LLMs Struggle to Detect AI-Generated Text in Computing Education

This paper is important because it highlights the unreliability of current LLMs in detecting AI-generated content, particularly in a sensitive area like academic integrity. The findings suggest that educators cannot confidently rely on these models to identify plagiarism or other forms of academic misconduct, as the models are prone to both false positives (flagging human work) and false negatives (failing to detect AI-generated text, especially when prompted to evade detection). This has significant implications for the use of LLMs in educational settings and underscores the need for more robust detection methods.