Andrej Karpathy

AI researcher. Former director of AI at Tesla, former OpenAI founding member. Known for influential educational content on neural networks, deep learning, and AI systems.

Channels

• YouTube: Andrej Karpathy — deep technical AI education
• X/Twitter: @karpathy
• GitHub: github.com/karpathy — open-source projects and gists
• Blog: karpathy.ai

Content in This Wiki

• Andrej Karpathy Just 10x’d Everyone’s Claude Code (covered via Nate Herk) — Karpathy’s viral LLM Wiki Pattern gist; Nate Herk’s video is the primary source ingested, not the gist itself

Key Ideas

• Published the LLM Wiki Pattern as a gist (April 2026) — using LLMs to maintain structured markdown knowledge bases instead of relying on chat history or RAG
• His own wiki: ~100 articles / ~500K words; uses index files rather than embedding-based retrieval
• Left the pattern intentionally vague for others to customize
• Connects the idea to Vannevar Bush’s Memex (1945)
• AutoResearch (autoresearch): Open-sourced auto_research repo — gives an agent a training setup, single GPU, and success metric → agent runs ~100 experiments overnight autonomously. Persistent log turns random search into intelligent convergence. Hit 50K GitHub stars in a single week per the Dubibubii ingest. Shopify CEO Toby Lutke used it to produce an agent-optimized model that outperformed a larger human-tuned model. Also inspired Nick Saraev’s application to Claude Code skill optimization and Hermes Agent’s self-improving loop concept.

Gaps

• Karpathy’s original gist has not been ingested directly — would add depth and primary-source details to LLM Wiki Pattern

See Also

• LLM Wiki Pattern
• AutoResearch and Evals — methodology originating from Karpathy’s repo
• Nate Herk
• Claude Code
• Source: Karpathy 10x’d Claude Code