TAIPEI (Taiwan News) — Two domestic AI models received full marks for societal values and ranked among the top 10 in the latest evaluation of 188 language models, the AI Evaluation Center said Wednesday.
AIEC evaluated the models in three areas: language, social and cultural context, and values, per CNA. Yating Intelligence’s Yating-FedGPT-1.10.18 ranked fourth overall, while Accelerate Private Machine Intelligence Company’s ACE-1-24B-2604 ranked 10th. Foxconn Research Institute’s FoxBrain v2.0 ranked 12th and Taiwan AI Cloud’s Llama3.3-FFM-70B ranked 48th.
Digital Affairs Minister Lin Yi-ching (林宜敬) said sovereign AI has two main goals: developing AI models that understand Taiwan, and building computing centers that allow them to operate under the law. The ministry uses questions from the past five years of the Chinese and Social Studies sections of the General Scholastic Ability Test to assess models’ understanding of social and cultural context, per Liberty Times.
Lin said companies performed strongly alongside major international AI developers including OpenAI, Anthropic, and Google. He said the four companies were impressive, adding that the results show the country can develop and train its own high-performing AI models.
Administration for Digital Industries Director-General Lin Chun-hsiu (林俊秀) said AI companies can submit their models for evaluation free of charge, with results made public only with their consent.