PAN
PAN is a series of scientific events and shared tasks on digital text forensics and stylometry.
Tasks
The PAN Lab features five different tasks:
Voight-Kampff Generative AI Detection
Given a (potentially obfuscated) document, decide whether it was written by a human or an AI.
Text Watermarking
Given a text, embed a watermark, then try to detect the watermark after a number of unknown perturbations.
Multi-Author Writing Style Analysis
Given a document, determine at which positions the author changes.
Generative Plagiarism Detection
Given a generated and a human-written source document, identify the passages of reused text between them.
Reasoning Trajectory Detection
Given a query, reasoning trajectory, and final answer, determine a) whether the trajectory is human-written or AI-generated or b) whether the trajectory or answer are safe.
Organizers
- Janek Bevendorff (Bauhaus-Universität Weimar, Germany)
- Benno Stein (Bauhaus-Universität Weimar, Germany)
- Martin Potthast (University of Kassel, Germany)