斯坦福研究:多Agent共享资源易冲突,单Agent协调更优
New Stanford paper finds that when each person's agent acts alone on a shared re…
Agent协作是当下热点,这篇研究用数据揭示了多Agent共享资源的致命缺陷,给做Multi-Agent系统的同学提供了关键的架构取舍依据。
New Stanford paper finds that when each person's agent acts alone on a shared resource, the group does worse than 1 agent serving everyone.
新的斯坦福论文发现,当每个人的智能体在共享资源上单独行动时,群体的表现不如由1个智能体为所有人服务。
A shared budget or calendar is handled better by 1 agent serving everyone than by 1 agent per user, across 5 frontier models.
在5种前沿模型中,由1个智能体为所有人服务的共享预算或日历处理效果优于每个用户配备1个智能体。
Each agent does a sensible job for its own user.
每个智能体都能为其各自的用户完成合理的工作。
Together they overwrite each other, stall as the team grows, and with no channel they collapsed outright in 2 environments.
但它们会相互覆盖,随着团队规模扩大而陷入停滞,且在缺乏沟通渠道的情况下,在两种环境中完全崩溃。
On a contested token budget, Opus 5 teams captured 30% of the achievable value against 64% for 1 coordinating agent. Agents invented facts about other users in more than half of Claude team episodes in the group-ordering environment.
在有争议的令牌预算下,Opus 5团队仅获得了可实现价值的30%,而1个协调智能体则达到了64%。 在群体排序环境中,超过一半的Claude团队会话里,智能体编造了关于其他用户的虚假事实。
Prefer 1 agent holding everyone's constraints, and if you run 1 per user, make reading peers a condition of committing.
倾向于让1个智能体掌握所有人的约束条件;如果你按用户分配智能体,请将查看同伴信息作为提交操作的前提条件。
更进一步:量化金融体系
看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力