arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2601.22290cs.AI
收藏

六西格玛代理:通过共识驱动的分解执行实现企业级可靠性

The Six Sigma Agent: Achieving Enterprise-Grade Reliability in LLM Systems Through Consensus-Driven Decomposed Execution

Lyzr Research(Lyzr研究)

Khush Patel, Siva Surendira, Jithin George, Shreyas Kapale

AI总结:

六西格玛代理通过共识驱动的分解执行实现企业级可靠性,显著提升AI系统可靠性并降低成本。

摘要:

大型语言模型展现出显著的能力,但本质上仍然是概率性的,这对企业部署构成了关键的可靠性挑战。我们引入了六西格玛代理,这是一种新的架构,通过三个协同作用的组件实现企业级可靠性:(1)将任务分解为依赖树中的原子操作;(2)微代理采样,其中每个任务在不同的LLM上并行执行n次以生成独立的输出;(3)共识投票与动态扩展,聚类输出并从得票最多的簇中选择答案。我们证明,采样n个独立输出,错误率p,系统错误率为O(p^{ceil(n/2)}),从而实现指数级的可靠性提升。即使使用更便宜的模型,每个操作的错误率为5%,使用5个代理的共识投票将错误率降低到0.11%;动态扩展到13个代理可达到3.4 DPMO(每百万机会缺陷数),即六西格玛标准。在三个企业使用案例中的评估显示,与单代理执行相比,可靠性提升了14700倍,同时成本降低了80%。我们的工作表明,AI系统中的可靠性源于原理化的冗余和共识,而非单纯模型扩展。

英文摘要:

Large Language Models demonstrate remarkable capabilities yet remain fundamentally probabilistic, presenting critical reliability challenges for enterprise deployment. We introduce the Six Sigma Agent, a novel architecture that achieves enterprise-grade reliability through three synergistic components: (1) task decomposition into a dependency tree of atomic actions; (2) micro-agent sampling where each task is executed n times in parallel across diverse LLMs to generate independent outputs; and (3) consensus voting with dynamic scaling, clustering outputs and selecting the answer from the winning cluster with maximum votes. We prove that sampling n independent outputs with error rate p achieves system error O(p^{ceil(n/2)}), enabling exponential reliability gains. Even using cheaper models with 5% per-action error, consensus voting with 5 agents reduces error to 0.11%; dynamic scaling to 13 agents achieves 3.4 DPMO (Defects Per Million Opportunities), the Six Sigma standard. Evaluation across three enterprise use cases demonstrates a 14,700x reliability improvement over single-agent execution while reducing costs by 80%. Our work establishes that reliability in AI systems emerges from principled redundancy and consensus rather than model scaling alone.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑