The artificial intelligence (AI) powered chat tools, like ChatGPT and Google Bard instantly got me hooked because I truly believe that these tools will definitely stay with us in the long run and I'm intrigued to see how ELT teachers, material writers, and teacher educators are going to use them in the future. I've been sharing my impressions and experiences on LinkedIn, and thought I'd compile them into one blog post.
As someone working in ELT, the first question that came to my mind was "How can these tools make my work easier, quicker, more efficient?" Of course, at the very beginning, I was also carried away by the immediate panic response wave, thinking that these LLMs (large language models) are going to take my job, destroy my future, etc. But the more I play around with them, the more I see that they are essentially are what they are - LLMs. They operate and manipulate language, which is of course one of the main working tools of language teachers, but they are (as of yet) not capable of much if you completely take out the human element of them.
That's why I love two images that I saw at the beginning of the AI-craze. One applied Dunning Kruger effect to ChatGPT, showing on a diagram that it's a perfectly normal first response to fall into panic, but the more we get to know what these tools can do, the more we can see their worth and their errors.
The other one shows the importance of the human element in the entire process. (For now) us, humans are writing and giving the prompts to these models, and our prompts define what the outcome will be. And even after getting the outcome, we're the ones who can decide whether those responses are factually correct, acceptable, to our liking, etc. Of course, it's perfectly possible to end up with responses that we don't wish to edit further, but my personal view on this is that I'm going to remain an "Agnes" (check out my webinar recording for IATEFL BESIG for further details).
Putting a lesson together with ChatGPT (LinkedIn link)
I really wanted to do a whole lesson with ChatGPT and AI, and today was finally the day! I'd love to hear your thoughts on it!
📓Topic of the lesson: Elevator pitches
📝Skills involved: reading, writing
📌Steps:
1. First, students read a short list of tips on how to prepare elevator pitches (created by ai), based on which they write their own versions.
2. I then prompt chatgpt to prepare the same thing.
3. Next, I ask my students to compare the two versions, and answer two questions: Which parts about your work did chatgpt get wrong? Which phrases would you like to copy over to your version?
4. Then we start focusing on the differences between written and spoken language. Chatgpt created responses in a written style, but at a conference (which is the context), we'd need more of a spoken style. So, we decide to cut out parts that make the pitch too wordy and not conversational enough.
4. Once the final versions are done, I copy them over into Twee, and select the "fill in the gap" activity. I choose the words I want to replace with gaps, and a short reinforcing controlled practice activity is done.
See the steps in the pdf below:
"Detecting" ChatGPT in student writing (my blog post)
I totally agree with those, including Jason Gulya and Leon Furze, who say that AI detection is not the way to deal with this new phenomenon. Instead of trying to catch them and find them guilty (which actually quite hard and unreliable), educators need to create meaningful assignments and let students integrate AI into their work. I also love Michelle Denis's design for AI assessment guidelines:
Although assessment is not such a big issue in general ELT classes, I also work in higher edication and teach EAP classes from time to time, where essay writing is one of the staple forms of assessment, and it's clear as day that AI tools can be a huge disruption in this area. This is where I find Philippa Hardman's posts highly valuable! One of my favourites:
My first steps with Google Bard (my blog post)
Since the main advantage of Google Bard is its connection to the web, I wanted to see how it can do more than Chat GPT. These are my findings so far:
1. First, I asked it to create a lesson plan with some recent info included. Here's the link to the chat:
💚 Things I liked:
- the lesson stages were logical
- the suggested activities and questions were totally average, but not terrible
- it included a differentiation section at the end
- the learning objectives laid out at the beginning got corresponding assessment points
- instant export to Google Docs and Gmail
- it did include recent info
- you can further tweak the given response, and I really like the suggested options (see attached image)
❌ Things that are not so good:
- there were no sources given for the news snippets, even if I prompted it further
- though I asked for a B2 lesson, I don’t think it understood what I wanted (maybe finetuning the prompt could help)
2. Then, I asked it to write a short essay. The prompt was: "Write me a 200-word essay on why and how are our summers going to get hotter in the future. Use recent information as source, and give credit to these sources, too"
The good thing: It did use and give sources at the end which are real (not in-text though)
The meh thing: I asked for an essay and it gave me notes
The bad thing: As soon as I ask for some regeneration, the previous response disappears :( [UPDATE: I found them! You need to click on the "history" icon in the top right corner next to help and settings]
* BUT *
I asked for a modified response - I wanted it to be more professional and Bard indeed produced more complex sentences with nicer linking words and fewer bullet points. The response still sounds more like a blog post than an essay but I'm sure it can be tweaked further. What I really liked is that at the end it summarized the modifications that have been made.