HyperCLOVA X, the generative AI of Naver Corp.’s cloud computing division Naver Cloud, outperformed top AI including OpenAI and Google in the recent evaluation of Korean AI performance KMMLU, which stands for Measuring Massive Multitask Language Understanding in Korean. The KMMLU system is best known for its comprehensive evaluation metric, led by a domestic research team HAE-RAE, which tests AI’s language and contextual understanding ability in Korean.
The evaluation consisted of 35,030 questions from 45 categories that require expert understanding from humanities to technology. Moreover, questions about all subjects that are internationally and generally required are included, and the questions for Korean was designed to comprehensively evaluate adaptability and depth of understanding, such as a Korean law and geography.
The fact that HyperCLOVA X surpassed one by one other world representatives makes it clear that they are making AI solution for Korean users. In particular, in a test of Korea-specific knowledge, it surpassed the future AI OpenAI GPT-4, not to mention the already released version. A Naver Cloud’s representative commented that they will upgrade it step by step to be the Sovereign AI solution that provides top-level security and performance.
