Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

MAGE: Machine-generated Text Detection in the Wild

About

Large language models (LLMs) have achieved human-level text generation, emphasizing the need for effective AI-generated text detection to mitigate risks like the spread of fake news and plagiarism. Existing research has been constrained by evaluating detection methods on specific domains or particular language models. In practical scenarios, however, the detector faces texts from various domains or LLMs without knowing their sources. To this end, we build a comprehensive testbed by gathering texts from diverse human writings and texts generated by different LLMs. Empirical results show challenges in distinguishing machine-generated texts from human-authored ones across various scenarios, especially out-of-distribution. These challenges are due to the decreasing linguistic distinctions between the two sources. Despite challenges, the top-performing detector can identify 86.54% out-of-domain texts generated by a new LLM, indicating the feasibility for application scenarios. We release our resources at https://github.com/yafuly/MAGE.

Yafu Li, Qintong Li, Leyang Cui, Wei Bi, Zhilin Wang, Longyue Wang, Linyi Yang, Shuming Shi, Yue Zhang• 2023

Related benchmarks

TaskDatasetResultRank
Single-target AI-generated Text DetectionM4
AUROC@158
25
AI-generated text detection and calibrationDetectRL Prompt Attack
AUC@1%75.8
20
AI Text DetectionMAGE in-distribution (test)
AUROC81
16
AI Text DetectionUnified RL corpus (test)
AUROC91.3
16
Machine-generated text detectionTELL benchmark (test)
AUROC91.32
16
Machine-generated text detectionMAGE Unseen Domains & Unseen Model (test)
AUROC0.93
11
AI-generated text detection and calibrationDetectRL (Paraphrase)
AUC@1%69.9
10
DetectionRAID Reviews
AUC@1%60.9
10
Detection and CalibrationRAID Reddit domain
AUC@1%58.2
10
Detection and CalibrationRAID News domain
AUC @ 1%0.529
10
Showing 10 of 79 rows
...

Other info

Code

Follow for update