Comparison
Persistent Wiki vs RAG
Compare classic RAG with the Karpathy Wiki pattern and see why persistent synthesis changes the economics of long-running knowledge work.
Why this page matters
RAG is retrieval. Karpathy Wiki is accumulation. The difference shows up in synthesis, contradiction tracking, and answer reuse.
Why RAG feels stateless
Classic retrieval systems can be extremely useful, but they often start every subtle question from zero. The model must locate the relevant pieces, reconcile them on the fly, and then throw the synthesis away once the conversation ends.
That is fine for transactional lookup. It is less ideal for ongoing research because the cost of integration shows up again and again.
Why a wiki compounds
The wiki pattern changes the unit of work. Instead of only producing an answer, the model updates the artifact that future answers will rely on. Better summaries, cleaner backlinks, and stronger comparison pages all survive into the next session.
This is why the model can act more like a maintainer than a search endpoint. Every ingest improves the environment for the next query.
When to combine both
Karpathy explicitly leaves tooling open, and that is important. The wiki does not replace raw sources. It sits on top of them. At small scale, an index file may be enough to navigate the wiki. At larger scale, search tools can help the model find the right pages or even search the raw corpus directly.
The practical hybrid is straightforward: keep immutable sources, maintain a persistent wiki, and add stronger search when the page count or source count grows past what manual navigation can handle.
Side-by-side view
| Dimension | Classic RAG | Karpathy Wiki pattern |
|---|---|---|
| Knowledge state | Mostly rediscovered at query time from raw chunks. | Persisted as linked markdown pages that keep evolving. |
| Cross-source synthesis | Repeated on every question. | Written into the wiki once, then refined over time. |
| Contradictions | Easy to miss unless the current prompt surfaces them. | Tracked directly on the relevant pages and revisited during maintenance. |
| Answer reuse | Good chats often disappear into history. | Useful answers can be filed back into the wiki as new pages. |
| Maintenance burden | Humans still need to organize and reconcile the material manually. | The model handles the bookkeeping across many pages in one pass. |