RDF and the future of LLMs
summary
Argues that RDF makes LLMs practically useful by feeding emit RDF into prompts and asking models to generate RDF queries, then executing those against an RDF store with entailment rules so the system returns precise answers (example: genealogical descent resolved by reasoning). Uses Transformers/attention models (hundreds of billions of parameters) and embeddings to project RDF concepts into latent space for soft inference and exploratory linking. Prototype work aims to automate the “data dishes” and shows reliable structured outputs when the underlying data and rules exist.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 3:55What RDF is
- 12:23RDF triples
- 18:18RDF knowledge system
- 20:07Language models as measurements
- 22:52LLM API and AI programming
- 25:21Prompt engineering
- 31:45RDF bridges data logic language
- 33:31RDF and natural language
- 36:37Semantics in LLMs and RDF
- 39:27Soft inference and neurosymbolic AI
- 40:50AI automating data tasks
related talks
29mThe Evolution of LLMs: From Imitation to Reasoning
53mFrom RAG to Riches: Secrets of Data-Augmented LLMs
22mPractical Strategies for Implementing Domain-Specific LLMs
44mGive Your LLMs a Left Brain
31mBeyond LLMs and Maturing RAG use cases
15mFrom Words to Data: Using LLMs to Turn Questions into SQL
52mGive Your LLMs a Left Brain
24mConversational Search with LLMs