The Decoder· Jonathan Kemper·· 3 小时前AI 评分54
Google 研究者提出 RRSI 方法,防止自改进 AI 智能体记忆测试任务
Google researchers find a way to keep self-improving AI agents from memorizing their tests
AI 导读
Google 研究者提出 RRSI(Regularized Recursive Self-Improvement of Agent Harnesses),通过逐步收紧的编辑预算和严格评审机制,约束 AI 智能体在自我优化 harness 时的过拟合。
来源:The Decoder · the-decoder.com