Abstract:

As technology continues to advance, the integration of artificial intelligence and natural language processing techniques has opened up new possibilities in various fields, including education. One such AI-powered tool that has gained significant attention is ChatGPT, a large language model capable of engaging in conversational interactions. This paper explores the potential applications of ChatGPT in computer science higher education.
We already integrated ChatGPT into several computer science lectures. For example, as part of the master's program in Computer Linguistics, students conducted an evaluation of ChatGPT's language capabilities. During their investigation, they identified several issues with ChatGPT, including gender and other biases, hallucination, coreference resolution, and domain-specific translation challenges. The students further delved into potential sources of these problems, considering factors such as training data limitations and model constraints. Additionally, they proposed potential solutions to address these issues and improve the performance of ChatGPT in these areas.
There are also several ideas how to further integrate ChatGPT into our courses. One new lecture concept is the integration of ChatGPT into networking courses. The main goal is to leverage its advantages and use it to better understand the configuration and operation of the networking devices. Students could try out to the recommended commands to solve the tasks about the setup and troubleshooting of networking devices. They can learn the right syntax and will have a deeper understanding due to the step-by-step instructions of this tool.
ChatGPT claims to be able to handle moderately complex codebases and refactoring tasks. To explore its limits, students confront ChatGPT with increasingly challenging assignments. This practices skills in code review and analysis, while also revealing the boundaries of what ChatGPT can achieve. In a "Prompt ChatGPT, Prompt Humans" approach, both the model and students are given the same task. Afterwards, the solutions are compared and discussed within the group. The results are then analyzed together.
Furthermore, AI can support computer science students in interdisciplinary topics. Artificial intelligence offers the opportunity to guide students along their previously learned domain to understand concepts of a new domain. Thus, the degree of individualization in learning new concepts can be increased. To implement this, predefined text blocks could be used, such as: act as an expert for operating systems and explain the term process using, for instance, a medicine example.
While ChatGPT and similar language models can be powerful tools for software development, they also bring certain challenges which shall also be addressed in our lectures. Examples are its lack of domain-specific knowledge and context understanding, potential security risks as well as ethical concerns and the dependency on internet connectivity. To mitigate these challenges and drawbacks, we advise to use ChatGPT as a supplementary tool and not rely solely on its outputs. We should combine its suggestions with our expertise, cross-reference with reputable sources, and maintain a critical mindset when interpreting its responses. Consequently, higher education will continue to play a crucial role for future software engineers.

Keywords:

Higer Education, Computer Science, Artificial Intelligence, ChatGPT.