@inproceedings{zhang-etal-2025-lists,
    title = "From Lists to Emojis: How Format Bias Affects Model Alignment",
    author = "Zhang, Xuanchang  and
      Xiong, Wei  and
      Chen, Lichang  and
      Zhou, Tianyi  and
      Huang, Heng  and
      Zhang, Tong",
    editor = "Che, Wanxiang  and
      Nabende, Joyce  and
      Shutova, Ekaterina  and
      Pilehvar, Mohammad Taher",
    booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingestion-acl-25/2025.acl-long.1308/",
    pages = "26940--26961",
    ISBN = "979-8-89176-251-0",
    abstract = "In this paper, we study format biases in reinforcement learning from human feedback (RLHF). We observe that many widely-used preference models{---}including human evaluators, GPT-4, and top-ranking models on the RewardBench benchmark{---}exhibit strong biases towards specific format patterns, such as lists, links, bold text, and emojis. Furthermore, large language models (LLMs) can exploit these biases to achieve higher rankings on popular benchmarks like AlpacaEval and LMSYS Chatbot Arena. One notable example is verbosity bias, where current preference models favor longer responses that appear more comprehensive, even when their quality is equal to or lower than shorter responses. However, format biases beyond verbosity remain largely underexplored. In this work, we extend the study of biases in preference learning beyond the commonly recognized length bias, offering a comprehensive analysis of a wider range of format biases. Additionally, we show that with a small amount of biased data (less than 1{\%}), we can inject significant bias into the reward model. Moreover, these format biases can also be easily exploited by downstream alignment algorithms, such as *best-of-n sampling* and online iterative *DPO*, as it is usually easier to manipulate the format than to improve the quality of responses. Our findings emphasize the need to disentangle format and content both for designing alignment algorithms and evaluating models."
}