Unverified50% confidenceSolutionExact time
对大语言模型而言，「不要做X」的负向约束有时比「要做Y」的正向指令更有效，因为负样本相当于在概率空间中设置了禁区
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
UnverifiedNegative constraints narrow the probability space by explicitly excluding undesired outputs, are equally important as positive instructions, and can reduce deviations caused by the model's creative improvisation.74% similarUnverified在代码生成Prompt中，否定性约束（说明不需要什么）能缩小模型搜索空间，比增加期望描述更能提高准确率73% similarUnverifiedPrompt engineering has inherent limitations: the effectiveness of instructions decays over conversation turns (instruction forgetting), and different model versions respond significantly differently to the same instruction.64% similarUnverified当前主流LLM在欧几里得空间推理（如判断两条线段是否相交）上的准确率远低于其在文本推理任务上的表现，这是自回归语言模型架构的固有局限64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/575492API
curl https://kongchang.com/api/v1/knowledge/claims/575492MCP
get_claim(id=575492)