BREAKING
Karpathy's LLM Wiki Goes Viral
LLM Wiki vs Classic RAG
RAGold way
Re-retrieves chunks each query
No knowledge accumulation
LLM Wikinew way
Compiles sources once
Knowledge compounds over time
Three Core Operations
1Ingest sources
2Query and answer
3Lint contradictions
Three-Layer Structure
1Raw sources
2Wiki (LLM-built)
3Schema / CLAUDE.md
Praise and Limits
Upsides
Setup in 5 to 15 minutes
Maintenance nearly zero
Limits
index.md fails past 100 articles
Tokens and hallucination risk
A Local RAG-Free Alternative
AI NEWS BLITZ
Andrej Karpathy's LLM Wiki pattern turns an Obsidian vault into a self-updating knowledge base.