In a groundbreaking collaboration, Atomic Canyon, a leading force in generative AI for the nuclear energy sector, and the Department of Energy’s Oak Ridge National Laboratory (ORNL) have unveiled the initial results of their joint venture. Within a mere six months, the team has successfully developed a sophisticated AI model capable of deciphering intricate nuclear terminology, harnessing the power of the world’s fastest supercomputer, Frontier.
This specialized, open-source AI model has established new benchmarks for accuracy, efficiency, and speed in AI-powered search within the nuclear domain. Designed for open access, the model will be readily available to ORNL, the nuclear national laboratory complex, independent researchers, and nuclear institutions. Furthermore, it will be seamlessly integrated into Neutron, Atomic Canyon’s AI search platform.
Based on a novel, industry-first evaluation benchmark derived from real-world user queries, the model consistently delivers the correct search result within the top ten results 98% of the time and within the top five results approximately 93% of the time. These independently verifiable initial results have solidified Atomic Canyon’s vision of an open-source search platform that will streamline nuclear processes and regulatory review.
“Atomic Canyon has developed an AI model that will help propel nuclear energy development,” asserts Thomas M. Evans, Group Lead and Distinguished R&D Staff at ORNL. “Their achievement is significant: it will be an invaluable tool for the research, engineering, and deployment of nuclear energy, paving the way for a future powered by sustainable energy.”
“Our collaboration with ORNL and the utilization of Frontier have enabled us to create a tool that is not merely a research milestone but a practical, commercial application that will transform data management within the nuclear sector,” remarked Kristian Kielhofner, co-founder and CTO of Atomic Canyon. “This AI model fundamentally reshapes how the nuclear industry interacts with vast, complex datasets, making data retrieval faster and more reliable than ever before.”
By leveraging state-of-the-art sentence-embedding algorithms, the model accurately interprets and processes complex nuclear terminology, achieving unparalleled performance metrics within the industry. The AI model has been engineered to handle larger contexts, facilitating more comprehensive processing of nuclear data and enabling faster, more accurate searches across extensive regulatory databases. Additionally, it employs the world’s first commercially usable SPLADEv2 sparse embedding model, setting new standards in AI-driven nuclear search.
“The use of AI technology to handle this essential, yet tedious, work is a meaningful first step along the lengthy path to licensing, and Frontier provides an exceptional resource to train these types of models,” remarked Bronson Messer, director of science for the Oak Ridge Leadership Computing Facility, home to the Frontier supercomputer.
Utilizing the Frontier supercomputer, Atomic Canyon trained the AI model on publicly available documents sourced from the U.S. Nuclear Regulatory Commission’s (NRC) Agency-wide Documents Access and Management System (ADAMS). By making the resulting AI model open source, Atomic Canyon aims to empower nuclear stakeholders, such as researchers and plant managers, to build upon their work and foster innovation for a safe, clean, and sustainable nuclear future.
In partnership with ORNL, Atomic Canyon will continue to refine and expand the model’s capabilities using Frontier over the next six months, with plans to release subsequent versions. For more information, visit Atomic Canyon’s website.
