Erik Körner Querela

Some remarks on Large Language Models

Yoav Goldberg, January 2023

Audience: I assume you heard of chatGPT, maybe played with it a little, and was imressed by it (or tried very hard not to be). And that you also heard that it is "a large language model". And maybe that it "solved natural language understanding". Here is a short personal perspective of my thoughts of this (and similar) models, and where we stand with respect to language understanding.

Intro

Around 2014-2017, right within the rise of neural-network based methods for NLP, I was giving a semi-academic-semi-popsci lecture, revolving around the story that achieving perfect language modeling is equivalent to being as intelligent as a human. Somewhere around the same time I was also asked in an academic panel "what would you do if you were given infinite compute and no need to worry about labour costs" to which I cockily responded "I would train a really huge language model, just to show that it doesn't solve everything!". We

Function refs vs. ref objects

Great question I got on Twitter: what are the benefits of returning a function ref from a Vue 3 composable, instead of returning the ref object directly?

Here are two SFC playgrounds to outline the two alternatives:

Installing Windows and Linux into the same partition

But WHY?

There was a reddit post about installing Arch on NTFS3 partition. Since Windows and Linux doesn't have directories with same names under the /(C:\), I thought it's possible, and turned out it was actually possible.
If you are not familiar to Linux, for example you've searched on Google "how to dualboot Linux and Windos" or brbrbr... you mustn't try this. This is not practical.

Pre-requirements

UEFI system
Any Linux live-boot CD/DVD/USB... with Linux kernel newer than 5.15
Windows installer USB

Backup and restore WSL2 vhdx, avoid 256GB vhdx limit for docker desktop WSL2 native

Review all steps to view the symptom this Gist addresses, customize a workaround to allow for >256GB Docker pulls safely on Windows, then perform a backup and recovery of the result (export and import). Please provide comments on any failures.

Determine the expected size of the docker-desktop-data volume

Get the base image size + additional + all docker volumes using docker system df
Add raw uncompressed size for alpine (50MB), docker for your version of windows (700MB) and linux (80MB)

docker system df

Offline-first browser apps and multiple tabs

For offline-first operation, browser apps cache data in an IndexedDb database. When the user makes a change while offline, they persist the change optimistically. When connectivity is restored, changes are synced to the server.

This pattern is well established today. However, given the possibility of opening the app in multiple tabs at the same time, we seem to be faced with a dilemma:

We allow users to use the app and to make changes in multiple tabs at the same time. But then two "threads" are writing to the same shared resource at the same time. Co-ordinating writes seems to be difficult.

To make things worse, while the localStorage API allows you to register a callback that fires whenever an item is changed outside the current tab, the IndexedDb API doesn't, at least not in a widely-available way.

	import logging

	# Configure logging level and format
	logging.basicConfig(
	level=logging.INFO, # options: DEBUG, INFO, WARNING, ERROR, CRITICAL
	format='%(asctime)s [%(levelname)s] %(message)s',
	handlers=[
	logging.StreamHandler() # prints to console
	]
	)

	call plug#begin()
	Plug 'drewtempelmeyer/palenight.vim'
	Plug 'vim-airline/vim-airline'
	Plug 'wlangstroth/vim-racket'
	Plug 'sheerun/vim-polyglot'
	Plug 'rust-lang/rust.vim'
	Plug 'preservim/tagbar'
	Plug 'universal-ctags/ctags'
	Plug 'luochen1990/rainbow'
	Plug 'vim-syntastic/syntastic'

	{
	"bip": "172.39.1.5/24",
	"fixed-cidr": "172.39.1.0/25",
	"runtimes": {
	"nvidia": {
	"path": "nvidia-container-runtime",
	"runtimeArgs": []
	}
	}
	}