An AI tutor is worth it once you can check its answers

A question this app already has a checked answer to

"Should I use RSpec or Minitest?" is the kind of question an AI coding assistant answers instantly, and on a comparison this well documented it's usually right. This app's course "Minitest: model & assertions" already worked out the answer for this codebase. RSpec is what this app runs: 206 spec files, gem "rspec-rails" in the Gemfile, and no test/ directory. Minitest is a valid alternative the app doesn't use, Rails' own built-in framework, still loaded with the app's Rails gems whether or not anything calls it. That course didn't guess. It ran bundle exec rspec spec/, read the installed minitest and activesupport gem source, and translated one of this app's specs line by line.

That's the point of this course. It doesn't re-teach RSpec or Minitest; "Minitest: model & assertions" already did that against running code. This course teaches something else: how to use an AI assistant as a tutor for a technical comparison like this one, and how to tell whether what it told you is right.

Why asking an AI is a skill, not a shortcut around learning

Prompting an AI assistant with "explain the difference between X and Y" produces an answer in seconds. For a topic as widely documented as RSpec versus Minitest, the general facts are usually accurate: what each framework is, roughly how it's used, what the basic syntax looks like. The risk isn't that AI assistants are bad at this question. It's closer to the opposite. They're right often enough that it's tempting to stop checking, and when they're subtly wrong, a smooth, plausible paragraph makes the mistake hard to catch.

This app's AI courses already build the general habit. "AI prompting: zero-shot Ruby" covers writing a prompt precise enough to get a useful answer. "AI safety: review before copying" covers reading an AI's generated code skeptically before it goes near a file. This course points the same instinct at a different kind of output, a generated explanation instead of generated code. It also has a useful shortcut for checking one, because this app already has a checked answer in its own catalog.

The check this course runs, lesson after lesson

Each of the next several lessons follows the same shape: a prompt you could type into an AI assistant, a realistic example of the answer it might give, and a check against what "Minitest: model & assertions" established by running the code. When the AI's answer holds up, the lesson says so; most of what a decent assistant produces on a well-known comparison does hold up. When it doesn't, that's the more useful lesson. The point isn't "AI is unreliable" in general. It's which kind of claim needs a second look, and how you'd catch it yourself without a finished course to check against.

A word on the AI answers quoted in this course

None of the AI responses in the next six lessons are captured verbatim from one tool's chat history. They're written to represent typical answers an assistant would give to the prompt shown: accurate where an assistant would usually be accurate, and, where the lesson says so, wrong in the specific, plausible way an assistant can be wrong. Treat each one the way you'd treat any AI output in a project: worth reading, not worth trusting without a check.

What this course covers

Seven lessons. Lesson 2 is a first comparison prompt and how to read what comes back. Lesson 3 asks an AI to translate a piece of this app's RSpec into Minitest, checked against the translation in lesson 6 of "Minitest: model & assertions". Lesson 4 is a narrower question about one Rails testing method, checked against the installed gem source that course quoted. Lesson 5 is a plausible AI answer that turns out to be wrong, caught by a fact that course established by running code. Lesson 6 is a short, reusable checklist for verifying an AI tutor's technical claims on any topic. Lesson 7 pulls the habit together.