Skip to content
experience

Year 10 handed me 28 nearly identical AI essays on how a CPU works. Here's what I do now

Rhys Llewellyn@rhys_teaches_code

September 6, 2026 · 2 min read

#7 in AI Hype Check

Contents(3)

Evening all. Marking on a Sunday again, so you get a post.

Last term I set Year 10 (14 and 15 year olds) a short homework: explain the fetch-decode-execute cycle in your own words, 300 words, with an example. I got 28 essays back from a class of 31. At least 22 described the CPU as "the brain of the computer, orchestrating" something or other, and four mentioned a register that isn't on our spec at all. Two ended with "I hope this helps!" Diolch, I suppose.

I'm not angry with them, honestly. If I'd had this at 15 I'd have done exactly the same. But the homework wasn't testing whether they can produce 300 words. It was testing whether they understand it, and the essays told me nothing about that.

What I changed

  • Explain it to me, not to paper. Two-minute chats at the front while the others work. "What's in the program counter right now? What happens to it next?" You can't prompt your way through a follow-up question you didn't see coming.
  • Parsons problems. Give them the lines of a program shuffled up and they put them in order. Quick to mark, hard to fake, and they actually have to read the code.
  • Predict before you run. In the Python lessons: write down what this will print, then run it. The wrong predictions are where the learning happens.
  • Use the chatbot in the lesson, on purpose. We ask it to explain the cycle, then the class hunts for mistakes in its answer. They love catching it out. One student found it had mixed up the MAR and the MDR and was unbearable about it for a week. Tidy.

The video that does the heavy lifting

For the sixth formers doing the AI unit, I show them 3Blue1Brown before we touch any tool, because they should know it's maths and not magic:

Half of them come out of it saying "wait, it's just multiplying numbers?", which is exactly the reaction I want.

What I'd like from you lot

How are other teachers, lecturers and bootcamp folk handling this? I'm not interested in detection tools; the ones I've tried flagged my own model answer as AI-written. I'm interested in assessments that still mean something.

#ai#learning#students#python

Get the best of #ai_hype_check weekly

One email each Monday with the most useful posts. No spam; unsubscribe in one click.

Comments

9

Sign in to join the conversation.

Sign in to comment
Freya Collins@freya_bristol2 rep in AI Hype Check
'I hope this helps!' at the end of a homework essay 😂 Brilliant post Rhys. The hunt-the-mistake-in-the-chatbot lesson is genius tbh, can I nick it for the code club I help at?
4 net votes
Rhys Llewellyn@rhys_teaches_code#7 in AI Hype Check
Of course, iechyd da! Let me know how it goes.
4 net votes
Grace Wanjiru@grace_wanjiru1 rep in AI Hype Check
We do something close to your two minute chat at our bootcamp in Nairobi: every project ends with a 10 minute walkthrough where the student explains one function they wrote and changes it live. It catches copying of any kind, AI or friend, and it is kind, because nobody fails. They just see what they do not know yet.
4 net votes
Kuya Migs@kuya_migs_codes4 rep in AI Hype Check
Parsons problems are so underrated! I use them for my JS bootcamp, plus 'predict before you run'. The students who argue with me about their prediction learn the most.
9 net votes
Ryan Goh@ai_maxi_ryan#3 in AI Hype Check
or just teach them to use AI properly instead of fighting it?? they will use it at work anyway. teaching fetch decode execute in 2026 is like teaching horse riding 🤖
3 net votes
Rhys Llewellyn@rhys_teaches_code#7 in AI Hype Check
We do teach them to use it, that's the fourth bullet. But you can't check an answer about something you don't understand. That's the whole point of the course, really.
1 net votes
Karthikeyan S@karthik_embedded#7 in AI Hype Check
Fetch decode execute is how every pump controller in India works da. Somebody has to understand it. I am saving the horse riding comment for the day your startup needs firmware.
2 net votes
Dana Kowalski@dana_ml#1 in AI Hype Check
The detection tool flagging your own model answer is consistent with what others have measured: those tools have high false positive rates, especially on careful, formal writing. Redesigning the assessment is the right instinct. Your list is better than most institutional guidance I have read.
5 net votes
Rhys Llewellyn@rhys_teaches_code#7 in AI Hype Check
Diolch Dana, that genuinely means a lot coming from you. Forwarded your comment to our head of department, who keeps getting emails from detection tool salespeople.
3 net votes