@inproceedings{bhandarkar-etal-2025-aaig,
    title = "{AAIG} at {G}en{AI} Detection Task 1: Exploring Syntactically-Aware, Resource-Efficient Small Autoregressive Decoders for {AI} Content Detection",
    author = "Bhandarkar, Avanti  and
      Wilson, Ronald  and
      Woodard, Damon",
    editor = "Alam, Firoj  and
      Nakov, Preslav  and
      Habash, Nizar  and
      Gurevych, Iryna  and
      Chowdhury, Shammur  and
      Shelmanov, Artem  and
      Wang, Yuxia  and
      Artemova, Ekaterina  and
      Kutlu, Mucahid  and
      Mikros, George",
    booktitle = "Proceedings of the 1stWorkshop on GenAI Content Detection (GenAIDetect)",
    month = jan,
    year = "2025",
    address = "Abu Dhabi, UAE",
    publisher = "International Conference on Computational Linguistics",
    url = "https://preview.aclanthology.org/fix-sig-urls/2025.genaidetect-1.23/",
    pages = "218--224",
    abstract = "This paper presents a lightweight and efficient approach to AI-generated content detection using small autoregressive fine-tuned decoders (AFDs) for secure, on-device deployment. Motivated by resource-efficiency, syntactic awareness, and bias mitigation, our model employs small language models (SLMs) with autoregressive pre-training and loss fusion to accurately distinguish between human and AI-generated content while significantly reducing computational demands. The system achieved highest macro-F1 score of 0.8186, with the submitted model scoring 0.7874{---}both significantly outperforming the task baseline while reducing model parameters by {\textasciitilde}60{\%}. Notably, our approach mitigates biases, improving recall for human-authored text by over 60{\%}. Ranking 8th out of 36 participants, these results confirm the feasibility and competitiveness of small AFDs in challenging, adversarial settings, making them ideal for privacy-preserving, on-device deployment suitable for real-world applications."
}