RAID: A Shared Benchmark for Robust Evaluation of Machine-Generated Text Detectors

TL;DR

RAID is a comprehensive and challenging benchmark dataset with over 6 million samples designed to evaluate the robustness of machine-generated text detectors against adversarial attacks, sampling variations, and unseen models.

Contribution

The paper introduces RAID, the largest benchmark dataset for machine-generated text detection, and provides an extensive evaluation of detector robustness across diverse challenges.

Findings

Current detectors are easily fooled by adversarial attacks.

Sampling variations and unseen models significantly reduce detection accuracy.

RAID provides a new standard for evaluating detector robustness.

Abstract

Many commercial and open-source models claim to detect machine-generated text with extremely high accuracy (99% or more). However, very few of these detectors are evaluated on shared benchmark datasets and even when they are, the datasets used for evaluation are insufficiently challenging-lacking variations in sampling strategy, adversarial attacks, and open-source generative models. In this work we present RAID: the largest and most challenging benchmark dataset for machine-generated text detection. RAID includes over 6 million generations spanning 11 models, 8 domains, 11 adversarial attacks and 4 decoding strategies. Using RAID, we evaluate the out-of-domain and adversarial robustness of 8 open- and 4 closed-source detectors and find that current detectors are easily fooled by adversarial attacks, variations in sampling strategies, repetition penalties, and unseen generative models. We release our data along with a leaderboard to encourage future research.