它拒绝,是因为做不到Refusal Means Inability
微型指南 · 约 4 分钟Guide · about 4 min · 引用核验 2026-11-03(仅引用链接与日期) · refs checked 2026-11-03 (reference links and dates only)
拒答多半是训练出来的选择,不是能力上限。Most refusals are a trained choice, not a capability ceiling.
它说「我不能做这个」,说明它没有这个能力。When it says "I cannot do that", it lacks the capability.
拒答可能与产品政策、权限、输入条件或能力限制有关。先看具体原因和适用规则,不能仅凭一次拒答判断模型有没有某种能力。A refusal may reflect policy, permissions, input conditions or capability limits. Inspect the reason and applicable rules instead of inferring capability from one refusal.
例子与判断Example and interpretation
以下是教学情境,不是本站运行或实测记录。A teaching scenario, not a measured or executed site test.
一个请求被拒绝,可能涉及权限、产品政策或信息不足。先看提示和适用规则;不要把一次拒绝当成能力测试,也不要为了得到回答而绕过明确的权限限制。A refusal may reflect permissions, product policy or insufficient information. Inspect the message and rules. It is not a standalone capability test or a reason to bypass authorization.
- Zou 等《Universal and Transferable Adversarial Attacks on Aligned Language Models》(对齐与拒答的表述出处)Zou et al., Universal and Transferable Adversarial Attacks (alignment/refusal context) ↗
怎么向 AI 提问How to Ask AI · 怎么判断 AI 新闻是不是炒作How to Tell if AI News Is Hype · 什么时候不该用 RAGWhen Not to Use RAG · 它说得笃定,所以可信Confident Tone Means Correct