I Replaced Karpathy's LLM Wiki with Something That Actually Works

Artem Zhutov ยท Intermediate ยท๐Ÿ’ป AI-Assisted Coding ยท1mo ago
๐Ÿ“š Obsidian x Claude Code Lab (May 5): https://lab.artemzhutov.com โ†’ Build your personal productivity system with Claude Code and Obsidian. Small group, you build your actual system. The top comment on the biggest LLM Wiki video says it's largely worthless for most people. I took the same sources, the same goal, and built Karpathy's wiki AND my system side by side in real time. Wiki ingested a single article for 13 minutes. NotebookLM loaded the same sources in seconds. 10 questions answered in under a minute with citations back to the raw transcript. ๐ŸŽ NotebookLM CLI skill. Upload sources, ask questions in parallel, get cited answers in Obsidian. Ready to use: โ†’ No API keys, no setup. Just drop the skill and go ๐Ÿ‘‡ https://notebooklm-skill-artemzhutov.netlify.app Follow me: Substack: https://artemxtech.substack.com/ X: https://x.com/ArtemXTech GitHub: https://github.com/ArtemXTech Timestamps: 0:00 "Largely worthless for most people" 0:30 Same goal. Same sources. Both systems. 1:00 Building Karpathy's wiki from scratch 2:32 NotebookLM: the parallel approach 3:49 Sources loaded in seconds 5:06 No middle layer needed 5:52 13 minutes for a single article 6:08 10 questions, all at once 7:23 Cited answers in Obsidian 7:54 Dalio's actual 5-step process 8:39 Concept extraction without the wiki 10:40 Same result, fraction of the cost 11:11 Mind maps vs graph view 12:26 Asking the wiki the same questions 13:12 44,000 tokens blown 15:00 Where scale breaks everything 16:00 When each approach wins 17:31 Where the wiki actually wins 19:35 "You created a wiki. Now what?" 20:36 Building a decision skill from Dalio 21:52 Side-by-side comparison 22:23 QMD: the local alternative 24:26 The loop that closes everything 25:11 What knowledge source would you pick?
Watch on YouTube โ†— (saves to browser)
Sign in to unlock AI tutor explanation ยท โšก30

Related AI Lessons

โšก
What Broke After 10M WebSocket Events (And How We Rewired Our Realtime AI Pipeline)
Learn how to optimize your real-time AI pipeline after hitting scalability issues with 10M WebSocket events
Dev.to ยท hamza qureshi
โšก
# Building EduForge AI โ€” Turning Lessons Into Interactive Games with AI ๐Ÿš€
Learn how to turn lessons into interactive games with AI using EduForge AI, a platform that generates quizzes and games from natural language prompts
Dev.to AI
โšก
I Wrote an MCP Server for My 3D Printer
Learn how to write an MCP server for a 3D printer and improve your printing experience
Dev.to ยท Nic Lydon
โšก
GitHub Copilot Prompts vs Skills vs Instructions: What's the Difference and When to Use Each
Learn to effectively use GitHub Copilot's Prompts, Skills, and Instructions to codify knowledge and improve coding efficiency
Dev.to AI

Chapters (24)

"Largely worthless for most people"
0:30 Same goal. Same sources. Both systems.
1:00 Building Karpathy's wiki from scratch
2:32 NotebookLM: the parallel approach
3:49 Sources loaded in seconds
5:06 No middle layer needed
5:52 13 minutes for a single article
6:08 10 questions, all at once
7:23 Cited answers in Obsidian
7:54 Dalio's actual 5-step process
8:39 Concept extraction without the wiki
10:40 Same result, fraction of the cost
11:11 Mind maps vs graph view
12:26 Asking the wiki the same questions
13:12 44,000 tokens blown
15:00 Where scale breaks everything
16:00 When each approach wins
17:31 Where the wiki actually wins
19:35 "You created a wiki. Now what?"
20:36 Building a decision skill from Dalio
21:52 Side-by-side comparison
22:23 QMD: the local alternative
24:26 The loop that closes everything
25:11 What knowledge source would you pick?
Up next
Why Casey Muratori stopped using C++
NeetCodeIO
Watch โ†’