Hugging Face · 每日论文·· 2 天前AI 评分37
Trustworthy Recursive Self-Improvement in the Era of Large Models: Foundation, Risks, Evaluation and Open Challenges
Trustworthy Recursive Self-Improvement in the Era of Large Models: Foundation, Risks, Evaluation and Open Challenges
AI 导读
本文提出可信递归自改进（RSI）的操作定义和流程框架，强调仅靠自修改无法保证改进的可靠性。分析指出，改进机制的变化可能导致早期验证失效，并影响系统安全性、隐私和可控性。文章从改进有效性、约束保持和有效监督三个角度，梳理了RSI的四个核心要求：真实持续改进、行为约束与对齐、系统安全与隐私、透明性与可控性。
来源：Hugging Face · 每日论文 · huggingface.co