Le co-fondateur d'OpenAI vient de changer Claude Code pour toujours (LLM Wiki)

Le co-fondateur d'OpenAI vient de changer Claude Code pour toujours (LLM Wiki)

The OpenAI Co-founder Just Changed Claude Code Forever (LLM Wiki)

Title history
  • Le co-fondateur d'OpenAI vient de changer Claude Code pour toujours (LLM Wiki) — observed on 20261112-0057 UTC (YouTube)

by Thomas Berton · View the channel on PSI

Applied Sciences & Engineering Computing & Cybersecurity UMComputer programmingUMZSoftware Engineering
🎙 Thomas Berton 👥 48K 📅 May 21, 2026 ⏱ 21 min 👁 8K 📄 tutorial 🧭 2026-11-12
Available in: English (current) Français

Keywords

LLM WikiClaude CodeRAGObsidianKnowledge base

Summary

The video presents a new method proposed by Andrej Karpathy, called ‘LLM Wiki’, which aims to replace traditional RAG (Retrieval-Augmented Generation) systems using interconnected Markdown files. The creator demonstrates how to set up this system using Obsidian and Claude Code, without the need for a vector database, server, or external costs. The process involves downloading Obsidian, creating a vault, opening Claude Code in the project folder, copying the entire LLM Wiki documentation from Karpathy’s GitHub, and providing a prompt to Claude, which then autonomously builds the wiki with files and links. The video shows a real-time construction of the wiki, including automatic scraping of a website (thomas-berton.com) and blog articles. Finally, a test in a fresh session shows that Claude Code can answer specific questions about the creator’s business using the wiki, demonstrating the effectiveness and simplicity of the method. The creator emphasizes the zero cost, lightweight nature, and potential for various use cases, such as client knowledge bases or content creation wikis.

163 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video offers a clear, step-by-step tutorial that is highly practical and easy to follow. The argumentation for the value of LLM Wiki is based on a live demonstration: the system is set up from scratch, automatically populates with real data, and successfully answers specific questions in a blank Claude Code session, proving its functionality. The creator also compares it to RAG, highlighting its simplicity and lower cost. However, the argumentation is largely anecdotal and based on a single use case (his own business). While the method seems promising, the video does not provide rigorous benchmarks or comparative analysis against RAG systems, and the claimed ‘95% token savings’ is mentioned without evidence.

Scientific Rigor, Source Quality, Title Accuracy

The tutorial is structured logically, with clear chapters and timestamps. The main source is the official LLM Wiki gist by Andrej Karpathy (linked in the description), which adds credibility. The creator also uses and shows actual tools (Obsidian, Claude Code) in action. The title accurately reflects the content. However, there is a commercial aspect: the creator promotes his own services, but this is not intrusive and does not affect the technical explanation. The public reaction (not provided) could not be analyzed here, but the content is reproducible and transparent.

216 words

Title / Content Match

The title accurately reflects the content: it focuses on Karpathy's concept and its impact on Claude Code, and the video delivers a practical implementation. Minor caveat: the title says 'co-founder of OpenAI' while Karpathy is often described as co-founder, but it's accurate.

Quality & Reliability

7/10

The video provides a hands-on tutorial based on a proposal by Andrej Karpathy (LLM Wiki), with working demonstrations and a test in a blank session. However, claims such as '95% token savings' are taken from Karpathy and not independently verified, and the evaluation is mainly anecdotal rather than systematic.

Chapters

Cited Sources

External References

Contribution & Novelties

The video introduces a novel concept (LLM Wiki) proposed by Andrej Karpathy that leverages human-friendly Markdown files instead of vector databases for AI context retrieval. This approach is simple, free, and transparent, contrasting with traditional RAG. It offers a lightweight alternative for personal knowledge management and AI agents, potentially reducing token usage and infrastructure costs. The tutorial itself is valuable as it provides a hands-on implementation in under five minutes, making the concept accessible.

Pour aller plus loin :

148 words

Radar Profile

The radar profile shows well-balanced scores across most dimensions, with a slight dip in fiabilité_globale due to the lack of independent verification. The high scores in information quantity and quality reflect the dense practical content, while technical level is adequate for a semi-technical audience.

Reliability 6/10