LLM Wiki
Source
Andrej Karpathy published the note on 4 April 2026. He proposes that a language model should gradually build a lasting wiki rather than look up fragments of the sources for every question.
The pattern has three layers: raw sources that the model does not change, a wiki written entirely by the model, and a schema that sets the rules the model works by. A person chooses the sources and asks the questions, while the model summarises the sources, links the pages and keeps the records.
In my articles
I describe this pattern in the second part of my second brain series. I have been working this way for several months, and this glossary follows the same approach.
Sources
Related articles
Related entries
Last checked:
All entriesAndrej Karpathy
Andrej Karpathy is a computer scientist, a co-founder of OpenAI and a former director of AI at Tesla. He has also described a way to have a language model maintain a wiki.
LLM (large language model)
An LLM, or large language model, is a program that has been trained on vast amounts of text and can generate answers and other writing.