Analysis of Plagiarism via ChatGPT on Domain-Specific Exams

Abstract

This work presents a case study, linguistic analysis and potential prevention methods on the use of large language models (LLM) for generating solutions for exams on cloud computing course that require domain-specific knowledge. The study involves analyzing the responses of three groups of students: a group who used ChatGPT to plagiarize solutions, another group who referred to external non-LLM resources (e.g., web search) to plagiarize solutions, a control group who generated solutions without any external assistance. Results show that solutions from groups that participated in plagiarism tend to be lengthy, use uncommon words, and are similar to each other compared to human-generated solutions. This study not only shows that it is possible to generate legitimate solutions for exams that require extensive domain-specific knowledge using ChatGPT, but also shows some potential signals one can use to detect plagiarism, thus providing potential of promoting academic integrity by curbing unethical use of AI in academic settings.

Ähnliche Arbeiten