Core Concepts
The author proposes using diagrams and videos to counter the over-reliance on large language models (LLMs) in object-oriented programming (OOP) exercises, aiming to foster problem-solving skills and deter unethical practices among students.
Abstract
The content discusses the challenges posed by LLMs like GPT and Bard in solving complex OOP exercises, raising concerns about academic integrity. The authors propose a novel approach using diagrams and videos to enhance student engagement, understanding, and motivation. Results from a survey of 56 students indicate positive responses to these new methods, with students showing less inclination towards relying on LLM-based code generation tools. The study also evaluates the limitations of LLMs in interpreting diagram-based exercises compared to text-based descriptions.
The proposed notation system includes various types of diagrams for different OOP scenarios, such as algorithmic function diagrams, state-change function diagrams, class declaration diagrams, inheritance diagrams, and state transition rules diagrams. Videos are suggested for more complex assignments requiring user interaction or input validation. The study highlights the benefits of these visual aids in promoting critical thinking skills and deterring unethical use of LLMs.
Experimental results show that students prefer diagram-based exercises over traditional text-based ones but find video-based projects more motivating. The content also explores the impact of these new formats on students' reliance on LLMs for code generation. Additionally, experiments with GPT-4 and Bard's vision capabilities reveal their limitations in accurately interpreting diagram-based exercises.
Overall, the content emphasizes the importance of innovative teaching methods to balance technological advancements like LLMs with authentic learning experiences in programming education.
Stats
Recent work has shown that LLMs can effectively solve a range of more complex object-oriented programming (OOP) exercises with text-based specifications.
Student perceptions were explored through a survey (n=56).
Students invested more effort in understanding the diagrams and felt more motivated to engage with video-based projects.
Experiments revealed that GPT-4 and Bard currently fall short in interpreting diagram-based exercises accurately.
Quotes
"Students responded positively to diagrams and videos."
"Video-based projects were better received than diagram-based exercises."