allison

Poet, programmer, game designer. Associate Arts Professor at NYU ITP. she/her 🏳️‍⚧️

Home page: https://www.decontextualize.com/

Public Key

npub1xs0zj7z50s9y0wv62atktwnzwkf95kyjhme5pknukcqz0anevakstnck5u

NIP-05 Address

[email protected]

Profile Code

nprofile1qqsrg83f0p28czj8hxd9w4m9hf38tyj6tzftau6qmf7tvqp87eukwmgprpmhxue69uhhyetvv9ujumt0d4hhxarj9ecxjmnt4z2qxl

Publishing to

relay.momostr.pink

Author Public Key

npub1xs0zj7z50s9y0wv62atktwnzwkf95kyjhme5pknukcqz0anevakstnck5u

Show more details

Last Notes

2026-05-16 17:30:27 UTC

regular folks: "measure twice, cut once"
vibe coders: "ALWAYS tell your saw to CUT IN THE RIGHT PLACE"

2026-05-14 19:34:12 UTC

re: https://social.coop/@cwebber/116522889714585276
the AI code bros' approach to open source lately seems to be "I'll take what I want and you can't stop me"—a snide dismissal of the very idea that they should be bound by anyone else's rules, or be expected to participate in an ecosystem as anything other than a parasite. it's pure trumpism enacted in software development

2026-04-29 17:24:37 UTC

kinda surprised that so many "fantasy consoles" are essentially just opinionated graphics/audio wrappers for interpreted languages that are otherwise running on the host machine (e.g., pico8/lua). i would have thought that there would be more fantasy consoles that specify a virtual machine (like uxn https://100r.co/site/uxn.html) or reuse an existing CPU architecture (like minicube64 https://aeriform.itch.io/minicube64)

2026-04-26 16:48:37 UTC

here i've been saying "i code without using llms" when i should have been saying "i code with maximum token efficiency"

2026-04-23 18:57:26 UTC

- reply

weird that they'll leave in error-riddled images like this *in the promo video*—e.g.,
- in "the solar system", planets are not following orbit circles; also uranus has weird pins coming out? (maybe supposed to be rings?)
- in "size comparison," earth is illustrated with... mars? and mercury is bigger than earth?
- inconsistent labels in "the moons" (i.e., "saturn" is illustrated with ganymede)
- "scale of distances" illustration is DEFINITELY not to scale as claimed
https://friend.camp/system/media_attachments/files/116/455/469/426/952/013/original/8748f757ef7c6742.png

2026-04-20 20:03:28 UTC

the nyc street tree census data is really fun and easy to work with! my favorite nyc street tree is the pin oak, and i made a little map of all of them (at least, as of 2015) (zoomed in here on brooklyn)
https://friend.camp/system/media_attachments/files/116/438/803/074/716/796/original/4eb98760870bfa19.png

2026-04-18 22:24:10 UTC

the OCR'ed text of books on the internet archive is truly a never-ending source of joy
https://friend.camp/system/media_attachments/files/116/428/047/315/037/648/original/a983a281eb090d7c.png

2026-04-18 14:52:33 UTC

- reply

hard for me to believe that people would independently come up with generative AI as a serious method for solving problems if the entire field of practice wasn't subsidized by investment ponzi schemes, real estate scams, and shovel sellers

2026-04-18 14:49:30 UTC

- reply

like... the cost of a single "state-of-the-art" llm training run would be a permanently institution-changing amount of money for your average computer science department (to say nothing of depts in CS-adjacent fields, especially in the arts and humanities). the claude mythos result, even if you take anthropopo's claims at face value (which: don't) seems indicative of this, like, yeah if you toss nine figures at security research, you're gonna find some exploits?

2026-04-18 14:44:15 UTC

i guess fundamentally i'm not convinced that whatever amount useful functionality has emerged from generative AI (and i think the jury's still out on whether that amount is non-zero) couldn't have been achieved in more cost-effective, equitable and sustainable ways by having taken the money and resources we've shoveled into gen AI and investing them elsewhere
("sustainable" in the sense of environmentally sustainable but also "can we keep this in a working state" sustainable)

2026-04-17 15:14:39 UTC

listen, our students are entering a workforce where their peers will be heavy cocaine users. so we would be putting our students at a disadvantage if we *don't* teach them how to use cocaine (responsibly and ethically of course)

2026-04-09 18:20:43 UTC

ooh you were so close
(from http://arxiv.org/abs/2604.04721)
https://friend.camp/system/media_attachments/files/116/376/123/388/981/568/original/7d90c470e0ff55cf.png

2026-04-08 17:41:38 UTC

- reply

love this thread, thank you for thinking through this and writing it out. i was wondering if you could explain the reasoning of "openness" requirement a bit more? why is it important that the full bus is exposed? (not challenging the idea, just interested to learn more about your specific reasoning)

2026-04-02 21:00:23 UTC

how claude code works, visualized
https://friend.camp/system/media_attachments/files/116/337/115/106/594/496/original/6e29ca09fa6dcf93.png

2026-04-01 18:19:26 UTC

my iphone insists on telling me that if i'm charging from my laptop's USB-A port, I'm using a "slow charger." i dunno, keep your opinions to yourself man. maybe you're just bad at charging, did you think of that. "waaaah i'm only getting 500mA" entitled little asshole

2026-03-31 15:09:51 UTC

ai slop sucks! but don't worry, this machine, which we're making out of slop, and which we're feeding lots of slop into, definitely won't produce more slop. i am the chief innovation officer

2026-03-27 21:26:35 UTC

the contract faculty union at NYU, of which I am a member, reached a tentative agreement with the university earlier this week (after a two-day strike). i have to say i'm pretty happy with the outcome of the agreement https://nyucontractfacultyunion.org/cfu-uaw-tentative-agreement-highlights/ in conclusion, do a union, you'll be glad you did

2026-03-25 14:20:59 UTC

- reply

i had also assumed it was in or near the strait! thank you for sharing this

2026-03-23 15:05:48 UTC

ios user interfaces have become truly nihilistic. buttons on top of buttons. text on top of text. multiple inscrutable hamburgers. nothing has any meaning and all human action is futile
https://friend.camp/system/media_attachments/files/116/279/081/280/548/919/original/dbb12e64da989ec5.png

2026-03-17 14:58:40 UTC

- reply

what they wrote: "The AI model is trained end to end to understand complex scene semantics such as characters, hair, fabric and translucent skin"
what we understand: "this model is going to do racist and sexist things on previously unimaginable scales"
etc etc

2026-03-17 14:34:38 UTC

- reply

developer: i consent / triangle: i consent / framebuffer: i consent / jensen huang: i don't

2026-03-17 14:25:25 UTC

we don't need your statistical opinions about this triangle, jensen. just put it on the screen, for the love of god

2026-03-10 19:46:10 UTC

- reply

the fuck?!

2026-03-06 22:37:05 UTC

- reply

simon, garfunkel and a series of NOR gates can be combined to generate any other 1960s american folk-rock act

2026-02-23 22:56:01 UTC

- reply

(this is actually my main concern about llms. i think people really underestimate how much llms reproduce the values and expectations in their corpus, their reinforcement learning tasks, their explicit engineering, and their product design. and they underestimate the effects that this will have on their understanding of code and the horizon of what's possible to do with code)

2026-02-23 22:43:42 UTC

- reply

i don't know that i'd trust these models for summarization or navigation. even when the outputs are technically correct, they can leave out certain information or frame the information in a misleading way, papering over whatever makes the code unique and materially suited for the task at hand

2026-02-14 15:44:36 UTC

- reply

(also the term "AI" as currently understood is 100% a marketing phrase and its uncritical use in this context entrenches the worldview and assumptions of that marketing. imagine if instead of "food safety" as a field, we had conferences on "Responsible Use of Cold, Refreshing Coca-Cola")

2026-02-14 15:30:01 UTC

- reply

... i mean i understand that the "[word with positive ethical valence] AI" branding is useful from a legibility standpoint, and opens up funding possibilities. but i don't like that the framing sorta excludes refusal as a possibility. and in a recommendation algorithm feedback loop-based economy, we run the risk of "get mad about AI" becoming its own genre/microculture that is financially parasitic on the existence of AI to get mad about

2026-02-14 15:18:25 UTC

i'm paranoid that communities of practice with names like "responsible AI" and "critical AI" and especially "ethical AI" (even when done in good faith) mostly serve as permission structures/fig leaves for big commercial AI ("yes, the tech is harmful, but we have people working on addressing that"), and at worst run the chance of entrenching commercial AI by creating a category of professionals whose careers depend on its continued adoption—so they have something to criticize

2026-02-09 21:23:24 UTC

i like writing assembly because it helps you realize that even when you do something really simple on a computer, like check if two numbers are equal, beneath the surface there's this old grandpa who's like "well hold on there, let me just load this little feller into the accumulator"

2026-02-06 22:29:52 UTC

hi fediverse, in your opinion which is the lisp-like language that best maximizes for (a) usefulness and (b) non-toxic community? (i'm trying to pick one to learn and maybe actually get good at?)

2026-02-03 17:04:34 UTC

- reply

— i wrote this computer program with an llm
— okay, does it work?
— probably?
— how do you know if it works?
— your kind will be cleansed from the cosmos when the singularity arrives
— i'm sorry, what?
— what

2026-02-03 17:01:58 UTC

— i wrote this report with an llm
— okay, is it accurate?
— probably?
— how do you know if it's accurate?
— 🤷
— okay can you do it again but in a way that lets you verify its accuracy?
— why do you hate the future

2026-01-31 18:04:16 UTC

just now realizing that while "Ozymandias" is conventionally read as a parable about the fleeting nature of power, it works just as well as a manifestation of the orientalist need for the middle east to be a history-less tabula rasa ready to be overwritten by western colonialist expansion. i would say "in this essay," but i looked it up and of course there is already a bunch of scholarship about this!

2026-01-19 21:39:29 UTC

- reply

it smacks of something that feels common in the vibe coding crowd: a kind of pick-me parasitic incuriosity, in which the vibe coder disdains the very process of knowledge creation that they're benefiting from—i.e., other people's acts of computer programming. as a side effect, this provides evidence that the vibe coder is "not one of THOSE nerds" that actually enjoys, like, learning things and building reliable technical infrastructure in community with one another

2026-01-19 21:37:47 UTC

- reply

honestly, I think this is a pretty even-handed piece over all, and is a lot more realistic about the potential and drawbacks of LLM-generated code than most AI booster posts I come across. but i've been thinking a lot about this particular paragraph. why is it "bad news" that this is "still programming"? why is a technical detail about vintage hardware "an absolutely terrible thing to learn"?

2026-01-19 21:37:15 UTC

a more appropriate way to phrase this, Mr. Vibe Coder Paul Ford, might be "I'm very grateful to the people who contributed to this C++ project [link to project]. Their code helped me to understand how the Yamaha DX7 synthesizer voice format works in context." (from https://aboard.com/triple-a-vibe-coding/)
https://friend.camp/system/media_attachments/files/115/923/585/696/709/783/original/b2e85d1debfc2cc9.png

2026-01-14 20:24:54 UTC

- reply

i'm honestly surprised—whenever i try to use chatgpt to produce game boy assembly code, it's kind of a nightmare (lots of invented opcodes, impossible uses of the memory map, etc., along with the usual subtle errors). but maybe there are enough examples of tracker engines out there, and information/example code on SID chips, to make this an easier task?

2026-01-14 18:22:19 UTC

i think another (maybe the primary?) factor driving the use of "ai" tools in software dev right now is that it simply doesn't make market sense right now for a business to care whether or not their software works. so what if it's broken? end users will stop using the software but they'll forget to cancel their subscription, or they'll grit their teeth and deal with the jankiness bc the business in question has a monopoly and there are no good alternatives

2026-01-12 21:17:13 UTC

I know that this is a citation but I can't help but read it as "Friendships are about a mutual concern for one another *side-eye* ARISTOTLE."
https://friend.camp/system/media_attachments/files/115/884/196/852/219/099/original/1bdcf5d61cd8fded.png

2026-01-12 18:02:45 UTC

thinking today about a double standard that seems built-in to the philosophy of AI coding tool boosters: writing code is both too menial for expert programmers ("my specification is good enough, and i can always check the code to make sure it does what i said") but too complicated for beginners or non-programmers (who should use AI coding tools instead)

2025-12-29 16:34:47 UTC

- reply

potential responses to this:
(a) "software can't enable racism or sexism" to which i respond *extended exhausted sigh*
(b) "software written without using LLMs can enable racism and sexism too" okay but does using an LLM make it easier to avoid doing that? how?
(c) "it's good that software written with LLMs enables racism and sexism" cool cybertruck bro

2025-12-29 16:29:59 UTC

re: software development with LLMs—if you don't know how the code works (as so many vibe coders brag), how do you know it's not subtly enabling racism or sexism? or overtly, for that matter

2025-12-13 21:21:55 UTC

i'm trying to figure out how this works, especially the stuff in the lower half. it's part of the Game Boy Color's LCD power circuit (schematic adapted from https://wiki.console5.com/tw/images/e/e6/Nintendo_GBC_Schematic.png). COM is a ~4.6kHz 50% duty square wave, 0–5v; the cap (C35) turns it into ±2.5v. the NPN and PNP BJTs appear to use the pot's voltage to add a variable DC bias to this square wave (ranging from 0v to ~2.5v). but i don't get WHY this works and/or why they did it this way (continued)
https://friend.camp/system/media_attachments/files/115/714/276/278/353/126/original/6045b965b2417148.png

2025-12-05 16:33:00 UTC

the timer app on ios has a "recents" list, which is admittedly helpful and good UX, but it's also kind of fundamentally absurd, like "here are some durations that have been of interest to you lately"

2025-12-03 21:01:15 UTC

you know i'd never stopped to consider how annoying tulip mania must have been for folks who just wanted to grow a few pretty flowers in their front garden https://arstechnica.com/gadgets/2025/12/after-nearly-30-years-crucial-will-stop-selling-ram-to-consumers/

2025-10-10 17:56:23 UTC

this is such a strange post! the "also" in "I'm also pro open source" leaves open the implication that being "pro-LGBTQ" and "pro-immigrant" is separate from being pro-foss, and that sustaining the latter might require compromise on the former. personally, i'm in favor of foss *because*—and frankly *only to the extent that*—it brings about autonomy and liberation for LGBTQ people, and immigrants, and the marginalized people of the world in general. otherwise wtf are we even doing here
https://friend.camp/system/media_attachments/files/115/351/116/990/787/856/original/7e035d2d0fad39bc.png

2025-10-09 18:14:35 UTC

so i have an array with several thousand elements, ~50% of which are empty. the empty elements aren't uniformly distributed over the length of the array. (it's a hash table.) i want to select a non-empty array element at random. is there a clever way to do this that (a) is constant time (or close); (b) doesn't require a second data structure; and (c) isn't just sampling at random in a loop until you get a hit? (this is for an embedded application so RAM is at a premium)

2025-09-30 14:27:26 UTC

- reply

the goofy insouciant grin on the guy's face makes it all the worse

2025-09-25 21:54:31 UTC

so here's a question: are there any text editors out there that (a) run on bare-metal microcontrollers and (b) meaningfully support unicode? the closest i can think of is like... the lvgl textarea and the circuitpython repl (lol). (also interested in such text editors on other low-resource systems—basically anything where you can't "just install linux")

2025-09-12 18:50:58 UTC

"exnomination" is a really useful word and I'm surprised it isn't used more often in the context of theory and politics. exnomination is the process by which a group works to depoliticize their worldview and values, so that those views and values come to be considered "natural," "apolitical," and "common sense" (the term is from Barthes' _Mythologies_ but I first encountered it in Robin Lakoff's _The Language War_)

2025-09-10 02:03:47 UTC

- reply

this is the best explanation i've read for why anyone who is monetarily backed by the tech industry is feeling pressured to include generative models in everything https://gauthierroussilhe.com/en/articles/how-to-use-computing-power-faster

2025-09-09 16:50:04 UTC

i love living in the future, in which i have to hold my web browser very gingerly so it doesn't replace the thing i'm reading with the carbon farts of a stochastic white supremacy machine https://blog.mozilla.org/en/firefox/shake-to-summarize/

2025-09-04 22:59:00 UTC

- reply

the Berman years of Star Trek get a bad rap but personally i at least love how they killed off Kirk in such a small story with such small stakes, so much so that hollywood did a half-billion dollar temper tantrum over the next two decades trying to bring him back

2025-09-04 16:54:38 UTC

- reply

can't wait to catch the latest episode of "Star Trek: Ship of Chosen Ones"

2025-08-28 14:51:00 UTC

- reply

ridiculous premise, obviously. because if computers were extensible, efficient, easy-to-use machines that people can use to make things and solve problems, people wouldn't need to buy new devices every year to keep up with the system requirements of their software subscriptions

2025-08-28 14:44:24 UTC

what if instead of investing $500 billion in GPUs and data centers, we'd invested $500 billion in like... HyperCard

2025-08-22 15:58:33 UTC

- reply

this happens all the time with google search: the algorithm seems like it's set up to drop search terms that would make the results usefully specific, in order to present results associated with more common and/or generic queries. it's not just annoying (having to go back and put quotes around words just to make sure they're not ignored), it's normative—what it communicates is "lol sure. here's what non-weirdos are searching for"

2025-08-22 15:38:55 UTC

me: google search, can you help me find commentary on Walt Whitman's poetry written by Black scholars?
google search:
https://friend.camp/system/media_attachments/files/115/073/150/443/864/773/original/82d6ad92e9c7e0cb.png

2025-07-22 18:48:33 UTC

so here's a thing i've been working on for a while: https://abc.decontextualize.com/ "Allison's Bootleg Cart"! it's a versatile, low-power flash cartridge for the Game Boy (and gb-compatible systems) based on the RP2040 microcontroller. i'm not currently selling pre-assembled cartridges, but the source code and design files (firmware, PCB layout/schematic, and ASM) are all available online under copyleft licenses, if you want to make your own!

2025-07-09 05:52:29 UTC

- reply

(to be clear to any future commenters in the thread, I understand the reasons behind the change from 5v to 3.3v, and the benefits/tradeoffs of using lower voltages. that is all easy to learn through web searches. what i meant to ask for is *specific examples* of early uses of 3.3v-only microcontrollers and processors, especially in consumer electronics. a history question, not an electrical engineering question)

2025-07-09 00:28:42 UTC

- reply

(my main points of reference are: game boy in 1989 was 5v, game boy color in 1998 was 5v but with 3.3v sram for some reason, game boy advance in 2001 was 3.3v. but when i was learning arduino etc in 2006, pretty much everything was still 5v)

2025-07-09 00:25:12 UTC

ok fediverse i need help with a question that has been surprisingly resistant to web searches: when did it begin to become common (or at least not surprising) for microcontrollers and processors to run at 3.3v, rather than 5v (or more)—especially in embedded and portable contexts? who was the first vendor in this area who was like "sure, let's take a chance on 3.3v"? (did it happen in industrial applications before it happened in consumer electronics?)

2025-07-07 21:01:18 UTC

- reply

oh they do have a bunch of photos on the main project site https://www.hybridbody.human.cornell.edu/#/ecothreads/ and in their recent paper https://dl.acm.org/doi/10.1145/3715336.3735409

2025-07-07 18:29:26 UTC

this is a very cool project: an open source, inexpensive design for a wet-spinning machine, designed to help you make biodegradable yarns for e-textile prototyping https://hybridbodylab.github.io/open-wet-spinning/ (example yarn recipe: https://hybridbodylab.github.io/open-wet-spinning/docs/recipes/thermo/)

2025-06-09 19:02:35 UTC

- reply

hmm. you know what would be good for this, is a set of conventions for clearly and unambiguously describing the way that a system should function. like some kind of language...? for programming...?
https://friend.camp/system/media_attachments/files/114/654/935/942/864/832/original/d72a26c25e543246.png

2025-06-09 18:56:22 UTC

i wonder if someone at anthropic is working on a prompt improver prompt improver (i.e., it improves the prompt improver by improving the prompts that they prompt the prompt improver with) https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/prompt-improver

2025-06-08 19:22:22 UTC

i just finished reading _The AI Con_ by @npub15xv…0jms and @npub1e2z…n97d and it's pretty great—a brisk read and compellingly argued. it's handy to have all of their arguments and notes and references gathered in one place, even if you feel like you're already familiar with the AI/LLM anti-hype discourse. very easy to recommend, perfect for the policymakers in your life https://thecon.ai/

2025-06-02 14:29:18 UTC

so now that bangbangcon and strange loop are gone, where are people doing their fun not-quite-tech/not-quite-art/not-quite-academic talks?

2025-05-14 19:25:35 UTC

hi everyone! this sunday (may 18th) at 4pm eastern, we're doing an ONLINE release show for my new book of poetry, Two of Pentacles. here's the place to sign up (free!) https://www.eventbrite.com/e/two-of-pentacles-zoom-release-show-tickets-1342086941169 (the two other readers—Holly Melgard and John Cayley—are both AMAZING btw) more info about the book here: https://shop.nothing-to-say.org/products/two-of-pentacles

2025-04-25 21:18:23 UTC

- reply

learning how to program is learning how to do the real work of understanding systems and caring for people

2025-04-25 21:05:27 UTC

hi everyone! my new book _Two of Pentacles_ comes out tomorrow (april 26th), and you can order a copy here: https://shop.nothing-to-say.org/products/two-of-pentacles if you've ever wanted to read a long experimental poem-essay(?) on the topic of a single tarot card that was composed using cut-up methods, randomness and combinatorial optimization, published in such a way that no two copies are alike, boy howdy are you in luck!

2025-04-17 01:30:07 UTC

- reply

AI bros: we have finally achieved AGI! whoops there's a little smudge on my screen, let me clean that off and... oh. API. we have achieved API

2025-04-17 01:23:31 UTC

- reply

(perfectly predictable in hindsight that the ultimate teleology of llms is to be unreliable rube goldberg devices that process "natural language" into http requests and co2)

2025-04-17 01:15:15 UTC

- reply

we were prompt engineers even way back in the 90s. it's just that our prompts looked like
GET /stuff HTTP/1.1

2025-04-17 01:11:18 UTC

i am reading this and i feel like i am losing my whole mind https://modelcontextprotocol.io/introduction imagine that your text extruder is SO GOOD AND RELIABLE at dealing with unstructured data that you have to reinvent REST with schemas so people can interact with it

2025-04-16 19:31:11 UTC

sure, you're doing posts, but are you taking advantage of all the antiquarian imperial post sizes?
https://friend.camp/system/media_attachments/files/114/349/288/103/172/101/original/c287ba7d81c09517.png

2025-03-26 16:12:21 UTC

who decided to call it the myth of theseus and the minotaur and not "bull-terminated string"

2025-03-24 19:34:22 UTC

- reply

#3 definitely imo. they'll always displace the blame and signal's the nearest target

2025-03-20 16:07:03 UTC

reminder that whenever the gop complains about "censorship" what they always always mean is "you're making me feel bad about being a white supremacist." when they say that universities shouldn't be able to "indoctrinate" students, what they always always mean is "you shouldn't be able to tell people that white supremacy is bad." it is never a good faith argument and it should never be treated as such

2025-03-03 18:50:01 UTC

i think you could make a pretty good cross-disciplinary multi-year high school math curriculum around the question of how to make decent offset curves from cubic bézier splines

2025-02-24 01:13:18 UTC

here's how i got my kindle books into calibre on linux, without a current kindle serial number (1/n)

2025-02-13 17:20:48 UTC

[in the faculty meeting] listen, this new technology is here to stay, and it's useless to fight against it. if we want to prepare our students for the jobs of the future, we *must* teach them how to (responsibly and ethically) purchase and consume up to five Taco BellⓇ Naked Chicken Chalupas™ each day

2025-02-04 02:09:26 UTC

okay actual question for the fedi. what's the best "protect trans youth" sticker and what's the best place to buy it from (does not need to be that exact wording)? also my macbook air is one of the macbooks that lights up the apple on top shell w/the lcd backlight—what's the best way to make sure that you can't see the apple shape through a sticker?

2025-01-23 19:19:44 UTC

just received my rp2350 stamp xl from solder party (cc @npub13ew…pw5d). it's very cute!
https://friend.camp/system/media_attachments/files/113/879/275/466/065/457/original/bb638138b9b5c595.jpeg

2025-01-01 21:25:35 UTC

by some celestial alignment of luck, meds and rest i had the energy to take a long walk in prospect park today (which i used to do every week or so before long covid) (i didn't write that note btw—i found it near a bunch of others, tied to a fence, near what i assume to be the site of one of the brush fires in the park a few months ago)
https://friend.camp/system/media_attachments/files/113/755/174/446/703/704/original/48767672ba120d20.jpeg
https://friend.camp/system/media_attachments/files/113/755/174/447/512/303/original/07275d7d9fee48e2.jpeg
https://friend.camp/system/media_attachments/files/113/755/177/006/893/099/original/0eb7f588b2b7fa9c.jpeg
https://friend.camp/system/media_attachments/files/113/755/177/008/804/996/original/3ab73c8bef6cf6d6.jpeg

2024-12-03 18:42:25 UTC

i'd like to draw attention to this lil indie gameboy game by @npub1ydn…ypsp https://objetdiscret.itch.io/tai-fab-museum it's a collection of nonogram (picross) puzzles framed within a breezy surreal adventure game. it's polished and quite substantial (~200 puzzles) and plays great on real hardware. highly recommended!
#TaiFab

2024-11-16 22:15:11 UTC

i'm so exhausted with studies like this https://www.nature.com/articles/s41598-024-76900-1

2024-10-30 16:29:59 UTC

lately whenever i try out new foss software, it's like i can see the next ten years of my life unfold. the group producing the software launches a startup for hosting/"enterprise support"; the startup gets acquired; the acquiring company does lawyer stuff to repossess the code from the commons; there's a community fork under a different name; the fork builds up cruft that encourages a *different* group to make an incompatible "stripped-down" version; that group launches a startup...

2024-08-30 21:15:58 UTC

i got so angry after reading this paper on LLMs and African American English that i literally had to stand up and go walk around the block to cool off https://www.nature.com/articles/s41586-024-07856-5 it's a very compelling paper, with a super clever methodology, and (i'm paraphrasing/extrapolating) shows that "alignment" strategies like RLHF only work to ensure that it never seems like a white person is saying something overtly racist, rather than addressing the actual prejudice baked into the model