Unverified50% confidenceBenchmarkExact time
当提示中包含展示标签对比的少样本样例时，所有大语言模型的表现都超越了零样本提示
1
Sources
50%
Confidence
Long-term
Relevance
9/23/2026
First Seen
Sources
Related Entities
Related Claims
Unverified大模型在审查和修改已有文本时的表现通常优于从零生成72% similarUnverified大语言模型通过指令微调和RLHF能够在零样本条件下完成高质量的语气转换，只需在提示词中描述目标语气72% similarUnverified零样本分类依托预训练大语言模型如BART、RoBERTa的NLI变体的语义理解能力69% similarUnverified零样本（Zero-shot）提示指不提供示例，完全依赖文字描述69% similarUnverified实验数据显示，语法约束解码使小型语言模型生成Bash代码的语法正确率大幅提高，引号不匹配和结构不完整等错误几乎被完全消除69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/942794API
curl https://kongchang.com/api/v1/knowledge/claims/942794MCP
get_claim(id=942794)