What is it about?
Advanced AI language models can now generate high-quality paraphrases that are hard to distinguish from original writing. This poses a significant threat to academic integrity. Our study examined how well two AI models (T5 and GPT-3) could paraphrase scientific articles, student theses, and Wikipedia content. We tested six automated plagiarism detection tools, one commercial software, and conducted a study with 105 human participants. The results show that humans struggle to identify AI-generated paraphrases, with only 53% accuracy. Experts rated GPT-3's paraphrases as high-quality as original texts. The best AI detection model achieved 66% accuracy in identifying these paraphrases. These findings highlight the growing challenge of detecting AI-assisted plagiarism in academic settings.
Featured Image
Photo by Aaron Burden on Unsplash
Read the Original
This page is a summary of: How Large Language Models are Transforming Machine-Paraphrase Plagiarism, January 2022, Association for Computational Linguistics (ACL),
DOI: 10.18653/v1/2022.emnlp-main.62.
You can read the full text:
Contributors
The following have contributed to this page
