Hello everybody. Good afternoon, good morning. And I am delighted to be the one chosen to introduce to you this really distinguished group of people that we've got here sitting around the table.
Six I think of the most brilliant, most consequential people on the planet today. And I don't think that's an overstatement. So these are the winners of the 2025 Queen Elizabeth Prize for Engineering.
and it honors the laureates here that we see today for their singular impact on today's artificial intelligence technology. Given your pioneering achievements in advanced machine learning and AI and how the innovations that you've helped build are shaping our lives today, I think it's clear to everyone why this is a really rare and exciting opportunity to have you together around the table. For me personally, I'm I'm really excited to hear you reflect on this present moment that we're in, the one that everybody's trying to get ahead of and understand and your journey, the journey that brought you here today.
Um, but also to understand how your work and you as individuals have influenced and impacted one another and the companies and the technologies that you've built. And finally, I'd love to hear from you to look ahead um and to help us all see a bit more clearly what is to come, which you are in the best position to do. So, I'm so um pleased to have you all with us today and and looking forward to this to the discussion.
So, I'm going to start going from the zooming out to the very personal. I want to hear from each of you your a personal kind of aha moment in your career that you've had that you think has sort of impacted the work that you've done or was a turning point for you that brought you on this path to why you're sitting here today whether it was kind of early in your career in your research or or much more recently what you know what was your personal moment of awakening um that that has impacted the technology do we should we start here with you Yeshua >> thank you yes uh with pleasure. I would um go to two moments.
One when I was a grad student and I was looking for something interesting to research on and I read some of Jeff Hinton's early papers and I thought wow this is so exciting. Maybe there are a few simple principles like the laws of physics that could help us understand human intelligence and help us build intelligent machines. And the second moment I want to talk about is two and a half years ago after chat GPT came out and I realized uhoh what are we doing?
Uh what will happen if we build machines that understand language uh have goals and we don't control those goals? What happens if they are smarter than us? Uh what happens if people abuse that power?
So that's why I decided to completely shift my research agenda and my career to try to do whatever I could about it. >> That's that that's two kind of very you know diverging things very interesting build what tell us about your moment of like kind of building the infrastructure that's that's fueling what we have. >> I'll give you two moments as well.
So the first was you know in in the late 90s I was at Stanford trying to figure out how to overcome what was at the time called the memory wall. fact that accessing data from memory is far more costly in energy and time than doing arithmetic on it. And it sort of you know struck me to organize computations into these kernels connected by streams.
So you could do a lot of arithmetic without having to do very much memory access. That basically led the way to what became called stream processing and ultimately GPU computing. Um and we we originally built that thinking we could apply GPUs not just for graphics but to general scientific computations.
So the second moment was I was having breakfast with my colleague Andrew Ing at Stanford and at the time he was working at Google finding cats on the internet you using 16,000 CPUs in this technology called neural networks >> which fay had something to do with those >> and uh um he uh he he basically convinced me this is a great technology so I with Brian Kenzo repeated the experiment on 48 GPUs in Nvidia and when I saw the results of that I was absolutely convinced that this is what Nvidia should be doing. we should be building our GPUs to do deep learning because this has, you know, huge applications in all sorts of fields beyond finding cats. And that was kind of an aha moment to really start working very hard on specializing the GPUs for deep learning and and to make them more effective.
>> And when was that what year? >> Um, the breakfast was in 2010 and I think we repeated the experiment in 2011. >> Okay.
>> Yeah. >> Jeff, tell us tell us about your work. One very important moment was when I in about 1984 I tried using back propagation to learn the next word in a sequence of words.
So it was a tiny language model and discovered it would learn interesting features for the meanings of words. So just giving it a string of symbols it just by trying to predict the next word in a string of symbols it could learn how to convert words into sets of features that captured the meaning of the word and have interactions between those features predict the features of the next word. >> So that was actually a tiny language model from 1980 late 1984 um that I think of as as a precursor for these big language models.
The basic principles were the same. It was just tiny. We had 100 training examples.
It took 40 years to get us here though. >> And it took 40 years to get here. And the reason it took 40 years was we didn't have the compute and we didn't have the data and we didn't know that at the time.
We couldn't understand why we weren't just solving everything with back propagation. >> Which takes us cleanly to to Jensen. We didn't have the compute for 40 years and here now you are building it.
Tell tell us about your moments that of real kind of clarity. Well, for my career, um, I was the, uh, first generation of chip designers that was able to use higher level representations and design tools to design chips. and and uh that that discovery um uh was helpful when I learned about a new way of developing software uh around the 2010 time frame simultaneously from three different labs uh what was going on in uh uh University of Toronto researchers uh reached out reached out to us at the same time that uh researchers at the NYU reached out to um as well as uh in Stanford reached out to us at the same time and I I I saw the early indications of what turned out to have been deep learning around the same time uh using uh a framework uh and a structured design to uh create software and that software turned out to have been incredibly effective.
Uh and that second that second observation uh is seen again using frameworks rep higher level representations structured types of uh structures like the deep learning networks. I uh was able to develop software uh w was very similar to designing chips for me and the patterns were very similar and I realized at that time maybe we could develop software uh and capabilities that that scale very nicely as we've scaled uh chip design over the years and so that was that was a quite a quite a moment for me >> and when do you think was the moment when the chips really started to help scale up today's sort of the the LLMs that we have today because you you said 2010 that's still a 15 year. >> Yeah.
The the thing about about Nvidia's architecture is is once you're able to get something to run well on a GPU because it became parallel, you could get it to run well on multiple GPUs. that same sensibility of scaling uh the algorithm to run on many processors on one GPU. This is the same logic and the same reasoning that you could do it on multiple GPUs and then now multiple systems and in fact you know multiple data centers and so that once we realized we could do that effectively then then the rest of it is about about uh imagining how far you could extrapolate this capability.
you know, how much data do we have? How large can the networks be? How much dimensionality can it capture?
What kind of problems can it solve? Uh the all of all of that is is really engineering at that point. You know, the the observation that that uh the deep learn deep learning models are so effective uh is is really quite the the the spark.
The rest of it is really engineering extrapolation. Fei, tell us about your your moment. >> Yeah, I also have two moments to share.
So around 2006 and 2007, I was transitioning from a graduate student to an a young assistant professor and I was among the first generation of machine learning graduate students um reading papers from young Yoshua uh Jeff and I was really obsessed in trying to solve the problem of ob uh visual recognition which is the ability for machines to see meaning in objects in everyday pictures and uh we were struggling with this problem in machine learning called generalizability which is um after learning from certain number of examples can we recognize something a a new example new sample and I've tried every single algorithm under the sun from baset support vector machines to neuronet network and the missing piece that my student and I realized is that data is missing that uh uh you know if you look at the evolution or development of uh intelligent animals like humans we were inundated with data in the early years of development but our machines were starved with data. So we um decided to do something crazy at that time to create a internet scale data set uh over the course of three years called imageet that uh uh in uh included 15 million images handcurated um by by people around the world across 22,000 categories. So, so for me the aha moment at that point is big data drives machine learning >> and it's now it's it's the limiting factor the building block of all of the you know algorithms that we're seeing with >> yeah it's part of the scaling law of today's AI and the second aha moment is um 2018 I was the first chief scientist of uh AI at Google cloud uh part of the the work we do is serving all vertical industries under the sun, right?
From healthcare to financial services, from entertainment to uh manufacturing, from agriculture to energy. And that was a few years after the the what we call the image that Alex moment, a couple of years after Alph Go, and I realized >> Alph Go being the algorithm that was able to beat humans at playing the Chinese board game Go. Uh yes and as the chief scientist at Google I realized this is a civilizational technology that's going to impact every single human individual as well as sector of business and uh if humanity is going to go enter an AI era what is the guiding framework so that we not only innovate but we also bring benevolence to uh through this powerful technology.
technology to everybody and that's when I returned to Stanford as a professor to uh co co-found the human center AI institute and and uh propose the human- center AI framework so that we can keep humanity and human values in the center of this uh technology. >> So developing but also looking at the impact and what's next which is where the rest of us come in. >> Um Yan do you want to round us out here?
What's what's been your highlight? >> Yeah, probably go back a long time. Um, I realized when I was in undergrad, I was fascinated by the question of AI and intelligence more generally and discovered that people in the 50s and 60s that worked on training machines instead of programming them.
I was really fascinated by this idea probably because I thought I was either too stupid or too lazy to actually build an intelligent machine from scratch, right? So it's better to let itself be um like train itself or self-organized and that's the way you know intelligence in in in life uh builds itself. It's uh it's selforganized.
So I I thought this concept was really fascinating and I couldn't find anybody when I graduated from engineering. I was doing chip design by the way um wanted to go to grad school. I couldn't find anybody who was uh working on this but connected with some people who kind of were interested in this and discovered Jeff's papers for example uh and uh he was the person in the world I wanted to meet most in 1983 when I started grad school and we eventually met two years later um [clears throat] and >> and today you're friends would you say?
>> Yes. Oh, we we we we [laughter] had lunch together in 1985 and we could finish each other's sentences. Basically, he had uh um I had a a paper written in French at a conference where he was a keynote speaker and and managed to actually kind of decipher the the math.
It was kind of sort of like back propagation a little bit to train multi-layer nets. It was known from the 60s that the limitation of machine learning was due to the fact that we could not train machine with multiple layers. So that was really my obsession and it was his obsession too and um and so I had a paper that kind of proposed some some way of doing it and he kind of managed to read the math.
So that's how we hooked up and >> and that's what has set you on this path. >> Right. So and and then after that you know once you can you can train complex systems like this you ask yourself questions.
So how do I build them so they do something useful like recognizing images or things of that type? And at at the time Jeff and I had this debate when I was a postoc with him in the late 80s. Um I I I thought um the only machine learning paradigm that was well formulated was supervised running.
You you show an image to the machine and you tell it what the answer is, right? And he said no no no like the only way we're going to get to make progress is through unsupervised running. And I was kind of dismissing this at the time.
Um, [snorts] and what happened in you know the mid 2000 when he Yosha and I sort of start getting together and restart the interest of the of the community in deep learning. We actually kind of uh made our bet on unsupervised learning or self reinforcement loop. Right?
>> This is not reinforcement. So this is basically discovering the structure in data without training the machine to do any particular task which is by the way the way LLMs are trained. So an LLM is trained to predict the next word but it's not really a task.
It's just a way for the system to learn a good kind of uh representation or capture the >> is there no reward system there that sorry to get geeky but is there no nothing to say this is correct and therefore keep doing it because >> well this is correct if you predict the next word correctly right >> from the rewards in reinforcement learning where you say that's good >> yeah okay >> um and so in fact uh I'm going to blame it on you [laughter] uh it turns out produced this big data set called imageet and uh which is which was labeled and so we could use supervised learning to train the systems on and that turned out to work actually much better than we expected and so we temporarily abandoned the whole program of working on self-supervised unsupervised learning because supervised learning was working so well we figured out a few tricks [laughter] >> Joshua stuck with it >> I said I didn't >> no you didn't I didn't either but [laughter] uh but it it kind of refocus the entire industry and and the research community if you want on sort of deep deep learning supervised learning etc. Mhm. >> And it it it took another few years maybe around 201617 to uh tell people like this is not going to tell take us where we want.
We need to do self-s supervised learning now and that's what LLM really are the best example of this. >> Okay. >> But uh what we're working on now is applying this to other types of data like like video sensor data which LLM are really not very good at at all.
Um and that's a new challenge for the next few years. So that brings us actually to the present moment and I think you know you'll all have seen this crest of the interest from people who had no idea what AI was before who had no interest in it and now everybody's flocking to this and this has become more than a technical innovation right that's a huge business boom it's become a geopolitical strategy issue um and you know everybody's trying to get their hands around what this is so or their heads around it Jensen I'll come to you here first to I want you all to reflect on this moment now here Nvidia in particular has it's basically in the news every day hour week you know and you have become the most valuable company in the world so there's something there that people want >> you'll be to hear that >> yeah [laughter] you know tell us about do are you worried that we are getting to the point where people don't quite understand and we're all getting ahead of ourselves and there's going to be a reckoning that there's a bubble that's going to burst and then it will write itself self and if not what is the kind of biggest misconception about demand coming from AI that is different to say the dotcom era or that people don't understand you know if if that's not the case >> uh during the dotcom era during the the bubble the vast majority of the fiber deployed were dark meaning the industry deployed a lot more fiber than it needed Mhm. >> Today almost every GPU you could find is lit up and used.
And so uh the reason why I think it's important to take a take a step back and understand and understand what AI is, you know, for a lot of people AI is Chad GBT and it's image generation and and it that's all true. That's one of the applications of it. Um, and AI has advanced tremendously in the last several years.
The ability to not just memorize and generalize, but to reason and effectively think and ground itself through research. It's able to produce answers and do things that are much more valuable now. It's much more effective.
and the number of companies that are able to build businesses that are that are helpful to other businesses. For example, a software programming company, an AI software company that that we use called Cursor, uh they're very profitable and we use their software tremendously and it's incredibly useful. uh or a bridged or open evidence who are uh serving the healthcare industry doing very very well producing really good results and and so so the AI capability has grown so much and as a result we were seeing these two exponentials that are happening at the same time on the one hand the amount of computation necessary to produce an answer has grown tremendously on the other hand the amount of usage of these AI models are growing also exponentially these two exponentials are causing a lot of demand on compute.
Now when you take a step back, you ask yourself fundamentally what's different between AI today and the software industry of the past. Well, software in the past was pre-ompiled and the amount of computation necessary for the software is not very high. >> But in order for AI to be effective, it has to be contextually aware.
It has to it can only produce the intelligence at the moment. You can't produce it in advance and retrieve it. That's you know that's called content.
AI intelligence has to be produced and generated in real time. And so as a result we now have an industry where the computation necessary to produce something that's really valuable in high demand is quite substantial. We have created an an industry that requires factories.
That's why I I remind ourselves that AI needs factories to produce these tokens to produce the intelligence and this is this is a a once you know once in a it's never happened before where the computer is actually part of a factory and and so we need hundreds of billions of dollars of these factories in order to serve the trillions of dollars of industries that sits on top of intelligence. You know, you go come back and take a look at at software in the past. Software in the past is they're software tools.
They're used by people. For the first time, AI is intelligence that augments people. And so, it addresses labor.
It addresses work. It does work. >> So, you're saying no, this is not a bubble.
>> I think this we're we're well in the beginning of the buildout of intelligence. And and the fact of the matter is most people still don't use AI today. And someday in the near future, almost everything we do, you know, every moment of the day, you're going to be engaging AI somehow.
And so between where we are today where the usage is quite low to where we will be someday where the usage is basically continuous, that buildout is is you know what >> and if even if the LLM runway runs out, you think GPUs and the infrastructure you're building can still be of use in a different paradigm and then I want to open up to others to talk. LLM is a is a piece of the AI technology. You know, AIS are systems of models, not just LLMs and LLM are big part of it, but there are systems of models and and uh the the technology necessary for for AI to be much more productive from where where it is today irrespective of what we call it.
Um we still have a lot of technology to develop yet. >> Can who wants to jump in on on this? >> Um I don't think >> especially if you disagree.
I don't think we should call them LLMs anymore. Um they're not language models anymore. They they >> right >> start as language models at least that's the pre-training but but more recently there's been a lot of advances in making them agents.
In other words, uh go through a sequence of steps in order to achieve something interactively with an environment with people right now through a dialogue but more and more with a computing infrastructure. And the technology is changing. It's not at all the same thing as what it was three years ago.
I don't think we can predict where the technology will be in two years, 5 years, 10 years. U but we can see the trend. So one of the things I'm doing is trying to uh bring together a group of international experts to keep track of what's happening with AI where it is going um what are the risks how are they being mitigated and and and and the trends are very clear across so many benchmarks now you know because we've had so much success in improving the technology uh in the past it doesn't mean that's going to be the same in the future.
So then then there would be financial uh consequences uh if the expectations are not met but in the long run I completely agree. Um >> but currently what about the rest of you? Do you think that the valuations are justified in terms of what you know about the technology the applications?
>> So I think there are three trends that sort of explain what's going on. The first is the models are getting more efficient. If you look just at attention for example, going from straight attention to GQA to MLA, you get the same or better results with far less computation.
And so that then drives demand in ways where things that may have been too expensive before become inexpensive of now. You can do more with AI. At the same time, the models are getting better and you know, maybe they'll continue to get better with transformers or maybe a new architecture will come along, but we will we won't go backwards.
We're going to continue to have better models that also >> they still need GPUs even if >> absolutely transformer based >> um in fact it makes it makes them much more valuable compared to more specialized things because they're more flexible and they can evolve with the models better but then the final thing is I think we've just begun to scratch the surface on applications so almost every aspect of human life can be made better by having AI you know assist somebody in their profession help them in their daily lives and you know I think we've you know started to reach maybe 1% of the ultimate demand for this. So as that expands, you know, the, you know, number of uses of this are going to go up. So I don't think there's any bubble here.
I think we're, like Jensen said, we're riding a multiple exponential and we're at the very beginning of it and it's going to just keep going. >> And in some ways, Nvidia is in to that because even if this paradigm changes and there's other types of AI and other architectures, you're still going to need the the atoms underneath. So that makes sense for you.
Did you want to jump in Fay? Uh yeah, I do think that um of course from a market point of view, it will have its own um dynamics and sometimes it does adjust itself, but if you look at the long-term trend, let's not forget AI by and large is still a very young field, right? We walk into this room and on the wall there were equations of physics.
Physics has been a more than 400 year old uh discipline. Even if we look at uh modern physics and AI is less than 70 years old if we go back to Alan Turing you that's about 75 years so there is a lot more new frontiers that is to come uh you know Jensen and Yoshua talk about LLMs and agents those are more languagebased but even if you do uh self uh introspection of human intelligence there's more intelligent capabilities is beyond language. I have been working on spatial intelligence which is really the combination or the lynchpin between perception and action where um where uh you know humans and animals have incredible ability to perceive reason interact with and uh and create uh worlds that goes far beyond language.
And even today's most powerful language-based uh or LLM based models uh fail at rudimentary spatial intelligence uh tests. So from that point of view as a as a discipline as a science there's far more frontiers to conquer and to uh open up and that brings the applications uh you know opens up more applications. >> Yeah.
and you work at a company and so you have the kind of dual perspective of being a researcher and working in a commercial space. Do you agree? Do you do you believe that this is all justified and you can see where this is all coming from or do you think we're reaching an end here and we need to find a new path?
>> So I think there are several point of views for which uh we're not in a bubble and at least one point of view suggesting that we we are in a bubble but there is but it's a different thing. So we're not in a bubble in the sense that um there are a lot of applications to develop based on LLMs. LLM is the current dominant paradigm and there's a lot to uh milk there.
This is you know what Bill was was saying to kind of help people in the daily lives with current technology that technology needs to be pushed and that justifies all the investment that is done on the software side and also on the infrastructure side. uh once we have you know smart wearable devices um in everybody's hands assisting them in their daily lives the amount of computation that would be required as as Jensen was saying to uh to serve all those all those people is going to be enormous so in that sense the investment is not is not wasted but there is a sense in which there is a bubble and it's the idea somehow that the current paradigm of LLM would be pushed to the point of having human level intelligence which I personally don't believe in and you don't either And we we need kind of a few breakthroughs before we get to machines that really have the kind of intelligence we observe not just in humans but also animals. We don't have robots that are nearly as smart as a cat, right?
Um and so we're missing something big still. Which is why AI progress is not just a question of more infrastructure, more data, uh more investment and more development of the current paradigm. It's actually a scientific question of how do we make progress towards the next generation of AI >> which is why all of you are here right because you actually sparked the entire thing off and I feel like you know we're moving much towards the engineering application side but what you're saying is we need to come back to what brought you here originally um on that question of human level intelligence we don't have long left so I just want to do a quick fire I'm curious can each of you say how long you think it will take until we do reach that point where you believe we're you know equivalent machine intelligence to a human or even a clever animal like an octopus or whatever.
How far away are we just just the years? >> It's not going to be an event. >> Okay.
>> Okay. Because the capabilities are going to expand progressively in various domains. >> Over what time periods?
>> Over, you know, maybe we'll make some significant progress over the next five to 10 years to come up with a new paradigm. >> F and then maybe, you know, progress will come. But it'll it'll take longer than we think.
Okay. Parts of machines will supersede human intelligence and part of the machine intelligence will never be similar um or the same as human intelligence. They are they are they're built for different purposes and they will >> when do we get to superseding?
>> Part of it is already here. How many of us can recognize 22,000 objects in the world? So part of >> do you not think an adult human can recognize 22,000 objects?
>> Um the kind of granularity and fidelity. No. How many adult humans can translate a 100 languages?
>> That's harder. Yeah. >> So yeah.
>> So I think we should be nuanced and grounded in scientific facts that uh just like airplanes fly but they don't fly like birds. and u machine-based intelligence will do a lot of powerful things but there is a profound um place for human intelligence to to always be critical in our human society. Jensen, do you have >> we have enough general intelligence to uh translate the technology to an enormous amount of uh society useful applications uh in the next coming years and with respect to >> Yeah.
>> Yeah. Yeah. We're doing it today.
>> Yeah. And so I think I think uh one we're already there >> and two the the other part of the answer is it doesn't matter >> because at this point it's a bit of an academic question. We're going to apply the technology to and the technology is going to keep on getting better and we're going to apply the technology to solve a lot of very important things from this point forward.
And so okay >> I I think the answer is it doesn't matter >> and and it's now as well. >> Yeah you decide. Right.
If you refine the question a bit to say how long before if you have a debate with this machine it'll always win. >> I think that's definitely coming within 20 years. We're not there yet but I think fairly definitely within 20 years we'll have that.
So if you define that as >> AGI it'll always win a debate with you. >> We're going to get we're going to get there in less than 20 years probably. >> Okay.
Bill, do you have >> Yeah. Well, I'm sort of with Jensen that it's the wrong question, right? Because our goal is not to build AI to replace humans or to be better than humans.
>> But it's a scientific question. It's not that we'll replace humans. The question is could we as as a society build something?
>> But our goal is to build AI to augment humans. And so what we want to do is complement what what humans are good at. Humans can't recognize 22,000 categories or most of us can't solve these math olympiad problems.
Um so we build AI to do that. So humans can do what is uniquely human, which is be creative and be empathetic and and understand how to interact with other people in our world. And I think that it's not clear to me that AI will ever do that, but AI can be huge assistance to humans.
>> So I'll beg to differ on this. Uh I don't see any reason why at some point we wouldn't be able to build machines that can do pretty much everything we can do. Um, of course, for now on the spatial and you know, robotic side, it's lagging, but there's no like uh conceptual reason why we couldn't.
So on on the timeline, I think there's a lot of uncertainty and that we should plan accordingly. Um, but there is some data that I find interesting where we see um the capability of AI to plan over different horizons to grow exponentially fast in the last six years. And if we continue that that trend, it would place roughly the level that an employee has in their job to uh AI being able to do it within about five years.
Now this is only one category of engineering tasks and there are many other things that matter. For example, uh one thing that could change the game that is that many companies are aiming to just to focus on the ability of AI to do AI research. In other words, to do engineering, to do computer science, and to design the next generation of AI, including maybe improving robotics and spatial understanding.
So, I'm not saying it will happen, but the area of ability of AI to do better and better programming and understanding of algorithms that is going very very fast and that could unlock many other things. We don't know and we should we should be really agnostic and not make big claims because there's a lot of possible futures there. M so so our consensus is in some ways we think that future is here today but there's never going to be one moment and the job of you all here today has helped to guide us along this route um until we get to a point where we're working alongside these systems.
Very excited personally to see where we're going to go with this. If we do this again in a year it'll be a different world. But thank you so much for joining us for sharing your stories and for talking us through this this huge kind of revolutionary moment.
Thank you. Thank you. >> Thank you.
