基础模型生成文本更像真人，指令微调版反而易被识破
Base Models Look Human To AI Detectors
- 用迭代改写让基础模型文本更难被检测
- 在多个大模型上提升人类似度，效果稳定
- 适合想绕过检测但保内容质量的研究者
随着AI生成文本大规模进入现实场景，教育和学术诚信领域越来越多使用商业文本检测工具。我们发现一个意外现象：在GPTZero和Pangram检测器下，基础模型生成的文本常被判定为高度人类化，而指令微调后的版本则不然。基于此，我们提出无需依赖检测器的「迭代改写人类化」（HIP）方法，通过最小微调使基础模型成为改写器并反复应用。相比测试的基线方法，HIP在语义保持与检测规避之间取得更优平衡。在Llama-3和Qwen-3系列（0.6B至70B参数）上，该方法持续提升检测器判断的人类似度。结果表明，当前检测器更关注指令微调带来的特征和局部上下文模式，而非稳定的机器生成痕迹。这提示需设计更显式建模这些因素的检测系统。
原文摘要 · Abstract (English)
As AI-generated text enters the real-world at scale, institutions increasingly use commercial AI-text detectors, especially in education and academic-integrity workflows. We report a surprising empirical finding about such systems: when evaluated by GPTZero and Pangram, generated text from base models is often judged overwhelmingly human, whereas text generated by their instruction-tuned counterparts is not. Building on this observation, we propose Humanization by Iterative Paraphrasing (HIP), a detector-agnostic pipeline that minimally fine-tunes a base model into a paraphraser and applies it iteratively. Compared with the baselines we test, HIP yields a stronger trade-off between semantic preservation and detector evasion on commercial detectors. Across Llama-3 and Qwen-3 families, spanning model sizes from 0.6B to 70B, HIP consistently improves detector human-likeness. Our findings suggest that current detectors are tracking artifacts of instruction tuning and local context more than any invariant notion of machine-generated text. This, in turn, calls for detector designs that model these factors more explicitly.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。