GenAI Content Detection Task 3: Cross-Domain Machine-Generated Text Detection Challenge
Date:
Abstract
In this talk I discuss the results of the “Cross-Domain Machine Generated Text Detection Challenge” shared task. In this task we seek to understand whether or not AI detectors can be robust to many domains and LLMs simultaneously. We show that detectors do surprisingly well on RAID, a highly challenging benchmark of generated text, and conclude that it may not be as challenging as we originally thought. We call on researchers to improve the difficulty and quality of future benchmarks and also give an overview of the various modeling techniques employed by the winning teams.
Location
This talk was given on January 19th 2025 at the Abu Dhabi National Exhibition Centre (ADNEC) in Abu Dhabi, UAE as part of the GenAI Content Detection Workshop at COLING 2025.