"The central claim of the tech companies selling LLMs is that any work that people do that results in text artifacts is just "text in-text out" and can therefore be replaced by their synthetic text-extruding machines. The best response to that claim is not "oh no, we can't keep up" but to take pride in one's work... and push back"
Nick Montfort's poetics: "Writing very small-scale computational poems allows me to learn more about computing and its intersection with language and poetry. Not computing in the abstract, but computing as embodied in particular platforms, which are intentionally designed and have platform imaginaries and communities of use and practice surrounding them."
"Open source, open data, open training code, fully reproducible and auditable text embedding model"
references, articles, publications on the history of fax machines
I define "writing" as a heuristic or a way of making educated guesses, letter by letter, until a writing task is considered complete. Users can experience this with whichever file they choose to upload."
"a speculative research project exploring the use of machine learning for the evolution of language. Large language models (LLM's) are fantastic at capturing our language as it currently is - but language is constantly evolving and adapting. Can machine learning help us create something truly new and unbounded by its training data?"
"a collection of full article texts extracted from historical U.S. newspaper images [that] includes nearly 20 million scans from the public domain"
"We need to ensure that human creators are compensated, not just for the sake of the creators, but so our books and arts continue to reflect both our real and imagined experiences, open our minds, teach us new ways of thinking, and move us forward as a society, rather than rehash old ideas."
intuitive and fun blackout poetry interface, using snippets from project gutenberg
"The chatbot’s answers sound extremely specific to the current context but are in fact statistically generic. The mathematical model behind the chatbot delivers a statistically plausible response to the question. The marks that find this convincing get pulled in." this is really good but i wish it approached the topic of psychics with a bit less bro-ey skepticism
"[E]ducational technology is overly dominated by psychological conceptions of individual learning... AI-based personalized learning systems [are] based on notions of mastery and... statistical measurement," reflecting an "assumption that human intelligence is an individual capacity, which can therefore be improved with technical solutions — like tutorbots — rather than something shaped by educational policies and institutions."
(via data is plural): "every quotation from 22 novels, plus who speaks each line, who they’re addressing, the characters they mention, and more. With 35,000+ quotations, the corpus 'is by an order of magnitude the largest dataset of annotated quotations for literary texts in English.'"
"12px pixel art font that resembles textura/textualis writing from late medieval manuscripts and early printed books"
"Snelson plays a wandering bard – misusing the system to produce the most unlikely of scrawls: small poems scattered across the game’s landscape. The book is a documentation of that performance in a prosody marked by the poetics of fandom. They are recorded here as movie captures, static images, and poetic texts, arranged in four parts spelling a newly coherent object."
computer-generated books ("a non-human reading")
"... an experimental typeface that is carved into the bark of a tree. As the tree grows, it deforms the letters and outputs new design variations, that are captured annually"
"Atkinson Hyperlegible font is named after Braille Institute founder, J. Robert Atkinson. What makes it different from traditional typography design is that it focuses on letterform distinction to increase character recognition, ultimately improving readability." looks good too imo
generating from a markov model by hand
"Alphabetical approaches upended accepted hierarchies"
replaces dialogue with detailed phonetic descriptions of the dialogue. very cool
"a website that produces event scores for performance. The material objects, locations and activities within each score are based on the performance archives of Nathan Walker between 2009-2014 and work towards shuffling and redistributing the archival record to create an anarchive."
zach whalen's notes! very thoughtful!
'I started to ask myself the question – how long will it take before we start seeing “documentary photojournalism” that has no other basis in reality than the photographer’s fantasy and a powerful computer graphics card? Will we be able to tell the difference? How hard is it to do? How skilled will our own community of photographers and editors be in sniffing out what are deep fakes and what is real?'
"While these languages are obviously not in common use today, we find it fascinating to think about the world that might have been. Even more surprisingly, it happens that many of these other options include features which developers would love to see appear in CSS even today."
"BLABRECS is a rules modification for the wordgame SCRABBLE that swaps out the dictionary of real-if-obscure English words for a capricious artificial intelligence. In BLABRECS, real English words aren't allowed! Instead, you have to play nonsense words that sound like English to the AI. These nonsense words are called – you guessed it – BLABRECS."
"Vosk is an offline open source speech recognition toolkit. [...] Vosk models are small (50 Mb) but provide continuous large vocabulary transcription, zero-latency response with streaming API, reconfigurable vocabulary and speaker identification." Bindings for various languages, "scales from small devices like Raspberry Pi or Android smartphone to big clusters."
"...the method here is quite different. DALL-E is trained end-to-end for the sole purpose of producing high quality images directly from language, whereas this CLIP method is more like a beautifully hacked together trick for using language to steer existing unconditional image generating models." good history of the emergence of CLIP art
"At the top of the machine, an LED panel endlessly regurgitates its own new neoist verses into the eyes of the audience, equally brainwashing humans, cyborgs, robots, and other technobiological systems. Anyone can directly hack into the system's artificial neural synapses by unplugging, replugging, and criss-crossing jack cables directly on the machine, thus deconstructing, reconstructing, and even destroying the generative capabilities of the system in real-time."
"Application for the sonification of text which can be transformed according to various triggers and parameters to facilitate the learning and analysis of literacoustics, reading by listening."
"Voluntarily provocative, The Hater Box transforms the principle of old split flap displays into a random generator of contestations, cold and impersonal." 2018 – Wood, motor, cardstock, print, 3D print
"Rainbow Zero is a... toy? widget? thingy? that allows you to explore a part of the space defined by the GloVe word vectors."
some text to image stuff
similarity of images based on semantic similarity between automatic captions
"a recurrent neural network that generates little stories about images"
"TextOCR provides ~1M high quality word annotations on TextVQA images allowing application of end-to-end reasoning on downstream tasks such as visual question answering or image captioning."
okay I had no idea this existed: an interactive CD-ROM of various Oulipo texts
char-rnn trained on ansi artwork
common format for representing syntax trees of html, markdown, etc.
allennlp's version of the c4 dataset
spacy 3 tutorials
"Icons are not primitive or rudimentary attempts to duplicate the physical world; they are nuanced and complex attempts to embody the spiritual world."
"How many words can we form by making folds in the straw-paper slogan? I could not have answered that question in 1967. I couldn’t have even asked it. But times change. Enumerating all the foldable messages now strikes me as an obvious thing to do when presented with the straw wrapper. Furthermore, I have the computational means to do it—although the project was not quite as easy as I expected."
"a rule-based sentence boundary detection that works out-of-the-box"
"graphs the usage of words (whether in description or dialogue) over time, distinguishing that usage both by the gender of the fictional characters the terms are associated with, and by the gender of the authors who used them"
"Alt-text is an essential part of web accessibility. It is often overlooked or understood through the lens of compliance, as an unwelcome burden to be met with minimum effort. How can we instead approach alt-text thoughtfully and creatively?" (presented at wordhack dec 2020)
"Karrik is an open source typeface designed by Jean-Baptiste Morizot and Lucas Le Bihan. This font was originally commissioned by ‘Cercle’ magazine for their 2020 issue—dedicated to the topic of ghosts. The design started in March 2019 and ended in October of the same year. [...] Karrik is rooted in vernacular typography. The weight disadjustments, the lack of optical corrections, the uneven width of the letters are some of the features of early sans serif typefaces that inspired us..." (hey plus it's open source!)
nice bold sans serif with variable features
lovely ofl serif with many features
"a collaborative effort to improve how NLP handles complex morphology in the world’s languages. The goal of UniMorph is to annotate morphological data in a universal schema that allows an inflected word from any language to be defined by its lexical meaning, typically carried by the lemma, and by a rendering of its inflectional form in terms of a bundle of morphological features from our schema."
cool détournement of gpt2
bglkjawbflablfhbawefjh
"The extended largest dataset in first-person (egocentric) vision; multi-faceted, audio-visual, non-scripted recordings in native environments - i.e. the wearers' homes, capturing all daily activities in the kitchen over multiple days. Annotations are collected using a novel 'Pause-and-Talk' narration interface."
python interface for the HTRC Extracted Features dataset
generates (via markov chain) and speaks made-up words from various corpora
"StereoSet is a dataset that measures stereotype bias in language models. StereoSet consists of 17,000 sentences that measures model preferences across gender, race, religion, and profession."
"The exhibition OPEN SCORES brought together a series of practices through which artists articulate their specific forms of digital commons. From online archives to digital tools/ infrastructure and educational formats, the projects envision a (post-)digital culture in which notions of collaboration, free access to knowledge, sustainable use of shared resources, and data privacy are central. For the exhibition, each of the projects created a unique score to present their practice."
"[a] dictionary and graphical data for over 9000 of the most common simplified and traditional Chinese characters. Among other things, this data includes stroke-order vector graphics for all these characters." (via gábor ugray's !!con 2020 talk)
"freely available for research with the condition that the research be used for the benefit of children"
should make sure I know all this stuff
"[T]o avoid the censorship, people have converted parts of the interview into Morse code, filled it up with emojis, or translated it into fictional languages like Sindarin from The Lord of the Rings or Klingon from Star Trek. In one particularly creative example, someone inserted it into the iconic opening crawl of Star Wars."
sam and tega. very good
'...very boring stories that did not even satisfy my youngest children... I tried these stories on my very small children but after some minutes they grew very irritable, because nothing actually happened. This shows that even small children of three can measure entropy'
"Face lets you edit both the text and the font it is rendered in. In text mode you can type and edit text normally. Press escape to enter font mode, where you can select a character to edit. Any changes to a character are visible immediately."
"CCMatrix is the largest data set of high-quality, web-based bitexts for training translation models. With more than 4.5 billion parallel sentences in 576 language pairs pulled from snapshots of the CommonCrawl public data set, CCMatrix is more than 50 times larger than the WikiMatrix corpus that we shared last year."
"Conclusion: We hypothesized that radical swings in affective posture would make the writer more emotionally flexible. Likewise, we hypothesized that attempting to discern the emotional valences of a machine learning model derived from achingly sensitive Tumblr posts would make the writer more empathetic. Unfortunately, no conclusions could be drawn from a single poem."
important
python-based mush/moo thing! could be fun to play around with
sesame street on phonaesthetics
"A simple resource for finding and trying variable fonts"
ranjit!
"an esoteric programming language that closely follows the grammar and tone of classical Chinese literature. Moreover, the alphabet of wenyan contains only traditional Chinese characters and 「」 quotes, so it is guaranteed to be readable by ancient Chinese people." (from one of Golan Levin's students)
"a large-scale dataset of misspellings and grammatical errors along with their corrections harvested from GitHub. It contains more than 350k edits and 65M characters in more than 15 languages, making it the largest dataset of misspellings to date."
this looks promising
"a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English" it warms my heart to see an ngram baseline in there, haha
"OpenMoji is an open source project of 53 students and 2 professors of the HfG Schwäbisch Gmünd and external contributors" (CC BY-SA 4.0)
"I run datasets of iconic feminist texts through a simple textRNN, generating new feminists texts in the legendary words of bell hooks, Simone De Beauvoir, Betty Friedan and Audre Lorde. Some are funny. Some are poetic. Some make no sense at all and some are way too real. Information about the model and settings can be found under each post."
another tutorial from emnlp-19
"In the movie-oriented CCPE dataset, individuals posing as a user speak into a microphone and the audio is played directly to the person posing as a digital assistant. The “assistant” types out their response, which is in turn played to the user via text-to-speech. [...] The Taskmaster-1 dataset makes use of both the methodology described above as well as a one-person, written technique to increase the corpus size and speaker diversity—about 7.7k written “self-dialog” entries and ~5.5k 2-person, spoken dialogs. For written dialogs, we engaged people to create the full conversation themselves based on scenarios outlined for each task, thereby playing roles of both the user and assistant."
"On October 19th of 1955, Pulitzer Prize-winning poet, Marianne Moore, was approached by a Mr. Robert Young of the Ford Motor Company and asked to assist them in naming a new series of cars."
"ways to make huge models like BERT smaller and faster": quantization, pruning, distillation
"We’ll never be able to read all of these documents. What’s unique about this text compared to all the rest? My eyes sting from searching these images for the same thing. We need to find more records like these in a huge pile of data. I could really use a heads-up before this happens again. (Post to come.)" I *reeeeeally* appreciate approaches to ml like this that start with problems to be solved (instead of just taking for granted that ai/ml is useful)
"Create a unique bitfont from a vast space of glyphs generated by a neural network." yacht/counterpoint
"Two neural nets learn to communicate through their own emergent visual language."
gpt-2 on svg for generated emoji and letterforms
a "talk about the history and environment of Hershey’s creation, and touch on the current state of resurrection."
"Universal Dependencies (UD) is a framework for consistent annotation of grammar (parts of speech, morphological features, and syntactic dependencies) across different human languages. UD is an open community effort with over 200 contributors producing more than 100 treebanks in over 70 languages."
"there are no consistent right answers, everything is way more important than you think, and everything affects everything else"
"The father/daughter team of Trevor F. Smith and Sparks Webb have gathered all available documentation about the memex and carefully fabricated the Memex #001 to match Dr. Bush's specifications."
"a free, non-commercial project with the goal of preserving selected paper-based cultural artifacts for future generations of readers, in the form of cover images in JPG format, and, where available, complete cover-to-cover scans in PDF format"—in this case, a bunch of old Analog magazines
I assume this is the full text of chapter one of the landow book?
"Recommendation engines like the ones powering the endless feeds on Twitter, Facebook and YouTube, are designed to maximize ad revenue, and therefore to keep you online for as long as possible. In doing so they promote the most reactionary content on their platforms. Yet, these recommendation systems are nothing more than sorting mechanisms. Other Orders provides an alternate set of sorts, optimized for other outcomes."
"Artists aim differently than sharpshooters. They are not typically trying to take something out, but to draw something out. The mark Holzer hits in this case is the mark in the most cave-drawing sense: the effort to leave (or find) a trace of something that is not an opinion, but a register of some kind, certifying a lived experience. There may be no such thing as a permanent record, but the fact that the Washington Post contributor found Holzer’s work dangerous is a sign in and of itself that it has achieved one of its goals: it has carved a deep enough mark to leave a strong impression (for that writer, a menacing one). That’s the most any language or other kind of mark-making can hope to accomplish."
well this looks like a dream come true?
"[A] dataset of UN Security Council debates between January 1995 and December 2017... split into distinct speeches" with metadata on "the speaker, the speaker's nation or affiliation, and the speaker's role in the meeting" and "the topic of the meeting." 65393 speeches extracted from 4958 meeting protocols (!). via data is plural
"a collection of pre-trained subword embeddings in 275 languages, based on Byte-Pair Encoding (BPE) and trained on Wikipedia. Its intended use is as input for neural models in natural language processing"
"The Transformer is nothing more than an architecture where the core functional unit is attention. You stack attention layers on top of attention layers, just like you would do with CNN or RNN layers."
"a community-driven, freely-licensed binary distribution of Microsoft’s editor VSCode"
wiktionary word frequency lists
amazing
css stylesheet that displays html as markdown. brilliant
down the rabbit hole of fantasy/sf authors using wordstar
uses my poetry corpus! though points out some shortcomings.
detailed explanation of an unbelievably bad-ass build, fuck
"Yulia Tsvetkov's research group at Language Technologies Institute of Carnegie Mellon University. Our work focuses on natural language processing, particularly cross-lingual approaches, low-resource settings, and social good."
"In this interactive installation participants enter the first word that comes to their mind in one of two input terminals in any language. These words are then the seed of a generative process that develops a poem, bifurcating and mutating, merging languages, poetic styles, sense and nonsense. Poems overlap and degrade over time, eventually fading away. Phonetics are remapped to a new alphabet of sound referencing the body and incidental noises, creating a unique expression for each word and making literal the arbitrariness of the language. This installation was projected on a massive scale covering the walls and ceiling and filling the hall of the old imperial castle in Poznan, Poland. This video shows a demonstration of the generated poetry."
kyle mcdonald's take
"a visually forensic tool to detect text that was automatically generated from large language models"
dan jurafsky intro lecture
observation on how avant-garde experimentation requires the materiality of the medium as a frame, or something
from what crpg addict calls the "establishing era," this reminds me a bit of 80 Days
"I love reading postmortems. They're educational, but unlike most educational docs, they tell an entertaining story."
"an international organization of scholars working in textual studies, editing and editorial theory, electronic textualities, and issues of textual culture across a wide variety of disciplines."