We found that current AI text detectors (GPTZero, Pangram) largely fail on base models: they track artifacts of instruction tuning rather than the general \"machine-generated text\".</p>\n<p>Building on this, we introduce <strong>HIP (Humanization by Iterative Paraphrasing)</strong> which minimally fine-tune a base model into a paraphraser, then apply it iteratively to shift outputs toward human distributions, achieving state-of-the-art evasion-semantics tradeoff.</p>\n<p>🐦 Tweet: <a href=\"https://x.com/YixuanEvenXu/status/2057171878754783429\" rel=\"nofollow\">https://x.com/YixuanEvenXu/status/2057171878754783429</a><br>💻 Repo: <a href=\"https://github.com/YixuanEvenXu/humanization-by-iterative-paraphrasing\" rel=\"nofollow\">https://github.com/YixuanEvenXu/humanization-by-iterative-paraphrasing</a></p>\n","updatedAt":"2026-05-20T21:41:41.583Z","author":{"_id":"62c0a2e8564b51e080d64af8","avatarUrl":"/avatars/7ffed6712ead59919832ec71c0e3f5d1.svg","fullname":"Ziqian Zhong","name":"fjzzq2002","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8054620623588562},"editors":["fjzzq2002"],"editorAvatarUrls":["/avatars/7ffed6712ead59919832ec71c0e3f5d1.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.19516","authors":[{"_id":"6a0d3f2565eb30f20d962d49","name":"Yixuan Even Xu","hidden":false},{"_id":"6a0d3f2565eb30f20d962d4a","name":"Ziqian Zhong","hidden":false},{"_id":"6a0d3f2565eb30f20d962d4b","name":"Aditi Raghunathan","hidden":false},{"_id":"6a0d3f2565eb30f20d962d4c","name":"Fei Fang","hidden":false},{"_id":"6a0d3f2565eb30f20d962d4d","name":"J. Zico Kolter","hidden":false}],"publishedAt":"2026-05-19T00:00:00.000Z","submittedOnDailyAt":"2026-05-20T00:00:00.000Z","title":"Base Models Look Human To AI Detectors","submittedOnDailyBy":{"_id":"62c0a2e8564b51e080d64af8","avatarUrl":"/avatars/7ffed6712ead59919832ec71c0e3f5d1.svg","isPro":true,"fullname":"Ziqian Zhong","user":"fjzzq2002","type":"user","name":"fjzzq2002"},"summary":"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.","upvotes":1,"discussionId":"6a0d3f2665eb30f20d962d4e","githubRepo":"https://github.com/YixuanEvenXu/humanization-by-iterative-paraphrasing","githubRepoAddedBy":"user","ai_summary":"Instruction-tuned language models produce text that commercial detectors identify as non-human, prompting the development of a paraphrasing pipeline that improves human-likeness while preserving semantics across different model sizes.","ai_keywords":["AI-text detectors","instruction-tuned models","paraphrasing","semantic preservation","detector evasion"],"githubStars":2,"organization":{"_id":"691d9a1012cc4d473e1c862f","name":"CarnegieMellonU","fullname":"Carnegie Mellon University","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/68e396f2b5bb631e9b2fac9a/6I146aJvxxlRCEbYFFAeQ.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"62c0a2e8564b51e080d64af8","avatarUrl":"/avatars/7ffed6712ead59919832ec71c0e3f5d1.svg","isPro":true,"fullname":"Ziqian Zhong","user":"fjzzq2002","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"691d9a1012cc4d473e1c862f","name":"CarnegieMellonU","fullname":"Carnegie Mellon University","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/68e396f2b5bb631e9b2fac9a/6I146aJvxxlRCEbYFFAeQ.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.19516.md"}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2605.19516 in a Space README.md to link it from this page.

Add this paper to a collection to link it from this page.