The Ministry of Science and ICT is reviewing the restructuring and integration of existing AI projects, including the Independent AI Foundation Model (IndepAI), to better respond to the development of global frontier-level artificial intelligence (AI) models. The third phase of the IndepAI evaluation will proceed as scheduled, selecting two out of three elite teams, while a new competitive framework for developing frontier-level AI models is being planned.
On August 18, the ministry announced the results of the second phase evaluation of the IndepAI project. As a result, three elite teams—Upstage, SK Telecom (SKT), and LG AI Research—will advance to the next stage. However, Motif Technologies, which joined through an additional application, did not pass the second phase evaluation.
Along with the evaluation results, the ministry indicated it is considering restructuring the competition and support system for the IndepAI project. This decision stems from concerns that the current approach cannot keep pace with the rapid advancements in global frontier AI models.
Deputy Minister Ryu Je-myung stated, "The performance of frontier models from global companies is advancing exponentially," adding that discussions with businesses have focused on whether the current scale of support and competition can keep up. He noted, "There is a consensus on the need to shift from the current competitive model to one that allows us to compete effectively with frontier companies."
The ministry confirmed that the third phase of the IndepAI evaluation will proceed as originally planned, with no changes to the selection of two teams from the three elite teams.
Potential restructuring will be considered after this phase. The government is exploring ways to link and restructure existing projects, including the IndepAI, while reviewing plans for developing frontier-level AI models. The continuation of the fourth phase of the IndepAI project is also under consideration. The ministry plans to announce specific directions for the frontier-level model development project soon.
Deputy Minister Ryu mentioned, "We are discussing new competitive methods, resource concentration and distribution, and consortium formation with businesses," but noted that specific plans are difficult to disclose until the budget is finalized.
In the second phase evaluation, not only technical capabilities but also usability and applicability influenced the outcomes. The elimination of Motif, which performed well in global benchmarks, was attributed to these factors. Ryu explained, "Motif demonstrated high performance in the AAII global benchmark, proving its innovative algorithms and AI capabilities. However, it received lower evaluations in usability and applicability compared to other companies."
However, the ministry clarified that no single evaluation criterion determined the outcomes. The score gap between the first and fourth places in the benchmark evaluation was only four points, with expert evaluations showing a difference of 2.4 points and user evaluations differing by five points. The top companies varied across different evaluation areas. Ryu noted, "The results were a cumulative outcome of minor score differences across evaluations."
Additionally, the newly introduced public evaluation did not impact the final outcomes. About 1,400 people applied for the public evaluation panel, with 200 selected, and after conflict of interest checks and actual response verification, 185 ultimately participated. Ryu stated, "After thorough review, we found that the public evaluation did not influence the outcomes."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.