生成式大语言模型能有效实现对话式教学吗  

Can Generative Large Language Models Effectively Implement Dialogic Teaching

作　　者：颜士刚 胡修磊 李文光[2] YAN Shigang;HU Xiulei;LI Wenguang

机构地区：[1]天津师范大学教育学部,天津300387 [2]深圳大学教育学部,广东深圳518060

出　　处：《现代远程教育研究》2025年第2期52-61,72,共11页Modern Distance Education Research

基　　金：2021年度深圳市教育科学“十四五”规划课题“慕课教学中批判性思维能力培养的理论与实践研究”(ybzz21002)。

摘　　要：生成式大语言模型从形式上实现了人与机器的自然对话,使古老的对话式教学成为现实可能,因此其自问世以来便受到教育界广为推崇。为了更好地了解生成式大语言模型在对话式教学应用中的适用性,有必要从其工作原理出发探索其在教育应用中的固有困难,以寻求其赋能对话式教学的适切路径。作为一种基于大规模语料库训练而具有自然语言对话能力的人工神经网络应用,生成式大语言模型本质上是一种基于数据处理的“聊天机器人”,本身存在缺乏理解能力、知识立场不坚定、语言仅是生成而非创造、难以满足个别化学习需求等固有困难。因此,无论从知识传授还是情感培养方面看,生成式大语言模型自身均因受到特定价值取向影响、缺乏策略性引导、难以做到因材施教而无法达成对话式教学的理想效果。即便如此,生成式大语言模型仍能为对话式教学提供适宜的应用场景,如为客观性知识类教学问题提供高效信息咨询服务,基于海量大数据为语言教学提供精准模拟和纠错服务,以及针对常规教学活动提供一般性的参考框架、范式或提纲等。Generative large language models have,in form,enabled natural conversation between humans and machines,making it a realistic possibility for the ancient dialogic teaching method.Therefore,since their advent,they have been widely praised in the education field.In order to better understand the applicability of generative large language models in the dialogic teaching,it is necessary to explore the inherent difficulties in the application to education based on their working principles,so as to find appropriate paths to empower dialogic teaching.As an application of artificial neural networks with the ability of natural language conversation based on the training of large-scale corpora,generative large language models are essentially“chatbots”based on data processing.They have inherent difficulties such as a lack of comprehension ability,an unsteady knowledge stance,the fact that language is merely generated rather than created,and the difficulty in meeting individualized learning needs.

关 键 词：人工智能 生成式大语言模型 对话式教学 个别化教学 

分 类 号：G434[文化科学—教育学]