One skill at a time, tested before it is scaled
The platform I tried to learn from was a catalogue: plenty of courses, loosely related, with nothing to say where a person in my position should start. Building a bigger catalogue is not the answer to a catalogue. So the first thing I am making here is one unit, on one skill, short enough to finish in a sitting and sequenced so the learner ends up doing the thing themselves rather than watching it done.
Each unit follows the same shape, which is a well-evidenced one rather than my invention: a worked example first, then guided practice, then the learner working alone, then a return to it a week later. Units are written to be read without buying anything or signing up to anything.
What exists so far
Effective prompting
What is really happening when you hit send, and the five things a useful prompt carries. Includes a table for working out what went wrong from the answer alone, and three exercises — two of which need no AI tool at all. Between forty-five minutes and an hour.
Reading a drive's health report
How to read what a hard drive says about itself, which numbers actually predict failure according to the largest published dataset on it, and the one manufacturer whose figures mislead you if read literally. Written while building the project's own server.
Building the pilot AI workstation
The build this project is running on its own desktop, written as a unit: what the hardware has to do, the two software routes and why the choice between them is not obvious, and the fixed measurements that make one run comparable to the next.
What is planned
The learner trial, and what it changes
The prompting unit goes to a small number of real learners before anything else is built on it. This is where that gets written up: what people actually did with it, where they got stuck, and what had to change as a result. If it does not work, that will be here too. Testing a unit and then only reporting the flattering half would undo the point of the rest of this site.
Handling your own data sensibly
The second unit, if the first one survives its trial. What actually happens to something you paste into a chat window, what the terms you agreed to say about it, and how to decide what should never go near one. Practical rules a small organisation can adopt, rather than a policy nobody reads.
Units delivered by the machine itself
The longer-term destination for all of this. Rather than sitting on a website waiting to be found, units are carried by the workstation described elsewhere on this site, offered when they are relevant to what someone is doing and shaped by what they have already worked through. Nothing here is built. The workstation page sets out how far that has got.
Why so little, so slowly
The failure this project is a response to was a large catalogue that had never been tested on anyone. Producing a large catalogue faster would repeat it. One unit, tried on real people, tells me more about whether the approach works than twenty units nobody has opened.
The drafts are public while they are still rough, which is the same standard the research pages hold to. It is easier to show the work than to describe it.