| title | Base Models Look Human To AI Detectors |
|---|---|
| authors | Yixuan Even Xu, Ziqian Zhong, Aditi Raghunathan, Fei Fang, J. Zico Kolter |
| date | 2026-05-19 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2605.19516 |
| affiliation | Carnegie Mellon University |
| upvotes | 1 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf code |
| abstract | 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. |
cite
@misc{xu2026base,
title={Base Models Look Human To AI Detectors},
author={Yixuan Even Xu and Ziqian Zhong and Aditi Raghunathan and Fei Fang and J. Zico Kolter},
year={2026},
eprint={2605.19516},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2605.19516}
}