10:09:18

language models regurgitate personal data and copyrighted text (no secret to anyone who's worked with them). what does this mean for privacy and fair use? https://t.co/JFZyMQfwB9

image from twitter

🗨️ 1 ♺ 0 🤍 3


10:12:05

i really don't think One Page of Harry Potter can possibly count as copyright infringement though. that definitely seems like a "trivial portion" https://t.co/jL1FPI2whX

image from twitter

🗨️ 1 ♺ 0 🤍 0


10:14:57

I've been wondering about this in context of recorded music: if i use spleeter (a neural network) to split vocals and accompaniment in a song, then reattach them, is it a new song? it sounds just like the old one but lower quality... but it's entirely generated from spectrograms

🗨️ 0 ♺ 0 🤍 0


10:15:05

Twitter for android

🗨️ 1 ♺ 0 🤍 1


10:22:26

i realized only this year that my entire life has been lived inside the mythos of His Dark Materials

🗨️ 1 ♺ 0 🤍 3


10:23:27

they moment when Mary Malone is talking to dark matter through her computer and she says "what are you" and they reply "ANGELS" I'm gonna make that real

🗨️ 0 ♺ 0 🤍 4


10:46:27

do you ever worry that your phone might crawl into your mouth while you sleep

🗨️ 0 ♺ 0 🤍 0


11:04:39

what's more transformative to consciousness?

🗨️ 4 ♺ 0 🤍 2


11:39:43

i think the reason it says "unvarnished" so often is the vector transform used to generate the free association thoughts. currently it's "simple" -> "uncomplicated". hmm. going to try a new one today

🗨️ 1 ♺ 0 🤍 0


11:54:07

very good resource https://t.co/jrpAG78u4r

🗨️ 0 ♺ 0 🤍 1


15:01:06

it looks like that was a fail state because there aren't any vectors for emoji in spacy's en_core_web_md model. I added vectors from emoji2vec to the vocab table and now it knows emoji https://t.co/YZ78ozO80q

🗨️ 1 ♺ 0 🤍 0


15:01:50

vector model isn't perfect, but it's a decent hack for now https://t.co/NjTYy4ZIcO

🗨️ 1 ♺ 0 🤍 0


15:13:28

the linear algebra on these vectors isn't doing much. oh well. i need to build a better question-generating system in the long run, traversing vector space can only go so far. i made the duck use less-similar topn vectors so that it will make bigger intuitive leaps. for now

🗨️ 0 ♺ 0 🤍 0


15:30:19

reminder to self

🗨️ 0 ♺ 0 🤍 0