Going Meta S03E03 – Ontologies, LLMs, Property Graphs – Do we really have a “Jaguar Problem”?

Season 03 Episode 3 of Going Meta – a Series on Semantics, Knowledge Graphs and All Things AI
Topic: Ontologies, LLMs, Property Graphs – Do we really have a “Jaguar Problem”?

Jesús Barrasa: https://twitter.com/BarrasaDV
Repository: https://github.com/jbarrasa/goingmeta
Knowledge Graph Book: https://bit.ly/3LaqE6b
Previous Episodes: https://neo4j.com/video/going-meta-a-series-on-graphs-semantics-and-knowledge/

Niklas Emegard on LinkedIn: https://www.linkedin.com/posts/niklasemegard_graphrag-knowledgegraphs-semanticweb-activity-7390898141820092416-0ojW/
GraphRAG Github Repo: https://github.com/nemegrod/graph_RAG
Going Meta S02E03 – Blueprints: https://www.youtube.com/watch?v=cPzy61odKCg

0:00 – Intro
4:41 – What is the “Jaguar Problem”? Context for the 6 claims
8:52 – Inaccuracy #1: “LLMs cannot deal with lexical ambiguity without help”
14:28 – Inaccuracy #2: “Feeding an ontology into the context window guarantees schema alignment”
23:02 – Inaccuracy #3: “Extracted triples require zero post-processing”
27:58 – Inaccuracy #4: “LLMs can run OWL reasoning (formal semantics)”
30:58 – Inaccuracy #5: “LLMs perform better when prompted with RDF/OWL vs natural language”
36:49 – Experimental results comparing NL vs OWL guidance
41:02 – Inaccuracy #6: “This cannot be done with LPG / property graph databases”
48:55 – Takeaways: Ontologies + LLMs + Property Graphs and how they *really* work together
57:51 – Closing Q&A

#graphdatabase #neo4j #graphrag #knowledgegraphs #ontology #data #llm