Unverified50% confidenceBenchmarkExact time
SAG2 uses no embeddings, adopting SQLite plus BM25 together with a lightweight "System One" model to make contextual evidence decisions at query time; indexing about 11,600 passages takes only 2.36 seconds with zero model calls, matching the recall level of dense retrieval on MuSiQue, 2Wiki, and HotpotQA.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAt very large scales (above tens of millions of vectors), pgvector's index build speed and query throughput are generally inferior to dedicated vector database solutions.59% similarUnverifiedThe model does not look up information in the context the way a database retrieves rows; every token in the window competes for a limited "attention budget."57% similarUnverifiedCompared to dedicated vector databases such as Pinecone and Weaviate, pgvector requires no additional operations for a separate system, but its retrieval performance at very large data scales is generally inferior to dedicated solutions.57% similarUnverified相比RAG提交prompt有硬上限，图谱工具调用允许模型按需调用任意次数以挖掘所需信息56% similarUnverifiedOn both datasets, the number of queries in which schemagate actually read a forbidden column was 0.56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1067949API
curl https://kongchang.com/api/v1/knowledge/claims/1067949MCP
get_claim(id=1067949)