Skip to main content Skip to navigation

AI Content Self-Detection for Transformer-based Large Language Models

Project Overview

The document examines the role of generative AI in education, particularly addressing the challenges posed by plagiarism detection when students utilize AI-generated content. It introduces the idea of self-detection, which allows AI models to identify their own generated texts, and assesses a range of generative AI tools for their effectiveness in this context. The findings reveal significant disparities among different AI models regarding their ability to differentiate between AI-generated and human-written texts, underlining the necessity for enhanced detection methodologies in academic settings. Overall, the document emphasizes the importance of understanding the implications of generative AI in educational environments, particularly as it relates to integrity and the need for robust mechanisms to ensure the authenticity of student work.

Key Applications

Self-detection of AI-generated text

Context: Academic context, particularly for students and educators

Implementation: Using transformer-based models to determine if the text matches their own generation patterns.

Outcomes: Improved understanding of AI's ability to detect its own generated content, with varying success across models.

Challenges: Models like Claude struggle with self-detection; the effectiveness of existing plagiarism detection tools is inadequate.

Implementation Barriers

Technical Barrier

Current plagiarism detection systems are not equipped to detect AI-generated content accurately.

Proposed Solutions: Development of self-detection methods using AI systems themselves to identify their output.

Ethical Barrier

The inappropriate use of AI tools may hinder student learning and academic integrity.

Proposed Solutions: Reconsidering definitions of plagiarism and enhancing detection techniques.

Project Team

Antônio Junior Alves Caiado

Researcher

Michael Hahsler

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Antônio Junior Alves Caiado, Michael Hahsler

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies