aishipped.lol0 min · 0 cards

$ cat /feed

the feed_

an endless scroll that isn't farming you.

you are the algorithm. Press tune: each slider is how often a kind of card shows up, and your taste is a profile you write yourself. That is the whole ranking.
every card tells you why it's here. No hidden engagement score. The line at the top of each card is the full reason, and delivery log lets you download every one.
any ranking runs on your computer. Taste matching uses your own words, or a model on your own machine through Ollama. No server sees your profile.
nothing is tracked. This page has no analytics. Your weights and saves stay in this browser.
everything is credited and linked. All of it comes from museums, Wikipedia, NASA and people who share their work openly.
it shows you the clock. The time you've spent is at the top, and every chapter ends with a place to stop.

keys: j/k next/previous card · s save · m more like this

loading…

tune

uncurated: every card is a random pick from a source's whole collection, not from anyone's selection (no featured lists, no most-voted, no trending, no museum highlights, no topic lists chosen by me). Sources that are nothing but someone's picks switch off. Expect it to be stranger and plainer: that is what an unedited archive looks like.strict: a card with too little text to check against your never-show list (an untitled picture, say) is not shown. Everything downloaded ahead of time was also screened against the whole list, every field, before it got here.

How often each kind of card shows up. 0 turns it off.

art 3public-domain paintings, prints and objects from the Art Institute of Chicagonature 3the most-loved wildlife photos on iNaturalist, identified by people who knowwikipedia: best of 3featured articles, pictures of the day, "did you know" and on-this-day, from a random day since 2016wikipedia: random 2a truly random article. most are small. some are wonderfulpoems 0off by default: short public-domain poems, mostly old English-language poetsphotos 3featured pictures on Wikimedia Commons, voted the best of 100 million filesold photos 2photographs from the Library of Congress, mostly 1860s–1960sbooks 2classic and well-loved books from Open Library, by subjectthe planet right now 1earthquakes from the last day, live from the US Geological Surveyfediverse 0off by default: posts people are boosting on mastodon.social right now, often tech and news chatterold films 2public-domain films from the Prelinger Archives: ads, newsreels, how-to films, 1910s–1970s. tap to playthe met 2public-domain art and objects from the Metropolitan Museum of Art in New York: 5,000 years of itfree books 2whole books you can read free on Project Gutenberg, from its 79,000old records 278rpm records from the 1900s–1950s, digitised by the Internet Archive. tap to playsecurity 2cybersecurity: vulnerabilities attackers are really using (CISA), and how attacks work (MITRE ATT&CK, CAPEC)today 0off by default: what the world is reading on Wikipedia right now, and today's newsdesign 2textiles, posters, ceramics, jewellery and dress from the V&A in Londoncuriosities 2old medical, botanical and scientific images from the Wellcome Collectioneurope’s museums 2open-licensed images from thousands of European museums, libraries and archives via Europeanatravel 2a random place in the world, from the Wikivoyage travel guideamour 0off by default: love poems and the nude in art, from Cleveland museum and PoetryDBpeople talking 0the best recent links on Hacker News, with their comment threads. off by default: mostly tech-industry talkchapter length 25cards between each stopping pointnever showOne word or phrase per line, lines starting with # are notes. Any card whose text mentions one is dropped before you see it, including "more like this". It reads text only: it can't see what is in a picture.

your taste runs on your computer

more of thisless of thistaste strength 00 is off: pure random draw. Each step compares more of the cards waiting in a source and shows the one closest to your taste.learn from my saves: cards you've starred count as "more of this" too.use the model on my computer: match by meaning instead of exact words, with through Ollama at localhost. One time: ollama pull nomic-embed-text, then let this site reach Ollama by settingOLLAMA_ORIGINS=https://aishipped.lol where Ollama starts: in its service settings if it runs in the background (on NixOS, services.ollama.environmentVariables), or as OLLAMA_ORIGINS=https://aishipped.lol ollama serve if you start it by hand.

The delivery log is every card shown on this device: when, from where, the link, the image address, the reason and the taste score, plus every server this page has contacted. It is kept in this browser and downloaded as a file you can open and check.

What decides the order: a random draw using the weights above, then, only if you turn taste strength up, a comparison against the profile you wrote, done in this browser or by your own computer's model. No server-side model, no profile anyone else holds, no history of what you clicked; the code is here. Some sources hand over their own popular lists (wildlife and Hacker News by votes, Mastodon trends, Wikipedia's most read), ranked the same for everyone. Set those to 0 for a feed with no popularity in it at all.