[The National AI Team Trapped by the Independence Dilemma] "Reference Allowed"... Government Proposes a Compromise ①

"Independent K-Architecture Is Not Easy"
Realism Emerges
Government: "Open Models Are Allowed... Must Prove Unique Technological Capabilities"

Editor's NoteThe government's "Korean Proprietary AI Foundation Model (Dokpamo) Project," often referred to as the national team selection for artificial intelligence (AI), is embroiled in controversy over its "proprietary" nature. There is growing criticism that the "from scratch" requirement-meaning everything must be developed in-house from the ground up-was not clearly defined during the first round of evaluations for the proprietary AI project. This has led to increasing calls for the government to clarify its position. The Asia Business Daily has examined the industry's perspectives on the controversy from multiple angles and explored ways for the Dokpamo Project to ultimately serve as a stepping stone for the development of Korea's AI ecosystem.

"It is permissible to use global open models. However, during the development process of artificial intelligence (AI) models-including pre-training-our own differentiated development capabilities must be clearly demonstrated." (Senior official at the Ministry of Science and ICT)

As the government’s "Korean Proprietary AI Foundation Model" project faces backlash within the AI community over the definition of "proprietary" during the first round of elite team selections, the government has issued a clearer interpretation of the proprietary standard. Amid criticism from developers and AI startups-such as accusations that some elite team candidates copied Chinese models-demands for clarification of the proprietary standard have grown louder, prompting the government to provide a kind of guideline.

Even now, ahead of the second round of selections, debate continues over the proprietary standard. Some argue that, since the final selected teams will receive substantial government funding and GPU resources, there must be a principle that only a truly "proprietary" architecture-unique to Korea-should be the foundation. However, others counter that, given the widespread use of proven large language models (LLMs) from the United States and China as open models, it is unrealistic for latecomers to design a completely proprietary K-architecture from scratch.

Should it be a completely proprietary design VS Open model use is standard in development... Government proposes a compromise amid controversy

Of the five teams that competed in the first round (LG AI Research Institute, Upstage, SK Telecom, NAVER Cloud, NC AI), none achieved a fully proprietary model design-only degrees of difference existed. Even LG AI Research Institute's "Exaone," which received the highest score, was not entirely free from controversy. This has led to calls, especially from the developer community, for the government to present firmer evaluation criteria. As the debate over proprietary standards persisted, the government took the unusual step of reopening the competition beyond the three elite teams selected in the first round (LG AI Research Institute, Upstage, SK Telecom), and Motif Technologies and Trillion Labs entered the race, emphasizing their proprietary architectures as differentiators.

On January 27, a senior official from the Ministry of Science and ICT emphasized, "Institutions applying for the competition must prove their capabilities and experience in AI model development-covering everything from model design, such as loss log, through pre-training, to global and up-to-date benchmark evaluations, and listings on major global leaderboards such as those by Artificial Analysis, an overseas AI research institute."

A representative from the AI industry, speaking on condition of anonymity, said, "It appears that all five companies that participated in the first round referenced different Chinese architectures as their starting point, and some models simply adopted those structures without much modification. However, even though LG AI Research Institute’s Exaone, which ranked first in the initial evaluation, cannot be considered a model with a completely original, proprietary, or entirely new concept, it did make significant optimization efforts compared to other companies, and there were attempts at new combinations."

President Lee Jaemyung also appeared to acknowledge the debate over proprietary AI in the industry. On January 25, he shared an analysis by Artificial Analysis on his X (formerly Twitter), which stated, "Thanks to Korea’s state-led AI initiative, several Korean AI research institutes have emerged with cutting-edge AI capabilities." He encouraged the industry, saying, "If you can’t avoid it, enjoy it." Lee Jehui, professor at Seoul National University's Department of Computer Science and Engineering, commented, "In the LLM industry, it is common for competitors to learn from and imitate each other’s work, continually updating performance. In a development environment where open model use has become standard, it is unreasonable to overemphasize proprietary standards without flexibility."

A senior government official stated, "The government fully recognizes that it is difficult to create a completely proprietary model. However, with foreign LLMs dominating the domestic market, building sovereign AI is extremely important. If fierce competition helps teams build their capabilities and this ultimately upgrades the competitiveness of Korea’s AI ecosystem, it is a meaningful endeavor."