Pinned post
allison boosted

With #Galactica and #ChatGPT I'm seeing people again getting excited about the prospect of using language models to "access knowledge" (i.e. instead of search engines). They are not fit for that purpose --- both because they are designed to just make shit up and because they don't support information literacy. Chirag Shah and I lay this out in detail in our CHIIR 2022 paper:

dl.acm.org/doi/10.1145/3498366

>>

allison boosted

announcement:

here is a new open source alternative to perlin noise: octavia noise

it is a lightweight fractal noise algorithm. it is generalizable to n dimensions. it is not based on gridded points, and does not exhibit linear artifacts.

this is suitable for terrain and other procgen work where organicity and reducing grid artifacts are important, especially situations where slope artifacts are important. the initial js implementation takes 4x as long as value noise.

github.com/lorenSchmidt/fracta

allison boosted
allison boosted

good news, fediverse! After a 6 month hiatus skull.supply/ is open again! Plenty of new objects and I'll be shipping in time for the holidays. Boosts welcome!

chatgpt output, linguistics misinformation, actually done now 

ah yes, "k," english's famous... glottalized alveolar lateral affricate (that is also somehow pronounced with your mouth closed)

Show thread

chatgpt output 

okay I'm going to stop but still lmao

Show thread

chatgpt output 

okay after getting a lot of prevaricating canned answers from questions about trans issues, it's kind of reassuring to get some garden-variety nonsensical misinformation

Show thread

chatgpt output, gender 

important tips, thanks chatbot

(my miniloom doesn't have shafts, of course, and I weave on it using a tapestry needle as a "shuttle." I don't know why it didn't occur to me earlier that I can do whatever weave structures I want on the miniloom, I just have to do them "by hand." I really like how the 3/1 twills came out, especially with the thicker silk)

Show thread

some recent samples i made on my miniloom. the warps are various silks (a slubby eri silk and a shiny bombyx) and weft is a bunch of different weave structure/yarn experiments: 3/1 and 2/2 twills with a variegated cotton, a zigzag twill with purple linen, and basket weave with a light yellow slubby cotton (along with bits of plain weave)

there should be a word for the sneaking suspicion that you actually already own a pair of the headphones you're thinking about buying. they're just in a storage box somewhere, under your bed probably?

allison boosted
allison boosted

tired: eat the marshmallow
wired: wait, and get two marshmallows
inspired: donate your worldly belongings to the marshmallow research institute so that in the distant future all the matter in the galaxy will be turned into 10^53 marshmallows

allison boosted

Over on #MetaFilter, I am a member of the MeFi Global BIPOC (Black, Indigenous, and People of Color) Advisory Board. We just posted a summary of our mandate, perspective, activities, scope, plans, etc.

metatalk.metafilter.com/26210/

Please reshare this with people who used to be #MeFites, especially those who left because behavior on the site kept making them feel unwelcome.

#MeFite

allison boosted

Digital narrative friends: What are examples of works which invite audience participation once they are live, and maintain the state of those audience changes for the next person?

A trivial example would be a non-linear hypertext story that retained the choices from a previous visitor, that the next visitor could override. But I'm interested in anything that broadly passes the test of "a story that changes over time as people engage with it."

(Googling for related terms mostly turns up digital storytelling tools for children, or platforms for collaborating pre-publication.)

#DigitalFiction #hypertext #GenerativeNarrative

machine learning, language models 

struggling to come up with a more rube goldberg-ey way of finding out the price of a stock than "train a large language model to write an API request"

Show thread

machine learning, language models 

if i'm reading this correctly, it seems to be claiming that the key to improving the performance of large language models is... access to lots of reliable structured data. which is hilarious and ironic jmcdonnell.substack.com/p/the-

Show older
Friend Camp

Hometown is adapted from Mastodon, a decentralized social network with no ads, no corporate surveillance, and ethical design.