Overview
AI researcher and educator. Former Director of AI at Tesla (2017–2022) and founding member of OpenAI. Known for influential technical writing and educational content on neural networks. Published the LLM wiki agent concept — a framework for using language models as active maintenance agents in personal knowledge systems — which has been widely adopted as an alternative to traditional note-taking systems.
Key Facts
- Former Director of AI at Tesla; left to return to OpenAI, then left OpenAI again in 2023
- Early OpenAI researcher; part of the founding team
- Published a gist describing the “LLM wiki agent” architecture: CLAUDE.md schema + index.md + log.md, with the LLM acting as the maintenance agent Obsidian Was Never the Problem
- Known for “micrograd” and “nanoGPT” — educational implementations of neural networks that demystify how LLMs work
- His framing: the bottleneck in knowledge systems is retrieval, not capture — and retrieval requires an active agent
Newsletter Relevance
Technology & AI: Karpathy occupies a rare position — technically credible enough to be cited by engineers, but writes for general audiences. His LLM wiki concept crossed from technical circles to productivity communities because it solved a genuine human problem (maintenance), not a technical one.
Connections
- LLM Wiki Agent — the concept he originated
- PKM Failure Pattern — the problem his framework addresses
- Vannevar Bush — intellectual predecessor; Bush’s 1945 Memex described the associative retrieval problem that Karpathy’s architecture actually solves
Source Appearances
- Obsidian Was Never the Problem — credited as originator of the LLM wiki agent architecture
Open Questions
- Has Karpathy written more extensively about how the LLM wiki architecture should evolve as models improve?
- What does Karpathy’s wiki actually contain — what domains does he use it for?