The government released additional details of the second-stage evaluation for the "independent AI foundation model project" whose disclosure scope had stirred controversy. It did not reveal the total scores and overall rankings of participating corporations, but as in the first evaluation, it disclosed the companies that received the highest scores in each category and their scores.
The Ministry of Science and ICT issued an explanatory note on the 20th and said it adjusted the disclosure scope in consideration of the fairness and transparency of the evaluation. Initially, it did not disclose category-by-category results, citing concerns that highlighting only the top corporation's recognition could stigmatize other corporations. However, after continued calls to transparently disclose the evaluation process of a project funded by the national budget, it reversed course.
This evaluation, with a total of 100 points, consisted of 40 points for benchmarks, 35 points for expert review, and 25 points for user evaluation. The weighting for originality and usability was raised from 10 to 15 points after the first evaluation.
In the AAII benchmark measuring global performance, Motif Technologies ranked first with 11.9 points out of 25. The average for the four teams was 9.48 points. The AAII assessment measured agents and coding, general performance, and scientific reasoning, and the external evaluation agency Artificial Analysis produced the results.
In the NIA benchmark covering seven areas including mathematics, knowledge, Korean, and safety, SK Telecom scored the highest with 13.4 points out of 15. In the evaluation where 10 experts reviewed development technology, future plans, and ecosystem contributions, LG AI Research Institute led with 29.5 points out of 35.
Results diverged in the real-use evaluation depending on the target group. Forty-nine expert users, including AI startup CEOs, gave SK Telecom the highest score of 11.6 points, while a national review panel of 185 people selected to reflect gender- and age-based demographics gave LG AI Research Institute the top score of 7.6 points. The Ministry of Science and ICT said the evaluation is a process intended to promote the technological advancement of homegrown AI and its on-site adoption, rather than to rank corporations.