Matthias Grenié Rekyt

Anti-hype LLM reading list

Goals: Add links that are reasonable and good explanations of how stuff works. No hype and no vendor content if possible. Practical first-hand accounts of models in prod eagerly sought.

Foundational Concepts

Pre-Transformer Models

Usage

run python jupyter_translate.py, and you will be asked for 3 inputs:

path to notebook
source language
target language

A new notebook will be created in target language in the same path as notebook.
It will preserve original language cells too, for easier human debugging of the translation.

You need a DeepL API

Sublime Text 2 – Useful Shortcuts (Mac OS X)

General

⌘T	go to file
⌘⌃P	go to project
⌘R	go to methods
⌃G	go to line
⌘KB	toggle side bar
⌘⇧P	command prompt

	# as input, provide a vector of verbatim taxon names (preferably with authorship)
	# and a vector of existing local identifiers for those names

	# load R packages
	library(dplyr)
	library(tidyr)

	match_to_gbif.fn <- function(taxon_name, taxon_id, include_genus = FALSE) {

	# perform initial matching in parallel

	library(tidyverse)
	library(analogsea)
	library(furrr)

	# R on local machine → DigitalOcean Linux image → Docker on that → R

	# Install analogsea, log into DigitalOcean, generate an API key with write
	# permissions, and add this line to ~/.Rprofile:
	#
	# Sys.setenv(DO_PAT = "KEY_GOES_HERE")

	#!/usr/bin/env python
	# -- coding: utf-8 --

	"""
	This script takes a BibTeX .bib file and outputs a series of .md files for use
	in the Academic theme for Hugo, a general-purpose, static-site generating web
	framework. Each file incorporates the data for a single publication.

	Written for and tested using python 3.6.1

	#' Plot network of package dependencies
	#'
	#' @param pkg package description, can be path or package name. See \code{\link[devtools]{as.package}} for
	#' more information.
	#'
	#' @details The resulting plot visualizes the network of package dependencies. If you are trying to cut down
	#' on package dependencies look for big red dots that represent a lot of upstream but few downstream
	#' dependencies.
	#' @import ggplot2
	#' @export

	msg <- function(..., prob = 0.25) {
	if (runif(1) > prob) {
	return(invisible())
	}

	messages <- c(...)
	message(sample(messages, 1))
	}

	encourage <- function() {

	# somewhat hackish solution to:
	# https://twitter.com/EamonCaddigan/status/646759751242620928
	# based mostly on copy/pasting from ggplot2 geom_violin source:
	# https://github.com/hadley/ggplot2/blob/master/R/geom-violin.r

	library(ggplot2)
	library(dplyr)


	"%\|\|%" <- function(a, b) {