Learning With AI

Use an AI assistant as a patient tutor without letting it do your thinking: try first, ask for hints, explain it back, and get quizzed.

14 minLesson 4 of 7
On this page
  1. Who does the thinking
  2. Set it up as a tutor
  3. Four habits that keep you in charge
  4. A tutor that is sometimes wrong
  5. Try it yourself
  6. Common mistakes
  7. Check yourself
  8. What to practise next

Two friends learn recursion with an AI assistant over the same week. The first pastes each exercise in, reads the solution and the explanation, and nods along. Everything makes sense. The second tries each exercise first, asks for a hint when stuck, and explains the answer back in their own words before moving on. It is slower and more frustrating.

At the end of the week they each get a recursion problem on paper, with no assistant. The second friend works through it. The first stares at the page. Reading a clear explanation felt exactly like understanding, but it was not practice at producing one.

Who does the thinking

You learn what you work to recall and produce. Following someone else's correct reasoning is easy and pleasant, which is why it feels productive. Building the reasoning yourself, getting it wrong and fixing it, is what makes it stick.

An AI assistant can sit anywhere on this line. It can do all the thinking for you, or it can make your own thinking faster and better aimed. The same tool, asked differently.

What you ask for How much of the thinking you do Write it for me Explain the answer Give me a hint Check my attempt Quiz me, no hints
The same assistant, asked five ways. The further down the list, the more of the thinking stays with you.

None of these is always wrong. Reading a worked answer is useful after you have tried, or for a topic you only need to recognise. The mistake is living at the top of the list for something you need to be able to do.

Set it up as a tutor

By default, assistants try to be helpful by giving complete answers. If you want a tutor, say so, and say what kind. A short setup at the start of the session does most of the work.

Tutor setup and a stuck momentIllustrative example. Real replies vary.
You
I'm learning recursion in Python. I know loops and lists well. Act as a tutor: when I'm stuck, give me one hint at a time, as a question if you can. Don't show me corrected code unless I ask for it. After I fix something, ask me to explain why the fix works. Here is my attempt at summing a list. It works on [4, 2, 5] but my test with an empty list crashes: def total(nums): if len(nums) == 1: return nums[0] return nums[0] + total(nums[1:])
Model
Good start: the recursive step is right. Let's find the crash. Trace total([]) by hand. Which line runs first, and what is nums[0] when the list is empty?
You
len is 0, not 1, so it skips the if and does nums[0]. There is no item 0, so that's the error.
Model
Exactly. Now: what should the total of an empty list be? And once you know that, do you still need the len(nums) == 1 case?

The tutor never wrote the fix. The learner found the cause by tracing, and the next question leads them to a better base case (an empty list sums to 0), which also makes the one-item case unnecessary. That is the lesson about base cases, and the learner will remember it because they worked it out.

Four habits that keep you in charge

  1. Bring an attempt. Even a wrong one. "Here is what I tried and where I got stuck" gets you a far more useful reply than the bare question, and it means you have already done the hard part of thinking.
  2. Explain it back. Write the idea in your own words and ask the assistant to point out anything wrong or missing. Gaps you cannot put into words are gaps you have.
  3. Get quizzed later. Ask for three questions on yesterday's topic and answer without looking anything up. Recalling something after a gap is one of the most reliable ways to make it last.
  4. Ask for a different angle. If an explanation does not land, ask for a concrete example with small numbers, a picture in words, or a comparison to something you already know. Then try the exercise again.

A tutor that is sometimes wrong

A human tutor rarely invents a fact. An AI tutor sometimes does, in the same calm voice it uses for everything else (see Why Models Hallucinate). When you are learning, this is harder to catch, because you do not know enough yet to notice.

Three safeguards help. Prefer well-trodden topics, where the model has seen many correct explanations; it is far more reliable on loops than on a niche library released last month. Keep a trusted source next to you, such as official docs or a course, and check anything surprising against it. And test claims you can test: if it says a function behaves a certain way, run it. Verifying AI Output has the full routine.

Try it yourself

Try it: Turn a question into a tutoring session

This request will produce a long lecture you will half-read and forget. Rewrite it so the assistant teaches you by making you think. Include your level, how you want to be helped, and how you will know you understood.

Show a sample answer
Rewritten requestIllustrative example. Real replies vary.
You
I want to learn dynamic programming. I'm comfortable with recursion and Big-O, and I've never seen DP before. Teach me by starting with one small problem I can solve by hand, not with definitions. Let me try it first. When I'm stuck, give one hint at a time. Only explain the general idea after I've solved the example. At the end, give me two new problems to solve on my own, and check my answers without showing me yours first.

Why it works: stating your level skips what you know and avoids what you are not ready for. "Problem first, idea after" means the explanation lands on something you have just struggled with. The final two problems are the real test: if you cannot do them alone, you have not learned it yet, however clear the explanation felt. The site's own DP introduction is a good trusted source to check against.

Common mistakes

Check yourself

Quick check

You are stuck on a practice problem. Which request keeps the most learning with you?

After reading a clear explanation of binary search, you feel you understand it. What is the best way to check?

Why is an AI tutor's mistake harder to catch when you are learning?

What to practise next

Pick one topic you are studying and run a tutor session with the setup above: one attempt, hints only, explain it back. Then continue with Privacy and Security, on what is safe to paste into an assistant in the first place.