Why Knowledge Should Survive the LLM
Why reusable knowledge should live outside any one LLM, preserve changes in meaning, keep multiple perspectives, and remain traceable over time.
技術ノート
ブログ
実際に使った技術、その可能性、限界、実装や運用で重要になる細部についての技術ノートです。記事本文は原則として英語です。
このサイトの記事
FB15K: Loading FB15K-237 into Neo4j
A practical Cypher import—and what a graph of MIDs and 237 relation names reveals about knowledge graph benchmark data.
FB15K: Understanding the Dataset
What FB15K and FB15K-237 contain, why inverse relations matter, and what these widely used benchmarks actually measure.
COVID-19: Same Patient, Different Data
How mixed CSV and Excel files, Japanese character encodings, source-specific patient IDs, and conflicting values turned COVID-19 data integration into an identity and provenance problem.
COVID-19: Validating Related Patient Data with Graphs
How Neo4j and Git history helped reveal suspicious relationships, report them to local governments, and trace later corrections.
Learning Neo4j: A Japanese Citrus Family Tree
Using Japanese citrus varieties to learn graph modeling with parent-child relationships, intermediate breeding lines, and path queries.
Migration Planning: Using Graphs
How graph modeling can help clarify dependencies, inconsistencies, validation points, and cutover risks in system and data migration.
Learning Neo4j: Okonomiyaki
Using a familiar food domain to learn graph modeling with shops, places, noren relationships, ingredients, and cooking steps.
外部メディアの記事
Knowledge Must Live Outside the LLM—and Carry Its History
Why knowledge should live outside individual LLMs and preserve its history as models, tools, and contexts change.