Volume 44

OpenAI and the U.S. Department of Energy (DOE) teamed up to push the boundaries of AI-driven science, with over 1,000 DOE scientists engaged as part of this effort. On February 28, the 1,000 Scientist AI Jam brought together researchers from across the DOE complex—including over 300 at Lawrence Livermore National Laboratory (LLNL)—to test OpenAI’s cutting-edge o3 model. 

This event was a chance for scientists to tackle challenging problems with a state-of-the-art AI tool, share expertise, and explore how AI can transform research. Attending solo and as part of teams, participants engaged directly with the o3 model, providing critical feedback to OpenAI while gaining hands-on experience with advanced reasoning technology. 

This initiative is part of a broader partnership between OpenAI and the DOE National Laboratories, aimed at strengthening America’s leadership in AI. By combining human ingenuity with powerful AI systems, the collaboration seeks to accelerate innovation in areas like energy and national security. 

Read OpenAI’s press release for more. 

The recently accepted paper by the American Astronomical Society, "On Finding Black Holes in Photometric Microlensing Surveys," highlights LLNL’s cutting-edge data science capabilities and advances our understanding of isolated black holes and Galactic evolution. 

This research, conducted under the Laboratory Directed Research and Development program and led by Will Dawson, focuses on developing a novel Bayesian classification framework to identify isolated black holes in photometric microlensing surveys. The work addresses the challenge of detecting stellar-origin black holes (SOBHs) using lightcurve data from the OGLE-III and OGLE-IV surveys, leveraging LLNL’s expertise in data analysis and simulation. 

The classification framework, implemented in the publicly available popclass Python package (hosted on LLNL’s GitHub: https://github.com/LLNL/popclass), combines posterior constraints on microlensing parameters with Galactic simulations to estimate the probability of a lens being a black hole. This approach is efficient, scalable, and adaptable, making it ideal for large datasets from current and upcoming surveys like the Vera C. Rubin Observatory and the Roman Space Telescope. The study identified 23 high-probability black hole candidates and demonstrated the potential for improving resource allocation for follow-up observations. 

The lead author, Zofia Kaczmarek, was a 2024 DSSI intern at LLNL when this work was completed, showcasing the student program’s impact in fostering innovative research. The DSSI program is gratefully acknowledged in the paper, a testament to LLNL’s investment in developing future scientific talent.  

The LLNL Decision Superiority team is pioneering a groundbreaking approach to real-time decision support by combining LLNL’s high-performance computing (HPC) capabilities with innovative data-driven algorithms. This effort aims to provide decision-makers with actionable insights in high-stakes scenarios, creating a strategic advantage for the U.S. 

Their goal is to operate classified HPC systems at LLNL and partner labs, fueled by real-time global data. These systems are designed to address both strategic and operational decision challenges. At the heart of this capability is a novel algorithmic approach that outperforms traditional methods, including modern AI, in handling the exponential complexity of multi-step decision-making. 

The fundamental challenge lies in the combinatorial explosion of decision possibilities over time. Traditional methods, such as dynamic programming and reinforcement learning, mitigate but do not eliminate this complexity. While modern machine learning and large language models offer tools for decision support, they lack the ability to optimize decision processes effectively. 

The Decision Superiority team is developing algorithms that aim to surpass these limitations, offering a potential breakthrough in computational decision-making. Early demonstrations of our tools and workflows have been deployed across LLNL networks with promising results. 

The Livermore Public Library’s annual Livermore Reads Together program returns this March, focusing on AI, Robotics, and Engineering. The DSI was proud to contribute to this community initiative (attended by over 500 individuals) by offering engaging, AI-related content designed to educate and inspire attendees of all ages. 

On March 1, 2025, from 1:00–3:00 p.m., DSI hosted a variety of demonstrations and presentations at the Livermore Library (1188 S. Livermore Avenue), including: 

• Virtual Reality Demonstration: “Explore Additive Manufacturing in Virtual Reality,” led by Vuthea Chheang and Haichao Miao (for ages 8+) 

• Teen Talk: "A Brief History of AI and How You Can Use It to Be Awesome," presented by Paige Jones and Brent Deaver (for teens) 

• Storytime: “Where the Science Things Are,” hosted by Elisa Esme Abadi (for kids up to age 10) 

• Q&A Session: “How AI May Shape Science (And Everything Else),” led by Peer-Timo Bremer (for all ages) 

DSI’s contributions were part of a larger day of activities, including hands-on exhibits from Quest Science Center, Livermore Flying Electrons, and System Overload Robotics.  The event also featured “The Future is Now” art show in the gallery. 

Visit the library website to learn more about this month-long programming. 

As AI continues to transform industries and society, the need for informed legislation has never been greater. With California playing a leading role in AI innovation, LLNL is stepping up to provide expert guidance to lawmakers. In 2024, the Lab bolstered its government relations efforts by hiring Andrew Kosydar as Senior Advisor for State Government Affairs. A former California Council on Science and Technology Fellow, Kosydar is well-known in Sacramento and helps elevate conversations about LLNL’s mission. 

One key effort is AMALT (AI/ML Advisory Lightning Taskforce), a multidisciplinary group from LLNL’s Data Science Institute. AMALT provides unbiased technical reviews and guidance on AI legislation, ensuring proposed policies are grounded in scientific rigor and practical feasibility. The group has already reviewed one proposed bill and expects more in the spring. 

On February 4, AMALT members and Andrew visited the Capitol. Brian Spears presented to the Assembly Privacy Committee on LLNL’s role as the public conscience in AI development, emphasizing public-private partnerships and the Lab’s unbiased expertise. Lawmakers posed thoughtful questions on topics like OpenAI’s reasoning models, quantum computing, bias removal, AI safety, and LLNL’s training data. 

The group also met with key legislators, including Assemblywoman Rebecca Bauer-Kahan and Senators Roger Niello, John Laird, and Dr. Akilah Weber-Pierson.  

Lawmakers expressed gratitude for LLNL’s expertise, with Dr. Weber-Pierson noting, “To create better legislation, you have to have the right people at the table.” LLNL remains committed to supporting California lawmakers through AMALT. (Pictured at left: DSI director Brian Giera, Bremer, Spears, and Kosydar.) 

Uttara Tipnis is a staff scientist in the Computer Vision Group within the Computational Engineering Division at LLNL, where she applies machine learning to areas such as biological threat detection and traumatic brain injury identification from MRI data. With a PhD in industrial engineering from Purdue University, she joined the Lab as a postdoctoral researcher in 2021 after interning in 2019 and 2020. In addition to her own work, she mentored a student in 2023 who sought to find a connection between neural imaging and genetic biomarkers for Alzheimer’s. “I enjoy research for research’s sake,” Tipnis said, “but I really love applying data science to tangible problems in the bio field with very real-world implications.” Tipnis was the chair of the climbing networking group for two years, which she says has broadened her involvement in the Lab community outside of research. She has also been helping to organize the Women in Data Science (WiDS) event since 2023 and is happy to be contributing to the 2025 event on March 12.