LLM Wiki: Persistent Knowledge Over RAG

• Core argument: RAG retrieves from raw documents on each query but still rediscovers knowledge from scratch every time; a persistent wiki is better.
• Persistent compounding: The wiki is a “persistent, compounding artifact” — synthesis becomes durable rather than disposable.
• Human role: The human curates; the LLM handles the bookkeeping burden that humans typically abandon over time.

Entities: Andrej Karpathy