Darío Gil Technical Keynote—Genesis Mission Summit 2026

I want to invite a series of short talks by brilliant scientists to tell you that this is happening across all fields of science. So you are going to see how this is touching AI, implications for mathematics. You're going to hear in a second the implications for imaging. We're going to see what is possible to model the electric grid in new ways, and you're going to see what is possible to also transform national security through AI and advanced manufacturing. So, first, I want to invite Terry Tao. Professor Tao is a field medalist winner and, without a doubt, one of the great mathematicians in the world, and his contributions to so many areas of mathematics are legendary. One thing I greatly admire about Professor Tao is that also he has been a pioneer of exploring the implications of modern AI to mathematics, and you're going to hear about the radical changes that are happening in the field as a consequence of the progress of this technology. So it's my great pleasure to welcome on stage Terry. Here you are. Thank you. Thank you very much. I'm very pleased to be here. It's a very exciting event and very timely, certainly in mathematics. Sorry. Do we have? Yeah. Mathematics is really being transformed almost by the week as there's an amazing new development in our subject, and it's really a combination of many things. So one is AI, which is providing the raw power, providing all kinds of new solutions to math problems that have stood open for decades, but also paired with formal verification, which is a complementary technology that allows you to check these answers because AI sometimes makes mistakes, but we can now put them out, and also modern collaborative tools to allow humans and AI and these formal verifiers to work together seamlessly. So it's transforming our field very unevenly, but ultimately in a very promising way. Some aspects of mathematics are accelerating very, very rapidly. Problem solving, co-generation, automatically formalizing proofs, refactoring, and searching literature. These are extremely, they've been almost tenfold, hundredfold accelerated. Other parts of mathematics are catching up. We are only now inventing new ways to collaborate and to take full advantage of these new technologies. Traditionally we work in small groups, one or two, I still use blackboards. It's pen and paper, but we are beginning to modernize our workflows. But we do have many new challenges. All these new technologies are creating new traffic jams, bottlenecks. Peer review is now extremely bottlenecked. We need to rapidly modernize our math education system and support our junior faculty more. But we will solve these problems, and we will be in a completely new era. I don't have much time, so I'll just give you one or two data points of how things are being accelerated. This is a graphic illustrating the Erdős Problems. It's a set of 1,200 very influential problems that poor Erdős posed in the 20th century. I solved a dozen of these myself. A year ago, about 300 of these problems, at least 1,200 were solved. Now it's almost 500. It's not entirely due to AI, but if you know this graph, every time a new model is released there's a notable spike in the number of solved solutions. It's been playing a huge role in this benchmark. I'd love to talk more about this, but I only have four minutes. Another place in which we have really, really transformed one aspect of our work is automatically verifying our answers. For centuries, when you write a proof, it can only be verified by hand. People had to check with pen and paper, referees, that the argument is correct. We have had automatic computer verification for decades, but it used to take years or even decades to actually enter in a proof correctly into a computer and check that everything worked. But it's accelerated amazingly. In the 70s, it took decades to formally verify a theorem. There was a major open problem solved by an AI last month, and within days multiple formalizations appeared. We can completely be confident that these things are correct. Those are two places where things have been really accelerating. Actually, DARPA is launching a massive challenge to auto-formalize all the math literature. This is really what's making AI input usable. Otherwise, we'd have a flood of proofs and arguments which we can't trust. The pairing of AI and formal verification is absolutely essential. What we have to do now is identify new ways to collaborate and really take full advantage of all these new capabilities. We are launching massive new collaborative projects, either to formalize huge amounts of mathematics or to generate huge amounts of new datasets organized on modern internet platforms. We are launching new internet competitions where unlimited AI use, but we verify every single submission in an extremely stringent way. As part of the Genesis mission, we are going to try to really initiate highly interdisciplinary collaborations with science and industry, really bring AI-powered mathematics to the rest of the sciences. If I want to summarize where we are in mathematics, for centuries we were in an era of proof scarcity where we were always short on proofs. We always wanted to prove. We were always searching for proofs to answer problems. We are going to enter an era of proof abundance where we have more proofs than we know how to deal with. We have other problems in mathematics, other challenges, but this is where we are right now. Thank you. Thank you, Terry. My next guest that I want to invite on stage is Laurent Chapon, who is Associate Lab Director of Photon Sciences at Argonne National Laboratory. Under his leadership, the Advanced Photon Source in Argonne became the world's brightest synchrotron in 2024. He and his team have been pioneering the implications of AI for imaging. He's going to show some amazing results. Thank you, Dario. Thanks very much. It's super exciting. I'm going to switch gears and tell you about how AI is accelerating science at the scientific user facility that we have in the complex. Many of you know the US is home to the best scattering facilities. I'm talking about light sources, neutron sources in the world. They generate a huge amount of data. This large machine is actually one of the entry points into the macroscopic world, down to atomic precision. They're wonderful. They produce beams of X-rays or neutrons that are billions of times brighter than what you can do conventionally. This is really, really amazing. I'll show you what we can do nowadays through an example using a microchip. What you see on this screen here is a microchip. The field of view is about 50 microns. That's actually smaller than the thickness of a human hair. With advanced X-ray technology nowadays, you can go in three dimensions non-destructively and inspect such microchip to the finest detail. You see here all in red, all of the metal lines. You can go deep. You can probe different layers and things like that. This is really amazing, down to the smallest feature at the nanoscale. The sensors that we use are not recording like a conventional camera. We're not recording directly a representation of reality. We are recording a very complex pattern that encodes this reality. Most of the game is actually to go from what we measure to this wonderful image and video that I show you here. How do we do that? Typically, in a conventional experiment without AI, which was the case not so long ago, to do an experiment, you collect data. By the way, this facility together, collectively, measure the equivalent of 150 million movies every year. That's a huge amount of data. Then you use advanced computing, mathematics, solvers, analysis to actually get the insight that you want to get from this structure. Where the experiment side takes only a few minutes or hours, the rest of the data analysis pipelines can take days and, in the most extreme case, weeks. With AI, everything is changing. You do a lot of experiments. You create large databanks, very like the protein challenge that we heard this morning. From these large databanks, you can build a vision model. This vision model can help do real-time analysis of these massive structures in real time. This is an example of the work we've done for Genesis. This is a correlation between Berkeley, Brookhaven, Slack, Oak Ridge, and Argonne. What you see on the top left-hand side, the circular pattern, is the image that we see on the detector. There's nothing to do with the representation of the chip that you see on the right-hand side. As we scan the X-ray through this specimen, the AI model is able, in real time, to reconstruct the data. Something that was impossible only a few months ago. You go from something that takes hours to something that takes, essentially, real time. This is done at 200 images per second. There's at least a factor of 100 in speedup, and it impacts everything we do. I'll finish by saying one thing. I've never seen a transformation like that in 25 years in a scattering facility. This is unbelievable, but this is only the start. Once you have a real-time application like that, what you do, you can have machine-actionable representation of the rest, which is finding features, extracting features, finding anomalies in a sample like that. Once you do that, everything becomes agentic and becomes completely autonomous. That's the direction of travel in this field. It will impact most of the things we do, whether you look at manufacturing processes, whether you want to accelerate material discovery, guiding a new enzyme in biotech. The applications are immense, and the speed is absolutely unbelievable. Thank you again for the opportunity. From the very small to the very large. I am going to invite now Dr. Hendrik Hamann, who is the Chief AI Scientist at Brookhaven National Laboratory and also a professor at Stony Brook. He has spent his career working at the intersection of high-performance computing and AI and big data, and he's going to tell you about the amazing results of using AI to model the electric grid. Okay, so I'm very excited to be here. The electric grid is arguably the most important critical infrastructure, powering literally everything we depend on. The electric grid is also one of the largest and most complex systems humanity has ever built. There are 200 million access points in the US, 15 million transformers, 10 million energy resources, 7 million miles of wires, and much more. It's all interacting nearly at the speed of light. Now, to control this grid, we must very carefully control, balance the flow of electricity through this vast network. If we don't, disaster strikes, outages can occur. However, we must not only control these flows in the moment, but also in hours, in days, in weeks, in months, and even years ahead. And that is really where the challenge starts, because the future has many, many possibilities. Demand, generation, weather, it's all uncertain. So when you combine all these possibilities, the number of possible scenarios which we must consider can reach billions and even more. However, today, we only analyze a few, which is crazy, considering that we are in the midst of a massive and unprecedented expansion of our grid. Now, here's where AI can come in, where AI actually must come in. So instead of doing it the old way, taking a few scenarios, each is being solved from the beginning iteratively with complex and somewhat slow physics solvers, we've built actually grid foundation models, which learn the direct relationship between topology, demand, and energy consumption. And because it learns it directly, it's much faster. And that's what we need. Now, unlike language foundation models, which are trained on text, our models are trained on the language of the grid. That means voltages, power, frequency, across a broad set of conditions. So the model learns how to predict the physics of the grid, enabling us to evaluate billions of scenarios rapidly, not only rapidly, but also accurately and robustly. Because as you can imagine, grid AI modeling is a national scale challenge. We're collaborating across the board with utilities, hyperscalers, universities. It's really a community effort. Now, we've already achieved speedups of up to 1,000, but don't worry, we're not done. Without AI, for example, data center interconnection simulation studies take months, sometimes years. Now we can do it literally in minutes. Without AI, we plan and operate the grid scenario poor. But with AI, we can remove the blindfolds. Without AI, the grid's complexity, the grid's uncertainty is a challenge. In fact, it's a threat. It's a nightmare. With AI, it can actually be a huge opportunity. So I want to leave you with this. AI is often considered to be a burden on the grid, but with Genesys, the opposite can actually be true. It is actually a huge opportunity to build the most secure, affordable, and reliable grid for the American people. Thank you, Hendo. All right. And my last guest for this plenary talk is Sivar Rajamanickam, who is a distinguished member of the technical staff at Sandia National Laboratories, and he is going to show you an amazing project in the field of national security. So, Sivar. Thank you. Excited to be here. Let us look at how AI and advanced manufacturing can transform national security applications. I want to draw your attention to this project called ARISTAR. ARISTAR is a Genesys mission demonstrator. The focus of ARISTAR is flight test vehicles. Flight test vehicles help us study and measure the heat and vibrations experienced by a payload. They're quite complicated to build. Takes years to build. Very, very expensive to build. Takes tens of millions of dollars. When we started this project last November, our goal was to have this done in a matter of months. Now, we have the advantage of using AI. We have the advantage of using advanced manufacturing, but we also had the foundation that we have built for the past several decades. We have our scientific software, our data, our on-prem HPC compute, which is the best in the world, our manufacturing facilities, and above all, the wealth of domain expertise our scientists have. Now, with that, three NNSA labs, Sandia, Livermore, and Los Alamos joined forces with our manufacturing partner in Kansas City to execute Genesys mission. Now, I want to focus on one aspect of Genesys mission first. The problem here is to focus on designing a payload structure. Given a set of objectives, a set of constraints, we want to build a payload structure. We developed an agent-driven topology optimization approach to do this. Agents go and explore this vast search space to find the best design. Then humans pick the best design out of this to manufacture. Here is an example. The insert shows you 16 different designs that the tool was able to produce. We chose one to manufacture that is zoomed in there. You can see how non-intuitive this design is. This is not something a human would produce, and that is number one. Number two is you won't even consider this design if you don't have advanced manufacturing. That's number two. So it is this combination that helps you, and you can go look at this design in the vehicle that is standing outside in the hallways if you are interested. Now, I want to zoom out. We didn't only design the payload structure. We used AI to design the payload itself. We used AI to place sensors on this flight test vehicle. We actually did a flight test and that way proving grounds in Utah last month. And then finally, we used some innovative manufacturing techniques where this entire thing can be built one shot. It is like a Russian nested dolls where the whole thing is nested, and you can build this in one week. So what is the result of this project? Genesis mission scientists brought the AI expertise. Aries tied brought the domain expertise. From a clean sheet of requirements, we were able to do the first production in three months and the first flight test in eight months. Now, you might wonder how does this affect you, actually? You come from a different domain. I'm not from national security. What we have done is taken all these tools, the way to access models, and put it into the Genesis platform. And you'll be able to use the same platform we used to build this Aries toy. Now there is this notion that if you have a complex problem like this, you have to choose two out of three, better, faster, or cheaper. With a technology like AI, Aries tied is one example where you can go for better, faster, and cheaper. Thank you. Wow. I find that really compelling and persuasive. It tells you that this is happening right now. And it brings to bear that sometimes when we give the historical analogy to previous moments of mobilization, like Apollo program or the Manhattan Project, of why we have chosen not to do one thing but many. Because all these fields are going to be transformed by science and engineering. So I want to take the opportunity now to share with you as plainly as I can, what is the Genesis mission? What are we trying to do technically? And how will we succeed? What is it? We are doing three things. The first thing is we have constructed a portfolio of national science and technology challenges to activate the best scientists and engineers in America to solve those problems, create new ways with new tools. Second, we are building collaboratively a platform to accelerate scientific discovery. And third, we want to invest in the next generation of scientists that are educated with these dual competencies. Earlier this year, we announced 26 national challenges. I don't expect all of you to read them. They're all being published for a while. But they span the full range of the mission of the Department of Energy. Energy from both nuclear, fusion, the electric grid you saw examples right now, discovery science, and national security. Today, Director Kratzios announced that now the Genesis mission has 20 federal agencies that are coming together. This is an outstanding moment of uniting the full strengths of the federal science and technology ecosystem. As a result of this expansion and partnership, we are expanding a subset of the national challenges. You see here highlighted in green areas in which we're expanding the scope of the original challenge as a consequence of the collaboration with other agencies. And we're adding new ones now spanning space, human health, defense, and veterans' health. We're incredibly excited. This is a new set of the revised set of now 33 national science and technology challenges. We will keep refining and expanding as appropriate. But this now reflects the full strength of these 20 different agencies and departments coming together. Now, with just the original set of 26, earlier this year, we selected 21 of those and put a call for proposals. The RFA that Secretary Wright announced the results of. Today, we're announcing 278 awards with over $250 million of funding. There are 342 participating institutions. You see in the color coding that they reflect national labs, higher ed, companies, nonprofits, and that they represent all 50 states. It's a true national mission representing talent from across the country. I expect all of you to memorize this. This, when later you go and see the poster session that is next to us, you'll have an opportunity. This is just a list of all the participating institutions. And you're going to get a chance to also see some of the highlighted projects that are being executed. But it gives you a sense of the scale. Gives you a sense of the talent that we're bringing to bear into this space and the capacity that we have in our great country to be able to innovate. I want to just briefly highlight an example that Secretary Wright highlighted, Project Prometheus, which is how to bring the power of AI to accelerate and deliver nuclear energy that is faster, safer, and cheaper. I highlight it not only because it's the first example of the phase two of the awards, I want to also share with you that we will be announcing the remaining of the phase two awards by early fall, hopefully early to mid-September. So we're like thick on that process. But this project involves 32 partners, uniting national laboratories, industry, and universities. And it is a tour de force of bringing all the frontier capability in terms of agentic orchestration, how do you do persistent digital threads, and physics-informed surrogates. A surrogate, an example of a surrogate is what you saw Hendrik discuss in modeling the grid, where you can go and instead of having those sort of initial physics-based simulations, you can do something much more efficient than before, as well as provable cyber assurance. So this is going to be a magnificent project whose end goal is to dramatically lower the cost and the time to realize the nuclear renaissance in the United States. If you're a PI, one of the things I want to do for you, one of my gifts I wanted to give to you is each one of you, I'm giving you a challenge coin. Because you are the inaugural teams of what will be many teams to come. So if you could just take a minute, if you're in the audience, and maybe raise your hand and stop up if you're a PI or a participant in one of those projects, we would love to celebrate you for a minute. Okay, there you are. Fantastic. We're so full of joy. You went through the most competitive process that we've ever done. You know, there were many also additional outstanding teams, and we wish we could have supported many more of them. But we're very proud of what you proposed to us. So I want to, second, explain what are we trying to do in building this platform to accelerate science. This is a science first strategy. So the underpinning of everything we do is based on the scientific method. And we are going to exploit every capability we have on theory, on measurement, bringing our national user facilities, and the distributed facilities we have built over decades across America's science and technology ecosystem, the world of simulation, both are high performance computing or classical supercomputers. And we're also within Genesis, we announced Quantum Genesis to bring forth the first generation of fault-tolerant quantum computers by 2028. We have a North Star also to bring and create a first generation of user facilities to support the scientific community. So this is a huge aspect of what we're going to push for. We're bringing AI models and agents. And the capability that they bring to bear is that they can encode and semantically connect scientific knowledge and realize a dream that has been in the field for over 300 years since Leibniz. That we can take all knowledge and encode it and represent it efficiently in a new representation. These models now can reason and efficiently predict. You saw examples that we experience every day with LLMs, but you saw what Hendrik and others showed with surrogate models, where we can dramatically accelerate simulation. And interestingly and fascinatingly, we can create an act. We have agents now that can write code and can take action and execute complex scientific workflows. Now with this capability, we are envisioning a future where we create an internet of science. In the original internet, the end nodes of the computing of the devices were computers. And the orchestration was a TCP IP protocol. In the modern world of an internet of science, the end nodes are both computers and scientific instruments, particle accelerators and telescopes and microscopes and additive manufacturing. And the orchestration is enabled by this agentic capability that can pass through complex scientific understanding and workflows and orchestration. Now you saw the example a minute ago of Aries Tide, where multiple national laboratories and a manufacturing environment allowed agents to use high-performance computing and simulation codes and additive manufacturing to execute a project that was multiple times faster than what you could do before. Well, if you look at a graph like this, kind of reminds you a little bit of what happened 50 years ago. In 1969, there was a diagram that looked like this. There was just a few nodes, a few computers being connected to one another. This was ARPANET. In 1971, there were 15 nodes that were connected. In 1980, there were 50 nodes that were connected. And we know where the story goes. So one of the key things we are going to do in the Genesis mission is to connect our national laboratories and our user facilities and our university colleagues to a federated approach and our industry partners in creating an internet of science. And if we look ahead a few years from now, what would success look like across these dimensions? So I'll ask it as a series of questions. Do we, the United States, have the most comprehensive ecosystem of high-performance computers for simulation? Have we built the world's most powerful AI supercomputers for science, engineering, and medicine? Have we realized the first generation of fault-tolerant quantum computers? Does the United States, through its private sector partners, have the best open and proprietary frontier AI models? Have we built the most powerful and capable science and engineering models on top of those proprietary and open models? And have we built the most comprehensive and fastest-growing internet of science? These questions will be answerable. They're quantitative, they can be measured, and it is our goal as a nation and with our partners to realize and making sure that we lead and we deliver this capability for the scientific and engineering community of our country. And finally, and arguably perhaps most importantly, we gotta commit again to motivate and inspire a whole new generation of Americans to go into science and engineering. In 1957, following Sputnik, there was a piece of legislation that was passed called the National Defense Education Act. And when combined also with the Apollo program, a whole generation of Americans entered science and engineering. Enrollment in PhD programs following the passage of NDEA increased 10% per year for a decade. So it is time for us to also consider the ambition of what will we have to do to motivate a whole new generation of Americans to go into science and engineering. And for that, we need deep partnership with our university colleagues and deep partnerships with Congress. Today, we're also announcing the expansion of the Genesis Mission Consortium beyond industry and philanthropy and our national laboratories to our university institutions. So I wanna encourage for universities listening to go to genesismissionconsortium.org. We have prepared documents for us to begin that dialogue on how to engage. We have spent months working collaboratively with many universities to design this. And if you have an interest in participating, you can join the interest form. And I also wanna give a plug to attend the congressional panel at 345 where they're gonna be some exciting dialogue on this topic of what are the right legislative actions also to motivate the next generation of Americans in science and engineering and also for the Genesis Mission. And I wanna close by the how. If that was the what, a national portfolio, a platform and programs to support participation in science and engineering, equally important is that we mobilize the full strengths of the nation's trillion dollar a year R&D ecosystem. Federal government spends 200 billion a year, private sector spends 700 billion a year and the rest is spent by the states, universities and philanthropies. So from the very beginning, we have created a consortium that can bring an aggregate investment and alignment to purpose and mission through the Genesis Consortium. We have also designed explicitly into our programs that collaboration is required. So if you saw that map before, you saw how many participating institutions were there, labs, universities, the private sector coming together. And then fundamental to the mission is the partnership with our allies. You're gonna hear very briefly a discussion also with the ambassador from Japan that is gonna have a discussion with Harriet Greene and you're gonna, Harriet Kong, and you're gonna see how powerful it is to have allies that have a shared sense of mission and purpose. So in the Genesis Mission Consortium, we have well over 40 industry participants and nonprofits. You can go to this website and see it on aggregate over $800 million worth of contributions towards the Genesis Mission have been put forth to phenomenal recognition of the power of collaboration. This is the partnership with Japan. You're gonna hear more about it, a billion dollar commitment over five years to invest both in platform, the research infrastructure, and national challenges to collaborate to solve problems that really matter to our respective countries. So I'll close now with a call to action. Look, there are few times where we have an opportunity to come together as a country to do something that is much larger than ourselves. It is in that context that we give the historical parallels to previous moments of mobilization. But this time, it's even larger and bigger on the number of participants that need to come together. And here's what I would say too. I am not arguing that this is the only national effort that we should do in science and technology. Far from it. But if we cannot agree on this one, that this computing revolution is transforming the practice of science and engineering, what is so obvious that we're all gonna agree on as the second and the third and the fourth? Let's demonstrate the potential where we see plenty of evidence that this is transforming every single field. And this is a moment also for us to come together to work across sectors, to work collaboratively and realize this premise. And once we learn how to do this again, that we can have a decentralized approach, but moments of unity and moments of working and doing really ambitious things together, we will learn that we can do a second and a third and a fourth and a fifth at a scale that we could never do before. Thank you. Congratulations to all the award winners. We're very proud of you. And we wish you all the success in the world. Thank you.