Unverified50% confidenceDecision RuleExact time
确定性的信息应走Context，允许Agent探索的走Query；放错会导致Context塞入过多Token增加干扰，或Query配置过窄导致LLM编造
1
Sources
50%
Confidence
Long-term
Relevance
7/6/2026
First Seen
Sources
Palantir Agent与Ontology交互机制深度解析
bilibili零点未来7/4/2026
Related Claims
Unverified由于Perplexity需要实时检索和生成答案，其延迟通常高于简单返回链接列表的传统搜索API65% similarUnverifiedContext window limitations in LLMs cause them to struggle with schema information spanning hundreds of columns64% similarUnverifiedRAG架构通过在推理时将外部知识库的相关片段动态注入上下文，缓解LLM的知识截止问题，但也扩大了间接提示注入的攻击面64% similarUnverified将推理链隐藏在API后端并不等于真正保护它，只要模型行为可被外部观测和大量查询就存在被逆向工程的空间63% similarVerified自注意力机制通过Query、Key、Value三个矩阵的点积运算，突破了RNN/LSTM因顺序计算导致的并行化瓶颈63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/113733API
curl https://kongchang.com/api/v1/knowledge/claims/113733MCP
get_claim(id=113733)