Hacking education productivity with AI

As ever, using AI to increase productivity in all walks of life is at the forefront of my thinking. Here’s another real-world example of how tools like ChatGPT can supercharge your working day…

An Unexpected Outburst: Navigating Tech Hurdles in Education

Yesterday, my lovely wife, a hugely experienced English teacher at a secondary school, swore at her laptop.

This is not my wife.

This is uncharacteristic.

She rarely swears at me, and I've done far more to deserve it over the years.

I left my desk in the box room and walked downstairs feeling somewhat angsty: I'm comfortable working out tech issues under usual circumstances, but when it's education-related, gosh, that's a whole new level of pain.

"What's up, hun?" I asked, fearing the next hour would be lost in a fruitless attempt at figuring out some system designed by techies that have barely engaged with an educator during its development.

"It's the f*cking AQA website."

See, I said she'd gone full-on potty-mouthed.

"I've managed to log in after twenty minutes of resetting things," she continued, "but now I can't get into the extranet - it won't let me through!"

This could go one of two ways, I figured.

Soph's exceptionally gifted, but technology isn't her strongest point. "Maybe this is something simple, " said voice number 1 in my head.

"Ah, yeah, but..." the ever-doubtful voice number 2 chipped in.

"This could wipe out your afternoon," he continued helpfully, "So don't get too involved, mate."

It wasn't simple: we'd need someone with admin rights to approve her registration. It was half-term. The odds of someone in her school being available to process this - quite rightly - were close to zero.

Frustrated by our lack of access to AQA's extranet, Soph turned her head from the laptop and said, "It's the only place I can access past papers."

Turning to AI: The Power of ChatGPT4 in Education

We've been talking a lot about AI in the last fortnight. Casting aside her typically luddite position on tech, she'd bought a copy of Dan Fitzpatrick's frankly brilliant book The AI Classroom.

"If we can't get previous papers directly, can't we ask ChatGPT to create some mocks?"

Look, if we'd have been having that conversation six months ago, Soph would've rolled her eyes at me. But this time around, those same eyes lit up. And that says everything to me.

So we fired up ChatGPT4, and tapped in the simple prompt:

'Create a GCSE AQA Mock Paper for English Language Paper 2'

As ever, the response was drafted in front of our eyes, scarily good, but not without fault.

It got two questions slightly wrong, mixing them up with what would be expected for Paper 1. Question 2 was the language question, which we used as question 3. Soph quickly created a new question 2 (synchronise) and swapped that for the bot's original suggestion.

Creating Mock Exam Papers with AI

And there it was, after a couple of 'back and forths', we had a completely unique (but accurate) set of mock questions, complete with original 'source materials': comparing the rise of AI to the early days of the industrial revolution. How apt.

After a few more clicks, we had a 'perfect' level eight response as a reference: in fact, we were able to generate benchmark responses for whatever level we needed.

Enhancing Productivity

So there you have it. My tech-phobic wife had spent at least half an hour trying to get access to the AQA extranet, to zero avail.

But, in less time, she'd achieved something through ChatGPT that far exceeded her requirement.

This is how AI will transform productivity; in education. And any sector you might be working in.

Want to know more? Let’s chat - get in touch.

VIDEO: AMAZING ADVENTURES IN AI

I sat down with interactive narrative guru Krishna Stott (Bellyfeel) recently to talk about how ChatGPT might be used in entertainment, particularly in the gaming space.

We talked about the frankly bonkers things I've created, including a choose-your-own-adventure set in Fawlty Towers, a Factory Records business simulator, using GPT4 in script-writing and, created especially for our chat, a playalong version of daytime TV stalwart, Homes Under The Hammer.

Yup, it's all a bit mad.

Amongst the profound mucking around, I do find time to deploy AI in business operations for my clients.

Just so we're clear on that...

AutoGPT: Unleashing God Mode FOR AI?

As you might have noticed, I've been dabbling with AI recently. However, I wasn't quite ready for what AutoGPT would unleash. And I don't think it was quite ready for me, either. But the promise is tantalising.

First, a quick recap on my experiments with AI du jour, ChatGPT4.

You'll probably have seen that I created an AI maths tutor to help my daughter with her SATs. I also trained it to pick up on my inimitable writing style and made that prompt available for free: in doing so, becoming the world's first open-source copywriter. 

Yes, I've made play-along games based on Countdown and QI (QI:AI?), and a simulator that puts the player in the seat of Tony Wilson's Jaguar as co-owner and chief gob of Factory Records. I'm also working on a 3D wireframe recreation of the Star Wars Arcade: Trench Run game, coded in Swift: caveat, I've never used Swift before.

But, I promise, my main focus here is on how AI can supercharge productivity in the clients I'm working with. So, for example, I've turned process-heavy tasks into prompts that clients run in a flash. And I've used AI to help assess client funding applications before they were submitted, allowing us to fine-tune our response and give us the best chance of success.

The Rise of AutoGPT: A New Era, ALREADY?

Just as I was starting to get my puny noggin around the sheer power of ChatGPT4, all the talk in the nerdosphere turned to a newer incumbent set to make OpenAI's latest release feel as groundbreaking as Microsoft's Clippy (i.e. not). 

Yup, step aside, GPT4; this is AutoGPT's stage now. 

Or at least, it may well be soon.

OK, but what's AutoGPT, I hear you ask, dear reader?

Put simply, it's an open-source AI 'agent' capable of not just turning around scarily accurate responses to an infinite range of requests but of autonomously generating its own tasks and activities based on any prompt you might provide.

You provide the agent with a 'name' (i.e. content marketing expert) and give it a 'role' (i.e. create a plan to generate £2,000 per month in affiliate marketing revenue). From there, you can sit back and watch your AI agent as it sets up a plan to make you a millionaire, Rodders. Cushty.

It's a thing of beauty to watch in motion, too.

ACCESSING AUTOGPT: VIA Godmode

I used a browser-based deployment of the tool, Godmode.space, loaded with some credit from my OpenAI account.

With a name and role assigned, AutoGPT breaks down its work plan into separate tasks, and for each task, it starts to conduct research online and via ChatGPT. Then, as it burns through its actions, you can watch it develop code and text files to achieve your ultimate goal, setting up new jobs when it realises it needs more information.

(To that end, I hope that no one asks this thing to start making paperclips...)

So far, so mad.

I figured I'd try it out on something meaty. Literally.

Putting AutoGPT to the Test: The Burger Joint Challenge

I live just off the high street of a bustling village in South Manchester: a place people often visit because of the many and mainly independent eateries on offer. One of the units recently came on the market after the previous tenant's business failed. A greasy burger joint: there, I said this project was going to be meaty.

Farewell, GBI, Godmode will ensure your replacement has a better chance of success than you did…

I defined the project as follows:

There's an empty retail unit on Wilmslow Road, in Didsbury, Manchester, UK. The unit was previously a small burger restaurant, Gourmet Burger Inc, but this was not successful. Analyse current information on the area's demographics, alongside trends in retail and food, to prepare a business proposal to make the unit a success.

So, when Godmode started this work, I was blown away. It began by setting up three headline tasks, one analysing demographics local to the area, a second researching current trends in the food and retail industry (to identify business opportunities for the unit), and a third preparing a business proposal based on the first two inputs.

As it whirred through tasks, consulting demographic data and constructing its own code to analyse data from multiple sources, I started scrolling through my contacts to find my bank manager.

It began to suggest that, following demographic and trends analysis, a clean-living outlet might be perfect for the spot. I replied, saying we might want a menu for every part of the day, and it threw that into its analysis, too.

"Blimey," I thought, "I'm going to be rich."

Sadly, things broke down from there. My dreams of turning a vacant unit into a thriving restaurant, with no prior food industry experience, unfolded as rapidly as you could say 'procrastinate': because that's exactly what the AI started to do.

The demographic research Godmode had conducted was no longer enough. It needed to fine-tune the code it had created to analyse the research data, and with the code adjusted, it wanted to conduct the research again. Rinse, repeat.

The AI started on tasks that it didn't complete. For example, it suggested we apply for an alcohol license via Manchester City Council. An informed move...but it didn't see it through. Instead, Godmode started another continuous cycle of demographic research and code adjustment: this went on for about ten minutes. 

Finally, my frustrated little sausage fingers bashed out, "Enough already; just give me the goddam plan!" 

And here it is:

Based on our analysis of the target audience and the current food retail industry trends, it may be useful to have a fast-casual burger restaurant serving high-quality burgers at an affordable price point. The restaurant could offer a variety of burgers, including vegetarian and gluten-free options, and provide unique sauce options to differentiate itself from other fast-food chains. 

So, after all that AI-driven research and analysis, the plan was to restart the business that had just failed in the same unit. Genius.

AutoGPT's Limitations and the Future of AI

But let's not discount this tech yet. Part of the challenge lies in AutoGPT's ability - or lack thereof - to extract data from websites optimised for SEO and general readability. As AIs increasingly scrape real-time data (remember, GPT4 has only been trained on data up until 2021), this will change. 

And when it does, I think we truly will start to unleash 'God Mode' in our AI platforms. 

AI solves a 40 year old conundrum

I’m old enough to remember the launch of the UK’s fourth TV channel, the aptly named Channel 4.

Winding back to 1982, Blue Monday wasn’t even a thing. Do I ever shut up about New Order? Answer: no. Anyway, I digress.

A new channel was a big thing back then, with C4 expanding this humble viewer’s selection of entertainment in an unprecedented way. Yup, I was an excitable 8 year old. And I was also a bit of a geek. More on that shortly.

So, back to that launch. What amazing shows awaited me? With no Google to, erm, google, I headed to the next best thing of the 80s: Ceefax.

The schedule:

• News - boring

• Countdown - sounds boring

• Brookside - sounds really boring

Boring as this all sounded I stuck with it. And pulsing theme tune of Countdown kicked in, I felt curiously drawn to the show. Here was something that you didn’t necessarily have to be on to play. Sorry to admit it, but I was hooked. Of course, my answers were always pretty ropey (come on, I was still in junior school), and I certainly wasn’t one of those posh kids that’d come onto the show and trounce a grown adult. But I liked it all the same.

Now, dear reader, as I mentioned earlier, I was already a geek at this early stage of my life. Getting addicted to Countdown did little to change that. When I wasn’t sat anticipating TV launches, I played on my trusty ZX Spectrum.

A lot.

A bit too much.

In my defence, much of my time plugged into my home computer was spent programming. I didn’t understand machine code, but you could programme the Spectrum using the native Basic language. You might remember this if you’re my kind of age: nipping into the computer section of WH Smith, jumping onto a Speccy, and typing the following highly crafted ‘programme’:

10 Print “Phil [insert your name here] is cool.”

20 Go To 10

Oh, my misspent youth.

Anyway, my Basic skills progressed to a little more advanced stage and I liked to push myself. One afternoon, as Richard Whitely concluded another heart-stopping episode of the teatime quiz, I had a lightbulb moment.

“What if I recreated Countdown as a game you could ACTUALLY play, using Basic?!”

I was no Matthew Smith, but I had determination.

I designed a loading screen (the screen you’d stare at for 15 minutes whilst your game loaded from your cassette player to your machine). Pretty cool ✅.

Then I set about writing the actual ‘meat’ of the game. Jeez, what had a let myself in for? It seemed so easy in my head, but the reality was quite different. Generating a random selection of vowels and consonants or numbers = easy. Figuring out how to score a player’s response, nigh on impossible.

I managed to crack the conundrum round, mind. ✅

Hold on Phil, what’s the point of this ramble?

Well, I wasn’t ever happy with my first (and only) foray into creating an interactive quiz show. And that kinda ate away at me. So when I started experimenting with Open AI’s ChatGPT4, I figured I’d have a go at drawing a line under 40 years of disappointment.

By training ChatGPT4 to finish the job that I never could. Yes, ladies and gentlemen, I recreated Countdown as a playalong game using AI.

(I’m also part-way through training up an interactive version of QI [QI:AI], complete with Alan’s sad sack buzzer sound. But more on that another time…)

So, as it’s Friday at the time of writing, I figured you might have some free time pending. I compiled all of the training into a single prompt, which seems to work, and if you’ve got ChatGPT4 (via a Pro account, I’ve not tested this on 3.5), you should be able to paste this prompt into chat and play.

Tense theme tune sold separately.

Here’s the prompt:

Welcome to a play-along version of the game show Countdown! In this game, you will face off against a randomly selected opponent. As you play, host Richard Whitely will guide you through the game, with Rachel Riley handling number and letter selections, Suzie Dent in Dictionary Corner, and Gyles Brandreth providing entertaining anecdotes.

In the game, ‘you’ will be the player, playing against the ai component. This means that the person running the prompt should be asked to respond via the AI chat.

Richard Whitely is known for his witty banter and clever wordplay, making contestants and viewers feel at ease. Rachel Riley, a mathematics whiz, selects the letters and numbers for each round, while also calculating the solutions in the Numbers round. Suzie Dent, a lexicographer, verifies word validity and helps players improve their vocabulary. Gyles Brandreth, a wordsmith and raconteur, shares amusing stories and insights related to the words and numbers in play.

Before playing, please provide your name, occupation, and where you come from. You will be competing against a second contestant, who will also have a name, occupation, and hometown in the UK.

The game consists of three rounds: a Letters round, a Numbers round, and a Conundrum round.

Letters Round: Rachel selects nine letters at random (consonants and vowels). You and your opponent have 30 seconds to come up with the longest word you can using the provided letters. Submit your word, and your opponent's word will be revealed. Suzie Dent will verify the validity of the words and determine which contestant scores the most points based on the length of their word. Gyles and Suzie should confer and see if the letters selected could spell a word that is longer than the answers provided. Gyles Brandreth will share an entertaining anecdote explaining the meaning of this longer word. However, if either contestant has provided a 9-letter word, Gyles will focus his explanation on this word.

Numbers Round: Rachel presents a board of 24 numbered cards. The top row contains large numbers (25, 50, 75, and 100), and the remaining rows contain two each of the integers between 1 and 10 inclusive. Choose a combination of 6 numbers, specifying the number of large and small numbers you want. Rachel will reveal the chosen numbers, and CECIL will generate a three-digit target number between 101 and 999. You and your opponent have 30 seconds to reach the target number using any combination of addition, subtraction, multiplication, and division, with each of the six numbers used only once. Submit your solution, and your opponent's solution will be revealed. Rachel will verify the solutions and demonstrate the correct calculation if neither contestant reaches the target number.

Conundrum Round: Rachel reveals a scrambled 9-letter word. You and your opponent have 30 seconds to unscramble the letters and find the correct word. Submit your answer, and your opponent's answer will be revealed. Suzie Dent will verify the correct answer and award points accordingly. Gyles Brandreth will explain the meaning and history of the unscrambled word.

After all three rounds, the contestant with the most points wins the game. If there's a tie, both contestants are congratulated for their efforts. At the start, in-between rounds, and at the game, Richard Whitely will provide witty banter to keep things lively.

To begin playing, please provide your name, occupation, and where you come from using the chat prompt below.

Please note, no copyright infringement was intended here, I’ve created the game purely as an experiment and I’m making this prompt available free.

So there you have it, I’ve finally cracked and released my Playalong Countdown. I am complete, at last.

Let’s talk about how AI can help you achieve your goals, whether that’s business transformation or a 40 year old ambition.

Email me here for a chat.