martin12333

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.

martin12333

tags	zero width space zero width non-joiner zero width joiner soft hyphen

Invisible "blank" Characters: How to write invisible characters (zero width space, zero width non-joiner, zero width joiner & soft hyphen

I think only 2 of these are "copy-able":

1. Zero-width space: ‎
2. Zero-width non-joiner: ‌‌‌

martin12333

Search Chrome Devtools history

1. Undock the console (click on the icon in the bottom-left corner, ![undock icon][1]).
   (if you don't see ![the undock icon][2], but ![][3], then hold the mouse pressed for a few seconds to get the desired icon)
2. Press Ctrl + Shift + J to open the console for this console. (On OSX use Cmd + Option + i)
3. Use the following snippet to get an array of matches for your search term:

const searchHistory = query => {console.dir(JSON.parse(localStorage.getItem('consoleHistory')).filter(function(item){ return ~item.indexOf(query);}))}
searchHistory('token')

martin12333

Segmentation 101, part 1: Why your strategy matters

I recent did some more exploring with a local LLM tool that would import your documents into a vector store.  Given the promising initial results with a handful of docs I wanted to see how it handled more / different data.  I decided to copy over the text files containing Expanse trivia and answers I use as a regression suite to test my own "Q&A over documents" process. I wanted to see what types of questions it could answer from that content...

The Problem With Generic Segmentation

The strategy employed by this tool used double newlines as their segmentation boundary condition. A strategy that works well for many types of content however for this content that was a terrible choice as the text in the files are formatted with numbered questions followed by their answers like this:

1. Long winded question with establishing context

 

martin12333

$ori = @{}
Try {
  $i = 0

  # Loading .env files
  if(Test-Path $args[0]) {
    foreach($line in (Get-Content $args[0])) {
      if($line -Match '^\s*$' -Or $line -Match '^#') {
        continue
      }

martin12333

C:\Users\user\Desktop>cscript //E:{1b7cd997-e5ff-4932-a7a6-2a9e636da385} repl.js
Microsoft (R) Windows Script Host Version 5.812
Copyright (C) Microsoft Corporation. All rights reserved.

> if(true){\
> 1\
> }
1
> var a = 10
undefined

martin12333

So we need people to have weird new
ideas ... we need more ideas to break it
and make it better ...

Use it. Break it. File bugs. Request features.

- Soledad Penadés, Real time front-end alchemy, or: capturing, playing,
  altering and encoding video and audio streams, without
  servers or plugins!
   

martin12333

/*
feidlambda v0.4.0 - LOGIC / UTILITIES FUNCTIONS BY FIAKO ENGINEERING
OFFICIAL GIST (feidlambda v0.4.x): 
    https://gist.github.com/taruma/b4df638ecb7af48ab63691951481d6b2
REPOSITORY: 
    https://github.com/fiakoenjiniring/feidlambda
CONTRIBUTOR: @taruma, @iingLK
TESTED: Microsoft Excel 365 v2304
*/

martin12333

md failed

dddddd

intent

martin12333

// ==UserScript==
// @name         jQuery For Chrome (A Cross Browser Example)
// @namespace    jQueryForChromeExample
// @include      *
// @author       Erik Vergobbi Vold
// @description  This userscript is meant to be an example on how to use jQuery in a userscript on Google Chrome.
// ==/UserScript==

// a function that loads jQuery and calls a callback function when jQuery has finished loading
function addJQuery(callback) {