BREAKING
Karpathy Publishes "LLM-Wiki" Pattern
How LLM-Wiki Works
1
Raw sources kept immutable
↓
2
LLM agent extracts entities
↓
3
Builds linked wiki
↓
4
Knowledge compounds
RAG vs LLM-Wiki
RAG
●
Re-synthesizes per query
●
Temporary knowledge
●
Hard to compound
LLM-Wiki
●
Uses compiled wiki
●
Persistent and structured
●
Grows in value over time
Obsidian as the IDE
0
+
pages converted
A Free, Open Knowledge Pattern
AI NEWS BLITZ
Andrej Karpathy has shared a new design pattern for building persistent AI knowledge bases.