BigBlueHat BigBlueHat

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.

Case 1: Match literal followed_account_id, Finds 3 Rows
Case 2: Where followed_account_id = SUBQUERY, Finds 0 Rows
Case 3: Join via CTE Finds 3 Rows
Case 4: Join via Temp Table Finds 3 Rows
Case 5: Where followed_account_id = SUBQUERY (of Temp Table), Finds 0 Rows
Analysis

Knowledge Space

— Draft —

Link for sharing: https://w3id.org/knowledge-space/

Vision

Contributing and using scientific knowledge is hard and slow. It takes a lot of time and effort for researchers to publish their results as papers, and smaller contributions lack recognized channels. Moreover, connecting findings and following discourse if tedious and time-consuming too. AI can help, but basically faces the same kinds of problems with publishing and consuming knowledge: There is no global channel for sharing AI contributions and related feedback, and high-precision AIs lack the precise and structured input data.

Highlight Selection Ranges

Code here takes a DOM Selection Range instance and starts traversing a DOM starting from startContainer up to endContainer and wraps Text elements with elements that have semi-transparent background & swaps img elements with clones that are styled to have semi-transparent yellow overlay.

Issues

Only handles selecting text and images & would not cover divs with backgrounds for instance.

Kinesis Freestyle (Terrible key switches. Mushy and un-lovable)
Kinesis Freestyle Edge (Traditional layout with too many keys, mech switches, proably too big to be tented easily/properly)
Matias Ergo Pro (Looks pretty great. Have not tried.)
ErgoDox Kit (Currently, my everyday keyboard. Can buy pre-assembled on eBay.)
ErgoDox EZ (Prolly the best option for most people.)

	{
	"vc-data-model-2.0": {
	"introduction": [],
	"conformance": [
	"A conforming processor is any algorithm realized as software and/or hardware that generates and/or consumes a conforming secured document according to the relevant normative statements in Section 4. Algorithms. Conforming processors MUST produce errors when non-conforming documents are consumed."
	],
	"basic-concepts": [
	"The type property MUST be present. It is used to express the type of verifiable presentation. One value of this property MUST be VerifiablePresentation, but additional types MAY be included. The related normative guidance in Section 4.5 Types MUST be followed."
	],
	"contexts": [],

	const jsonld = require('jsonld');

	const input = {
	"@context": {
	"@vocab": "http://vocab.example/"
	},
	"age": {
	"@value": "27",
	"@type": "http://www.w3.org/2001/XMLSchema#integer"
	}

	{% assign relBase = '' %}
	{% assign tempDepth = page.url \| append: 'hackish-solution' \| split: '/' \| size \| minus: 2 %}
	{% for i in (1..tempDepth) %}
	{% assign relBase = relBase \| append: "../" %}
	{% endfor %}
	{% if relBase == '' %}
	{% assign relBase = './' %}
	{% endif %}

	#!/bin/bash

	echo "#############################################################"
	echo "#############################################################"
	echo "## Infoblox to bind export script ##"
	echo "## This script should be run on ##"
	echo "## a server allowed transfer the zones ##"
	echo "## the zones by transfer ##"
	echo "#############################################################"
	echo "#############################################################"