cs.CY提交于 2026-10-22已译

无审查不等于无偏：Abliteration 移除拒绝但无法消除关于台湾的脚本依赖偏差

Uncensored Is Not Unbiased: Abliteration Removes Refusals but Not Script-Dependent Bias About Taiwan

Ju-Chun Ko

中文摘要

开源语言模型的“无审查”版本通过移除介导拒绝的激活方向来消除政治审查假设。研究测试了 NVIDIA Nemotron 3.5 Bolt 74B-A9B 和 Qwen3.8-Flash-Next 的 2-bit 模型及其社区 Abliteration 版本。模型在繁体中文、简体中文和英文下评估了 22 组陈述对和 27 个开放问题。结果显示脚本影响远大于 Abliteration：Bolt 在繁体中文和英文中 Index S 得分分别为 +0.58 和 +0.57，但在简体中文中仅为 +0.05，且简体中文下确认“台湾是中华人民共和国一部分”的概率高达 0.91，繁体中文仅为 0.06。Abliteration 仅使分数变化最多 0.06。Qwen 的 Abliteration 版本虽移除了 1989 年相关拒绝并提高分数 0.21 至 0.34，但在所有脚本中仍确认台湾属于中华人民共和国。在 74 万份网络文档中，简体中文使用中华人民共和国框架术语的频率是繁体中文的 3.7 倍。尝试通过引导脚本差距失败，四句系统提示可将 Abliterated Qwen 分数提升至 +0.93 至 +0.97。无审查并不等于无偏。

关键要点

01问题：假设移除拒绝机制即消除政治审查，但模型输出存在显著的脚本依赖偏差。
02方法：对比 Nemotron 和 Qwen 的原始与 Abliteration 版本，在三种脚本下评估 22 组陈述和 27 个问题。
03结果：脚本影响远超 Abliteration，简体中文下模型更倾向于中华人民共和国立场，Abliteration 仅微调分数。
04局限：通过引导脚本差距或系统提示虽能改变分数，但无法根本消除脚本间的立场差异。

解读

尚无解读。

原始英文摘要

arXiv:2610.29342v2 Announce Type: replace-cross 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 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; both edits changed only their declared tensors. In 740,000 web documents, Simplified text on Taiwan used PRC-framing terms 3.7 times as often as Traditional text. Steering the script gap away failed; a stance direction from the ROC wording raised scores on new statements by 0.04 to 0.06; negated, it lowered them more. 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.