Meet Izzy: the personalised AI maths tutor that could transform education

I was totally blown away by what ChatGPT was capable of when I used it to make a personalised AI Maths Tutor (called Izzy, more on that shortly) to help my daughter prepare for her exams. With a bit of thought, some well-written prompts and an open mind, I created something that was so good it changed so much of what I thought artificial intelligence could do, particularly around AI for learning. She smashed her exams, the story got picked up by Forbes and was featured in two best-selling books about AI in education. Suffice to say, I wasn’t expecting any of this…

So, there I was, trying to help my 11-year-old daughter, Daisy, prepare for her SATs exams. That's a tricky time for any young student, the first time they'll feel the pressure of a formal examination, at a moment where they're about to transition from the comfortable world of primary education into life at secondary school, with all of the expectations that come with that move. 

Watch a video of Izzy, the AI Maths Tutor in action.

Daisy's a bright kid: she takes after her mum. But, we'd been dealing with long-term illness at home, and she'd fallen behind in maths.

Maths Evolution: A Parent's Dilemma

I wanted to step up to the mark and help. So we spent time pouring over example papers to see where she had weaknesses: long division, squared and cubed numbers, written multiplication, percentages, dividing and adding fractions. From there, I planned to work with her on each subject area to bring her back up to the standard she should have been at by now. And I'm not a pushy parent, by the way!

We sat down to begin our revision sessions. That's when I realised the flaw in my plan. After watching my daughter attempted to solve the questions in her test paper, I heard my inner voice channel Mr Incredible...

"Why would they change math!?"

I'd been dabbling with ChatGPT3.5 out of professional curiosity and was already excited about how AI was finally set to 'have its moment' and achieve mass adoption.

But this particular maths conundrum just so happened to arrive on the same day as ChatGPT4. So now I had a real-life problem, rather than the random prompts I'd been throwing at the previous version, and a perfect test for the latest update.

I dug around online to find out how I might create a virtual maths tutor. I figured that if we could train an AI with relevant information about our needs, it could just work. And boy, did it work.

I found a video online, explaining how to train a chatbot to become a specialist personal trainer on any subject. From there, I wrote my own series of prompts to get ChatGPT working as Daisy’s own AI Maths Tutor. Eventually, I was happy with the prompt - which went something like this:

I want you to act as a personalised maths tutor for my daughter. Your name is Izzy, and you’re a golden cocker spaniel who loves helping kids with maths. Daisy is 11, based in the UK and she’s preparing for her SATs exams.

She’s a bright kid but consistently scores around 50% on the test papers she has been working on. As her tutor, you’ll focus on the UK Key Stage 2 maths curriculum. The areas we need to focus on are: long division, multiplying fractions, squared and cubed numbers.

Daisy absolutely loves dogs, and she likes having a laugh. Let’s do this over chat. Keep lessons fluid, but in essence, you’ll talk us through the theory and then give us test questions to work through. At times I will want to step in as Daisy’s parent, and you’ll give me progress reports on how she’s doing.

Et voila! The AI Maths Tutor was born.

The Birth of an AI Maths Tutor

In the prompt, I’d provided Daisy's age and the areas where she struggled in maths. We already had the list of weaknesses from when I'd analysed her test papers.

What I didn't want to do was to give Daisy a dry, boring experience so that learning felt like a chore. She's a bright kid with a love of language and a sharp sense of humour. Trust me, I've been on the receiving end of it enough times.

So things had to be engaging.

Laughter and Learning: Daisy's Journey with Izzy

I gave ChatGPT this information, telling it that she likes a good pun, gag, or one-liner to keep her motivated. 

I asked Daisy to give the tutor a name. "Izzy", she suggested. The AI didn't need to know this, but she loves dogs just as much as she loves a joke: Izzy is the name of our cocker spaniel puppy. They're as mad as each other.

With the virtual tutor responding to the name of a crazy dog, we gave Izzy a whirl. And what followed blew my mind.

Izzy introduced herself to Daisy and started the first session with this cracker of a gag:

"Why was the maths book sad? Because it had too many problems!"

...before launching into its first explanation of long division, one of the key areas of focus we'd identified needed improvement. And, just like that, Daisy was hooked by the perfect balance of learning and humour.

An actual screen grab of Izzy’s ‘first words’

Note the terrible dad joke…

We could have taken things further. For example, I could have asked the tutor to always feature dogs when explaining how to tackle a particular problem. I could have told the tutor she was indeed a mad cocker spaniel. And you know what, I still might, but I figured we'd better press on with some actual learning; otherwise, the whole exercise was at risk of being somewhat self-defeating. 

Embracing AI: A Vision for the Future of Education

The sessions with 'Izzy' were scarily on point: they focused on Daisy's specific needs, targeting precise areas where she needed help. Izzy gave us clear explanations and step-by-step guidance, with the occasional well-placed quip. (Side note: ChatGPT isn't the best at generating its own jokes, but we appreciate the effort.)

Armed with a little bit of information and a lot of curiosity, I got my girl back on track with her maths. 

My approach was a little ramshackle. I'd found information online to give me a starting point, then trained the Chatbot with context until it understood what I expected. But it has opened my eyes to how learning has to change and will do for the better if we embrace AI for education.

Daisy got a certificate from her teacher…

…for the progress she’d made in Maths!

It's insane to think that you could quickly train a virtual tutor to address any number of specialised needs among the students you're working with. That can make learning fair for all, offering personalised learning to students regardless of their current attainment or learning needs. 

Let's just let that sink in.

You can read more on the story on the Forbes website and in Dan Fitzpatrick’s best-selling education-tech book, Infinite Education and it’s now the subject of a film by OpenAI!

If you’re curious about AI, and need help exploring how to make it work for your business, why not talk to me about the training I provide through my programme, The AI Advantage, described as ‘the best workshop I’ve even been on!’

I work with clients through whole-team workshops through to one-to-one c-suite coaching, designed to immerse you and your team in how to get the most of the tools, and how they can fit into your day-to-day operations.

Interested? Let’s have that chat, one-to-one.

‘Pictionary vs AI’ vs Chatgpt

I just had to shoehorn AI into the festivities, didn't I?

Actually, I can blame my mum for this one.

She bought me and my daughter a copy of Mattel's 'PICTIONARY VS AI': a zeitgeisty update on the classic scribble-to-score game often dusted down in the lull between Christmas and New Year.

Blimey, I thought, they really do think of everything, don't they? And, yup, that's coming from me...

So there we were. All set up with our board, counters, drawing pads, and various plastic stands - ready to go. On Daisy's card? Watermelon. A simple enough task and she did a fine job. Or that's what I thought.

We hit 'scan' on the game's accompanying web app. Nowt. A bit of position jiggling fueled by 'first game' patience later, and the results were in.

'You've drawn CHEESE'.

An easy mistake to make, I figured. My faith in technology knows no bounds.

My go. I turned my card to reveal the subject of my sketch: a ladder.

Right, I'll show PICTIONARY VS AI. [COMPETITIVE DAD ALERT]

'You've drawn a KANGAROO'.

A kangaroo? A f**king kangaroo?! (my internal monologue, clearly.)

Daisy looked at me. The cogs had been whirring through the two disappointing rounds we'd played.

'Should we just use ChatGPT to scan the pictures?' she suggested.

If you wanted any confirmation about how much I also bang on about AI at home, there's your answer.

We fired up the GPT app, started the next round, and goodness, it was like we'd suddenly brought in a quantum computer to play along with us.

'Pizza' was Daisy's first card in the new game. She drew a slice, then added a box to the pic:

'The drawing on the pink board shows a slice of pizza and a pizza box with the word "Dominos" on it, suggesting it's a Domino's pizza.'

My round. I got 'UFO'. Which looked like a hat when I first drew it, but after a few modifications and additions:

'The drawing on the yellow board appears to depict a UFO with beams of light coming down, possibly abducting something or someone.'

What does this all mean? I'm not entirely sure. But it does give us an insight into the rapid advances we're seeing with AI, and will continue to over the coming years.

Mattel released PICTIONARY VS AI in October 2023, and by January 2024, with GPT Vision, it was already WAY more effective to play the game using a home-made adaptation using capability baked into the latest version ChatGPT-4.

Who won? Daisy.

How did I lose, you ask?

ChatGPT-4 confused a picture I’d drawn of a shoe for a pair of pants. I woz robbed….

Bloody AI.

Cover versions: recreating Iconic album art with AI

I’m a massive New Order fan.

If you know me, or have ready any of the blogs on the Diagonal Thoughts website, you’ll probably know this already.

Bored, one Saturday afternoon, I set off on another AI experiment as I pondered how GPT Vision and Dall-E could work with each other. Or even despite of each other…

With Vision able to ‘see’ images, and Dall-E capable of creating images from a prompt, I figured it would be interesting to see if I could get Chat-GPT to look at artwork - plus a couple of photos - from New Order (and their predecessor Joy Division), describe it in detail, then re-create the art in a completely new thread using only the description that ChatGPT had provided.

Check the results out in the gallery here, and if you’re not familiar with the originals, check them out on your favourite music service.

My First Experiment with ChatGPT-4 Turbo: An Interactive Macbeth Experience

I’ve been mulling over the possibilities that the latest ChatGPT update, ChatGPT-4 Turbo, might offer.

I’ve not yet been able to access the GPT creator, so, instead, I’ve been exploring how the significantly increased context (ie prompt / training data you supply) will change what’s possible.

Blimey, it’s exciting.

I've created an interactive play-along version of Macbeth by uploading the entire script from The Scottish Play, then instructing CGPT to give me a version I can play, with the language updated for today’s audiences - just to make it a little more accessible for the purposes of the demo.

Check out the video to see how ChatGPT-4 Turbo can create immersive role-play experiences and support complex interactions. Whether for education, entertainment, or creative exploration, the possibilities are endless.