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.

rich-iannone

library(tidyverse)
library(paletteer)
library(gt)

pizzaplace %>%
  mutate(type = case_when(
    type == "chicken" ~ "chicken (pizzas with chicken as a major ingredient)",
    type == "classic" ~ "classic (classical pizzas)",
    type == "supreme" ~ "supreme (pizzas that try a little harder)",
    type == "veggie" ~ "veggie (pizzas without any meats whatsoever)",

mkahn5

def import_to_big_query(data, context, dataset='UPDATE_DATASET_HERE', table='UPDATE_TABLE_HERE', verbose=True):
    def vprint(s):
        if verbose:
            print(s)

    vprint('Event ID: {}'.format(context.event_id))
    vprint('Event type: {}'.format(context.event_type))
    vprint('Importing required modules.')

    from google.cloud import bigquery

zross

• Eurostat, World Bank and others: https://ikashnitsky.github.io/2017/data-acquisition-two/
• Star Wars data: in dplyr, http://dplyr.tidyverse.org/reference/starwars.html
• Baby names data: babynames package, https://cran.r-project.org/web/packages/babynames/index.html
• Movies data: https://cran.r-project.org/web/packages/ggplot2movies/index.html ggplot2movies package
• Game of Thrones screen time: https://github.com/Preetish/GoT_screen_time
• Open Bike Data: https://github.com/ropensci/bikedata
• Tons of data through 538: https://cran.r-project.org/web/packages/fivethirtyeight/vignettes/fivethirtyeight.html
• Public health data England: fingertipsR, https://cran.r-project.org/web/packages/fingertipsR/
• Financial data via Quandl: https://www.quandl.com/tools/r
• Cyclones: https://github.com/ropensci/rrricanesdat

ppetraki

#!/bin/bash

SOURCE="$1"
if [ "${SOURCE}" == "" ]; then
    echo "Must specify a source url"
    exit 1
fi

DEST="$2"
if [ "${DEST}" == "" ]; then

arose13

# Setup Ubuntu
sudo apt update --yes
sudo apt upgrade --yes

# Get Miniconda and make it the main Python interpreter
wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda.sh
bash ~/miniconda.sh -b -p ~/miniconda 
rm ~/miniconda.sh

export PATH=~/miniconda/bin:$PATH

manashmandal

# Day 0: Mean, Median, and Mode

# Selecting standard input
con <- file('stdin', open='r')

# We don't need the first input 
data_line <- readLines(con)[[2]]

# splitting the data into individual string
split_data <- strsplit(data_line, " ")

alysonla

-- Active issues
-- Count of total active issues in the specified time frame
-- Source: githubarchive public data set via Google BigQuery http://githubarchive.org/

SELECT
	COUNT(DISTINCT JSON_EXTRACT_SCALAR(events.payload, '$.issue.id')) AS events_issue_count
FROM (SELECT * FROM TABLE_DATE_RANGE([githubarchive:day.],TIMESTAMP('2015-09-01'),TIMESTAMP('2016-08-31')))
 AS events

-- 10,723,492 active issues

MarkEdmondson1234

#' Email a user a report is ready
#'
#' Requires an account at Mailgun: https://mailgun.com
#' Pre-verification can only send to a whitelist of emails you configure
#'
#' @param email Email to send to
#' @param mail_message Any extra info
#'
#' @return TRUE if successful email sent
#' @import httr

pvictor

# BB-8 --------------------------------------------------------------------

# Inspired by Brian Hough in http://codepen.io/bhough/pen/wawrPL



# Packages ----------------------------------------------------------------

library("dplyr")
library("ggplot2")