LLM Wiki

Appearance

Built 26/04/19 07:55commit 447639a

LLM Wiki ​

中文 | English

Summary ​

LLM Wiki is a pattern where an LLM maintains a persistent markdown wiki that sits between stable raw sources and downstream questions, so synthesis compounds over time instead of being recomputed from scratch on every query.

Core Idea ​

  • Most document workflows behave like RAG: raw files are searched at query time and the answer is rebuilt each time.
  • LLM Wiki moves the synthesis step forward in time. New sources are read once, integrated into markdown pages, and kept current as the corpus grows.
  • The maintained wiki becomes the default answer surface, with raw files reserved for verification and deeper re-reading.

Key Claims ​

  • The wiki is a persistent, compounding artifact rather than a disposable query result.
  • Cross-links, contradictions, and evolving summaries should be maintained directly in markdown.
  • The abstract pattern can be packaged as a reusable skill bundle, as long as the scripts remain narrow helpers and the model still owns synthesis and editorial judgment.
  • Periodic consolidation should repair contradictions at the source, convert relative dates to absolute ones, and keep top-level index files compact enough to stay cheap to load.
  • Raw sources stay stable in substance, but their directory layout can still evolve if moved paths, local links, and source-page references are repaired together.
  • Raw translation siblings should remain faithful to the source; summaries and editorial condensation belong in the maintained wiki layer instead.
  • Screenshot-heavy sources should be turned into textual evidence through OCR and editorial synthesis instead of remaining opaque image dumps, with Chinese-capable OCR enabled when the source material demands it.
  • AGENTS.md is not metadata; it is the schema that turns the model into a disciplined maintainer.
  • index.md, index.zh.md, and log.md are first-class operating files that make the corpus navigable without heavier infrastructure at small scale.

Typical Uses ​

  • Personal knowledge management across journals, articles, and self-tracking notes.
  • Research programs that accumulate papers, reports, and evolving theses over weeks or months.
  • Book or media companion wikis that track characters, themes, and plot threads.
  • Team or business knowledge bases fed by documents, conversations, and project artifacts.

Sources ​