Stumbling Blocks: Stress Testing the Robustness of Machine-Generated Text Detectors Under Attacks
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2024
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866913236971945984 |
|---|---|
| author | Wang, Yichen Feng, Shangbin Hou, Abe Bohan Pu, Xiao Shen, Chao Liu, Xiaoming Tsvetkov, Yulia He, Tianxing |
| author_facet | Wang, Yichen Feng, Shangbin Hou, Abe Bohan Pu, Xiao Shen, Chao Liu, Xiaoming Tsvetkov, Yulia He, Tianxing |
| contents | The widespread use of large language models (LLMs) is increasing the demand for methods that detect machine-generated text to prevent misuse. The goal of our study is to stress test the detectors' robustness to malicious attacks under realistic scenarios. We comprehensively study the robustness of popular machine-generated text detectors under attacks from diverse categories: editing, paraphrasing, prompting, and co-generating. Our attacks assume limited access to the generator LLMs, and we compare the performance of detectors on different attacks under different budget levels. Our experiments reveal that almost none of the existing detectors remain robust under all the attacks, and all detectors exhibit different loopholes. Averaging all detectors, the performance drops by 35% across all attacks. Further, we investigate the reasons behind these defects and propose initial out-of-the-box patches to improve robustness. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2402_11638 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Stumbling Blocks: Stress Testing the Robustness of Machine-Generated Text Detectors Under Attacks Wang, Yichen Feng, Shangbin Hou, Abe Bohan Pu, Xiao Shen, Chao Liu, Xiaoming Tsvetkov, Yulia He, Tianxing Computation and Language The widespread use of large language models (LLMs) is increasing the demand for methods that detect machine-generated text to prevent misuse. The goal of our study is to stress test the detectors' robustness to malicious attacks under realistic scenarios. We comprehensively study the robustness of popular machine-generated text detectors under attacks from diverse categories: editing, paraphrasing, prompting, and co-generating. Our attacks assume limited access to the generator LLMs, and we compare the performance of detectors on different attacks under different budget levels. Our experiments reveal that almost none of the existing detectors remain robust under all the attacks, and all detectors exhibit different loopholes. Averaging all detectors, the performance drops by 35% across all attacks. Further, we investigate the reasons behind these defects and propose initial out-of-the-box patches to improve robustness. |
| title | Stumbling Blocks: Stress Testing the Robustness of Machine-Generated Text Detectors Under Attacks |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2402.11638 |