Pangram at GenAI Detection Task 3: An Active Learning Approach to Machine-Generated Text Detection

Bradley N. Emi, Max Spero, Elyas Masrour


Abstract
We pretrain an autoregressive LLM-based detector on a wide variety of datasets, domains, languages, prompt schemes, and LLMs used to generate the AI portion of the dataset. We aggressively employ several augmentation strategies and preprocessing strategies to improve robustness. We then mine the RAID train set for the AI examples with the largest error based on the original classifier, and mix those examples and their human-written counterparts back into the training set. We then retrain the detector until convergence.
Anthology ID:
2025.genaidetect-1.40
Volume:
Proceedings of the 1stWorkshop on GenAI Content Detection (GenAIDetect)
Month:
January
Year:
2025
Address:
Abu Dhabi, UAE
Editors:
Firoj Alam, Preslav Nakov, Nizar Habash, Iryna Gurevych, Shammur Chowdhury, Artem Shelmanov, Yuxia Wang, Ekaterina Artemova, Mucahid Kutlu, George Mikros
Venues:
GenAIDetect | WS
SIG:
Publisher:
International Conference on Computational Linguistics
Note:
Pages:
347–351
Language:
URL:
https://aclanthology.org/2025.genaidetect-1.40/
DOI:
Bibkey:
Cite (ACL):
Bradley N. Emi, Max Spero, and Elyas Masrour. 2025. Pangram at GenAI Detection Task 3: An Active Learning Approach to Machine-Generated Text Detection. In Proceedings of the 1stWorkshop on GenAI Content Detection (GenAIDetect), pages 347–351, Abu Dhabi, UAE. International Conference on Computational Linguistics.
Cite (Informal):
Pangram at GenAI Detection Task 3: An Active Learning Approach to Machine-Generated Text Detection (Emi et al., GenAIDetect 2025)
Copy Citation:
PDF:
https://aclanthology.org/2025.genaidetect-1.40.pdf