> [!bot-text] AI-generated
> This text was drafted by [[Jake]] and has not been reviewed.
> [!info] Source
> [LLM Knowledge Bases](https://x.com/karpathy/status/2039805659525644595/?rw_tt_thread=True)
> by <!-- IQ: =this.creator -->[[Andrej Karpathy|Andrej Karpathy]]<!-- /IQ -->
> Related: <!-- IQ: =this.related -->[[Concepts/AI.md|AI]], [[ia|ia]], [[Concepts/semantic web.md|semantic web]], [[Concepts/knowledge graph.md|knowledge graph]]<!-- /IQ -->
## Summary
Andrej Karpathy uses large language models (LLMs) to build personal knowledge bases by compiling data into markdown wikis. The LLM writes, maintains, and answers complex questions from the wiki, which is viewed and managed in Obsidian. This method helps organize research efficiently and suggests new insights without manual editing. *(Summarized by Ghostreader)*
## Significance
Directly relevant to the PSKA knowledge garden approach — a prominent practitioner running essentially the same architecture (LLM + markdown wiki + Obsidian).
## Claims & Insights
Claims that cite this summary (claims link here via `source`):
<!-- QueryToSerialize: TABLE WITHOUT ID link(file.link, file.title) AS "Claim", confidence AS "Confidence" FROM #claim WHERE contains(file.outlinks, [[Summary - LLM Knowledge Bases (Karpathy)]]) SORT file.mtime DESC -->
<!-- SerializedQuery: TABLE WITHOUT ID link(file.link, file.title) AS "Claim", confidence AS "Confidence" FROM #claim WHERE contains(file.outlinks, [[Summary - LLM Knowledge Bases (Karpathy)]]) SORT file.mtime DESC -->
| Claim | Confidence |
| ----- | ---------- |
<!-- SerializedQuery END -->
## Concepts touched
```dataview
LIST WITHOUT ID
link(file.link, file.title)
FROM #concept
WHERE contains(this.file.outlinks, file.link)
SORT file.mtime DESC
```