[KongchangAI]
Unverified50% confidenceBenchmarkExact time

「自我精炼」（Self-Refinement）由 Madaan 等人在 2023 年系统性提出，研究表明可在无需额外训练数据的情况下将任务完成质量提升 5%-25%

1
Sources
50%
Confidence
Long-term
Relevance
8/24/2026
First Seen

Sources

Claude Code 入门实战教程：安装配置到自动化开发完整指南

Related Entities

Self-RefinementconceptMadaanperson

Related Claims

Unverified智能体的偏好受训练数据、提示词、版本迭代等多重因素影响，今天有效的优化手段可能在下一次模型更新后失效65% similarVerifiedDeepMind's Chinchilla research (2022) pointed out that optimal training should keep data volume and parameter count growing at roughly a 1:1 ratio63% similar

Cite This Claim

Stable URI
https://kongchang.com/claim/793885
API
curl https://kongchang.com/api/v1/knowledge/claims/793885
MCP
get_claim(id=793885)