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
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% similarUnverifiedMultiple studies in 2024 found that the strong performance of certain models on standard benchmarks resulted from benchmark test questions being mixed into the training data (data contamination).64% similarVerifiedDeepMind's Chinchilla research (2022) pointed out that optimal training should keep data volume and parameter count growing at roughly a 1:1 ratio63% similarUnverifiedThe cleanliness of a benchmark is time-limited—once test questions are publicly circulated, future model training will very likely absorb them, causing fairness to decay accordingly.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/793885API
curl https://kongchang.com/api/v1/knowledge/claims/793885MCP
get_claim(id=793885)