BREAKING
Karpathy's LLM Wiki Method
Three Core Actions
1
Ingest sources
↓
2
Query & synthesize
↓
3
Lint contradictions
LLM Wiki vs RAG
LLM Wiki
●
Persistent vault
●
No vector DB
●
Fully local
Conventional RAG
●
Re-retrieves chunks
●
Resets context
●
Weak synthesis
0
x
faster than grep
0
files in vault
0
published skills
Setup in 5 to 15 Minutes
Local, Compounding Knowledge
AI NEWS BLITZ
Andrej Karpathy pairs Claude with Obsidian to build a local knowledge base.