KISTI Releases LLM for Research Institutes, SMEs, and Public Institutions

The Korea Institute of Science and Technology Information (President Kim Jae-su, hereinafter KISTI) announced on the 20th that it has developed KONI (KISTI Open Natural Intelligence, Korean name Goni) 13b, a generative large language model (LLM: Large Language Model) specialized in science and technology data and easy to use in government-funded research institutes and public institutions.

KONI was developed to facilitate the application of Retrieval-Augmented Generation (RAG) technology, which utilizes data held by existing information service systems to provide answers, in order to minimize hallucinations commonly occurring in LLMs.

KISTI plans to support LLM adoption for government-funded research institutes, public institutions, and small and medium-sized enterprises that have faced difficulties in adopting LLMs due to computational costs, security, and information leakage concerns. Additionally, the model’s performance will be continuously improved by increasing the model size beyond the current 13 billion parameters and securing additional training data.

Kim Jae-su, President of KISTI, stated, "We will establish an AI information service system at the level of an engineering doctorate that can be utilized by both the general public and professional scientists through KISTISMS, and in the future, we plan to expand its use to actual research by supporting hypothesis generation and scientific discovery through AI."

This content was produced with the assistance of AI translation services.