A pattern for building personal knowledge bases with LLMs. Not RAG — a compiled, persistent wiki that the LLM builds and maintains incrementally.
"The wiki is a persistent, compounding artifact. The cross-references are already there. The contradictions have already been flagged. The synthesis already reflects everything you've read." — Andrej Karpathy
Standard setup (NotebookLM, ChatGPT file uploads, generic RAG): drop documents in, retrieve on every question, the LLM rediscovers your knowledge from scratch each time. No accumulation, no compounding. A subtle question that synthesizes five sources starts at zero, every time.