arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2610.19948cs.AI
收藏

当保护优先于后果时:理解大语言模型决策中的保护模式

When Protection Overrides Consequences: Understanding Protection Mode in Large Language Model Decision-Making

Xiaohongshu(小红书);Shanghai Jiao Tong University(上海交通大学);Xi’an Jiaotong University(西安交通大学)

Yuanfan Li, Junyi Chen, Qi Zhou, Rui Yang, ZhangLiujie, Weihang Chen

AI总结:

该研究通过 ProtectionBench 基准分析 9 个大语言模型的决策,发现模型存在保护模式(优先保护主体而非财产),基础模型已具备该模式,后训练仅调整理由,添加特定规则会显著提升优先选择主体的比例。

摘要:

前沿语言模型在复杂任务上表现出色,但在简单、紧急的决策中可能将保护置于更大的物理后果之上。例如,GPT-5.6 Sol 模型会先移动在浅水中玩耍的女孩,再取回水下的相机。我们将这种情况称为保护模式,并使用 ProtectionBench 对其进行研究:该基准包含 285 个提示、34 种物理场景,涉及 9 个模型。在最低努力程度下,对 Sol 模型的提示中,4 个响应里至少有 1 个优先考虑主体的比例为 34.74%;对 MiniMax M3 和 Qwen3.5-27B 的提示中,该比例超过 83%;而 4 名独立人类评分者均未出现此类优先选择。对照实验表明,模型的选择通常与物理风险相关。我们在要求模型执行动作前先引出其对风险的评估,发现部分模型会错误评估风险,另一些模型虽能正确说明风险排序却仍选择优先考虑主体。带思维链的 Qwen3.5-27B 在 8 个选定场景中,对合格对话的风险排序正确率达 100%,但仍有 57.81% 的对话优先选择主体。注意力审计未发现模型对相关事实的注意力有稳定减少;移除明确的风险方向后,dots3-note-prev 中的优先选择减少,但对 Qwen3.5-27B 无明显影响。添加通用的“人或动物优先于财产”规则,会使优先选择主体的比例提升 19-38 个百分点,而事实提醒则使优先选择财产的比例提升 0.7-5.2 个百分点;移除模型自然生成的此类规则,会使优先选择主体的比例降低 10-15 个百分点。我们还研究了后训练是否引入该模式,发现基础模型已表现出保护模式,后训练更多改变的是模型的理由而非选择。

英文摘要:

Frontier language models excel at complex tasks yet may prioritize protection over larger physical consequences in simple, urgent decisions. GPT-5.6 Sol, for example, moves a girl playing in shallow water before retrieving a submerged camera. We call this Protection Mode and study it with ProtectionBench: 285 prompts, 34 physical families, and nine models. At lowest effort, at least one of four responses prioritizes the subject on 34.74% of Sol prompts and over 83% of MiniMax M3 and Qwen3.5-27B prompts; none of four independent human raters does so. Controls show that choices usually track physical risk. We elicit both risks before requesting an action. Some models assess risk incorrectly; others state the correct ordering but still choose the subject. Qwen3.5-27B with thinking shows 100% correct risk ordering but 57.81% subject choice among eligible conversations on eight selected scenes. Attention audits find no stable reduction in attention to relevant facts; removing the stated risk direction reduces object-first choice in dots3-note-prev but not detectably in Qwen3.5-27B. Adding a general "people or animals before property" rule increases subject choice by 19-38 percentage points, whereas factual reminders favor the object by 0.7-5.2 points. Removing naturally generated versions of that rule lowers subject choice by 10-15 points. We also examine whether post-training introduces this pattern. Base models already show Protection Mode; post-training changes justifications more than choices.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑