📄 中文摘要
在第一部分中，提出了五个原则以解决问题：明确书写、使用层次、编目幻觉陷阱、标记静默崩溃、保持网关。然而，作者故意遗漏了三个重要方面：如何构建不相互干扰的智能体、构建系统时的禁忌以及如何验证系统的有效性。文章指出，单一强大的智能体架构并不理想，因为它可能导致医源性问题，即在治疗过程中产生的负面影响。强调了多智能体系统的必要性，以避免单点故障和提高系统的整体鲁棒性。
你的知识，你的模型 — 第二部分：智能体与医源性问题
出处: Your Knowledge, Your Model — Part 2: Agents, Iatrogenics
发布: 2026年3月31日
📄 English Summary
Your Knowledge, Your Model — Part 2: Agents, Iatrogenics
The first part outlined five principles to address the problem: write everything explicitly, use layers, catalog hallucination traps, mark silent collapses, and stay the gateway. However, three crucial aspects were deliberately omitted: how to build agents that do not interfere with each other, what not to do when constructing the system, and how to verify that the system works. The article argues against the architecture of a single powerful agent, as it can lead to iatrogenesis, which refers to negative effects arising from treatment. It emphasizes the necessity of a multi-agent system to avoid single points of failure and enhance the overall robustness of the system.