Tutor or impostor:Exploring computer practice teaching based on large language models
[Objective]Large language models(LLMs)have demonstrated remarkable performance in various fields,including natural language understanding,multimodal generation,and human-computer interaction.These advancements have led to profound changes in conventional educational paradigms,particularly in computer-related education.LLMs can serve as"personalized tutors"for students,tailoring their educational journey to meet each student's unique needs.However,the improper use of LLMs may lead students to rely too much on them,potentially diminishing their independent thinking and practical skills.This study elucidates the nuanced influence of LLMs on computer practice teaching,including the promise they hold and the challenges they pose.[Methods]This paper elaborates on the landscape of computer practice education,dissecting it into five levels:course experiments,course design,graduation projects,professional internships,and extracurricular and social engagement.Each level plays a critical role in the holistic development of students'skills and knowledge.This exploration continues with an in-depth examination of the working principles and core technologies of LLMs,demonstrating their advantages over traditional AI techniques in terms of content generation,logical reasoning,and domain adaptation capabilities.Based on these advantages,this study analyzes how LLMs can revolutionize computer practice teaching.It depicts a dual-edged sword:on the one hand,LLMs can act as an inexhaustible repository of knowledge,offer real-time,intelligent guidance,and provide personalized teaching services.However,they can engender a culture of cheating,foster cognitive inertia,and introduce cognitive biases among learners.[Results]The narrative is further enriched by a case study of the"Algorithm Design and Analysis"course taught by the authors,highlighting the tangible benefits and challenges of integrating LLMs into practical education.This course is the cornerstone of computer science education and serves as an ideal backdrop for evaluating the practical implications of these technologies.This study explores a set of application patterns emphasizing process reports,reverse engineering thinking,and centralized assessments.We show the initial benefits of this innovative teaching approach.Of the 46 students in the class who were free to choose whether to use LLMs,84.8%opted to use LLMs,and 51%of those who used them believed that LLMs were of great help.Notably,up to 93%of the students completed all practical assignments,achieving a higher completion rate than the previous semester.However,during the closed-book exams,the students'average scores were lower than their average homework scores,indicating that their independent programming ability could be negatively affected.[Conclusions]Although LLMs can serve as"personalized tutors"providing individualized guidance,they may also act as"impostors",undermining the development of students'problem-solving skills.The benefits of the strategic application of LLMs in enhancing students'learning experience are demonstrated through the lens of the"Algorithm Design and Analysis"course while cautioning against over-reliance risks.The findings suggest that when used judiciously,these technologies can enrich the learning process and foster a more engaging,effective,and tailored educational journey for computer science students.
large language modelspractice teachingalgorithm designonline judgecode generation
NETL
NSTL
万方数据
