Adaptive difficulty

Have the difficulty adjust to your performance in real time, so you stay at the edge of what you can do.

When to use this: While you practise
Learning impact: Strong change
Learning evidence: Credible with caveats
Where you are with the topic: Strengthening memory
How often to use it: Build and keep as a habit

When to use it

You are mid-way through a topic and fixed problem sets feel too easy then suddenly too hard.
You want practice time spent at the edge, not in comfort or in chaos.
You finished an adaptive session and need a diagnosis of what makes items hard for you.

What to ask AI

Ask to be tested on the topic with difficulty adjusting: harder when you are right, easier when you are wrong.

Ask it to aim for roughly 75-80% correct and to announce when difficulty changes.

After about twenty questions, ask what difficulty you settled at, what feature makes items hard for you, and where your ceiling is.

How it helps

Too easy and you are practising what you already have: pleasant, low return.

Too hard and you fail without learning, because you cannot connect the feedback to anything.

The productive zone is narrow and it moves as you improve, which is why a fixed set is the wrong shape.

Aim for roughly 70-85% success: high enough that feedback is interpretable, low enough that you are stretched.

Below about 60% you are usually missing a prerequisite rather than needing more practice at this level.

How to run it

Use this when you already have some foothold, not on first contact with the material.
Ask for adaptive questions with difficulty changes announced.
Work in short blocks (about twenty items) and log when it got easier or harder.
Ask the diagnostic question: what specific feature of a question makes it hard for you?
If difficulty keeps sinking, stop and diagnose prerequisites instead of enjoying the descent.
For exam rehearsal, match exam difficulty on purpose, do not let adaptation replace that standard.

Example exchange

You

Test me on the chain rule and adapt as we go, aim around three-quarters correct.

Question 1, baseline: differentiate sin(3x²).

You

6x cos(3x²).

Correct. Stepping up, question 2: differentiate cos(e^{2x}).

You

−2e^{2x} sin(e^{2x}).

Right. I'll keep adjusting. After a longer set we can name which feature raises the difficulty for you.

Copyable prompt

Test me on [topic], adapting as we go: harder when I get one right, easier when
I get one wrong. Aim to keep me at about 75-80% correct. Tell me when the
difficulty changes and in what way, so I can see where my ceiling is.
After twenty questions, tell me: what difficulty did I settle at, what specific
feature of a question makes it hard for me, and where exactly is my ceiling?

The Tell

You have done several adaptive sessions and cannot say what makes a question hard for you.

You have been scored, not diagnosed.

A session that drifts steadily downward is often a prerequisite gap wearing a difficulty costume.

Principle evidence

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

The underlying learning idea is rated moderate. Matching challenge to performance is the promise of intelligent tutoring and computerised adaptive testing. There is credible literature with boundary conditions; the matching idea is solid.

AI delivery evidence

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

Chat-AI adaptivity is crude compared with true adaptive testing. Treat difficulty labels and “levels” as approximate coaching, not psychometrically validated scores.