Robust and Fine-Grained Detection of AI Generated Texts
Ram Mohan Rao Kadiyala, Siddartha Pullakhandam, Kanwal Mehreen, Drishti Sharma, Siddhant Gupta, Jebish Purbey, Ashay Srivastava, Subhasya TippaReddy, Arvind Reddy Bobbili, Suraj Telugara Chandrashekhar, Modabbir Adeeb, Srinadh Vura, Hamza Farooq
2025-04-17
Summary
This paper talks about new models that are really good at spotting whether a piece of text was written by an AI, by a human, or by both working together, even if someone tries to trick the system.
What's the problem?
The problem is that as AI-generated writing becomes more common and convincing, it's getting harder to tell if something was written by a person or by a computer. This is especially tricky when humans and AI work together, or when people try to hide the fact that AI helped. This can cause issues in areas like education, journalism, or online safety.
What's the solution?
The researchers developed a set of advanced models that look closely at each part of the text to figure out if it was made by an AI, a human, or both. These models work well across different topics and types of AI, and they can still spot AI writing even if someone tries to change the text to fool the detector.
Why it matters?
This matters because it helps keep things honest and transparent in a world where AI writing is everywhere. Being able to reliably detect AI-generated content protects against cheating, misinformation, and other problems that can happen when people can't tell what's real and what's not.
Abstract
A collection of models for token classification effectively identifies AI-generated content, including human-LLM co-authored texts, across various domains and generators, even with adversarial inputs.