The Korea Atomic Energy Research Institute (KAERI) announced on the 11th that it has developed “AtomicGPT,” the world’s first AI large language model (LLM) specialized for the nuclear sector. As AI technology is broadly applied in bio-industry research, such as protein studies, the institute is proactively introducing AI into the nuclear industry to enhance R&D efficiency. AtomicGPT, which supports nuclear R&D, has been made available on the open-source platform Hugging Face.
AtomicGPT comes in two versions, one with 8 billion parameters and the other with 70 billion. The number of parameters is an indicator of AI model performance, with larger numbers representing more advanced AI capabilities. The smaller version is designed for fast and lightweight usage, while the larger version is intended for responding to complex queries. The model aids researchers by utilizing extensive knowledge accumulated in the nuclear field and training on data like open-access papers, dictionaries, regulations, and reports. Experts with over 10 years of experience in nuclear R&D have enhanced the model’s vocabulary and instructions to improve its practicality.
AtomicGPT is expected to be used for key tasks in the nuclear industry, including report writing, regulatory compliance review, technology verification, standardization processes, and configuration management. KAERI is also researching the application of AtomicGPT to the “Reactor AI Operator,” which controls nuclear power plant simulators. This would involve predicting accident scenarios based on real-time reactor data and assisting human experts. Yoo Yong-kyun, head of the AI Application Research Lab at KAERI, and Professor Lim Kyung-tae from Seoul National University of Science and Technology participated in the development of AtomicGPT.