Unverified50% confidenceTradeoffExact time
在多Agent编排场景下，每个Agent需要获得'恰好够用'的上下文，信息不足导致理解偏差，信息过载则稀释注意力增加幻觉风险
1
Sources
50%
Confidence
Long-term
Relevance
7/18/2026
First Seen
Sources
Related Claims
Unverified推理能力越强的模型生成内容越详尽，单次交互token消耗随之攀升，形成「能力越强、消耗越大」的技术约束72% similarUnverifiedHuman annotators exhibit a cognitive bias toward more information being better when evaluating answers, causing reward models to implicitly link length with quality, which is one of the causes of model verbosity.72% similarUnverified长上下文模型在自我审查场景中因锚定效应可能带来更大偏差，早期生成内容对后续注意力计算影响越显著69% similarUnverifiedMemory extractors tend to have higher recognition rates for explicitly declared information, while being prone to missing or misjudging user preferences or habitual patterns that are implicit in context.69% similarUnverified当任务包含天然独立的认知模式（如生成与批判）或存在明确质量把关需求时倾向于使用多Agent架构69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/550672API
curl https://kongchang.com/api/v1/knowledge/claims/550672MCP
get_claim(id=550672)