LLMs Change — Where Should Knowledge Live? (Lessons From SBOM)
summary
LLMs produce useful answers but those answers change with model, prompt, and context, so knowledge needs to live as structured, persistent data outside the model. The claim: store facts, relationships, provenance and history in a graph so RAG returns traceable evidence — demonstrated with a Wikipedia example of 500+ revisions and an implementation using Neo4j and Cypher to record sources, model/prompt versions and generated outputs. Built components include a knowledge-bomb graph that traces dependencies, supports compare/version queries, and exports to SPDX3.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 0:45Knowledge beyond simple answers
- 3:17Knowledge needs structure
- 5:30Wikipedia and knowledge changes
- 7:45LLM evolution and limits
- 9:38RAG limits and traceability
- 11:07What is an SBOM
- 13:22AI as knowledge supply chains
- 15:18Extending SBOM to KnowledgeBOM
- 17:12Neo4j graph database
- 20:29Knowledge outside LLMs
- 23:59Key takeaways
- 26:37Q&A
related talks
52mGive Your LLMs a Left Brain
44mGive Your LLMs a Left Brain
32mTruth-seeker: Using LLM agents to build and verify knowledge bases
23mKnowledge Graphs and LLMs in Action
47mGive Your LLMs a Left Brain
25mEnhance LLM Applications' Trustworthiness and Explainability
30mBuilding Knowledge Graphs With LLMs
20mPractical GraphRAG: Making LLMs smarter with Knowledge Graphs