Unverified50% confidenceFactExact time
Sparse attention reduces the complexity to O(nkd) by having each query interact with only k keys, but it introduces the indexing cost of selecting those k keys.
1
Sources
50%
Confidence
Long-term
Relevance
11/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedFull-context injection avoids index staleness drift and retrieval missing key passages, but the cost is consuming tokens on every call, so it is only suitable for small-scale knowledge bases; large and dynamic knowledge bases must rely on retrieval.64% similarVerifiedThe self-attention mechanism uses dot-product operations among three matrices—Query, Key, and Value—to overcome the parallelization bottleneck caused by the sequential computation of RNNs/LSTMs.62% similarUnverifiedThe choice of embedding model, chunking strategy (chunk size), and Top-K parameter tuning during retrieval all directly affect the answer quality of the knowledge base.62% similarUnverifiedThe bottleneck of single-pass retrieval stems from vector similarity search compressing the user query into a single high-dimensional vector; when a question contains multiple semantic directions, this vector becomes an average of multiple semantics, causing sub-intents to be insufficiently represented.60% similarUnverifiedMulti-hop queries need to traverse multiple information nodes in a step-by-step manner, and vector retrieval is ill-suited for such scenarios due to its lack of a relational dimension.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1126133API
curl https://kongchang.com/api/v1/knowledge/claims/1126133MCP
get_claim(id=1126133)