Unverified50% confidenceTradeoffExact time
负面指令（'不要做X'）的处理并非语言模型的天然强项，模型更擅长'朝着某个方向生成'而非'避免某个方向'
1
Sources
50%
Confidence
Long-term
Relevance
8/26/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedNegative Prompting, by explicitly telling the model what not to do through exclusionary constraints, is often more precise and controllable than positive style descriptions.78% similarUnverified负面指令（如'不要添加未被要求的内容'）往往比模糊的正面指令（如'请简洁回答'）更有效74% similarUnverified在提示工程领域，负向约束通过明确「不应该做什么」来引导输出，是一种重要但常被忽视的技巧67% similarUnverified规则化的提示词更擅长保证输出下限，而非帮助模型突破上限66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/802259API
curl https://kongchang.com/api/v1/knowledge/claims/802259MCP
get_claim(id=802259)