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.

karpathy

"""
The most atomic way to train and run inference for a GPT in pure, dependency-free Python.
This file is the complete algorithm.
Everything else is just efficiency.

@karpathy
"""

import os       # os.path.exists
import math     # math.log, math.exp

minimaxir

Agent Guidelines for Rust Code Quality

This document provides guidelines for maintaining high-quality Rust code. These rules MUST be followed by all AI coding agents and contributors.

Your Core Principles

All code you write MUST be fully optimized.

"Fully optimized" includes:

FelikZ

#!/usr/bin/env bash

# Script to force RGB Color Output on M1 and M2 based Macs for a selected display

# Function to display an error message and exit
function error_exit {
    echo "Error: $1" >&2
    exit 1
}

thesamesam

FAQ on the xz-utils backdoor (CVE-2024-3094)

This is a living document. Everything in this document is made in good faith of being accurate, but like I just said; we don't yet know everything about what's going on.

Update: I've disabled comments as of 2025-01-26 to avoid everyone having notifications for something a year on if someone wants to suggest a correction. Folks are free to email to suggest corrections still, of course.

Background

swarn

Semantic Highlighting in Neovim

• What is Semantic Highlighting?
• Default Highlighting
  • Tokens to Highlights
  • Changing Highlights
  • Disable Semantic Highlighting
• Using LspTokenUpdate for Complex Highlighting
  • Highlighting Based on More Than One Modifier
• Dealing with Ambiguity

romkatv

# Runs "$@" in a subshell with the caller's options, sets reply=(stdout stderr)
# and returns the status of the executed command. Both stdout and stderr are
# captured completely, including NUL bytes, incomplete UTF-8 characters and
# trailing LF, if any.
#
# Example:
#
#   % zsh-run-command ls -d ~ /garbage
#   % printf 'status: %s\nstdout: %s\nstderr: %s\n' $? "${(q+)reply[@]}"
#   status: 2

Linerre

• TL;DR
• Zsh initializations
  • Zsh way of $PATH
  • the special macOS
• /usr/libexec/path_helper
  • What is it?
  • What does it do?
  • What does it break?
  • Situation explained

• The .*shrc file

runiq

What are these?

Functions that allow you to call a function not more than once in a given timeframe.

Throttling on the leading edge

This can be illustrated with timing diagrams. In the following diagrams, f designates call to the throttled function, t is the period where the timer is running, and x shows you the actual execution point of the throttled function.

f 1  2  3  4  5  6

calexandre

#!/bin/bash
# Inspired on https://david-kerwick.github.io/2017-01-04-combining-zsh-history-files/
set -e
history1=$1
history2=$2
merged=$3

echo "Merging history files: $history1 + $history2"

test ! -f $history1 && echo "File $history1 not found" && exit 1