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.

n8henrie

macOS Seatbelt/Sandbox Trace Script

macOS sandbox profiles used to be able to include a trace command that would write all the denied operations to a sandbox profile, allowing a profile to be iterativley built up. Apple removed that functionality for reasons explained below.

trace.sh examines the kernel log for the denied operations and creates the relevant allow rules in a sandbox profile, just like the sandbox profile trace command used to.

shrink.sh tries to reduce a sandbox profile to the minimum lines necessary.

It's very rough and ready at the moment (check the sed regex'es in the script to see what I mean) and needs more testing with a wider set of use cases.

petergs

Entra ID First Party Microsoft Applications

csv of pre-consented first-party applications listed via the servicePrincipals endpoint

The list (microsoft-first-party-applications.csv) can be generated via

mgc service-principals list --all --filter "appOwnerOrganizationId eq f8cdef31-a31e-4b4a-93e4-5f571e91255a" --count true --consistency-level eventual --select appId,displayName | jq -r '.value[] | [.appId, .displayName] | @csv'

This list is more extensive, but is missing entries compared to this page from Microsoft Documentation: Verify first-party Microsoft applications in sign-in reports. These applications are captured in applications-from-docs.csv.

otaviog

WSL export/import cheat sheet

Reference: Setting up multiple WSL distribution instances

List distros

\> wsl --list
Windows Subsystem for Linux Distributions:
docker-desktop (Default)

adbertram

## No
Do-Thing | Out-Null

## yes
$null = Do-Thing
[void](Do-Thing)

MartinMiles

#Run as Admin

# https://gist.github.com/jermdavis/6fb0a6e47d6f1342c089af4c04d29c35#file-1_install-docker-ps1

param(
    [string]$dockerEnginePath = "C:\",
    [string]$dockerInstallPath = "C:\Docker",
    [string]$dockerEngineUrl = "https://download.docker.com/win/static/stable/x86_64/docker-24.0.6.zip",
    [string]$dockerZip = "docker.zip",

netfantom-spb

WSL list

Get-ChildItem "HKCU:\Software\Microsoft\Windows\CurrentVersion\Lxss" -Recurse

Move WSL to another drive/PC

wsl --export Ubuntu .\Ubuntu\ext4.vhdx --vhd
wsl --unregister Ubuntu
wsl --import-in-place Ubuntu .\Ubuntu\ext4.vhdx

monsat

import { useStorage } from '@vueuse/core'

export const useStoredState = <T>(key: string, initValue: T) => {
  const shared = useState<T>(key, () => initValue)
  const stored = useStorage<T>(key, initValue)

  const state = computed({
    get: () => shared.value,
    set: (value: T) => {
      shared.value = value

Omerr

log --graph --pretty=format:'%Cred%h%Creset -%C(yellow)%d%Creset %s %Cgreen(%cr) %C(bold blue)<%an>%Creset' --abbrev-commit

To configure as an alias git lol:

git config --global alias.lol "log --graph --pretty=format:'%Cred%h%Creset -%C(yellow)%d%Creset %s %Cgreen(%cr) %C(bold blue)<%an>%Creset' --abbrev-commit"

chgeuer

#!/bin/bash

access_token="$( az account get-access-token --resource-type arm | jq -r '.accessToken' )"

# echo "${access_token}" | jq -R 'split(".")|.[1]|@base64d|fromjson' 

subscriptionId="706df49f-998b-40ec-aed3-7f0ce9c67759"

# IFS='' read -r -d '' azureResourceGraphQuery <<'EOF'
# SpotResources