Methods · Generative practice & retrieval

Practice-problem generation

Get problems matched to a named skill at a named difficulty, so practice aims at what you are actually building.

When to use this: While you practise
Learning impact: Strong change
Learning evidence: Strong research tradition
Where you are with the topic: Strengthening memory
How often to use it: Worth repeating often

When to use it

The textbook jumps difficulty too fast, or stays on the same easy pattern.
You know the chapter title but cannot name the skill you are weak on.
You want three more problems like the one you just missed, not a random worksheet.
You are drilling near variants and need a clean difficulty dial.

What to ask AI

Type something like: “I need practice on [named skill] at [easy/medium/hard] level. Give me one problem only. Do not show the solution until I attempt it.”

Name the skill in verbs (“differentiate composite trig”, not “chapter 4”).

Ask for one problem at a time so you cannot skim ahead to the next answer.

How it helps

Practice works when it targets a skill you can name and a difficulty you can feel.

Vague requests (“more calculus”) produce random difficulty and teach little.

The AI’s usefulness here is generation and pacing, problems on demand, while you keep the attempt.

Pair this with feedback on your working, not with an instant full solution.

How to run it

Name one skill in concrete language and pick a difficulty.
Ask for a single problem with no solution attached.
Attempt it fully in writing, including uncertain steps.
Ask for the first error or a hint only after you commit.
Request the next problem one notch harder or easier based on that outcome.

Example exchange

You

I need one medium chain-rule problem with composite trig. Please give the problem only, no solution until I attempt it.

Differentiate y = sin(3x²). Show your working when you are ready.

You

I got y′ = cos(3x²)·6x. I treated 3x² as the inner function.

First check only: what is the derivative of the inner function 3x²? Confirm that factor before we raise the difficulty.

Copyable prompt

Skill: [named skill in verbs]. Level: [easy/medium/hard].
Give me ONE problem only. Do not include the solution, hints, or next problems
until I submit my attempt.

The Tell

You will know the method has flipped when you ask for a problem set and read the solutions in the same breath.

Then generation became a worksheet download, and the learning event never started.

Demand one problem, close the chat until your working is on paper, then reopen.

Principle evidence

Strength of the underlying learning idea, not a claim about AI products.

Deliberate practice aims at a defined skill with feedback. Generating well-matched problems supports that loop; the learning still lives in the attempt and the correction.

AI delivery evidence

Whether an AI tutor delivers this method well is a separate question.

Problem quality and difficulty calibration from AI vary. If a “medium” problem is trivial or impossible, say so and recalibrate, do not assume the label is accurate.