LG announced on the 18th that LG AI Research Institute’s “K-EXAONE” has passed the second-stage evaluation of the “Independent AI Foundation Model Project” overseen by the Ministry of Science and ICT.

Nicknamed “Dokpamo,” the Independent AI Foundation Model Project is a public-private cooperation initiative to develop a Korean-style large-scale AI model at the national level in order to reduce dependence on global big tech companies for AI utilization and safeguard Korea’s technological security. The project is being carried out in a survival format, in which the selected teams (companies) undergo step-by-step evaluations approximately every six months through next year. The two teams finally selected will receive various forms of government support, including ultra-high-performance graphics processing units (GPUs) and personnel expenses.

LG AI Research Institute stated that in this evaluation it focused on establishing a foundation for developing a flagship ultra-large foundation model capable of competing in the global market going forward. To this end, it expanded the scale to more than three times that of the previous model and took on technical challenges across the entire process of training and inference for ultra-large models. Through this, it is assessed to have secured a basis for competing with big-tech AI models such as Gemini, ChatGPT and Claude.

An LG AI Research Institute official emphasized, “This second phase is meaningful in that we secured a technology base that enables stable training and operation of ultra-large models, while at the same time improving the response quality and practical performance that users actually experience.”

Based on the technologies secured in this stage, LG AI Research Institute plans to further develop K-EXAONE into Korea’s representative frontier-level expert AI model. It also aims to secure a technological foundation that can compete with globally leading models, beyond merely providing AI optimized for the domestic environment.

Lim Woo-hyung, co-head of LG AI Research Institute, said, “Significantly scaling up the model size to compete on equal footing with global frontier-level models was an inevitable strategic choice,” adding, “With only small models, it is difficult to narrow the technology gap with global big tech, so we will continue to pursue the goal of realizing equivalent or better performance at the same scale tier.”

He continued, “The improvement compared with the first phase may appear somewhat modest, but this is a technological verification process that must be completed for LG to leap to the frontier-level AI it is pursuing,” and added, “In particular, this second phase focused on optimizing the training of an ultra-large model within a limited period, and based on the technical assets and know-how accumulated in this stage, we will successfully achieve our final target performance.”

LG AI Research Institute plans to hold a talk concert in about a month in relation to this achievement, where it will provide a detailed explanation of its future technology development roadmap.

Meanwhile, the Ministry of Science and ICT held a briefing on the 18th and announced that, in the second-stage evaluation of Dokpamo conducted for four teams, three teams — LG AI Research Institute, SK Telecom (SKT) and Upstage — advanced to the next evaluation. The evaluation was conducted by combining benchmark scores (40 points), expert assessment (35 points) and user assessment (25 points). Scores and rankings for each team were not disclosed.

Popular News