Effective Utilization of AIGC in University Computer Education

Education, as a key element of social progress and individual growth, has been constantly evolving and reforming. The traditional education system primarily focuses on knowledge transmission, carried out through standardized and scaled classroom teaching activities^[1]. In such a mode, the teaching content is relatively uniform, making it difficult to meet individual differences among students and not conducive to stimulating students’ creative thinking. With the rapid development of technologies such as artificial intelligence, the degree of digitalization in various fields is gradually deepening, making the digital transformation of higher education an inevitable trend.

AIGC (Artificial Intelligence Generated Content) is a generative method based on artificial intelligence technology. It learns and trains on a large amount of data, discovering rules within and automatically generating content with a certain quality and creativity. The scope of knowledge in computer science is broad, and it also requires students to have certain mathematical abilities and logical thinking skills. In addition, computer majors are highly practical, and students often have a large number of algorithm design and project development tasks. Based on these characteristics, the emergence of AIGC may provide significant support for learning in computer majors, offering students broad and deep knowledge support and bringing new methods and means to teaching activities. However, in this process, AIGC also presents some challenges, including how to ensure that the content generated by AIGC meets educational standards and how to balance the collaborative work between artificial intelligence and human teaching. As AIGC technology continues to improve, computer educators also need to more actively guide students in the correct use of these tools, making them truly beneficial auxiliary means for education, bringing positive and sustainable changes to the educational cause.

In November 2022, OpenAI released ChatGPT, a multi-turn dialogue AI robot fine-tuned based on the large language model GPT-3.5. With its outstanding text generation capabilities and interactive experience, the launch of ChatGPT immediately sparked widespread discussion, from basic natural language processing tasks to comprehensive virtual assistants, from personal daily conversations to applications in professional fields such as education and healthcare, all expected to achieve more efficient and intelligent solutions through ChatGPT. The GPT series models that ChatGPT is based on have undergone multiple updates since their first release in 2018, and many companies and research institutions have released large models and related products, providing a solid foundation for the development of AIGC technology.

AIGC technology has brought many new opportunities, and how to apply these AIGC technologies in various fields of production practice to enhance work efficiency has become a question explored by more and more people. In the field of education, many efforts have already been made, mainly adopting the data + fine-tuning paradigm^[2], and relevant studies have already realized the implementation of AIGC in education. EduChat^[3] is a large-scale bilingual educational dialogue model that can provide automatic question generation, homework grading, and other functions in educational scenarios. Literature^[4] focused on the shared task of BEA 2023, generating better AI teacher dialogues through fine-tuning open-source large models. Literature^[5] developed and open-sourced the G-LLaVA model to enhance the large model’s ability to solve geometric spatial mathematics problems.

2.1 Knowledge Acquisition

The emergence and development of AIGC technology have opened up a brand new era of knowledge acquisition. Generative large models trained on massive data possess a strong knowledge reserve, enabling deep understanding of semantics, considering context more comprehensively, and understanding the questions posed by users to generate answers in a more intelligent manner. The development speed of the computer field is very fast, facing knowledge updates almost daily. Connecting large models to the internet for knowledge search can provide students with computer knowledge that is both deep and timely, effectively assisting them in acquiring and understanding the latest developments in computer science in a rapidly changing technological environment.

In addition, the computer field encompasses a large number of highly specialized academic terms and expressions. The introduction of AIGC provides students with the convenience of interacting in natural language. During the questioning phase, they can express doubts or explore new knowledge in natural language without focusing too much on specific keywords or search techniques. After receiving students’ questions, large models can intelligently conduct information retrieval and integration, returning results in natural language, presenting intuitive and comprehensible answers to students, which aligns better with people’s communication habits. Such natural language interaction can eliminate technical barriers in the knowledge acquisition process, improve user experience, and provide students with learning support that is both intelligent and closely aligned with actual needs.

For example, using two computer science knowledge topics to demonstrate the effect of obtaining knowledge from ChatGPT. In Example 1, asking “Briefly compare the advantages and application scenarios of traditional relational databases and NoSQL databases,” ChatGPT’s response is shown in Figure 1 (a). In Example 2, asking “Please give examples of how to convert binary original code, inverse code, and complement code,” ChatGPT’s response is shown in Figure 1 (b).

2.2 Knowledge Extraction and Organization

Organizing and summarizing knowledge can help students quickly grasp the key points of a subject. Common tools include mind maps, structured tables, knowledge graphs, etc., but manual construction is time-consuming and labor-intensive. To address this issue, generative technology provides students with efficient solutions.Large models possess a certain level of semantic understanding, capable of identifying knowledge entities and relationships, transforming a large amount of textual information into structured information, and automatically organizing key information and summarizing knowledge points. Furthermore, large models themselves have a wealth of common knowledge, which can supplement common knowledge to book-based maps, allowing students to not only organize book knowledge in a short time but also understand more related knowledge in the discipline. The subject knowledge in the computer field often has a certain structure or context. For knowledge extraction and organization tasks, AIGC can play a significant advantage. Taking the textbook “Big Data Storage: NoSQL” published by the author as an example, AIGC can organize knowledge as shown in Figure 2. First, the text content of the course textbook and related materials are input into the large model, and the demand for organizing knowledge is raised. The large model returns the knowledge structure in Markdown syntax. Markmap is a tool that visualizes Markdown into mind maps. After obtaining the Markdown statements generated by the large model, students can improve upon them, and then use the Markmap tool to convert them into mind maps, providing great assistance in organizing knowledge during the learning process.

For example, asking the question “Briefly summarize the knowledge points in the data storage section, using Markdown format to answer” demonstrates the effect of knowledge extraction and organization by the large model, as shown in Figure 3.

2.3 Intelligent Creation

In the learning process, students often need to create content, and during their study of computer science, they also face a large amount of code writing. Large models provide creative assistance to students through their powerful generation capabilities.

Academic research reports, project documents, and other files often have commonly used writing structures. For students writing related documents for the first time, large models can help them follow this generic structure, ensuring the professionalism of the documents. In addition, large models can help students generate properly formatted emails based on specific contexts, from greetings to the body and closing, ensuring that the emails are clear and properly formatted to meet the needs of different occasions.

For programming tasks, large models can also play a significant role. Large models possess the ability to write in various programming languages; students only need to provide basic requirements, and the model can generate code that complies with syntax norms. This greatly lowers the threshold for students who are just starting to learn programming, and the professional naming conventions and modular code structure also help students develop good coding habits. When errors occur during code execution, large models can also assist students in modifying code and fixing bugs, significantly improving coding efficiency. Furthermore, large models can analyze and explain code snippets, enabling students to better understand and learn from open-source code.

For example, two examples demonstrate the assistance of ChatGPT in creative tasks. In Example 3, asking “In the MongoDB database, there exists an employee information table (id, empId, name, depId, pos, entryTime, ownMess), where ownMess contains four fields: sex, phone, mail, hobbies. Please query the names of employees who like both table tennis and basketball, requiring an aggregation query,” ChatGPT’s response is shown in Figure 4 (a). In Example 4, asking “I want to apply to participate in the offline presentation event held by XX University in China, and state that the session I am participating in is the Beijing session. Please help me write an application email in English, politely, around 100 words,” ChatGPT’s response is shown in Figure 4 (b).

2.4 Personalized Learning

Computer major courses have a certain level of difficulty, and students’ mastery of knowledge can vary significantly. Teachers often face hundreds of students and cannot provide targeted teaching. At this time, large models can play a great role. When students encounter difficulties in their after-class studies, they can ask large models for timely and effective guidance, with the “smart assistant” outside of class complementing the teacher’s instruction in class, providing each student with a more beneficial learning path. LangChain is a framework that helps build applications based on large language models. The content shown in Figure 5 is a non-relational database course assistant system built based on LangChain, capable of answering students’ questions outside of class and helping to construct mind maps to assist students in their review process.

Arizona State University (ASU) has reached a technological cooperation with OpenAI to bring the advanced features of ChatGPT Enterprise Edition into higher education, setting a new benchmark for how universities can improve learning, creativity, and student outcomes. Meanwhile, ASU has also launched a course called “Prompt Engineering” to help students deeply learn about prompts, thereby asking ChatGPT more precise questions to obtain high-quality answers. Such attempts enable students to receive more targeted support when thinking and solving problems, which may further promote the integration of advanced generative technologies into higher education and facilitate their broader application in the field of education in the future.

3.1 Uncertainty in AIGC Knowledge Generation

Large models are trained on a vast amount of data and may encounter contradictory or inaccurate information during the learning process^[6]. The content they generate may be influenced by errors or biases present in the training data. Since the model cannot effectively judge which information is correct or credible, dirty data from the training set may mix into conversations with users, and this inaccuracy may mislead students into learning incorrect knowledge and information, leading to misconceptions and even misleading behaviors in academic or practical applications. Moreover, the knowledge reasoning process may also have issues. When students have limited understanding of a particular content area, they may not be able to accurately assess its accuracy. Over time, incorrect knowledge may become deeply rooted in students’ cognitive frameworks, making it difficult to correct and impacting their future learning and development.

Inaccuracies in knowledge may also exacerbate confusion in academic fields. Students may learn correct knowledge in class, but when using generative large models, they may encounter incorrect information, leading to doubts about the consistency and norms within the discipline. This confusion may hinder students’ comprehensive understanding of the disciplinary knowledge system and affect their ability to establish correct associations within the knowledge structure. To address this issue, the education sector needs to implement a series of measures emphasizing the cultivation of students’ critical thinking and information verification skills, as well as introducing more human guidance and supervision in teaching.

3.2 Student Dependency

The widespread use of generative large models in education may lead to excessive dependency among students on these models. Students may habitually rely on the intelligent outputs of the models, affecting their ability to engage in self-directed learning, problem-solving, and creative thinking, leading them to focus more on immediate answers while neglecting in-depth understanding of foundational concepts in the discipline. At the same time, over-reliance on such models for completing assignments may increase the risk of plagiarism, as students may simply integrate generated content into their assignments or papers without a deep understanding and critical thinking about the materials used. Early reports indicated that 89% of American college students used ChatGPT to complete assignments. Furthermore, some students even attempted to use it to cheat on exams, letting machines do the work that should have been done by themselves, violating the essence of education.

In a digital learning environment, more and more students are becoming overly reliant on personal learning and AI tools, neglecting the importance of research communication with teachers and peers. The collective wisdom and collaborative power often surpass individual efforts. Engaging in research discussions with those around them not only promotes knowledge sharing and collision but also helps expand thinking and stimulate innovation. AIGC should serve as an aid in learning and education rather than replace collective discussions and teamwork. Students should recognize the balance between independent learning and teamwork, as this comprehensive learning approach benefits not only their academic pursuits but also lays a solid foundation for future teamwork and careers.

In computer education, excessive reliance on generative large models is most notably reflected in students seeking model-generated code first when facing programming problems instead of attempting to solve them themselves. The practical nature of programming is strong; by personally writing code, debugging, and improving programs, individual programming skills and problem-solving abilities can be enhanced. While large models can provide significant help during programming, when students become overly dependent on generative large models, they may abandon personal thinking and attempts, focusing more on obtaining correct code rather than deeply understanding program structure, algorithms, and design principles. This phenomenon not only affects the enhancement of students’ independent thinking abilities but may also lead to misunderstandings about programming practice. Such a learning approach lacks depth and is unlikely to cultivate students’ solid programming foundation and practical problem-solving abilities.

ChatGPT has indeed gained wide recognition and application in a short time, but as its use increases, tighter regulation and guidance are also needed. Educators should guide students to use AI tools reasonably, treating them as auxiliary tools rather than dependencies, to ensure that the core values of academic integrity and independent thinking are not compromised.

AIGC has brought rich opportunities to the field of computer education, including knowledge acquisition, knowledge summarization, intelligent creation, and personalized learning, providing students with a more efficient, innovative, and personalized learning experience. However, challenges such as knowledge inaccuracies and students’ excessive reliance on artificial intelligence have also emerged. With the rapid updates and iterations of AIGC technology, it is expected that AIGC will have broader application prospects in computer education in the future. In this regard, both educators and students should maintain an open and rational attitude, proactively facing challenges and solving problems to maximize the opportunities that AIGC brings to the field of university computer education, shaping a more innovative, efficient, and beneficial learning environment, and ensuring that AIGC truly becomes an aid to education, enabling students to better embrace the knowledge era of the future.

References:

Funding Project: National Natural Science Foundation of China General Project “Key Technologies for Generative Adversarial Network Space Scene Generation with Embedded Spatial Constraints” (42371476); Basic Research Business Expenses Fund Project of Central Universities “Key Technologies for Intelligent Prediction of Chemical Equipment Health Driven by Big Data” (buctrc202132).

First Author Profile: Guo Danhuai, Male, Professor at Beijing University of Chemical Technology, research areas include spatiotemporal data analysis, big data processing, data intelligence, [email protected].