[KongchangAI]
Unverified50% confidenceFactExact time

2023年微软研究论文《LIMA: Less Is More for Alignment》表明仅用1000条精心策划的高质量样本就能让基座模型展现出与数万条样本微调相当的对话能力

1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen

Sources

Related Entities

Related Claims

Cite This Claim

Stable URI
https://kongchang.com/claim/892733
API
curl https://kongchang.com/api/v1/knowledge/claims/892733
MCP
get_claim(id=892733)