The Ministry of Science and ICT's elite team from Motif Technologies, which joined as a wildcard in the independent AI foundation model project to select the "national AI," was eliminated in the second-stage evaluation. The Motif elite team benefited from the wildcard after the government reversed its plan in Jan. to eliminate only one team in the first-stage evaluation.
The ministry said it will still select the final two teams in the third-stage evaluation scheduled for the end of this year, but it is completely reexamining how to provide support after the final selection. In industry and academia, there are concerns about policy continuity and predictability.
Ryu Je-myeong, the second vice minister of the Ministry of Science and ICT, held a briefing on the "results of the second-stage evaluation of the independent AI foundation model project" at the Government Complex Seoul in Jongno-gu, Seoul, on the 18th and said, "We are completely reexamining whether to continue support for the final two teams in 2027 in the current form," adding, "It may not be a structure of whether it is LG, SK Telecom or Upstage, and in reviewing frontier-level AI model development, we are considering linking or reorganizing with existing independent AI foundation model project, the artificial general intelligence (AGI) project, and other related programs."
An industry official said, "There were even rumors that the elite teams should be merged, but from the perspective of corporations, the criteria for competition and what support can be received until when must be clear to make long-term investment decisions," adding, "If the government keeps changing the competition method and support system without understanding the position of corporations that must pursue profit, corporations may find it hard to trust and follow government policy." The official added, "Because the independent AI foundation model project has no clear results, they keep changing the approach, and it feels like it is fizzling out."
◇ Even after two "national AI" teams are chosen, it's not over
It is not easy to question the fairness of the evaluation solely because the Motif elite team was eliminated. The second-stage evaluation compared LG AI Research's "KxOne 2.0," SK Telecom's "adot X K2 (A.X K2)," Upstage's "Solar Open 2," and Motif's "Motif-3," and proceeded with △benchmark (40 points) △expert evaluation (35 points) △user evaluation (35 points). Although the Motif elite team scored high on the Artificial Analysis (AAII) index, a global AI performance evaluation body, the government's evaluation this time was not simply a competition of model performance, but also looked at practical industrial applicability and the potential for ecosystem expansion. For this reason, the industry does not view the elimination of the Motif elite team as unexpected.
The issue is what happens after this elimination. The government had initially presented a broad framework of selecting the final two teams in the third-stage evaluation after the second stage and continuing support for them. However, with the government saying it will completely reexamine the method of support after 2027, even corporations selected as the final two teams now face uncertainty about how they will receive government support going forward. A senior official at a participating corporation said, "Before changing the staged evaluation method, the government does ask for opinions in advance, but frankly, even if you don't want to agree, who can raise an objection?"
Of course, the government also has its concerns. As the development pace of global AI models has accelerated faster than expected, it is seen as a judgment that it will be difficult to narrow the gap with advanced models from the United States and China with only the existing independent AI foundation model project. The Ministry of Science and ICT said, "In leading countries, moves to make AI models national security strategic assets are becoming a reality," adding, "The ministry also deeply recognizes the need to develop AI models and is discussing with relevant ministries a new support system that goes beyond existing approaches, and we will soon prepare and announce specific details."
Lee Kyung-jeon, a professor in the Department of Big Data Application at Kyunghee University, said, "Since the AI industry is growing fast, policies also appear to be changing in step," adding, "However, so far it is not so much that Korean elite teams are exceptionally strong, but rather that the pace of performance improvement in U.S. foundation models has slowed environmentally, allowing us to catch up to some extent. Going forward, we need to strengthen competitiveness so that, beyond ecosystem building, corporations and the public can actively choose Korean foundation models."
◇ Consensus on the need for independent AI models… "We shouldn't be overly optimistic about current results"
Experts agree with the independent foundation model project itself, but say the direction is somewhat disappointing. They say the current results should be viewed more cautiously.
Bang Hyo-chang, policy director at the Citizens' Coalition for Economic Justice, said, "We agree with the purpose of the independent AI foundation model project, but to secure competitiveness in the marathon of AI model development, we need a strategy to aggressively utilize open-source models that anyone can use within the scope of funding and permissions to break into the top tier," adding, "While the United States and China are locked in a race to develop ultra-large AI models with trillions of parameters, if the maximum parameter size of the four elite teams is only 750 billion, the government may end up patting itself on the back for running a good sprint in a marathon." He added, "Because there is a lack of confidence in the independent AI foundation model policy, it appears the government is supplementing policy and creating domestic demand for AI models by running the 'Everyone's AI' project to build a Korean AI chatbot that the entire public can use for free."