Most sites in the CARECOM family help you do something — find a hike, print a 3D model, take a moment to breathe. The newest member helps you understand something. lostcontext.ai is about artificial intelligence, told the way you rarely read about it: without the hype and without the panic, by someone who builds applications with these tools every day rather than just talking about them.
Even the name is a small joke with a double bottom. “Lost context” is a technical term — language models lose the thread in the middle of long texts; the context slips away. And at the same time it is a mission: to give the often overheated AI debate its context back. The tagline puts it in three words: AI, in context.
From practice, not from the hype
There is no shortage of AI talk. My angle is a practical one: I build applications with these tools — with agents, with the Model Context Protocol, with the very concrete questions of the EU AI Act from the point of view of someone who actually puts such things into operation. What I see while building is what lands here: no marketing, no fear, just what really happens in the day-to-day work — explained so that it clicks even without a computer-science degree, yet without making the subject smaller than it is.
Hands-on: the Lab
My favourite pieces are the ones you don't just read but try out. The Lab holds four small interactive demos, each making one abstract trait of AI tangible — all right in your browser, without a single word ever being sent to an AI model:
- What does a sentence cost? breaks your text into “tokens” live — the little pieces a model reckons and bills in. Along the way you'll see why German costs more than English.
- The dice in its head makes visible that a model doesn't know “the” answer but rolls for it — and how a temperature dial sharpens or flattens that die.
- Confidently wrong is a little quiz: seven statements, all equally self-assured, some true and some pure invention. Two minutes to grasp what “hallucinating” means for AI.
- The context window to touch shows how the start of a conversation “falls out of the window” as a model runs on.
Food for thought: the articles
On top of that come thirteen articles — from the basics (What exactly is “context rot”? What might the EU AI Act mean for you?) through the workshop (the three dials the quality of an AI answer really hangs on; why an AI blindly draws an “ugly cat”) to a few more reflective pieces on what all of this does to us. With sources you can look up — and where I'm not sure myself, that is exactly what it says.
Cache, the squirrel
A mascot scurries through the site: Cache, a squirrel. That is no accident. Squirrels are “scatter-hoarders” — they hide their stores in hundreds of places and never find some of them again. From exactly those forgotten caches, new trees grow. I couldn't find a lovelier picture for “from lost context, something new grows”. And should you ever land on a page that isn't there, Cache apologises with a small in-joke: “I've lost this context.” — a cache miss, of course.
Honesty stays the rule
One thread runs through all of it: honesty. A site that explains AI mustn't tell itself fairy tales. That is why even the mascot is fact-checked — the famous “three-second goldfish” is a myth, so it was dropped. That is why the site remembers, if you like, what you've “started” and what you've deliberately “read”, because scrolling isn't reading yet. Drop by the next time you feel you're losing the thread with AI — here you get it back.