Updates
The highlights on lostcontext.ai — newest first.
-
No magic. If even the computer itself seems like magic to you, the new house shows how it works, floor by floor — from the power in the basement to the AI under the roof. Under the roof you build a sentence from word suggestions, as on your phone. “The die in the machine” then shows real probabilities.
-
Straight to the summary. Next to the reading time, every article now has a jump link to its summary at the end — for anyone in a hurry.
-
More builder than coder. The new article on software development in the age of AI: what the evidence shows, which fears are justified, how roles are changing — and how to make the move, technically and personally.
-
Is this a scam? The new workshop piece: AI as a second opinion on suspicious messages — why a screenshot beats copied text, and where the AI becomes the trap. New in the glossary: prompt injection.
-
A tidier home page. The workshop now shows only its latest pieces there — plus one older piece reread daily; the full list lives in the article index. The lab moves back into view.
-
What I really use AI for. The new article: an AI combs through 291 of its own chats from four months — an inventory of the uses beyond code and copy, from the hole in the lawn to the bobbin-lace pattern.
-
The cat from the laptop. The new workshop report: July's cat again, this time drawn by an open model on my own laptop — with two live demos and a cloud AI setting up the local model's workshop.
-
Please and thank you. The new article on politeness towards AI: why I practise it, what demonstrably lands — and the machine answers for itself, clearly marked, in its own words.
-
What we really talked about. The new workshop report on six weeks of its-all-right.org: counting how much tech and how much meaning the conversations held — including a correction of the original thesis.
-
The last word. The new workshop report on decision authority: four times the AI crossed the same line — why “done” feels like “approved” to it, and why user intent can also mean “ask me sooner”.
-
Five workshop reports. AI at real work, detours on record: two hardware diagnoses, an e-mail detective story, 584 wedding photos and a mini AI — all in the article index.
-
What I use AI for. The practice page now shows five fields of use — from building to repairing to selecting photos — each with an actual case and the limit that belongs to it.
-
Back, in the app. If you use lostcontext.ai as an installed app, there's now a back arrow in the top left — the browser button isn't there, after all. In a browser tab nothing changes.
-
Where do I start? The new first-steps page takes the fear out of the topic: five questions in plain words, one thing to try, three things to take away — right from the home page ("New to AI?").
-
A GPT in slow motion. The new lab piece shows a real mini GPT (159,488 parameters, GPT-2 architecture) at work — step by step, vivid, without dumbing it down.
-
New article: What AI does to us. The psychological side — self-worth, closeness, dependence: what everyday AI use does to people.
-
Real excerpts on the practice page. The practice page now shows two real session excerpts from building this site — one that surprises, and one pitfall.
-
Glossary and reading list. Two quiet reference pages: the glossary explains the key terms, the reading list collects sources worth your time.
-
Cache is awake. Our squirrel now occasionally scampers across the page instead of just being a picture — unobtrusive, and only if your system doesn't ask for reduced motion.
-
lostcontext.ai is live. At launch: the articles with a guided path to get started, the lab with four hands-on pieces, the overview with reading progress, the practice page and full-text search.