Similar Different
This is a poetry collection I made. The poems are displayed in chronological
order, if you want to read from start to finish. You can click "similar" to see
poems that are similar to that particular poem, and you can click "different" to
see poems that are different. Go figure.
The similar pages show you all the poems that are most similar to the poem you
picked. It shows the first most similar, then second, then third, etc.
The different pages however show the poems that are most different to each poem
that comes before it - meaning, the second poem is the most different to the
first, but the third poem is different from both the first and second, triangle
style. The fourth poem takes into account the first, second, and third poem.
There's also a wordcloud, which looks at each occurence of every word in all the
poems and then finds the poems that are most similar to that word. Also there's
an image gallery for some pictures I either made or found on the internet, and
an index at the bottom of the image gallery if you want to jump to a specific
poem in the chronological pages.
What does it mean to be "similar" or "different"?
Computers see text as just a long string of numbers. It doesn't mean anything
unless we build meaning into it. We could go through and build out a massive
tree of word synergies and antonyms and such, but that's a lot of work. Besides,
someone has already done that for us!
There's this thing called a "word embedding" which is used in modern AI, with
Large Language Models specifically. It's like the first 20% of the process, and
I took that system and used it to sort my words. Here's how it works:
First, each word is turned into numbers. These numbers hold not just the shape
of the word (like, first there's an A, then an N, then A, then R, then C, then
H, then Y) but also the semantic meaning of the word. This semantic meaning is
not just randomly assigned, during the process of building these statistical
models they shuffle words around to sit next to other words that mean the same
thing. So "anarchy" probably sits next to "liberation" if I had to guess, but
also probably right next to "pearl clutching" which is two words but you get
the idea.
Anyway the system I made takes the "embedding" of each of the words in my poem,
then combines it into one single number. The neat thing about these embeddings
is that they're additive - meaning, you can get an average "meaning" of a poem
just by looking at the combined average of the embeddings of each word. Of
course, this isn't very accurate at all, because you might start out a poem
talking about the weather, spending the rest of it describing a cute girl, and
then it's revealed at the end that it was all a dream or whatever - the poem
would probably focus on the girl and less on the weather or the dream. That's...
fine. The system isn't perfect, but I wanted a way to put words that had similar
concepts next to each other, and I wanted people who felt like they were viewing
themselves into a rut to have a way to jump out and see things that were totally
different, for maximum diversity. To let them explore in a non random, yet fully
unpredictable way.
I used AI to make this. I am utterly unrepentant, I don't believe it's wrong.
WHAT IS HERE (counted at build time)
8510 poems across 5 sources
spanning 2021-04-27 to 2026-08-08
68 of them are image-only posts
sources:
fediverse 5977
messages 1558
notes 470
fediverse_boost 458
bluesky 47
LINKS
Start reading (chronological)
Word cloud
Image gallery
How the similarity actually works (the math)
Browse the source code
|