重新思考大语言模型中基于提示的去偏方法
Rethinking Prompt-based Debiasing in Large Language Models
AI总结：
本文通过 BBQ 和 StereoSet 基准系统分析大语言模型基于提示的去偏方法，发现其效果表面化且评估指标存在缺陷，呼吁重新思考偏见度量标准以构建真正可信赖的 AI。
摘要：
研究大语言模型（LLMs）中的偏见对于开发可信赖的人工智能至关重要。虽然通过提示工程进行基于提示的去偏方法很常见，但其有效性依赖于模型本质上理解偏见的假设。我们的研究使用 BBQ 和 StereoSet 基准，在开源模型以及商业 GPT 模型上系统性地分析了这一假设。实验结果表明，基于提示的去偏方法往往是表面化的；例如，Llama2-7B-Chat 模型尽管在 BBQ 数据集上识别偏见问题时达到了很高的准确率，却将超过 90% 的无偏见内容错误分类为有偏见。此外，偏见基准中的特定评估和问题设置常常导致 LLMs 选择“回避性答案”，无视问题的核心以及回答与上下文的相关性。而且，先前方法的表面成功可能源于有缺陷的评估指标。我们的研究揭示了基于提示的去偏努力中潜在的“虚假繁荣”，并强调需要重新思考偏见度量标准，以确保真正可信赖的人工智能。
英文摘要：
Investigating bias in large language models (LLMs) is crucial for developing trustworthy AI. While prompt-based through prompt engineering is common, its effectiveness relies on the assumption that models inherently understand biases. Our study systematically analyzed this assumption using the BBQ and StereoSet benchmarks on both open-source models as well as commercial GPT model. Experimental results indicate that prompt-based is often superficial; for instance, the Llama2-7B-Chat model misclassified over 90% of unbiased content as biased, despite achieving high accuracy in identifying bias issues on the BBQ dataset. Additionally, specific evaluation and question settings in bias benchmarks often lead LLMs to choose "evasive answers", disregarding the core of the question and the relevance of the response to the context. Moreover, the apparent success of previous methods may stem from flawed evaluation metrics. Our research highlights a potential "false prosperity" in prompt-base efforts and emphasizes the need to rethink bias metrics to ensure truly trustworthy AI.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
