Files
nexus/wiki/sources/multi-agent-system-reliability.md

44 lines
1.7 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
---
title: "Multi-Agent System Reliability"
type: source
tags: [multi-agent, reliability, architecture, LLM]
date: 2023-01-09
---
## Source File
- [[raw/AI/Multi-Agent System Reliability.md]]
## Summary
- 核心主题:多智能体系统可靠性架构模式
- 问题域LLM的不可靠性幻觉、逻辑谬误、上下文漂移导致多智能体系统崩溃
- 方法/机制:层级、共识、对抗辩论、淘汰制四种架构模式
- 结论/价值将LLM视为分布式系统中不可靠的组件通过架构设计强制正确性
## Key Claims
- 4种可靠性架构模式Hierarchy层级、Consensus共识、Adversarial Debate对抗辩论、Knock-out淘汰制
- 单一模型20%幻觉率下3个模型共识投票可将谎言概率降至0.8%
- 不要将LLM拟人化它们没有死亡恐惧或同理心
- 构建稳健系统需要停止要求模型"小心",而是强制它正确
## Key Quotes
> "We don't need AI that 'cares.' We need AI that is constrained, verified, pruned, and challenged."
## Key Concepts
- [[层级模式]]Supervisor规划器分配任务Worker执行Validator验证
- [[共识模式]]:多数投票机制降低幻觉概率
- [[对抗辩论]]Generator提出Critic攻击Judge裁决
- [[淘汰制]]多个agent竞争验证器淘汰最差者
- [[LLM不可靠性]]:幻觉、逻辑谬误、上下文漂移
## Key Entities
- [[Alex Ewerlöf]]:作者,可靠性工程专家
## Connections
- [[多Agent系统]] ← uses ← [[层级模式]]
- [[多Agent系统]] ← uses ← [[共识模式]]
- [[多Agent系统]] ← uses ← [[对抗辩论]]
- [[多Agent系统]] ← uses ← [[淘汰制]]
- [[可靠性工程]] ← applies_to ← [[多Agent系统]]
## Contradictions