A model, here, is a worked way of producing something with AI — a method a normal person can run themselves, written down exactly. Every model on this site was run for real, on a real problem, and every one carries an honest rating: Works, or Room for Improvement. The second shelf is not a junk drawer. It's where this site keeps its credibility.
Every model is documented the same way:
Google Timeline → Calendar Journal
My phone already knew where I'd been. This model turns a Google Timeline export into diary entries on Google Calendar — real stops only, home excluded, duplicates impossible. 579 stops processed; the calendar now journals itself.
Write-up in progress · Becomes Application 001
The Backup That Said “Finished”
A cloud backup of 700 GB reported success. Instead of trusting the progress bar, the backup was checked against the provider's own file records — which is how the gaps got found and closed.
Write-up in progress
Twenty Drives, One Catalog
A lifetime of photos and files spread across twenty old hard drives, deduplicated into one catalog of 268,000 files — with the AI keeping the ledger and a human checking its math.
Write-up in progress
The Fiction Model
How one person and two AI assistants produced novels that stayed consistent across hundreds of thousands of words: a written canon, separated roles (one drafts, one critiques, the human rules), and verification instead of trust.
Write-up in progress
The Sensor That Cries Wolf
An always-on home sensor, checked hourly for months. The sensor is fine. The reporting path lies sometimes — and telling a real outage from a reporting failure took longer than it should have. Honest status: still being improved.
In retest
Short, concrete how-tos that grow out of the models above — written for someone who has tried a chatbot once and wants it to be useful. The first three are in progress:
Your First Useful Conversation with an AI
Bring a real problem, not a toy one — and what to do with the answer.
In progress
How to Check an AI's Answer Before You Trust It
Verification basics: the habit this whole site is built on, in one short lesson.
In progress
What Never to Paste into a Chatbot
Privacy basics for normal people: what stays on your side of the screen.
In progress
Reading about a model is good. Running it is better. Successful models here graduate into small applications you can use yourself — the first is in development:
Application 001 — Google Timeline Journal
Upload your own Timeline export and get calendar-journal entries out, using the exact rules from the model above. It runs in your browser: your location file never leaves your computer.
In development
The discussion will live on Reddit, opening with the first published models: ask questions, challenge a result, and — best of all — suggest the next experiment. When a reader's suggestion becomes a model, the verified results get published here and linked back to the thread it came from.
Beginner Model is one person's bench notebook for AI: a working technical person testing the major AI assistants (Muse, ChatGPT, Gemini) on real tasks and publishing what holds up. It leans toward whichever assistant does the job — around here that's often Muse, because personal-assistant work is what I wanted from AI on the very first day I used it, and because it has done specific jobs the others couldn't.
Beginner Model is not affiliated with Meta, OpenAI, Google, or any AI company. There's no team and no hype department — just a PO box, below, on purpose.
One short email when a new model or experiment posts. No schedule, no spam, unsubscribe any time.
