 Hey everybody, it's great to be here and you know thank you very much for the introduction. I'm really happy to tell you about a very new program that we have launched in collaboration with many other organizations and we hope more soon including those here that are not yet part of our little endeavor. We call it the AI Alliance and I'm going to tell you about the program but first I'm going to tell you about why we came together to do this. If you remember the beginning of this year and it was a long time ago it was a strange time for AI, wasn't it? Because AI has been very open for many years, software, open science, research and it was very collaborative but in the last couple of years that started to switch and change and early this year it looked at risk of really changing, proprietary models risked kind of taking over. In fact if it was in the spring and you were a developer you wanted to build the Gen AI application you pretty much started with an API tying into probably one proprietary model set. Now that's a pretty striking contrast actually because the last months have been really great for open source. There's a couple of compelling examples. You've heard about a lot of this already but I just think that they're illustrative of what's happening. Lama, Lama 2 with the permissive license was only released in the summer, it's only been a few months and in that time it's been downloaded over 100 million times and there's a whole ecosystem of tools, of variants growing up around it. That's not like years, that's months, it's pretty incredible if you think about it. A few other examples, you've heard about some of this today but if you look at just these two application frameworks, helping developers that want to get started trying to build an application of Gen AI like 95,000 stars in again six to nine months. That's pretty unbelievable. You've heard from Jeff earlier, Hugging Face, right 300,000 stars, 400,000 open models. You all don't need to be convinced that the community wants more than just a managed service API. But there are lots of headwinds still, we shouldn't joke, we shouldn't kid ourselves that everything's good and be complacent. There are actually lots of challenges still to open source AI. It starts with the resources required. It's still the case that if you want to build especially at the leading edge of model capability, it takes a lot of resources, it takes a lot of money, that is excluding a lot of people, especially researchers that need to be part of this. Things have changed a lot too in terms of building applications and it changes all the time, right? The patterns are the thing today but there are a million different varieties of how to do that. Things that you optimize for one model family are not portable to another and so on. And then there are safety and trust concerns, a lot of really legitimate, serious concerns that need to be addressed with practical solutions and some hyperbole too of course, which we need to try to distinguish between. And then in terms of legal and regulatory environment, there are also challenges in terms of copyrights, challenges in terms of new regulation, which we'll get to in just a second, that threaten to restrict what you all do. And then there's this interesting, mostly productive but sometimes unproductive debate happening about what is open source AI? What's good enough to be called open source AI? Look, this is a very generally healthy thing but let's not get too caught up in what open source AI models should be just yet. Let's work this through and let's also celebrate increasing steps to be more open in more ways. Now, always fun to have a few numbers to attach to these statements, right? What does it take? Well, it takes still to build a model, a lot of compute. It takes compute measures in hundreds of thousands of GPUs. It takes time scales of order of weeks. Those are numbers that are millions of dollars. Now, that is simply hard for organizations and especially the research community to deal with. If you think about what it takes in terms of data, Chinchilla scaling laws show pretty clearly empirically that you need multiple trillions of tokens of data to train a model at the leading edge. Then there's a whole different variety of challenges, as I mentioned. There are, in case you haven't been paying attention, vigorous debates in many parts of the world, especially here in the US, about what we should do to regulate AI. Just last week, the EU has decided to move forward on their AI Act, which is actually reasonably encouraging, I think, from an open source standpoint. Now, we have to see how things evolve. But for a little while, there was a real risk that open source would be greatly restricted and that risk has not gone away. Here in the US, there are very active debates about what we should do to regulate AI. And there are many people that are threatening to really put a damper, to put a restriction on open source work in AI. Because what's happening is to some success, people are arguing that open source is much more dangerous than a closed source solution, which is kind of crazy if you think about it. But that argument is holding sway in places. And then there's this. This is a great one, right? Do we all have to confront the fact that we're contributing to an existential risk to humanity? No, no, we don't, okay? If we take a step back and you think about what's happening here, there are lots of legitimate, serious, important challenges and risks of AI to solve, but they're much better solved in the open. And these arguments that are getting put forth, somewhat successful in policy circles, that AI represents an existential risk to humanity, is frankly a lot of nonsense. Now, for all of these reasons, a number of us have come together to create an umbrella program we call the AI Alliance. And the Alliance mission is to support, is to enable open innovation, open science, open source in AI, and to do that inclusively, to do it globally, and to do it in a way that we can make reality continue for what we all have known for a long time. That open source provides far more benefit than it offers in terms of drawbacks. Now, the program is straightforward. We are bringing people, we're bringing resources, we're bringing data, we're bringing compute together to enable building and supporting open technology development with the simple goal of across the stack making sure capabilities in the open are just as good, if not better than any proprietary capability. We are going to be enabling developers, educating students, we're going to make sure that these technologies are accessible and usable and understood. And we're going to advocate, not from the sidelines, but very directly to make sure that every policymaker and every business leader and organizational leader understands clearly why open source in AI is so important and why it really is the solution to many of the practical problems we have and not the danger. Now, we just announced it. This is actually the first public talk other than some media announcements. This is the first public talk about the alliance. So I'm testing some messaging. If it's not resonating, I'm going to, you know, let me know and I'll tweak it next time. But we just announced this. My favorite headline is right here. Open source AI fights back. I love it. Lots of other headlines. It has gotten more attention than, you know, even an IBM marketing department could hope for. And that's a lot of attention. I'm going to tell you that. We're really happy that this is getting the world's attention. We're happier that a lot of the core messaging has actually come through here, right? Supporting open source, focusing on the benefits of open source, there's a fun little battle that, of course, has been set up, as you might imagine. But that's not entirely a bad thing. Now, as I mentioned, we wanted this, even from day one. And we plan to grow it. We plan to add many more. We plan to expand the program. But even from day one, we wanted to try to be representative of the diverse community that exists in AI and in open source. So we tried to do this quickly. We've got a number of organizations globally, as you can see here. They do span research organizations, universities, large companies, startups. We cover a number of areas of the world. There's more to do here. But we're happy with the starting point we had. And if you look through this list, there are many names that have been represented at this conference here. And so you've already seen a piece, many pieces, of what this alliance is building from. And if you think about the program vision, imagine we're going to be bringing things together and enhancing and growing. And we're going to do it with this base. And we're going to do it even broader as we move forward. Now, I'll just take a few minutes, or a minute or two, to go through the program itself. So what is this? This is a broad program. And it's structured in terms of an alliance of projects. And it is very self-driven. It's not a monolithic program. It's not modeled after a traditional open source organization necessarily either. This is a program that's based around self-driven and organizationally driven priorities. It will be composed of projects. Projects are going to be spanning the six categories on the right, which I'll mention in just a minute. And importantly, this is very much about what individuals or organizations would like to prioritize, what resources they're going to bring. And as such, different organizations can take part in different projects as they see fit. There's an agility built into the program and a lightweight kind of approach to the overall governance that we're going to try to maintain very strongly. Now, if you look through the categories, we are fairly ambitious here. We want to work across the stack and across the ecosystem that starts with education and skills building. So a number of our organizations have already made strong commitments and have a lot of activity here. We're going to be expanding what we do here. And in particular, bringing resources to the challenge of enabling exploratory research to continue at the leading edge. We want to, number two here, build solutions to the challenges of trust and safety in deploying AI. There are real issues here. And we want to build real solutions, right? We don't want to be focused on hyperbole. We want to build tools. We want to build standards and benchmarks. We want to do that collaboratively. And we want to then obviously show the world that, practically speaking, open source can solve these problems. We then want to make sure that we continue to build out the very important existing frameworks and new frameworks and software projects that are enabling model development and application development. We have an important thrust in number four here to enable choice, to enable the democratization of access to different forms of hardware that are rapidly advancing and becoming part of the scene. We absolutely are going to enhance and work toward better open foundation models. And I'll just point out one really important part of this, as I think you may realize. But just to be clear, there's a lot of languages and cultures in the world that are not included in the LLM revolution yet, whether because native language data sources are not good enough or not really available enough or because alignment and tuning principles don't aren't established for different cultural norms and so on. This is a really important thing, and it's a topic that many of the members in the broader, especially outside the English-speaking world are extremely interested in. And finally, the advocacy piece, super important. We need to make sure that our elected officials, that our policymakers understand the importance of open source AI and why it is essential to have a healthy and very open ecosystem. So I am going to leave it at that. If you'd like to learn more, check out the site. I'll be around afterward. I'd love to talk to all of you. As mentioned, many of you are kind of part of this already. And if you're not, let's talk. This is just the starting point. We have a lot of big things planned, and we'd love to have you contribute to it. So with that, thank you for your time. Hopefully I get a chance to meet some of you after.