danieldietrich

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.

danieldietrich

People

:bowtie:	😄 :smile:	😆 :laughing:
😊 :blush:	😃 :smiley:	☺️ :relaxed:
😏 :smirk:	😍 :heart_eyes:	😘 :kissing_heart:
😚 :kissing_closed_eyes:	😳 :flushed:	😌 :relieved:
😆 :satisfied:	😁 :grin:	😉 :wink:
😜 :stuck_out_tongue_winking_eye:	😝 :stuck_out_tongue_closed_eyes:	😀 :grinning:
😗 :kissing:	😙 :kissing_smiling_eyes:	😛 :stuck_out_tongue:

danieldietrich

Purpose

Bootstrap knowledge of LLMs ASAP. With a bias/focus to GPT.

Avoid being a link dump. Try to provide only valuable well tuned information.

Prelude

Neural network links before starting with transformers.

danieldietrich

{
    "version": "0.2.0",
    "configurations": [
        {
            "name": "Run Extension",
            "type": "extensionHost",
            "request": "launch",
            "args": [
                "--extensionDevelopmentPath=${workspaceFolder}"
            ]

danieldietrich

IsEmpty<T>

A value of type T is defined to be empty, if

• it is an array and has no elements
• it is an object and has no properties

Handling of universal types

• any can be seen as the union of all types. A union distributes in conditional types, so any may or may not be empty. true | false = boolean

danieldietrich

• code: Google & StackOverflow
• music:
  • https://soundcloud.com/aennigma/aennigma-at-70109-stunden-sudpol
• https://soundcloud.com/muybien-kollektiv/muy-bien-63-lukov

danieldietrich

Def. A conditional type with a checked naked type parameter T is called distributive conditional types.

Example:

// given U, X and Y
type DistributiveConditionalType<T> = T extends U ? X : Y;

Distributive conditional types are automatically distributed over union types during instantiation.

danieldietrich

#!/bin/bash

# Tom Hale, 2016. MIT Licence.
# Print out 256 colours, with each number printed in its corresponding colour
# See http://askubuntu.com/questions/821157/print-a-256-color-test-pattern-in-the-terminal/821163#821163

set -eu # Fail on errors or undeclared variables

printable_colours=256

danieldietrich

#
# Usage: `source <(curl -sL https://bashrc.danieldietrich.dev)`
#

export LANG=en_US.UTF-8
export LC_ALL=${LANG}
export TERM=xterm-256color
export GREP_OPTIONS='--color=auto' GREP_COLOR='1;32'
export CLICOLOR=1
export LSCOLORS=gxBxhxDxfxhxhxhxhxcxcx

danieldietrich

/*
 * Easing Functions - inspired from http://gizma.com/easing/
 * only considering the t value for the range [0, 1] => [0, 1]
 */
EasingFunctions = {
  // no easing, no acceleration
  linear: t => t,
  // accelerating from zero velocity
  easeInQuad: t => t*t,
  // decelerating to zero velocity