Llm-Wiki

RAG starts from scratch every time. Karpathy proposes having the LLM maintain a wiki directly so knowledge accumulates. DataNexus’s ontology catalog needs the same principle to avoid abandonment.