karpathy

LLM Wiki

A pattern for building personal knowledge bases using LLMs.

This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.

The core idea

Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.

schollz

// save as index.js
// npm install https://github.com/GoogleChrome/puppeteer/
// node index.js URL
const puppeteer = require('puppeteer');

(async () => {
  const url = process.argv[2];
  const browser = await puppeteer.launch();
  // use tor
  //const browser = await puppeteer.launch({args:['--proxy-server=socks5://127.0.0.1:9050']});

rspeer

/* This Rust code scans through the Common Crawl, looking for text that's
 * not English. I suspect I may learn much later that it's terrible,
 * unidiomatic Rust, but it would take me months to learn what good Rust is.
 *
 * We depend on some external libraries:
 * 
 *   - html5ever: an HTML parser (we only use its low-level tokenizer)
 *   - encoding: handles text in all the encodings that WHATWG recognizes
 *   - string_cache: interns a bunch of frequently-used strings, like tag names -- necessary to use
 *     the html5ever tokenizer

trekdemo

/*
 *
 * Mongo-Hacker
 * MongoDB Shell Enhancements for Hackers
 *
 * Tyler J. Brock - 2013
 *
 * http://tylerbrock.github.com/mongo-hacker
 *
 */

evernotegists

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE en-export SYSTEM "http://xml.evernote.com/pub/evernote-export3.dtd">
<en-export export-date="20130730T205637Z" application="Evernote" version="Evernote Mac">
    <note>
        <title>Test Note for Export</title>
        <content>
            <![CDATA[<?xml version="1.0" encoding="UTF-8" standalone="no"?>
            <!DOCTYPE en-note SYSTEM "http://xml.evernote.com/pub/enml2.dtd">
            <en-note style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space;">
                Hello, World.

didip

import unittest, os, os.path, sys, urllib
import tornado.database
import tornado.options
from tornado.options import options
from tornado.testing import AsyncHTTPTestCase

# add application root to sys.path
APP_ROOT = os.path.abspath(os.path.join(os.path.dirname(__file__), '..'))
sys.path.append(os.path.join(APP_ROOT, '..'))

tim-smart

--langdef=js
--langmap=js:.js
--regex-js=/(,|(;|^)[ \t]*(var|let|([A-Za-z_$][A-Za-z0-9_$.]+\.)*))[ \t]*([A-Za-z0-9_$]+)[ \t]*=[ \t]*\{/\5/,object/
--regex-js=/(,|(;|^)[ \t]*(var|let|([A-Za-z_$][A-Za-z0-9_$.]+\.)*))[ \t]*([A-Za-z0-9_$]+)[ \t]*=[ \t]*function[ \t]*\(/\5/,function/
--regex-js=/(,|(;|^)[ \t]*(var|let|([A-Za-z_$][A-Za-z0-9_$.]+\.)*))[ \t]*([A-Za-z0-9_$]+)[ \t]*=[ \t]*\[/\5/,array/
--regex-js=/(,|(;|^)[ \t]*(var|let|([A-Za-z_$][A-Za-z0-9_$.]+\.)*))[ \t]*([A-Za-z0-9_$]+)[ \t]*=[ \t]*[^"]'[^']*/\5/,string/
--regex-js=/(,|(;|^)[ \t]*(var|let|([A-Za-z_$][A-Za-z0-9_$.]+\.)*))[ \t]*([A-Za-z0-9_$]+)[ \t]*=[ \t]*(true|false)/\5/,boolean/
--regex-js=/(,|(;|^)[ \t]*(var|let|([A-Za-z_$][A-Za-z0-9_$.]+\.)*))[ \t]*([A-Za-z0-9_$]+)[ \t]*=[ \t]*[0-9]+/\5/,number/
--regex-js=/(,|(;|^)[ \t]*(var|let|([A-Za-z_$][A-Za-z0-9_$.]+\.)*))[ \t]*([A-Za-z0-9_$]+)[ \t]*=[ \t]*.+([,;=]|$)/\5/,variable/
--regex-js=/(,|(;|^)[ \t]*(var|let|([A-Za-z_$][A-Za-z0-9_$.]+\.)*))[ \t]*([A-Za-z0-9_$]+)[ \t]*[ \t]*([,;]|$)/\5/,variable/