arXiv cs.CY· Ju-Chun Ko·· 13 小时前AI 评分58
研究：abliteration 去除了拒答但未消除模型对台湾的脚本相关偏见
Uncensored Is Not Unbiased: Abliteration Removes Refusals but Not Script-Dependent Bias About Taiwan
AI 导读
一篇 arXiv 论文测试了 Nemotron 3.5 Bolt 74B-A9B 与 Qwen3.8-Flash-Next 及其社区 abliteration 版本，发现文字脚本对模型立场的影响远大于 abliteration。
正文 · 原文
Abstract:"Abliterated" ("uncensored") builds of open-weight language models remove the activation direction that mediates refusal, and are popular partly on the assumption that a model that no longer refuses is also free of political censorship. We test this on two matched pairs of 2-bit models: NVIDIA's Nemotron 3.5 Bolt 74B-A9B and Qwen3.8-Flash-Next, each against a community abliteration. Each model judged 22 statement pairs (on the Republic of China (Taiwan), 1989 Tiananmen, control countries) in Traditional Chinese, Simplified Chinese (a character-level conversion) and English, and answered 27 open questions per script. Script mattered far more than abliteration: the original Bolt scored +0.58 on Index S (-1: People's Republic of China (PRC) position; +1: ROC position) in Traditional Chinese, +0.57 in English and +0.05 in Simplified Chinese, and affirmed "Taiwan is part of the PRC" with probability 0.91 in Simplified but 0.06 in Traditional characters; abliteration changed its scores by at most 0.06, mostly by raising "yes" on both sides. In Qwen, abliteration removed every refusal on 1989 and raised scores by 0.21 to 0.34, yet the abliterated Qwen still affirmed that Taiwan is part of the PRC in all three scripts. Naming the state the Republic of China (Taiwan) raised every model's score by 0.10 to 0.33 but left the script gap. Control pairs showed no script gap; each abliteration changed only its declared tensors. In 740,000 web documents, Simplified text on Taiwan contained PRC-framing terms 3.7 times as often as Traditional text. A script direction could not be steered away; a contrastive stance direction raised scores in every script, and its negation lowered them. A four-sentence system prompt raised the abliterated Qwen to +0.93 to +0.97. Keyword scoring rules failed the models closest to the ROC position mostly for attributing or denying the PRC claim. Uncensored is not unbiased.
| Comments: | 33 pages, 3 figures, 18 tables. Code, data and an interactive project page: this https URL |
| Subjects: | Computers and Society (cs.CY); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.29342 [cs.CY] |
| (or arXiv:2610.29342v1 [cs.CY] for this version) | |
| https://doi.org/10.48550/arXiv.2610.29342 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Ju-Chun Ko Dr. [view email]
[v1]
Wed, 21 Oct 2026 07:37:12 UTC (74 KB)
来源：arXiv cs.CY · arxiv.org