On the 15th, the government knocked Naver Cloud and NC AI out in the first round among five elite teams participating in the "national AI model project," and pulled out a "consolation round" card. The initial plan was to eliminate only one of the five teams, but two teams were cut after Naver Cloud's originality issue surfaced. The government said it would give another chance to all corporations, including the elite teams eliminated this time as well as corporations that failed to make the elite team selection in the past. But despite the government's generosity, Naver, NC AI and Kakao, which already drank the bitter cup of elimination, declared they would not try again. It is questionable whether the consolation round can draw interest.
◇ Stricter originality bar, focus on in-house AI roadmap, lower reliance on government
Summing up industry officials' comments, the reasons corporations are reluctant to try again are the government's originality assessment that is stricter than global standards, concerns that a failed retry would damage corporate image, and a strategic judgment to focus on their own AI development roadmaps and reduce reliance on government projects.
In particular, the industry believes that the exclusion of Naver Cloud—which most experts did not expect to be eliminated in the first round—and Naver's statement that it has no intention of retrying show that there is little to gain from joining the project. With Naver Cloud choosing to focus on its own roadmap, large companies with relatively more capacity to secure resources are more likely to choose to strengthen their own capabilities rather than rely on government rules and a consortium format that requires mutual coordination. The idea is to create AI models that are chosen globally through autonomous innovation. Motif Technologies and KT, which were eliminated from the national team selection, did not make the finals, but later accelerated technology development and achieved top-tier results at "Artificial Analysis," a global AI performance evaluation body.
Right after the first-round results were announced, internet users evaluated corporations that said they would not try again by saying, "Isn't it better not to be swayed while taking government money," "Corporations should pursue their vision and maximize profits," "It's better not to participate in a national AI without proper standards," and "They can be free from government pressure, and this could be better for business."
◇ Ambiguous standards and shifting rules fuel uncertainty
In fact, the government's national AI project was a hit at first. In addition to Naver Cloud, Upstage, SK Telecom, NC AI and LG AI Research, Motif Technologies, Kakao, KT, Konan Technology, KAIST, Lunit, Bionexus, Scionic AI, Junction Med and Pion Corporation consortia rushed to compete to participate in the first round. The mood was not bad. That is because the national goal of reducing reliance on foreign AI models and securing base AI applicable to domestic industry and the public sector, large-scale policy and financial support, and the aim of promoting AI that people can feel in their daily lives aligned well.
The mood shifted after the new year. After the Dec. 30 presentation of the first round of the "independent AI foundation model project," controversies over borrowing Chinese models erupted for three of the five teams that unveiled models. Critics said the government should clarify the standard for "from scratch" (developing an AI model directly from the beginning). But the government did not specify clear standards, citing the ongoing evaluation period.
On the 15th, the day the stage-one results were announced, Vice Minister Ryu Je-myeong of the Science and ICT Ministry explained the from-scratch standard, about which doubts had been growing. Ryu said, "Technically, even if you use open source, initializing the weights and then conducting training and development is the minimum condition to secure a model's originality," adding, "You must develop the AI model entirely with our technology, or use open source without license restrictions to develop and advance it on your own, and you must be free from external control or interference arising from the use of open source."
It is true the standards became clearer than before. But corporations are hesitating to participate due to the government's elastic standards. Kim Kyung-man, director general for AI policy at the Science and ICT Ministry, also said of the second-round criteria, "For the from-scratch aspect, we will further gather opinions from academia, industry and experts, and make the gradations and scoring more specific." He said uncertainty would still remain in the second round.
The industry reaction is that it is hard to understand suddenly holding a consolation round in a cutthroat competition system. One participating corporation said, "We cannot actively protest to the government as the referee, but in the end, wasn't the consolation round intended to bring in Naver, which is capable but fell short on the originality standard?" adding, "We worry about becoming window dressing for elastic policy."
Lee Kyung-jun, a business administration professor at Kyunghee University, said, "I understand the government's intent to give a retry opportunity, but changing the rules of the game is not desirable," adding, "When rules change, it is easy to be misunderstood as being influenced by external actors."
◇ Self-praise for "activating AI competition"… corporations and government at cross purposes
While corporations' deliberations deepened, the government gave itself credit, saying that through the national AI project, a mood has formed across Korea's AI industry that Korean AI can do it, expanding all-around cooperation and healthy innovation competition among domestic AI corporations, academia and research circles. As a prime example, Clement Delangue, CEO of the world's largest open-source community, Hugging Face, introduced that three foundation models released by the (Korean elite teams) recently made the trending list, calling it "the result of national support for open-source AI." The government also cited as a key achievement that all five elite teams were listed among Epoch AI's "notable AI models." An industry official said, "It's not that the national AI project created a competitive atmosphere; rather, in the AI era, every corporation is going all-out to create opportunities in AI."
In fact, the government and participating corporations also have different goals. Most of the five elite teams in the first-round evaluation of the national AI project say they joined because of the "national AI" title. If selected as the final two elite teams, the marketing effect would be significant. One participant even said, "If we are eliminated, we can't use the 'K-AI' certification mark, so we should use it more while we can."