“Ask an AI to do extended thinking…” — and it tells you what it’s doing?! (Sort of.)

 

“Ask an AI to do extended thinking…” — and it tells you what it’s doing?! (Sort of.)

A Level Computing | Philip M Russell Ltd style | UK spelling

You’ve seen it happen:

You ask an AI a tricky question.
You click the button that says something like “extended thinking”.
And then—instead of just blurting out an answer like an overconfident Year 10—you get a response that sounds like the AI is narrating its brain:

“First I’ll break the problem down… then I’ll check edge cases… then I’ll verify…”

It feels like watching a student show their working. Which is oddly comforting.

But here’s the important bit for A Level Computing:

The AI isn’t “showing its thoughts” in the way you think

Most modern AI systems do not reveal their full internal reasoning (often called chain-of-thought). What you’re seeing is usually a summary of the approach: a tidy, human-readable explanation of the steps it took or would take.

That’s not a bad thing. In fact, for learning, it can be brilliant — but you need to understand what you’re getting.

---

What “extended thinking” usually means (in plain English)

When you request extended thinking, you’re generally asking the model to:

• Spend more compute/time on reasoning

• Break the task into sub-problems

• Self-check for contradictions and missing cases

• Explain the method more explicitly than usual

In A Level terms, it’s similar to switching from:

• “Give me the answer”
  to

• “Show me your algorithm, and then run it carefully.”

---

Why it looks like the AI is narrating its process

Because narration is useful.

A well-structured explanation often includes:

• Identifying inputs/outputs (specification thinking)

• Planning a method (algorithm design)

• Checking constraints (edge cases, assumptions)

• Verifying results (testing / validation)

That’s basically the Computational Thinking toolkit:
Decomposition, abstraction, algorithmic thinking, evaluation.

So the AI is doing what your teacher has been nagging you to do all along. (Annoying, isn’t it?)

---

The catch: “explanations” are not the same as “proof”

Even if the AI gives you a lovely step-by-step explanation, it can still:

• use a wrong assumption,

• miss a constraint,

• produce an answer that sounds correct but isn’t.

So treat it like a very fast study partner who sometimes confidently walks into lampposts.

A Level-friendly rule:

Use the AI’s explanation as a draft algorithm — then test it like you would test your own code.

---

How to prompt it properly (so it actually helps you learn)

Try these prompt styles:

1) Ask for a plan first (before the final answer)

Prompt:
“Give me a brief plan (like pseudocode / method) before the final answer.”

Why it helps: you can spot dodgy logic early.

---

2) Force it to state assumptions

Prompt:
“List your assumptions explicitly before solving.”

Why it helps: you can challenge the weak bits.

---

3) Ask it to check edge cases

Prompt:
“After answering, test your solution against 3 edge cases.”

Why it helps: that’s literally exam evaluation.

---

4) Ask for a marking-grid style response

Prompt:
“Answer like an A Level student: define terms, show method, give final result, then evaluate limitations.”

Why it helps: it mirrors how marks are awarded.

---

A quick example: “Explain how you’d search for the fastest route”

Instead of:
“Find the fastest route.”

Try:
“Explain how you’d approach this: identify the graph model, choose an algorithm, and justify it.”

Now you’re doing proper A Level:

• Graph representation (nodes/edges/weights)

• Algorithm choice (Dijkstra vs A* vs BFS)

• Justification (constraints, complexity, correctness)

---

So… should you trust the “thinking”?

Trust it the way you trust a calculator:

• Great for speed

• Great for structure

• Still your job to check it’s answering the right question

And if it gives you a neat method: brilliant. That’s basically revision.

Just don’t confuse “a convincing explanation” with “guaranteed correct”.