用治理优先框架解决智能体不可靠问题
From Craft to Constitution: A Governance-First Paradigm for Principled Agent Engineering
- 提出治理优先的智能体工程架构 ArbiterOS
- 将概率性大模型与确定性治理结合
- 适合高可靠性场景的智能体开发
强大的大语言模型（LLMs）开启了「智能体时代」，使自主系统能够应对复杂目标。然而，从原型到生产阶段仍面临普遍的「技艺危机」，导致智能体脆弱、不可预测，难以在关键任务中信赖。本文认为，这一危机源于根本性的范式错配——试图用传统软件工程的确定性思维去指挥本质上具有概率性的处理器。为解决此问题，我们提出一种治理优先的智能体工程范式，并以正式架构 ArbiterOS 予以实现。
原文摘要 · Abstract (English)
The advent of powerful Large Language Models (LLMs) has ushered in an ``Age of the Agent,'' enabling autonomous systems to tackle complex goals. However, the transition from prototype to production is hindered by a pervasive ``crisis of craft,'' resulting in agents that are brittle, unpredictable, and ultimately untrustworthy in mission-critical applications. This paper argues this crisis stems from a fundamental paradigm mismatch -- attempting to command inherently probabilistic processors with the deterministic mental models of traditional software engineering. To solve this crisis, we introduce a governance-first paradigm for principled agent engineering, embodied in a formal architecture we call ArbiterOS.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。