Zephyr On Call

Gwern.net is up there for me with zappa-analysis.com and CGP Grey’s first podcast with Brady Haran when it comes to fun Internet rabbit holes I fell into as a teen, the random quips and offhand remarks of which ended up significantly influencing my life decisions the decade afterwards. In Gwern’s case it’s less “offhand remarks”, more “lesser-known essays” - everyone knows how great spaced repetition is, and everyone in tech should know why we’re all trying to commoditize each another’s complements, but where are my homies taking nicotine lozenges to get addicted to Anki or using fake poison dart frogs to explain why the MIT license is worse is better?

This post owes its existence to one of these lesser known delights: “Resorting Media Ratings”, where he starts with the question “Why are the most mediocre things I’ve ever watched a 3.7/5 on Letterboxd?” and ends with putting together a tiny command-line program in R that pits titles head-to-head against one another until it has a reasonable approximation of a truly uniform standard distribution.

$ Rscript resorter.r --output out.csv --input in.csv

Comparison commands: 1=yes, 2=tied, 3=second is better, p=print estimates, s=skip, q=quit
Is 'Blinky' greater than 'Clyde'? 1
Is 'Clyde' greater than 'Inky'? 3
Is 'Inky' greater than 'Pinky'? 1
Is 'Pinky' greater than 'Inky'? 3
Is 'Inky' greater than 'Blinky'? 4
Is 'Inky' greater than 'Blinky'? 3
Is 'Clyde' greater than 'Pinky'? 1

Resorting complete
# see out.csv for example results
# out.csv
"Media","Quantile"
"Blinky","5"
"Inky","4"
"Clyde","2"
"Pinky","1"

Gwern’s whole website is under CC Zero, the Bourbaki way of saying in the public domain, and R isn’t the1 most straightforward thing to get up and running for someone who has never used it before, so after wiping my Github a few weeks ago, I decided to redo my repackaged resorter there with instructions on how to get it to work on Ubuntu 22.04 LTS.

Then I found out the best social media platform on Earth lets you export all of your data in CSV format, and import them to a separate list in CSV format too! I love it when people work with brain-dead simple formats, and the humble CSV is a Swiss Army knife I am particularly fond of thanks to the extensive tooling around it.

Gwern may have developed this for MyAnimeList originally. But when I found out Letterboxd was that accomodating of my weird power-user wishes, I knew what I had to do.

That’s it! The full resorted list can be found at FILMS RESORTED - 2022 - UNIFORM DISTRIBUTION, along with a link to this blog post.

Aaaaaaaaand for those of you who just want to know what I put where, here’s the final CSV. I stand by every hot take, including Howl’s Moving Castle being hot garbage.

"Title","Rating10"
"Apocalypse Now","10"
"Inherent Vice","10"
"My Neighbor Totoro","10"
"West Side Story","10"
"Joker","9"
"Skinamarink","9"
"The Cat Returns","9"
"Ponyo","8"
"Uncut Gems","8"
"Vampire Hunter D: Bloodlust","8"
"Whisper of the Heart","8"
"Black Swan","7"
"The Tale of the Princess Kaguya","7"
"The Wind Rises","7"
"Castle in the Sky","6"
"Nightcrawler","6"
"Requiem for a Dream","6"
"The Wicker Man","6"
"Everything Everywhere All at Once","5"
"The Secret World of Arrietty","5"
"When Marnie Was There","5"
"From Up on Poppy Hill","4"
"La Llorona","4"
"Nausicaä of the Valley of the Wind","4"
"Weird: The Al Yankovic Story","4"
"Pi","3"
"Pom Poko","3"
"The Northman","3"
"Fear and Loathing in Las Vegas","2"
"Parasite","2"
"Striking Daughter","2"
"Woodlands Dark and Days Bewitched: A History of Folk Horror","2"
"Annie Hall","1"
"Howl's Moving Castle","1"
"Mike And Melissa","1"
"V for Vendetta","1"

  1. That joke was too obscure to just drop as if you would instantly understand me. Nicholas Bourbaki “was” a “French” “mathematician” known for “his” heavily rigorous approach to the ars longa. Much like how “1 + 1 = 2” isn’t unambiguous enough for Bourbaki to accept on faith, apparently, just saying “I put X into the public domain” isn’t unambiguous enough for Germans to accept on faith. Hence, the “axiomatic” CC0, designed to degrade gracefully2 as it comes into contact with legal systems very different from our own↩︎

  2. Now that’s an interesting frame. You don’t normally think of legal constructs as things that degrade - and if you do, you definitely don’t think of them in terms of degrading gracefully. Often times, however, I’ve found that the flip-flop from centralized system thinking to distributed system thinking like this is a powerful sign you’ve moved productively from one abstraction to another, higher or lower on the epistemic stack. If it seems odd to you, consider non-SWEs don’t usually think of software as things that degrade, either - it’s just 1s and 0s! Those are always the same, right? So how on Earth can it degrade? The same way the concept of public domain degrades: Different communities disagree on the ways to do things, competing standards emerge, more communities emerge with newer ideas, more competing standards emerge, and eventually someone decides to put in the work to wrangle the inherent complexity of building a system that tries to provide a common interface between all of these communities to the same basic idea, minimizing surprises. ↩︎