I want the best open weights model in the world to be a US model. I think the critical point there is the frontier models are not open sourced.
Good afternoon everyone. For the ethics lawyers in the audience, I do want to clarify that Emil is the nominee for the Under Secretary of Defense for Research and Engineering. He is not yet the Under Secretary. I'm not sure how many of you got to catch my comments this morning, but I flagged AI as one of the fields where the United States has a lead, but we need to maintain that lead, and in particular ahead of China. Alphabet and OpenAI obviously are two of the leading research labs in the United States, Alphabet having even invented the transformer architecture. I first wanted to start off with a little bit of a bragging session to remind everybody how great American AI is. The two data points that I want folks to cover are: Kevin, if you can start off with just talking about how ChatGPT has a bajillion monthly active users and keeps growing and whatever you guys are up to next. And then Ruth, you guys won basically one of the first Nobel Prizes of AI with AlphaFold. So I want to hear you guys talk about that. But yeah, Kevin.
All right. So because Ruth is here, who's ridden in a Waymo? Some folks, they're all over San Francisco. They're coming to DC. You should definitely do it. My first ride in a Waymo was the first minute you're in there, you're gripping like, 'Oh my god, watch out for that bicycle.' And then your blood pressure comes down and five minutes later you're like, 'Oh my god, I am living in the future. I'm being driven around San Francisco by a robot.' And it's amazing. And then 10 minutes later you're bored looking at your phone. There's an analogy: all of a sudden this thing that was incredible to you 20 minutes ago is now just a boring part of your life, and of course robots drive you around San Francisco. There's an analogy there to AI in general and how we adapt. Who uses ChatGPT? Love it. There are over 500 million people using ChatGPT every week. That's like 7% of the planet that's used ChatGPT in the last seven days. And it's two years old. So you've gone from GPT-3, which at the time was mind-blowing, and now if I asked you all to use GPT-3 inside ChatGPT today, you'd be like, 'Ugh, what is this thing?' But ChatGPT and the intelligence of models is on a literal exponential curve. It's getting better incredibly quickly, which I think is one of the reasons that people are adopting it so fast, because every few months computers can do something that computers have never been able to do in the history of the planet. As a couple examples, we launched our frontier model o3 just recently. If you go back a couple versions, one of the ways we judge the intelligence of these models is on these benchmarks. There's a competitive programming benchmark. A couple models ago, this was like three or five months ago, the model was the millionth best competitive coder in the world. That doesn't sound great, but that's still top 2 or 3% in the world. So that's pretty good. People were using this all over the place to write code in production environments. The previous model, which was like three or four months ago, was the thousandth best competitive coder in the world. o3, which we launched a couple weeks ago, is the 175th best competitive coder in the world. So over the space of six months, you've gone from a millionth to 175th. Talking about exponentials, it's not going to be long before we have a model that is the best competitive coder in the world, and forever after that point computers will be better than humans at writing code. But I'm actually excited about that. If you think about the movie Hidden Figures, there were people helping do calculations to compute rocket trajectories to help get us to the moon. They're writing on pieces of paper, calculating rocket trajectories. We look at that as anachronistic. Of course people don't do rocket trajectories by hand. Computers do that way better. Coding is going to be the same way. We're going to look a few years from now and say, 'Well of course computers write code. They're better at it.' But by the way, we still have rocket scientists. Rocket scientists are hugely important, and engineers are going to be hugely important. We are just going to be able to move faster. We're going to be able to do more in our personal lives, in business, in government, in national security because of these models. And I could not be more excited.
Nobel Prize, AlphaFold. Incredible. So if I go back in time to your point that the US has led in technology, we need to continue leading in technology and we can't take it for granted. I think that's a really important point. To put this moment in history in context with AI, I'll go back to another one of our colleagues, Vint Cerf, who is the chief internet evangelist at Google. If you Google him, you will learn that he is described as one of the fathers of the internet. I asked him this question about how profound is this moment in history. He said he believes that AI has greater potential than even the internet did. This platform shift is going to be more important because with generative AI, with agents, the potential to augment human capabilities is so profound and endless. We are very excited. We have AI now with Gemini already in billions of people around the globe using it. Thank you for your comments on Waymo. We absolutely love Waymo and hopefully very soon everyone in DC will be enjoying Waymo. When we look at the implications, it goes to a number of areas. One, to your question, is about advancing science. We were extremely proud when Demis and John Jumper, Demis Hassabis and John Jumper, were awarded the Nobel last fall. As probably people in this room know, it was for AlphaFold, which is predicting the protein structure for all known proteins known to humanity. Previously it would take a couple of years just to predict one. They did all 200 million. That is being described as the single greatest contribution to drug discovery in our lifetime. That to us is the... Thank you. So what we're really excited about is this opportunity, as Demis has described it, to address many, many diseases in our lifetime. We're looking at early detection of disease. You can detect early stage metastatic cancer, breast cancer, lung cancer with AI in ways that augment physicians and radiologists everywhere. As someone who's been through cancer, I asked my oncologist and he said the only way to democratize healthcare for everyone anywhere in the globe, to get the same level of diagnosis, is with AI. So whether it's the economic uplift, the science benefit, healthcare benefit, it is stunningly transformative. And Kevin, you guys are also doing some work with some of the national labs, right?
Yeah, I think honestly it's what Ruth was saying, that the capacity for these models to drive fundamental science is one of the most exciting things. Why is America America? It's because we have entrepreneurship, we have innovation, we are driving scientific innovation that leads to GDP growth. Everybody wins. We have a partnership with the national labs where we're working with Los Alamos, Sandia, Lawrence Livermore folks like that. We've deployed frontier models on a high-side supercomputer. These are folks that are advancing material science, doing fundamental physics research, safeguarding our nuclear stockpile. We have seen them do unbelievable things. They're some of the most sophisticated users of our AI models out there. Humans, great scientists and researchers, plus AI models move scientific research faster. They discover more things. If we could invent things two times faster, five times faster, ten times faster than we do, think of the good that will do America and the world. We're talking about AI, but the next step is really all about quantum AI. We're extremely excited about the application of quantum computing to life sciences and many other applications. Continuing to maintain the lead of the US is key as this unlocks.
So about a quarter ago, the Chinese released DeepSeek, an open source model that was highly capable, maybe a generation behind the leading labs, but by default as an open source model it's effectively free for anybody to deploy. This gives them significant soft power if it now gets deployed all across the globe, whether in hardware systems that are used for the military or in consumer applications. If that AI is unwilling to recognize the fact that the Tiananmen Square massacre has happened, that gives China a huge advantage. So as two of the leading labs, I'm curious how both of you guys think about releasing open source models. How do you guys think about that relative to the things you keep closed source?
I can start if you want. We said maybe a month ago we're going to be releasing an open weights model. We're super excited about it. The way we think about it is, first and foremost, we want to bring AI to the world. We believe deeply in the ability of AI to improve all of our lives. So an open weights model contributes to that. Also, I want the best open weights model in the world to be a US model. I want it to be built on democratic values. I don't want the best open weights model in the world to be a Chinese model. So from that perspective, we think it's really important that we put a great model out there. Now it will not be our frontier model. The way we think about it is probably something like a generation behind, because putting a frontier model out is also accelerative to China. But we think we can put out a great open weights model that the entire world can adopt. So we have a model that the entire world is using, built on democratic values. We will also have our full frontier models that'll be a generation ahead that we offer like we do today. And the US will have the best models in the world, regardless of how they're used.
I think the critical point there is the frontier models are not open sourced, and it's the smaller models. But I think core to your question is we do need to be very mindful collectively about the conditions to keep the US in the lead, because we've been talking for quite some time that the gap between the US and China, whether it's on models or on chips, has been narrowing. Now we've seen it with DeepSeek, we see it with the Huawei progress on Ascend, on Harmony OS, you see it in a lot of the data around graduates, the ascendancy of universities. So that's a big concern, and see what they're doing on energy, which we should get to at some point. So this is not a foregone conclusion that the US will stay in the lead. Also, one of the reasons, Emil, why I'm excited for you to formally take on your role as soon as you do, is America has the best models in the world, but it does us no good if those models are sitting on the shelf. Everything that I understand from various conversations is that China is doing a better job today at integrating and deploying and using these models, not just in business but especially in government and in the military. It's not just about having the best models, we've got to actually use them. I think there's a huge opportunity for the US there.
Speaking of keeping the US ahead of Chinese models, Emil, how do we feel about the fact that a prominent US venture capital firm, which may or may not have also fired your former coworker, is choosing to invest into frontier Chinese AI models and funding those with US capital?
You know, it's very curious to say the least. It's not like there's not a plethora of US companies doing AI to invest in that are high quality. There's big companies that are doing some incredibly innovative stuff, which you don't usually associate with big companies. Some of our big companies are really innovating quite quickly, and there's a lot of small companies that are doing that. Invest in a Chinese AI company where by definition the products that they produce are going to be dual use for their Chinese Communist Party and their military is a curious sort of move to me. So it just doesn't make sense. But I do think it's surprising to us in a way because the era that hopefully we're at the end of, the last era was not to pick on Google, but when Maven was a thing of 'well, should we support the US military's use of our own technologies developed in the US?' I think that era is over, which is good. So I think you'll see a lot of cooperation not only from OpenAI but all the big companies here. That's a big positive. And then when our venture capital community, which is US, is trying to invest in innovation, ideally they're doing it here, at least in allied countries. Or else we're going to have some cross incentives that make things weird.
Yeah, I mean you clearly have such a unique perspective in that when confirmed, this won't be your first time serving for the department. You flip-flop back and forth between the department and tech. How do you think about just within your seat what you think needs to change within the Pentagon to ensure that you guys are adopting some of these phenomenal technologies as rapidly as possible?
