In the first-round evaluation to select a “national representative artificial intelligence (AI),” Naver and NC AI were eliminated. Accordingly, LG AI Research, SK Telecom, and Upstage advanced to the second stage.

The Ministry of Science and ICT, the National IT Industry Promotion Agency (NIPA), and the Telecommunications Technology Association (TTA) on 15 January announced the first-round results of the “Independent AI Foundation Model (Dokpamo)” project. The government stated that, based on a combined assessment of benchmark, expert, and user evaluations, NC AI, which received the lowest score, was eliminated. Although the original plan was to eliminate only one team, Naver, which used Chinese technology, was also eliminated over the issue of “independence.” The government decided to set up a “wild-card round” to select one additional elite team at a later stage.

● ‘Low borrowing ratio’ explained, but ultimately eliminated

According to the evaluation results released by the government, in the assessment using 13 benchmarks, LG AI Research scored the highest with 33.6 points out of 40. In the evaluation committee’s in-depth review as well, LG AI Research led with 31.6 points out of 35. In the user evaluation, in which 49 AI expert users, including AI startup CEOs, participated, LG AI Research also ranked first with a perfect score of 25 out of 25. The government stated that, based on these combined results, LG AI Research, Naver Cloud, SK Telecom, and Upstage were the top four teams.

However, what determined Naver’s fate was the analysis of independence. Earlier this month, suspicions were raised on an open-source platform that Naver had used the vision and voice encoders—equivalent to the eyes and ears of an AI—from “Qwen,” Alibaba’s AI from China. Analysis also suggested that the “weights” of the encoder, which correspond to the outcomes of training, had been directly borrowed from Qwen. Naver immediately acknowledged these facts, but maintained that since it had independently developed the engine that serves as the “brain” of the AI, borrowing the encoder, which it characterized as a type of “component,” did not infringe on independence.

The government’s view differed. It stated, “In the global AI ecosystem, it is common to utilize open source, but the basic condition for an independent AI model, as generally accepted across the domestic and global AI industries, is to develop AI by initializing the weights and then training.” In other words, using the trained results produced by others as they are does not meet the government’s criteria for independent implementation of AI.

● One vacant team to be newly recruited… Naver and NC also get another chance

Although Naver was eliminated, controversy over fairness continues as the government has announced an unscheduled additional call to newly select one consortium. In line with its original plan to select four teams in the first round, the government intends to select one more consortium and conduct the second-round evaluation with four teams. In the first half of this year, three consortia will be shortlisted in this manner, and then, through a final evaluation at year-end, two elite teams will be selected.

Any capable company, including consortia eliminated in this evaluation, may apply for the additional call. However, Naver stated after the first-round results were announced that it “respects the decision of the Ministry of Science and ICT” and is “currently not considering reapplying.”

An AI industry official expressed concern, saying, “If the government fails to clearly present standards for assessing independence in the second-stage selection process as well, there is a sufficient possibility that the same issues seen in this first-round evaluation will be repeated.”

Popular News