Unverified60% confidenceTradeoffExact time
任务拆得太粗等于没拆，拆得太细则会导致上下文传递成本爆炸
2
Sources
60%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
Unverified对于简单任务使用旗舰模型属于过度供给，切换到轻量模型可在几乎不影响体验的前提下将相关成本削减数倍73% similarUnverified状态离散化存在权衡：分桶粒度过粗导致信息损失，过细则导致维度灾难70% similarUnverified面向Agent的编程评测面临任务空间爆炸导致评测成本极高、多步骤交互引入随机性难以复现的难点70% similarUnverified针对某类任务的专项强化往往以牺牲其他任务的表现为代价，此现象被称为灾难性遗忘70% similarUnverified工具过于细粒度会导致规划步骤爆炸，工具过于粗粒度则会牺牲灵活性，这是Agent系统设计中最常见的权衡问题之一69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/896065API
curl https://kongchang.com/api/v1/knowledge/claims/896065MCP
get_claim(id=896065)