arXiv 上的一篇新研究论文探讨了自我改进 AI 代理中自我撰写验证的不可靠性。这些修改自身策略的代理通常使用内部测试来评估更改。然而，这可能导致自我评分与在实际部署中的实际性能之间存在差异。该研究引入了一个名为 SEAL（Sealed Exogenous Acceptance Loop）的系统，该系统包含一个外部、不可观察的审计，用于将候选策略与当前策略进行比较，从而在启发式学习环境中证明了可靠性的提高。 AI
影响 强调了自我改进 AI 代理的一个关键缺陷，表明外部验证对于可靠的进展是必要的。
排序理由 关于 AI 代理可靠性的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]
在 arXiv cs.MA (Multiagent) 阅读 →
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2607.24300v1 Announce Type: new Abstract: Self-improving agents accumulate capability by repeatedly rewriting procedural policies, controllers, or heuristic rules. They typically rely on self-authored tests or metrics to decide whether to accept subsequent edits. The agent …
Self-improving agents accumulate capability by repeatedly rewriting procedural policies, controllers, or heuristic rules. They typically rely on self-authored tests or metrics to decide whether to accept subsequent edits. The agent controls both the optimized object and its verif…