Andrej Karpathy’s LLM Wiki
Photo by Steve Johnson on Unsplash

Andrej Karpathy broke the AI internet with a zero-code GitHub Gist.

He introduced the LLM Wiki.

Problem: Standard RAG suffers from amnesia, re-discovering knowledge on every query.

But, Karpathy’s architecture shifts the paradigm using:

  1. immutable raw sources,
  2. an LLM-maintained markdown wiki,
  3. a behavioral schema.

The LLM acts as an autonomous librarian — summarizing, cross-referencing, and updating connections.

Seriously? No vector database is required?

The result? The community is split:

Did he kill RAG or just reinvent the cache layer?

Both sides miss the truth, to be honest.

Knowledge should compound, not evaporate.

LLMs are finally capable of the tedious bookkeeping required to actually maintain a real “Second Brain.”

And I know that it lacks enterprise scaling, but for personal knowledge management , it is an absolute weapon.