🚀 New Video: I Built an AI Second Brain That Fixes Its Own Mistakes (Karpathy's Method)
Karpathy posted his "LLM knowledge base" setup and 100k+ people saved it. Almost nobody built it. So I built a simpler, local version in Claude Cowork that files your notes, answers from your own stuff, and once a month audits and cleans itself. No code, no Obsidian, no database. - Three folders, one file. raw (your messy inbox), wiki (AI writes it, you never touch it), outputs (answers + reports), plus a CLAUDE.md rulebook that makes Claude your librarian. - One prompt builds the wiki. Claude reads everything in raw, finds the topics, writes a plain-English page each, links them, and indexes it. No RAG needed at this size. - Answers cite sources and save themselves. Ask a question, it pulls the exact pages, tells you which it used, and files the answer in outputs so the brain compounds. - The monthly check-up is the point. It hunts contradictions, unsourced claims, and gaps, web-searches to fact-check, and shows a report before changing anything. Save it as a Skill, put it on a timer, done. 📎 Full Second Brain guide PDF pinned below — every prompt, the exact CLAUDE.md rules, and the monthly audit skill. This is the kind of build we run inside The AI Accelerator (free, 19k+ members). Come grab the prompts and share your setup. 👉 What would you point your second brain at first — AI, your clients, or something else? [Watch the video here ▶️]