human existence? What's the meaning of life?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Beautifully put, yeah. Are there specific mechanisms you can think of of aligning AI and values to human values? Do you think about these problems of continued alignment as we develop the AI systems?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“But let me take a step back to that moment where you create the AGI system. I think this is a really crucial moment. And between that moment and the Democratic board members with the AGI at the head, There has to be a relinquishing of power. So as George Washington, despite all the bad things he did, one of the big things he did is you relinquished power. First of all, didn't want to be president. And even when he became president, he didn't keep just serving as most dictators do for indefinitely. Do you see yourself? Being able to relinquish control over an AGI system, given how much power you can have over the world, at first financial, just make a lot of money, right? And then control by having possession of the AGI system.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Wow, that's part of their, so it's not that just they can't help but be controlled, but They exist. One of the objectives of their existence is to be controlled.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Let me sort of that's actually okay, that's a beautiful vision, I think, as long as it's possible to press the reset button. You think it will always be possible to press the res”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Now, again, let me emphasize the fact that you truly are one of the people that might be in the room where this happens. So let me ask. Sort of a profound question about, I just talked to Stalin historian. I've been talking to a lot of people who are studying power. Abraham Lincoln said nearly all men can stand adversity, but if you want to test a man's character, give him power. I would say the power of the twenty first century, maybe the twenty second, but hopefully the 21st would be the creation of an AGI system and the people who have control direct possession and control of the AGI system. So, what do you think after spending that evening? Having a discussion with the AGI system, what do you think you would do?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“What kind of questions do you think would they be factual or would they be Personal, emotional, psychological, what do you think?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“You're one of the people that might be able to create an AGI system here, not you, but you and OpenAI. If you do create an AGI system and you get to spend sort of the evening with it, him, her, what would you talk about, do you think?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“The same is applied to autonomous vehicles. The same is probably going to continue to be applied to a lot of artificial intelligence systems. We find this annoying, this is the process of, in the 21st century, the process of analyzing the progress of AI is the search for one case where the system fails in a big way where humans would not. And then many people writing articles about it and then broadly as the public generally gets convinced that the system is not intelligent. And we pacify ourselves by thinking it's not intelligent because of this one anecdotal case. And this seems to continue happening.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“That's a really nice way to put it. I also just don't like that human instinct to criticize a model is not intelligent. That's the same instinct as we do when we criticize any group of creatures as the other. It's very possible that GPT 2 is much smarter than human beings at many things”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“So right now they make mistakes. They might be more accurate than human beings, but they still make a difference set of mistakes.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“So we talk about consciousness, but let me talk about another poorly defined concept of intelligence Again, we've talked about reasoning, we've talked about memory, what do you think is a good test of intelligence for you? Are you impressed by the test that Alan Touring formulated with the imitation game of natural language? Is there something in your mind that you will be deeply impressed by if a system was able to do?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“No, I know, I know I know. It's still an open question if there's not some magic in the brain that I don't mean a non materialistic magic, but that the brain might be a lot more complicated and interesting than we give it credit for.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Do you think that's an emergent thing that just comes from? Do you think consciousness could emerge from the representation that's stored within your networks? So that it naturally just emerges when you become more and more, you're able to represent more and more of the world.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Even if you're not able to physically interact with the world, and if you're not able to, I mean, I actually was getting at. Let me ask on the more particular, I'm not sure if it's connected to having a body or not, but the idea of consciousness and a more constrained version of that is self-awareness. Do you think an EGI system should have consciousness? We can't define whatever the heck you think consciousness is.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Let me ask sort of an embodied question staying on AGI for a sec. Do you think AGI system would need to have a body? We need to have some of those human elements of self-awareness, consciousness, sort of fear of mortality, sort of self-preservation in the physical space, which comes with having a body.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“That those are novel perturbations. Well, that's okay That's a clean Small scale, but clean example of a transit from the simulated world to the physical world.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, but okay, I understand that that's That's true. One of the criticisms of self play, one of the criticisms in reinforcement learning is one of the Its current power, its current results, while amazing have been demonstrated in a simulated environment, or very constrained physical environments. Do you think it's possible to escape them? Escaped 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?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Now, a lot of the self play mechanisms have been used in the game context, or at least in the simulation context. 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, whether it's the real world of digital real world data or real world like actual physical world of robotics?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Do you think self play will be involved? So, like you've spoken about the powerful mechanism of self-play where systems learn by Sort of 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?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Do you think it takes to let's talk about AGI a little bit? 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?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, I tend to believe in the better angels of our nature, but I do hope. When you build a really powerful AI system in a particular domain, that you also think about Potential negative consequences of yeah. An interesting and scary possibility there will be a race for AI development that would push people to close that development and not share ideas with others.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Right, of course. But is there some moral ethical responsibility when you have a very powerful model to sort of communicate? 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. Getting an answer to that might require that you talk to other really smart people that are outside your particular group. Have you please tell me there's some optimistic pathway for people across the world to collaborate on these kinds of cases? Or is it still really difficult from one company to talk to another company”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“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.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“An interesting question, though, that we know of. So, in your view, stage release is at least part of the answer to the question of how do we. How, what do we do once we create a system like this?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“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?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Another interesting thing that OpenAI has brought up with GPT too, which is when you create a powerful artificial intelligence system, and it was unclear what kind of detrimental, once you release GPT to what kind of detrimental effect it will have. Because if you have a model that can generate a pretty realistic text, you can start to imagine that it would be used by bots and some way that we can't even imagine. So like there's this nervousness about what it's possible to do. So you did a really kind of brave and I think profound thing, which is started 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.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Right. Like we're now past the stage where getting a result and amnest, some clever formulation of MNIST will convince people.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, listen, I love active learning. So let me ask, does the selection of data, can you just elaborate that a little bit more? Do you think the selection of data 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. In the near future, because there hasn't I feel like there might be private breakthroughs that companies keep to themselves because the fundamental problem has to be solved if you want to solve self-driving, if you want to solve a particular task. What do you think about the space in general?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, one of the beautiful things is that GPT, the transformers are fundamentally simple to explain, to train. Do you think Bigger will continue to show better results in language. Sort of like, what are the next steps with GPT to do that?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Just to check, you're not seeing a connection between driving and language. Be a poetic connection. I think there might be some, like you said, there might be some kind of unification towards a kind of multitask transformers that can take on both language and vision tasks. That'd be an interesting unification. Let's see, what can I ask about GPT2 more?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“So we were talking offline about translating Russian to English and how there's a lot of brilliant work in Russian that the rest of the world doesn't know about. That's true for Chinese. That's true 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?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“And yet we adapt really quickly and now there is a sort of Some cognitive scientist writes 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? For impressing us that Think that bar will continuously be moved.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Were you surprised how well Transformers worked and GPT2 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 surprised.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Is attention maybe too? Because I think that's the interesting idea, not necessarily sort of technically speaking, but the idea of attention versus maybe what recurrent neural networks”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“You take a step back and say, what is GPT2, which is one of the big language models that was the conversation changer in the past couple of years”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“With size, you quickly run out of syntax to model, and then you really start to focus on the semantics would be the idea.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“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.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Let's linger on that. That's where you and Noam Chomsky disagree. So, you think we're actually taking incremental steps, a sort of larger network, larger compute will be able to Get to the semantics to be able to understand language without what Noam likes to sort of think of as a... Fundamental understandings of the structure of language, like 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.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“That whole mortality thing is kind of a sticky problem that we haven't quite figured out. Maybe we'll solve that one. I think one of the fascinating things 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 kind of try to describe the recent history of using neural networks in the domain of language and text”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“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.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“The answers you get are sticky in the sense you already have a mental model, you already have an Yeah, metamata. a big conception of that human being how they think what they know how they see the world and then everything you ask you're adding on to that and that stickiness seems to be One of the really interesting qualities of the human being is that information is sticky. You seem to remember the useful stuff, aggregate it well, and forget most of the information that's not useful. That process, but that's also pretty similar to the process in neural networks do, is just that neural networks are much crappier at this time. It doesn't seem to be fundamentally that different. But just to stick on reasoning for a little longer. He said, why not? Why can't I reason? What's a good, impressive feat benchmark to you of reasoning?”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Find that the epitome of interpretability can you do better? Can you because you can't, okay, I'd 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'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 Fanil now had some self-awareness about it.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Precisely, I like the word precisely. So I'm thinking of the kind of compression of information the knowledge bases represent, sort of creating a I apologize for my sort of human centric thinking about what knowledge is because neural networks aren't interpretable necessarily with the kind of knowledge they have discovered. 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. Knowledge base. Obviously, not the actual Wikipedia or the language, but like a semantic web, the dream that semantic web represented. So it's a really nice compressed knowledge base or something akin to that in the... Neural networks would have.”
2020-05-08 · Lex Fridman Podcast · #94 – Ilya Sutskever: Deep Learning · IDENTIFIED FROM THE TRANSCRIPT · source