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.

brianloveswords

<audio id="alarm" src="alarm.ogg" controls></audio>
<script>
 const alarmElement = document.getElementById('alarm');
 const alarmTimes = [
   new Date('2015-11-13 07:00 -0500'),
   new Date('2015-11-13 08:00 -0500'),
   new Date('2015-11-13 08:30 -0500'),
 ];

 function playAlarm() {

jturkel

CREATE VIEW product_stats AS
SELECT organization_id, 
       SUM(CASE WHEN status = 'active' THEN 1 END) active_count, 
       SUM(CASE WHEN status = 'inactive' THEN 1 END) inactive_count
FROM products
GROUP BY organization_id

dhh

class PostsController < ActionController::Base
  def create
    Post.create(post_params)
  end
  
  def update
    Post.find(params[:id]).update_attributes!(post_params)
  end

  private

mattetti

module Excel
  module Formulas  
    def pmt(rate, nper, pv, fv=0, type=0)
      ((-pv * pvif(rate, nper) - fv ) / ((1.0 + rate * type) * fvifa(rate, nper)))
    end

    def ipmt(rate, per, nper, pv, fv=0, type=0)
      p = pmt(rate, nper, pv, fv, 0);
      ip = -(pv * pow1p(rate, per - 1) * rate + p * pow1pm1(rate, per - 1))
      (type == 0) ? ip : ip / (1 + rate)