About this series
Join Leon Furze as he navigates the ever-evolving world of artificial intelligence, examining its origins and significance within education. Learn how AI can support personalised learning, encourage student engagement, and foster the development of future-ready learners. Unearth the potential of AI in your school and understand its capacity to transform teaching and learning experiences.
Key takeaways include: the history and significance of AI in education; the potential for AI to support personalised learning and student engagement and strategies for implementing AI in your school to enhance teaching and learning experiences
Book an upcoming session with ACMI's Education team today
Transcript
Zoe McDonald:
Welcome everyone. I'd just like to begin by acknowledging the traditional owners, the Wurundjeri and Boonwurrung people of the Kulin Nation on whose lands we meet, share, and work here at ACMI, in the center of Naarm in Melbourne. I'd like to pay my respects to elders past, present, and emerging, and also acknowledge First Nations participants who might be joining us today.
My name's Zoe. I'm a producer of school programs here at ACMI, and it's my great pleasure to welcome you back for our second session in our series on demystifying AI. Today we're talking about using ChatGPT in education. If you missed the last session rather, you can catch up on the recording, which will have been emailed to you last week, along with any of the answers that we didn't get through to your questions from last session. And we'll do the same with the recording today. As with last week, we will have some time for Q&A at the end. We don't have a lot of time, so please try and get your questions into the Q&A and you can also upvote by clicking on a question that might be similar to one you have. So I'd like to welcome back our AI expert, author, speaker, and consultant Leon Furze. Welcome, Leon.
Leon Furze:
Thank you very much, Zoe. Good to be back. Okay, I'm just going to switch my screen sharing on. All right, well welcome back if this is your second session and hello, if this is the first time. As Zoe mentioned, do catch up on that first session which gave an overview of artificial intelligence and programs like ChatGPT. I'm just going to dive straight in to the session today, starting with a recap on session one, very briefly just going over what we touched on in that first session, but then quickly getting into slightly more practical approach this time around, and we'll be looking at a few ways that educators can use ChatGPT in their own workflow. We're focusing on ChatGPT. Obviously as I mentioned last week, last time, sorry, there are many other applications of artificial intelligence and we will actually cover some of those other applications in the following two sessions, when we start looking at image generation, audio generation and video and even 3D asset generation.
But this time around we're going to focus on ChatGPT because it's the easiest program to use really, it's got a very low point of entry, it's very simplistic to use and it's free to sign up for the free version. So hopefully what you'll take away from this session are some ideas for how to use prompts to build into your own workflow as an educator. And it doesn't really matter whether you're in K to 12 or even in tertiary, the kinds of prompts and techniques that we'll discuss are applicable pretty much across the board.
So just to recap on the content of session one, and the point of this whole series is demystifying AI, because AI isn't magic, no matter what some of the companies behind the AI models might have you believe. It's really important, I think, in education that we start to break down some of the barriers and some of the mythical and magical language that surrounds artificial intelligence. I don't think it's particularly helpful to speak about AI in that way because at the end of the day, it is a technology built by humans. There are, as we talked about in the first session, a few ethical concerns with AI. And when we start to see AI as magic, it puts it as a level of removed away from some of those ethical concerns. And I think really in education, we need to be as across those ethical concerns as possible.
ChatGPT is really the crux of this session and it is a language model developed by OpenAI. Now, last time around I used this analogy to talk about ChatGPT. So a very quick recap on this analogy because I think again, it's useful to have a technical grasp, without getting into too much detail about how these models work. And there's also been some interesting news since the last time we ran the first session, about the dataset component in particular, which I'll touch on in a moment. So just go back to the analogy, that data set is the huge thing under the waterline, which really powers the language model. That data set is scraped from many, many parts of the internet, mostly open access parts of the internet, so blogs, media sites, social media and so on. That powers the language model, which is the algorithm that crunches through all of that data and creates the predictive model. So sort of like a supercharged predictive text, that's a very simplistic way of looking at it. But really what we're doing is we are creating some kind of meaning from that huge pile of text data.
And then finally we've got ChatGPT, the little snowman sitting on the top there, as a refinement trained to act like a conversational chatbot. Now even in the brief window of time since our last session, there have been some developments along these lines and in terms of data sets and things like that, which I'll just flag now because it's always good to keep up to date even though this industry moves very, very quickly. So the first is an article from the Washington Post. I share a lot of these things on my LinkedIn profile. So if you are interested in following up, I've shared these links earlier this morning. But the Washington Post article spoke about an issue which I think is particularly relevant to education, which is the dataset that powers something like GPT is really quite a sort of an average example of all of the knowledge that's out there in the world.
And that's because it relies on all of that open source knowledge. So it doesn't have any knowledge in the dataset that's behind a paywall or that's behind in the ivory tower of academia where we find this research sometimes, doesn't have any proprietary information. And what we're seeing happen now is a lot of organizations in education, so I'm talking about big publishers like Pearson, I'm talking about companies like Duolingo, Khan Academy, Coursera, Canvas. They're using these big foundation models like GPT-4, which are trained on the kind of average data across the internet, but then they're applying their own proprietary data into training their own fine-tuned models, so their own little snowman essentially. And what this means is that we may end up in a situation where companies that have the most data are able to produce the most effective education chatbots or AI tutors, whatever you want to call them. Alternatives to that might be companies like OpenAI and Google purchasing data from those organizations and then building them into their foundation models. They haven't done that yet, but that's a possibility.
Or people open sourcing or making their knowledge open access. So we know a lot of knowledge gets open access, but we know that in an academic context for people to publish open access journals, generally costs the universities or the individuals publishing money as well. So there's a lot of complexities in pulling all of this data into these chatbots, which I think is something we've got to be prepared to talk about in education. So that's one aspect. Another interesting thing that's popped up in the news in the last couple of weeks is that the social media site, Reddit was a huge source of training data for GPT models.
Now, we haven't got any confirmation from OpenAI about exactly what it went into GPT for, but it's a safe bet that there's a big chunk of Reddit in there. And Reddit's actually just changed its pricing model and a lot of Reddit users are now protesting and locking down their Reddits and subreddits, their threads, putting them into private mode essentially. So what we've got now is we've got an economic imperative, we've got Reddits having to make money, we've got Reddit blocking these AI models from scraping all of that data. We've got users protesting that because they don't want to pay a service that they haven't paid for before, and knowledge, again getting put behind walls. So it's going to be really interesting, I think, to watch how this unfolds over the next few months and years as we're working with these models.
What I want to talk about today though is this idea of AI augmenting but not de-skilling teachers. And just in the way that the big wealth of data in that ChatGPT dataset is fairly generic and sometimes prone to errors, just like I spoke about in the last session, when we're using them, we need to be really mindful that we're using these technologies in ways which add to or increase our own creative capacities, which allow us to improvise on the fly or which augment our capacity for planning and collaborating very quickly with one another, eliminating redundant tasks, not just making redundant tasks more efficient essentially. So there is a bit of a narrative creeping out at the moment about AI replacing teachers. I don't think that's going to happen anytime soon. Even with AI chatbots like the ones I mentioned from Khan Academy and Duolingo, they still require a certain level of supervision.
They need somebody with expertise to help guide the students. A lot of them increasingly, they're very fun, they're very attractive to young people, they're deliberately marketed that way. So we still need to be there with our critical eye on things, making sure that these applications are used properly. I've no doubt that these applications will be used in education, so I don't think it's a case of just turning a blind eye to this and saying, "Oh, it won't happen, don't worry about it." I think we need to adapt, but I don't think it's replacing us any time soon. There's been a few suggestions in the media and around a couple of conferences recently, that AI could and should be used for marking student work or writing reports and things like that. And frankly, I think that's a bit pointless. I think one of the purposes of education and a teacher in the room with the students should be the teacher giving the students feedback.
We might need to change how we do that or the way that we assess the kind of work that we're getting. But if students are using AI to write their work and we're using AI to mark it, then probably nobody's learning anything along the way. When it comes to writing reports, I spoke at a conference recently and gave the scenario where we know that parents are very busy, many of us are parents, I'm a parent myself. We know that sometimes if a school sends home a lengthy report, then it might not all get read. And I can envision a few future where teachers again are using AI to generate reports, which parents are using AI to generate summaries of whilst they're on the drive to work. And again, nothing is really happening.
So we've got to go beyond those kinds of uses. We've got to use the technology in ways which are extending the normal way that we are doing things as teachers, because we're the professionals, we've got the expertise, we know how to do the job of teaching, we just use these tools to help us do that a little bit more easily and a little bit less friction there.
So I'm going to talk about three areas that I think we can use these technologies responsibly as teachers. I'm not talking about using them with students today. So this is purely about you using them as educators, whatever area of education you work in. And I'm going to go through three prompts with each section and then give an example. Now these prompts and examples, I've been running a similar session to this over the last few months that's been very popular with schools. And the reason I think it's gone down so well is they've quite simple prompts, and again, there's no magic behind these prompts, but you can adapt them in any way you see fit. So what I would encourage you to do after this session is grab a hold of the recording when that comes out and then just pause on the slides. You can copy the prompts in verbatim, but it would be even better if you adapted them for your own use and in your own context, and just straight away, start experimenting with what you can do with ChatGPT-3.5.
I can see there's a question in the chat which we'll save all of the questions to the end, about the paid version. The paid version does make the output slightly improved, but for most of the purposes, GPT-3.5 in the free version will do what you need it to do, because you're always going to have to go over and fact check and probably build out whatever you get from ChatGPT. So let's take a look at those areas now. The first one is planning, and we've got three prompts on the screen. I'm going to just go through them all one at a time and then explain sort of why and how they work the way that they do. So the first one lesson on this topic should meet the following outcomes, we copy and paste in the outcomes from curriculum documents, suggest learning intentions written in student-friendly language for an introductory lesson which covers whatever.
So one thing you'll notice a bit of a thread running through all of these prompts is I do a lot of copying and pasting from other sources in the prompt itself. That little chat bar suggests short prompts, but you can actually put around about two to 3,000 words worth of text into a ChatGPT prompt, which really opens up a lot of possibilities for when we're working with curriculum documents. So for example, you could take ACARA or Victorian standards from the curriculum, you could take achievement standards or performance descriptors, key knowledge and skills from the VCE. You can take assessment standards from HSC, any kind of area that you're teaching in, and we can drop that in as part of a prompt, like this one. This will give us learning intentions in student friendly language. You could adapt this prompt and ask for essential questions.
You could ask for a series of questions that might be used throughout a topic. You could ask for assessment tasks and assessment items, or even a suggested outline for an eight-week unit of work for a given year level. The more context you put into the prompt, the better the result will be when you get it out again. The second is a little bit more involved, role play. You're a teacher in our faculty, we are currently in a meeting to discuss a unit of work on this topic for this year level, you are knowledgeable but highly critical and a little cynical. Your role is to critique and question the unit plan and we will type our responses. Don't provide ours, only yours, here is the unit plan. So again, we're giving it a unit plan, but we're giving it a lot more in this prompt.
We're telling ChatGPT that it's a role play. We are saying this is the role that you are playing. You are a teacher. We're giving it context, faculty, year level, topic. We're giving it a persona, knowledgeable, cynical, critical, and then a job, critique and question the unit plan. The reason all of that works is because of that huge dataset, a lot of data in there which represents conversational language between people. A lot of data in there which is again, scraped from those social media sites. Also, a lot of data labeled with tone and the kind of language that's in there. So if you say critical and cynical, it's going to give you slightly harder edged responses than you might get if you just asked it to ask questions about a unit, for example. The final one there, suggest places where the lesson might bottleneck or become less engaging and suggest alternative approaches for the flow of the lesson, we give it a lesson plan.
It's just going to look again, at ways to sharpen up a lesson plan, to refine that lesson plan. Always remember it's not magic, it's not thinking, but there is a lot of data in that dataset that it can build a probabilistic response from, which includes data on good pedagogy and data on interesting lessons and probably chat threads from Twitter between teachers. So all of this kind of data that's in there allows a model like GPT to predict a pretty reasonable assimilation of a, approximation, sorry, of a human-like response. It's not human, not thinking, but it does a pretty good job of pretending. This is an example of that role play prompt, with the 10 critical questions of a unit of work. So what I've got here is a year 10 unit of work on biofuels. I just took again, an open access unit of work on biofuels from online and pasted a few chunks of it into ChatGPT and fed it that role play prompt.
So just take a look at a few of the questions here. So can you explain the reasoning behind only covering the case study of one specific biofuel production? How will this relate to the broader concept? So you can see they're quite critical of the unit and of the planning. What we have though is perhaps an extra perspective beyond our faculty group that we're working with. So I used to teach in a regional school, a few of the faculties were quite small, you might be in a similar situation, you might be a faculty of one and not have anyone to bounce ideas around with. You need to explore ways that you can use something like ChatGPT in participation to bounce a few ideas around, always keeping that critical eye on the responses. So those are a few examples of planning with ChatGPT.
What you could do to adjust those for your context is just have a think about the kind of planning that you do in a normal teaching year. Whether you do all of your planning at the end of the year in a staff week, or maybe you have a few staff professional learning days over the course of the year, or do you plan in your free periods? Do you get no time for planning? I know that if we've got any primary school teachers in the audience, you get very limited time for actual planning and preparation in between all of your teaching. And think of ways that you can take the jobs that you do in that time and offload some of that work onto ChatGPT. So are you creating rubrics? Are you creating lesson plans? Are you creating worksheets? Are you creating units of work or scope and sequence documents? Any of that can be sped up a lot by feeding information like curriculum documents into ChatGPT and developing a few resources in that way.
The next area that we're going to look at is this personalizing idea. And the reason I've put this one into this session is because using ChatGPT for personalizing is becoming a really sort of hot topic. One thing that I will point out at this stage is that we're finally starting to get statewide policies around generative AI. So New South Wales have come out with their policy, South Australia SACE have come out with their policy, Queensland QCAA have come out with theirs. In Victoria, there's a community of practice running at the moment through DET, which is cross-sectoral. So things are happening, we're getting a little bit more advice. And one note of caution, resources for students, we never put any identifying information into ChatGPT. That being said, there are many ways that we could use this AI model to personalize or differentiate work.
I know from experience administering parts of the curriculum side of NCCD, that those processes take a lot of time. So using them to create parts of goals or ILPs, PLPs and so on, is a great way of making those processes a bit tighter. So the first one, reproduce this text at a level six Flesch Kincaid grade level. Keep the original tone, style, ideas and structure, copy and paste in the text. So level six Flesch Kincaid is just a reading level which gives you an average readability of a US grade six student. And what we could do for this is if you've got a conversation article or you've got an academic journal article, a worksheet, something from a textbook and a student needs it to be brought down a level or two because they have literacy issues or maybe you just want to speed up the process of the lesson a little bit and you don't want to wade through a really lengthy text, then you could use this.
A student could also use a similar prompt like this themselves. So if they're 13 or over, according to the ChatGPT terms and conditions, they can use it themselves. At this stage I'm generally advising schools not to charge in and make students sign up for ChatGPT or anything like that. I think your school's got to have pretty clear and solid policies around the use of generative AI before you dive into that extent. But you could certainly furnish your students with these kinds of prompts and they could experiment with them themselves if they wanted to. Create a list of five alternative assessment tasks for a student who struggles with traditional assessments. These tasks should aim to assess the student's knowledge and understanding in a way that better suits their strengths and learning style. So we're creating a personalized assessment pathway for a student.
Maybe we've got a student who has anxiety and therefore doesn't do well under exam conditions, so we need to do an adjusted examination. Maybe we've got a student who's not very confident with public speaking and we've got an oral exam, we could use a prompt like this to generate a few alternate pathways through that exam. Or maybe you have a student with a disability, you may have a student who is autistic, you may have a student who is dyslexic and needs modifications for a given reason. And again, using this to suggest a few ways of modifying that task and then working with the student and determining which is the most appropriate for them.
The last one with my kind of NCCD hat on, create a personalized learning plan for students based on their strengths and interests. We don't need to give any identifying information there, we can just copy and paste in a dot point list of strength and interests with no names or anything. Outline the goals, activities and assessments that will support the students learning and development. So the example that I'm going to give now is similar to that. This comes from one of my earlier blog posts. So if you go to leonfurze.com/blog, you'll see all of the blog posts there. The most recent one is about those state policies, if you're interested in that.
But in a much earlier blog post I went through this whole process. You can see in the prompt here it says using the above. So earlier in that thread I'd taken a piece of research about inclusive education, I'd taken out the bits that I was the most interested in, dropped them into ChatGPT and said, summarize this research for me and then come up with three practical ways I could use this in my year seven classroom. Building on that thread a little bit, and I'll talk about building on chat threads in a moment, I've ended up with this prompt, using all of the above, all of that information about inclusive education create three goals from the student's perspective.
And what I'll come up with is a student written or student style, first person goals. They probably would still need a bit of work. They're a little bit too wordy for me. I think if I was a teacher looking at those sitting down with a students, I would say, "Oh, they're not exactly right." But this certainly gives me a prompt to sit down with the student, maybe the student and their parents in a parent support group meeting or an NCCD data collection meeting, and work through those goals alongside the students and the parents and with ChatGPT there in front of us. So one thing that ChatGPT is really good at in terms of personalizing resources, is taking a lot of information in, synthesizing a lot of information and then giving you some kind of output. So taking in all the information about a student's interests, their strengths, their limitations, and then turning that into a cohesive set of goals is a really good use of the technology.
Along that kind of personalizing narrative, we're starting to see that conversation I mentioned earlier around AI tutors and chatbots coming into that narrative, so they can potentially develop personalized learning pathways for students. Again, I still think it's the teacher's job to guide some of that process because we know the students and at the end of the day, the chatbots don't really know anything. They particularly don't know our students, they don't have relationships with the students. So the third and last area that I want to demonstrate a few prompts with, and then I think I'll have about five minutes to do a live demo, technology willing, it's always a bit hairy when we go for a live demo, is collaborating and we can collaborate with ChatGPT or we can use it to sort of augment our own collaboration with one another.
So there's a few prompts here to do this. So these are my notes from a PD on whatever topic, we copy and paste in the notes. Maybe you are taking notes on this PD or maybe you're going to go away with the video and take a few notes on this, and then turn them into an outline for a three-minute oral presentation for my faculty. So again, a really easy way of taking something in one form, in this case notes, and turning it into another form, a three-minute oral presentation, outline for a PowerPoint, article for a staff newsletter, whatever you like. And going backwards and forwards with ChatGPT in this way is again, it's taking some of the workload off you, but it's also helping us to collaborate with one another. It's helping us with sharing knowledge amongst ourselves as educators.
The second one, role play, you're a member of the school's senior leadership team and we're writing a policy about whatever, critically appraise the policy and then we throw in the policy draft. So working as a member of a leadership team, in an example I'll show you in a moment, you may be working on an AI policy or an academic integrity policy for example, and you just bounce around a few ideas there. And then the last one, maybe if we're collaborating with one another, we might be doing a faculty retreat or a staff PD day where we're doing a bit of scope and sequence work, a bit of team building, and we want an outline for the half day. So again, extending this out and just thinking about context where a prompt like this might be useful. You might be a faculty leader coming up with an agenda for a meeting.
You might have a couple of email threads between staff that you want to condense into some items for the agenda. So you take that email thread, drop the whole thing into ChatGPT and say, turn this email thread into an agenda for meeting for discussion. Or whilst you're working on these kinds of days, you might be setting up a staff retreat that's got a more formal part at the beginning of the day and you want an outline for the presentation that you're going to do in that session. If you're going on an excursion with students, you might throw a few details about where you're going on the excursion, come up with a bit of an itinerary for the day.
So those are three areas that I'd like you to think of and planning, coming up with differentiation, and then collaborating are three things which I know from my experience in secondary education for 15 years, are three things that are really important but that we get the least amount of time to do in schools because we're busy doing the actual job of teaching and giving feedback and all of the other business that goes along with working in a school.
So I'm just going to switch over my screen sharing now to bring up ChatGPT, so you just bear with me for a moment. And what you'll see is GPT-3.5 and GPT-4. So I've got the plus subscription, it's worth it if you use it extensively like I do. So I'm using it daily and using it quite a lot. It is $30 a month. There's no business license just yet, although that is on the horizon. Whether that extends to an education license, I'm not sure. However, there are alternatives. So Microsoft's Bing chat, which you can access through their Edge browser if you download the Microsoft Edge browser and get Bing, if you use the creative mode of Microsoft Edge, it's essentially GPT-4, the paid version of OpenAI's GPT, it's not quite as good as this version but it's almost there.
Google's Bard is also free and uses its own language model and of course, you can use GPT 3.5 with a free ChatGPT account. So I'll just point out a few features in case you're not familiar with it and even if you are, they are updating features all the time. So we've got a history down the side here where we've got all of our previous chats. We've got a new feature where if we are interested in a particular chat that we've been working on, so let's say, I'll go with one of my ones from a previous PD session where we were creating some, we created a fake student, Emily Johnson, and then we did some work on a personalized learning plan for Emily Johnson. If you look on the left-hand side of the screen here, we can change the name of that to make it something more recognizable or we can click this middle button here and as of fairly recently, we can actually share the whole chat.
So this is really good for collaboration. I can copy this link here and then in a new tab, if I'm a colleague, I can open up that entire chat, static at first, so just a sort of carbon copy of the whole chat thread. But then I can also click this continue this conversation button and it will load it up in my own version of ChatGPT. So that's a relatively new feature, but it's very useful for our context. You could also share chats with students if you were going down that path. In the settings, there's a few other things I'll point out. So settings, we've got data controls now, so we can turn chat history and training off, which means that our data doesn't train future models, it's essentially an incognito mode for ChatGPT.
We can manage the links that we've shared, turn them on and off. We can export the entire history of our chats and we can of course, delete the entire account or clear all of the chat data without deleting the account. If you've got the plus subscription, that's where you unlock these beta features of browsing and plugins. So let me just show you a very quick example that sort of feeds off on the planning conversation I had earlier. We go with a very generic plan, we'll get a very generic response. So generic prompt, generic response. Let's say creates a lesson plan for a health class on the UN goals. Okay, really generic, really broad, pretty useless and that's what we'll get in response. We'll get a very broad and generic response. So lesson plan, United Nations sustainable development goals, grade level, high school, ninth to 12 grade, very broad, subject health, education and objective.
It's given us a few kind of random things here. It warbles on, it's quite a long looking lesson for me. So we want to avoid that kind of thing if we are using this for planning, we want to do what I spoke about a moment ago, which is to use some really specific input. So in a new tab, I'm just going to very quickly Google, let's go with Victorian curriculum health, let's just go with that for now, health and physical education, scope and sequence. And let's look at level five to 10 in the PDF here.
And let's say we're at level seven and eight, so sort of year seven and eight level. And we're looking at healthy and active communities, planning use, health strategies, resources to enhance the health safety and communities, planning strategies for connecting to natural and built environments. Let's take these aspects here. So let's go with this one. And grab another one of those and drop them in there. And then I'll just say use these level seven, eight, Australian, Victorian curriculum standards for health and physical education to suggest five assessment outcomes for a unit of work. It's always good to give as much context as possible and then to perhaps ask for a few suggestions for outcomes so that you can choose which you think are the best outcomes here. So it's given us a range. Let's say that I wanted to focus in on this, reflect on personal community health practices and identify areas for improvement, and I want to build that one out in a little bit more detail. I want to say, provide more detail for number five and suggest four lessons to deliver the content.
And I'm just going on the fly here and I am not a health and physical education teacher. I've deliberately picked an area that I'm not familiar with just so that I can kind of go in cold. So it's given us assessment criteria, lesson one, lesson two, lesson three, lesson four. And let's say then develop lesson one into a complete lesson plan for a 45-minute lesson, a single 45-minute lesson, year seven students. And then it's going to give us much more detail with learning objectives, what materials we need and the procedure and so on. So that was a very condensed run through and I apologize for the speed of going through that, but you can obviously watch the recording here. But really what I've done there is just emphasize that it's a process. It's not just a one hit wonder, kind of ask for a lesson plan and you'll get magic in return.
It is a tool for working backwards and forwards. So I gave it specific direction from the curriculum. I could have gone in in more detail, I could have used the descriptors rather than the achievement standards. I could have used specific elaborations from the Victorian curriculum. I've asked for assessment outcomes and I've asked for a range of options and then chosen the one that I liked the most. Then I've asked for some lesson plans in brief and then I've said, okay, now we're going to develop this lesson plan in more detail. And I could go even further, so let's say we get to stage three sharing and discussion, for the sharing and discussion stage.
Suggest a visible thinking routine from Project Zero. Those project zero thinking routines, they're free, they're online, they're pre 2021, so they're part of the ChatGPT dataset in that below the waterline iceberg that I mentioned. And it's given a see, think, wonder from Project Zero. That's pretty generic. Let's go for another, suggest another circle of viewpoints, okay, that's a little bit more interesting in terms of a routine. And then create a worksheet to support this activity. Don't need to worry too much about typos because it just rolls over the top of them.
There we go. And we could take that, we could very easily copy and paste that into a Word document, a shared Google doc, wherever we like and use that to facilitate some of our discussion. So hopefully that very, very brief run through there gives you an indication of how I would use ChatGPT as lesson planning. So it is a dialogue, it is a back and forth. Remember that it's a chatbot. Remember that because it's trained to act like a chatbot, you're always best off following through some kind of dialogue. So we've got there 10 minutes now for Q&A and I'll let Zoe field the questions at me, rather than me trying to multitask and read the questions at the same time. If you have questions, please drop them into the Q&A part in Zoom.
Zoe McDonald:
Thanks, Leon. We have a question, I think you did answer it earlier, about the benefits of using the paid version. And I think you said you'd be pretty right with the free version for these sort of purposes. Is that right?
Leon Furze:
Most of the time, yeah. So I can see the question in front of me there. It does fall prey to high traffic sometimes. So you'll find best results on this side of the world when America's asleep because they're the really heavy users. So whatever that sort of 14 to, I don't know, 11 to 14 hour time difference is, if you can pick your window so that it's daytime here, then you can get away with a little bit more there. The paid version gives a few more features, so you've got the internet access, you've got the plugins, although I don't use the plugins, they're still very much beta, sort of early development phase and they're not so much good. But there are definitely times when you get a better use out of ChatGPT, the free version than when it's peak hour.
Zoe McDonald:
Next question's in relation to the prompts. Have you trialed such an AI informed lesson with face-to-face learners? And if so, how'd it go? Example the places where lessons might be bottlenecked or become less engaging.
Leon Furze:
Hi Mark, I know you. What I'm doing at the moment actually is quite a lot of consulting work with schools who are experimenting with AI in a lot of ways. So I've got a student forum coming up with one school, where we're going to get exactly that kind of feedback from the students, which is really important. I haven't tried an AI informed lesson plan, but what I have tried is running units of work through the artificial intelligence and getting it to suggest places where I might've missed things, so some of my own blind spots. And I've noticed that it is actually quite good at picking up on areas where I might have missed diverse voices of students or areas where I might need to modify tasks. So I imagine that would be a really good use in front of students. But by the time we do the third session, I will have done some student forums and I'll get the answer to that question directly from the students themselves, which I think would be much better than me just taking a punt on the answer.
Zoe McDonald:
Next question, what do you think of Google Bard?
Leon Furze:
Hi Lisa, I know you as well. Google Bard is not as competent as ChatGPT at the moment, possibly, I think, because of the guardrails that Google put in. So they've been really, really cautious and quite conservative with their models, releasing a model only when they're certain that it can't produce discriminatory outputs, where it's not as biased as as GPT was when it first came out. So in kind of computer geek language, they've nerfed it, they've taken some of the power out of it, which means it's just not as good as ChatGPT at the moment. But I mean it's Google, so to suggest that it won't be good at some point in the future I think would be a bit ridiculous.
I've used Bard because it's got a few features that work really well natively within Google Workspace. So you can export a Google Bard chat straight into a Google Doc. If you're a Google school, that's a really useful feature. You can export a chat straight into a Gmail, into a blank email, which is also quite useful. And even the free version of Bard has the live connection to the internet now, so that's a feature that the free version of ChatGPT doesn't have.
Zoe McDonald:
Next question is when ChatGPT doesn't work, I switch to Perplexity AI. Are there any other similar AI platforms you would recommend or others are using?
Leon Furze:
Yeah, I've used Perplexity AI, so that's built on the same model as ChatGPT, so it uses the same language model in the backend. Perplexity is quite good as a search engine with AI into it, so I have used that one. I've used Bing chat that I mentioned earlier, Microsoft's Bing, which again is the same model and Microsoft Windows 11 is going to come out with Copilot soon, which will basically put Bing chat style features into the sidebar of the operating system, meaning you can use it in whatever app you're using at the time. So that's very soon, I think they're set to release that this month.
I have experimented with a couple of others. So Claude from Anthropic, which is a huge language model, comparable again to GPT, but not as easy to access at the moment. And Poe, which is an app that you can use GPT-4, you can use GPT-3, you can use Claude, you can use a whole range of models. So that's Poe, like Edgar Allen Poe, you can download that as an app. Look, for me, because I've got the paid subscription, GPT normally works, but of all of those, I would recommend Perplexity and Bing as your next ports of call.
Zoe McDonald:
Next question, do you always need to add contextual details such as Australian, Victorian, et cetera? I suppose this detail generates more usable results.
Leon Furze:
I would, yeah, because the bulk of the data in the data sets, the English language data is from an American context. So it will default to a very American style response. You'll notice it coming out with language like grades rather than year levels. If you tell it that you're looking for Australian context, it's more accurate. And then if you give it the further context of the actual curriculum dot points, it's even better again. One thing it can't do is obviously ChatGPT has a dataset which ends in 2021. So anything post 2021 isn't in that data and you won't get curriculum that's been updated since then. But even with the internet connection with the plus version, I've found that it struggles to access curriculum pages and I think it's because of the layout of the pages, there's a lot of clicky buttons and dropdowns and PDFs and it just can't read that information. So for me, copying and pasting is still the way to go.
Zoe McDonald:
How do you integrate technology and digital tools into your writing instruction to enhance student engagement, foster creativity, and develop digital literacy skills?
Leon Furze:
That's a big one. I'm running an entire session on that at the AATE Conference in a couple of weeks in Canberra. So if you're an English teacher and you're coming to the AATE Conference, you'll see the answer to that question there. Basically what I've been arguing for is we treat writing as a process and particularly in the English classroom, we talk about idea generation, we talk about brainstorming, drafting, editing. So I'm looking for ways that prompts can be used at each stage of the writing process, rather than to replace writing entirely. Can a student use it to generate ideas and then they do a bit of writing themselves and then they use it to help.
And I think really it comes down to the question, what's the point of writing? For me, the point of writing should be to clarify, solidify your ideas, to help you organize your own thoughts. The point of writing shouldn't be to assess somebody else's knowledge. If we're getting students to demonstrate their knowledge through writing, they can do that in other ways, orally, through presentations, practically, through their practical application of the knowledge, in ways that ChatGPT can't do for them. If we're getting them to write, it's because we want them to develop the skill of writing. So I think there are ways that we can kind of work through that whole process of writing.
Zoe McDonald:
We've just got about one more minute, so just a couple of questions that came through the chat. Can you drop other formats other than text, example slides or images into ChatGPT as prompts?
Leon Furze:
Right now, no, but look, give it a few months. We know that GPT-4 that's sitting underneath ChatGPT, can be multimodal, so we know that it potentially can read images. It's got image recognition capabilities. They haven't released that publicly yet, but it's definitely on its way. Once that happens, you would be able to drop in a PDF of slides and images. Currently, there are things like chatpdf.com, which you can upload a PDF of text and it can read the PDF and you can interact with that as well. But the multimodal feature, it's definitely on a fairly near horizon.
Zoe McDonald:
All right, we might have to wrap it up there. If there's anything that didn't get answered today, we will follow up in the email with the recording from today as well. Thank you so much, Leon, for another fantastic session. Our next webinar you can see up on screen, is exploring media literacy in the age of AI. So that's the same time, 3:45 Melbourne Eastern Standard Time, on Wednesday the 19th of July. If you have any issues or questions in the meantime, please feel free to email us at education@acmi.net.au and we look forward to seeing you all then. Thank you.
Leon Furze:
Thanks for coming everyone. Thanks ACMI. See you soon.