2024 年 17 巻 p. 41-52
This study aims to generate medical language that is easy to understand and evaluates the feedback of GPT-4 on various tasks. Based on existing research, firstly, tests were conducted on simple and specific text generation instructions. GPT-4's feedback generally met the requirements, but the repetition rate of metaphor generation was relatively high. Subsequently, text templates were provided for GPT-4 and tests were conducted on imitation instructions. In response to imitation instructions, GPT-4's feedback generally met the requirements and generated ideal text for some content. However, it was found that some information generation was incorrect, and the stability of multilingual tasks was insufficient. In addition, this study believes that metaphor generation tasks may have certain value in assessing the language capabilities of artificial intelligence.