With Naver Cloud, which was eliminated in the independent AI foundation project competition, deciding not to join the repechage, Kakao also chose not to make another attempt.
Kakao on the 15th said it "decided not to review" whether to participate in the additional call for the independent AI foundation project.
The Ministry of Science and ICT on the day selected three teams—LG AI Research Institute, SK Telecom, and Upstage—to advance to the second round. Naver Cloud and NC AI were eliminated.
However, the Ministry of Science and ICT introduced a "repechage" system, deciding to give another chance to the Naver Cloud and NC AI consortiums that were eliminated in the first evaluation, as well as Kakao, Motif Technologies, KT, Konan Technology, and Korea Advanced Institute of Science and Technology (KAIST) consortiums that were eliminated when the five elite teams were selected. One additional elite team to be selected will be guaranteed computing and data support and the use of the "K-AI corporations" designation, just like the three consortiums that passed the first evaluation on the day.
Following this policy, the industry projected that Kakao could make another attempt. Kakao recently released the performance of its self-developed hybrid multimodal language model "kanana-v-4b-hybrid," which scored 92.8 on "KoNET," a Korean AI capability evaluation benchmark.
However, Kakao reached a final decision not to take part in this new call. In the industry, some said that the burden would not have been small if it had tried again and been eliminated again.
Earlier, Naver also said it would accept the government's first-round evaluation results and decided not to join the repechage. Naver Cloud said it "respects the Ministry of Science and ICT's judgment" and "will continue multifaceted efforts to enhance AI technology competitiveness."
Regarding the decision on Naver Cloud's elite team, the Ministry of Science and ICT said, "Even if an open-source model was used, it should have been verified that the weights were trained with data secured on their own, but issues were raised on the technical side regarding the part where the weights were used as is."