allison @aparrish@friend.camp

a gan't

apparently the answer to "why isn't my gan working" is usually "well why didn't you put more batch normalization in there, hotshot"

(currently trying to hack the dcgan model I've been using to generate conditioned on labels, with only sputtering success, woo)

"doing this" = using DistilBERT to gradually transform a sequence words picked at random from a word list into text that appears to make sense

okay I THINK I finally found a way of doing this that comes close to meeting all of my criteria for this project (i.e., each step shows visible and meaningful change; the change is gradual, but the result "converges" after relatively few steps): calculate the probability of token in source text vs. token sampled from the distribution of mask token at that position, then find "peaks" of improbable tokens, and replace w/sampled token at those peaks; stop when any output repeats

ANYWAY the goal of all this was to make poem things with the word image GANs I trained for NaNoGenMo last year—here's a result from the first working implementation! (now the question is: do I go easy on myself and generate a book with one poem per page, or do I let the poems stretch across pages, or do I try to group shorter poems on the same page, etc. etc. etc.)

hey fediverse, one of my (computer-generated) poems is on the cover of BOMB Magazine's Winter 2021 issue! there are also several new poems from the Compasses series inside https://bombmagazine.org/issues/154/

okay I guess today is a day where I'm just going through a bunch of nlproc papers?? I love this example output from "Controlled Affective Text Generation" (columns from left to right: text completion prompt for the language model; desired topic; desired emotion; "knob" value adjusting emotional intensity; text output) paper link: https://wordplay-workshop.github.io/modern/assets/pdfs/8.pdf

original paper here: https://arxiv.org/abs/1906.05317 it really sucks that this paper has pretty much no discussion of how the use of a pre-trained language model increases the potential for harm—as just one example, the ConceptNet node for "transgender" is likely curated and more or less informative http://conceptnet.io/c/en/transgender but putting "transgender" into COMeT... basically spits out a bunch of harmful stereotypes

it's like the AI says. I'm hot, drunk, cool, dead and cute and I need you to to drink beer and study hard

playing around with COMeT (language model trained on knowledge base tuples from ConceptNet) https://mosaickg.apps.allenai.org/comet_conceptnet/

it does get stuck with repetitive words a lot, e.g.

more fun with DistilBERT: iteratively expanding a sentence by inserting a mask token *between* random adjacent tokens and predicting the most likely token for that position with the model

this is a cool site for finding out what features your variable fonts support https://wakamaifondue.com/ easier than what I did, which is dig through the python freetype bindings to find the completely undocumented function that maps to the barely documented "multiple masters" header in freetype or whatever 😎