Using AI to Learn Without Letting It Do the Learning

An assistant that can explain anything, at any level, at two in the morning, without ever getting impatient with you.

For an adult teaching themselves something — no teacher, no classmates, nobody to ask when a paragraph makes no sense — that is genuinely new, and mostly good. The thing that stopped most self-teaching attempts was never lack of material. It was getting stuck on something small at half nine at night with nobody to ask, and quietly not going back.

There's a catch, and it isn't the one people usually name. The problem isn't that the answers are sometimes wrong, though they are. It's that they arrive already finished.

The trap, and why it's the same old one

Memory and skill are built by producing, not by receiving. Retrieving something from your own head, attempting a problem, explaining a thing in your own words — that's the work that changes what you're capable of. Recognising a good explanation when you read one feels almost identical and does almost nothing.

So an assistant that will produce a clear, well-structured, correct-sounding answer to any question you have is, from a learning point of view, a machine for doing your repetitions on your behalf.

That's the Input-Only Trap again, and this version is far more seductive than a highlighter. The explanation is tailored to exactly what you asked, pitched at your level, and it lands smoothly. Smoothness reads as understanding. You close the laptop feeling like you've learned something, and two days later there's nothing there — the same emptiness as finishing a book you can't summarise, arriving faster.

The tell is easy to check. Shut everything and explain the thing out loud. If you stall in the second sentence, you didn't learn it, you watched it being learned.

Four ways to put the work back

Answer first, ask second.

Before you prompt, write your own attempt — even a bad one, even three lines of what you think might be going on. Then ask. This one change does most of the work on this page, because now you're comparing your answer to a better one, which is a form of feedback, rather than receiving a finished product with nothing of yours to compare it to. It also makes you notice the specific place your model was wrong, which is the only bit worth remembering.

Make it examine you rather than tell you.

Ask it to test you on what you just read, without giving the answers until you've tried. Ask it to give you five problems slightly harder than the one you managed. Ask it to mark your explanation and point at what you left out. It's very good at this and almost nobody uses it this way, because being examined is unpleasant and being told is nice. This is also how you find the edge of awkward without having to guess where it is.

Explain it back with the tab closed.

Do the closed-book version out loud, then check. The stall is the useful part — it's the precise location of what you didn't take in, and it's invisible while the explanation is still on screen in front of you.

Ask why, and ask for the disagreement.

Push past the first answer. Why is it done that way, what breaks if you do it differently, what do people get wrong here, where do experts disagree. That kind of questioning is how you get from a working answer to actual understanding, and it's the thing this tool can do that a textbook can't — it will follow your confusion wherever it goes, for as long as you want, which is exactly what a good tutor does and what almost nobody has access to.

Seven prompts you can copy

Everything above is a habit rather than a trick, and habits are easier to keep when the wording is already written. These work on any of the main assistants — ChatGPT, Claude, Gemini, Copilot, Perplexity, whatever you already have open — because none of them depend on a feature only one product has. They are just instructions phrased so the thing stops handing you finished answers.

Copy one, swap anything in [SQUARE BRACKETS] for your own words, and send it. That is the whole operation. The brackets are the only part you need to touch.

I'm learning [TOPIC]. Before you explain anything, here is my own attempt at it:

[WRITE YOUR OWN ATTEMPT HERE - EVEN IF YOU THINK IT'S WRONG]

Do not rewrite it for me. Tell me which parts are right, which parts are wrong, and exactly where my understanding breaks down. Then give me the correct version of only the part I got wrong.

I have just been learning about [TOPIC]. Test me on it.

Ask me one question at a time and wait for my answer before moving on. Do not give me the answer until I have attempted it. Start straightforward and get harder depending on how I do. After [NUMBER] questions, stop and tell me which specific things I should go back over.

I am going to explain [TOPIC] to you from memory, with my notes and the page closed. Here is my explanation:

[YOUR EXPLANATION, IN YOUR OWN WORDS - WRITE IT BEFORE YOU LOOK ANYTHING UP]

Mark it. Tell me what I got wrong, what I left out completely, and what I said in a way that sounds right but isn't quite. Be blunt about it - I would rather find out now.

I understand the basics of [TOPIC]. Roughly: [ONE SENTENCE ON WHAT YOU THINK YOU KNOW].

Now take me past the basics. Why is it done this way rather than the obvious alternative? What breaks if I do it differently? What do beginners reliably get wrong here? And where do people who know this subject well genuinely disagree with each other?

I am stuck on this and I have read it three times already:

[PASTE THE PARAGRAPH, PROBLEM OR CODE THAT ISN'T LANDING]

Explain it to me three different ways: once as though I have never met the subject, once with a concrete real-world example, and once using an analogy from [SOMETHING YOU ALREADY KNOW WELL - COOKING, DRIVING, FOOTBALL, YOUR JOB]. Then ask me one question to check which version actually landed.

Give me [NUMBER] practice problems on [TOPIC], pitched slightly harder than this one, which I can already do:

[A PROBLEM OR EXAMPLE YOU CAN ALREADY HANDLE]

Give me the problems only. No answers, no hints, no worked examples underneath. I will come back with my attempts and you can mark them then.

You have just told me the following about [TOPIC]:

[PASTE WHAT IT TOLD YOU]

Go through that answer claim by claim. For each one, tell me how confident you actually are, what I should search for to verify it independently, and which parts are most likely to be out of date or simply wrong. If any of it touches UK law, tax, benefits, employment rights or health, say so explicitly and point me at the official source rather than summarising it.

Two things worth knowing about all of them. They work better in a fresh conversation than bolted onto the end of a long one, because a chat that has already been answering your questions tends to keep answering them. And if it slips back into explaining when you asked to be tested, say so plainly — stop giving me the answers, ask me the next question — and it will pick the thread back up.

Where it's genuinely brilliant

Worth saying clearly, because the cautious version of this article would undersell it.

Being stuck is the thing that ends self-teaching, and this dissolves being stuck. A paragraph that makes no sense can be re-explained five different ways in a minute. A concept can be pitched at whatever level you actually are rather than the level the book assumed. You can ask the question you're embarrassed to ask a person, including the very basic one you feel you should know by now, and nothing happens except that you get an answer.

It's also excellent at the scaffolding around learning: turning a chapter into a set of practice questions, generating examples until one clicks, or acting as the person you explain things to when there isn't one.

Where it will let you down

It is confidently wrong sometimes, and the confidence doesn't drop when the accuracy does. That's the property to keep in mind. Anything factual that matters — a figure, a rule, a date, a medical or legal or financial detail — needs checking against a proper source, and the fact that it sounded authoritative is not evidence of anything.

It's least reliable exactly where you're least able to catch it: at the edge of your knowledge, on niche topics, on anything recent. When you know a subject you spot the errors instantly. When you don't, you can't, which is the awkward shape of the problem.

For UK-specific things in particular — benefits, tax, allowances, employment rights, NHS pathways — check GOV.UK or the relevant official body rather than accepting a summary. Rules change, and a plausible answer about a rule that changed last April is worse than no answer.

And it can't tell you whether you've understood something. It only sees what you typed. You're still the one who has to close the laptop and find out.

This is Pillar 03: Knowledge

Attempt it first. Get examined. Explain it with the tab shut. Keep asking why.

Used the lazy way, this is the most efficient method ever invented for feeling like you're learning while learning nothing — infinite smooth explanations, zero retrieval, and a growing sense that you understand a subject you couldn't say three sentences about.

Used the other way, it's closer to what people have always meant by a good tutor: something that will meet you exactly where you're stuck, at eleven at night, and keep going until it lands. That was never available to most people. It's worth using properly.

More at https://thequietreset.uk