Leon Furze

I used Gen AI to help write Practical AI Strategies, but not how you might think…

Practical AI Strategies includes an entire section on GenAI policy and assessment. It is available from Amba Press

It’s release day! Practical AI Strategies is officially out today from Amba Press. You can also get the print and ebook copies at Amazon, Barnes and Noble, and Booktopia.

When I spoke to my publisher about putting together Practical AI Strategies, it was with a certain amount of hesitancy. I felt it would be important to provide teachers with a resource as soon as possible at the start of Term 1 2024 to prepare for Generative AI. But I also had to juggle the workload of my PhD and my consulting work with schools across the country. I also didn’t actually pick up the phone until November, so… I’d left it a bit late.

But I had the advantage that I’d been writing about Generative AI since before ChatGPT was even released, and had spent a lot of time working with the technology since then. So, I sat down and brainstormed a few ways that GenAI could assist in the creation of the book.

Obviously, I could have hammered out the entire book using ChatGPT, just like thousands of books on Amazon, and many of the current texts available educators. But honestly, I don’t use ChatGPT all that much for writing. It doesn’t sound like me, and it’s cliched turns of phrase and overuse of words like “delve” and “ever shifting landscape” frankly set my teeth on edge. Trained as it is on the corpus of the internet, it also tends towards technodeterministic, overly positive language.

But I could think of a few ways to use ChatGPT, Claude, and other models in getting this book together. I wrote about how I used AI in the “about this book” section. Here it is in full:

This is part disclaimer, part explainer and part guilty confession. ChatGPT was used to create some of this book. Shock! Horror! Disgust! But not in the ways you might think…

Since OpenAI kindly dropped ChatGPT on us from a great height in November 2022, there has been a lot of angst about cheating and plagiarism, which I’ll write about in Part 2. In the interest of transparency, I think that, for now, it’s helpful if people explain how they have used generative AI (GenAI). In the future, the technology will be so ubiquitous we probably won’t bother (imagine a disclaimer like this for I have used spellcheck in Word…).

I haven’t relied on ChatGPT to write the content of this book or my blog posts, but that’s mostly because I enjoy writing. But I do use ChatGPT and other GenAI tools daily, and that includes in the construction of this book. Here are some of the ways that GenAI was used in the process:

• I created a simple piece of software using ChatGPT that used a programming language called Python to ‘scrape’ my blog for articles in the AI category. The code took the content of those posts, stripped the website code and images, and copied them into individual Word documents. I then manually reorganised and edited those documents as the basis for many of the sections in this book.
• I used another piece of code – also written in ChatGPT – to compile the final sections into the draft manuscript. It saved me about six and a half minutes of copying and pasting.
• For the original articles and some of the new and updated sections (including the final sections), I recorded the drafts in a voice memo and used Otter.ai to transcribe them, and GPT-4 to remove transcription errors and repetitions.
• I used ChatGPT and occasionally Claude for working out the tangles in some of my clunky and overwritten paragraphs. I tend to write how I speak, with lots of punctuation and enthusiasm, so most of those prompts were phrases like: Take the following paragraph, and make sure it actually makes sense.
• I have used image generation in various places. The AI Iceberg on page XX is a Canva graphic, but most others were created in DALL·E 3 and Adobe Firefly. Even for those mostly administrative tasks, I still feel a little uneasy writing ‘I used ChatGPT while creating this book’. I think that’s something we’re all going to have to get used to.

Enjoy the article and want to stay up to date? Join the mailing list:

The GenAI programmer

I’m no software developer, and my heavy-handed use of ChatGPT to scrape my own website might horrify some. But it presented a solution to a problem which would have otherwise taken me a long time to solve, manually copying and pasting my earlier posts. One of the reasons I continue to write for my own website, and avoid platforms like Substack, is so that I have full control over how I use my content, and this time around it came in very handy.

In fact, I’ve always played around with code. When I was a teenager, I learned HTML and JavaScript and built websites. As a teacher, I taught basic Python in a STEM class using Raspberry Pi’s and other software and hardware. But I’ve never dedicated the time to really learning to code.

Ironically, since using apps like ChatGPT to do some of this work, I’ve started learning myself and have begun courses for Python, SQL, and HTML. I found that having just enough knowledge to pull together code in ChatGPT is fine, but I want to know more, and I want to be able to do it for myself.

Voice to text to text to text

The other method I mentioned – using voice memos in combination with GenAI – is probably how I use the technology most. I typically record directly into Otter.ai, or I use the Voice Memos app on my iPhone and later upload the audio for transcription.

The Voice Memos app is more accessible from home where I have poor internet access beyond my satellite. I can also record directly into Voice Memos if I’m out for a run/walk, including straight to an Apple Watch if I don’t have my phone. Once I have the transcript in Otter, I then take it to another platform.

Whilst Otter provides a (GPT-3.5 powered) AI summary, it’s generally not what I’m looking for. Instead, I tend to paste the entire transcript into ChatGPT Plus or Copilot with a prompt like the following:

Correct any transcription errors, repetitions, or unusual phrases in this transcript. DO NOT change the transcript beyond making minor transcriptions. Structure the resulting edited transcript in clear paragraphs. <copy/paste transcript>

Use markdown formatting to produce a summary of the following transcript. Make clear headings, subheadings and lists. Capture EVERYTHING from the transcript, even seemingly inconsequential comments. Create suitable sections for the summary. <copy/paste transcript>

Turn the following transcript into an outline for a book chapter. Identify if there are any points where more evidence is needed or specific websites or resources are mentioned. <copy/paste transcript>

So, while I didn’t use ChatGPT to generate swathes of enthusiastic text about delving into the next great wave of technology to enhance humanity in the ever evolving landscape of education, it was useful for a variety of other reasons. I hope you’ve enjoyed this “behind the scenes” and that you’ve learned a few tips for using ChatGPT in your own work.

If you’d like to get in touch to discuss GenAI, advisory, or professional learning, please contact me via the form below: