从语言模型到教学工具

ChatGPT是基于大规模数据训练的语言模型，其主要训练机制包括“下一个词预测” (Next-token prediction)、对齐 (alignment) 以及有人类反馈的强化学习机制 (Reinforcement Learning from Human Feedback, RLHF)。 本文从ChatGPT的工作原理出发，分析其在以当代中国文学与电影为选题的高年级中文课上的应用表现，包括课程设计、课堂讨论支持及对学生写作的反馈。本文认为， ChatGPT的工作原理决定了其在帮助学习者掌握词汇、语法、篇章方面的语言使用规则上可以发挥相当的作用。然而在中文学习的高级阶段，ChatGPT 的使用不利于深入文本内部的细读，其背后的伦理原则也有可能限制学生批判性思维的发展。此外，ChatGPT 设计的教学大纲有明显的随机性与事实性错误，仅适合作为课程设计初级阶段的参考。有鉴于此，本文提出了在高年级中文课上使用ChatGPT的策略，以期使之成为教学的有效补充与助力。

ChatGPT is a language model trained on large-scale data, with its main training mechanisms including “next-token prediction,” alignment, and reinforcement learning with human feedback (Reinforcement Learning from Human Feedback, RLHF). Based on ChatGPT’s operational principles, this paper analyzes its application in an advanced Chinese course focused on contemporary Chinese literature and film, including course design, classroom discussion support, and feedback on student writing. This article argues that ChatGPT’s operational principles enable it to play a significant role in helping learners master vocabulary, grammar, and discourse-level language usage rules. However, at the advanced stage of Chinese learning, the use of ChatGPT is not conducive to in-depth textual analysis, and its underlying ethical principles may limit the development of students’ critical thinking. Moreover, the teaching syllabus designed by ChatGPT contains clear randomness and factual errors, making it only suitable as a reference in the initial stages of course design. In light of this, the article proposes strategies for using ChatGPT in advanced Chinese courses, aiming to make it an effective complement and aid in teaching.