Retrieval-Augmented Generationconcept

RAG retrieves relevant context chunks for a query, assembles them into a prompt, and asks an LLM to generate an answer — fresh on every query.

RAG grounds LLM answers in external knowledge and reduces hallucination versus parametric-only generation. It is the dominant runtime-knowledge pattern SKCE contrasts itself against.

Where it appears

Descend

Historical evolution

Introduced 2020; became the default enterprise LLM pattern by 2023. SKCE argues it is structurally an interpreter — re-deriving meaning every query.

Implementation details

SKCE's explorer does zero RAG at runtime; the compiled curriculum is the static alternative.

Sources

View in graph →