Episode 149: This week on Thinking Deeply about Primary Education, I’m joined by Neil Almond, a teacher, leader, and co-creator of AI Education newsletter, Teacher Prompts, for a fascinating discussion about the role of artificial intelligence in education. As part of the chat we discuss… What is Artificial Intelligence? Is it likely to change the way teachers work? What are the chances it’s all smoke and mirrors like a lot of other technology? What are the three things all teachers need to know about AI? What’s your bold prediction for the next 12 months of AI? …and much, much more! Whether you’re new to the profession or an experienced senior leader, you won’t want to miss this one! Support the podcast via www.ko-fi.com/tdape or by subscribing to www.youtube.com/@TDaPE If you would like to submit a question, email us at thinkingdeeplyinfo@gmail.com or join the TDaPE discord, where we have a special channel for unseen question submission.
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
Welcome to Thinking Deeply About Primary Education, the podcast that makes time and space to think about pedagogy, teaching and learning professional development. Anything of interest to time. Poor but enthusiastic primary teachers join us as we delve into the murky depths of education in an attempt to help you enhance your classroom practice. In today's episode, I, Neil
Almond will be taking the reins. Welcome to Thinking DV by Primary Education, the podcast that makes time and space to think about pedagogy, teaching and learning, professional development. Anything of interest to Time, Pure but enthusiasm, rich primary teachers. This week I'm joined by Neil Almond. Needs to be back, Kieran. This week we're going to explore the potential of artificial intelligence and the impact it might have on the work of teachers.
But first, Dale, what you reading for? Hey, what you reading for? So I was recommended a book called Spelling, Structure and Strategies by Paul and Jean Hannah and Richard Hodges. It's quite an old book, 1971 quite so. It's getting on just over 50 years old now, quite enjoyable in the sense that it's affirming
many of the biases that I have into how I think we should teach spelling, which on one hand is really good and on another hand, you know, not so great. What was interesting about this one though is that they were kind of the first people there was a study and I had no idea about this. They did a study in the 60s and you know used the all the computing power that you know we had back in the 60s which obviously was enough to put a man on the moon. So fair amount of it to really kind of study the structure of the English language. And so they're kind of the main
appeal for this book for me was around that structure element because they're kind of in a very basic way, it would have to be updated for the modern era, kind of revealed the statistical properties of the English language. So you could teach children to a high degree percentage that if this sound, the sound L was in the end position, then there was
a I'm making this statistic up. So like a 95% chance that it's spelt OW and that's kind of something that has always kind of interested me and it's something that I felt has been missing from spelling instruction that kind of making that statistical elements explicit to students as quickly as possible. So yeah, haven't finished it yet, but really enjoyed kind of going through it and kind of gives me hope that there'll be some researchers out there who will create a larger corpus than the 17,000 words I think that
they used when they did this ideally, hopefully British English spelling as well. And obviously with more computing power, stronger algorithms, you know, we'll be able to produce something more up to date. So we'll see what about you? What are you reading for? Nice. Just the 50 years have passed. Yeah. So I mean might be implemented in classrooms any day now. I've got a maths one for
balance. It was shared with me. You think? Oh, Brendan Lee shared it. It's called The Need for Speed. Why Fluency Counts for Maths. Learning by Tony Hat and Rob very interesting read. I won't give away too many spoilers but I'm sure you can probably guess what the sort of the the main takeaways are. But yeah, well, well, we're checking out. So I think probably Neil, it was around about this time last year that you introduced me to Open AI and.
In that in those 12 months, a lot has happened. The rest of everyone has tried to catch up with you, so obviously you're that you're trail business in terms of artificial intelligence. So I thought we'd sit down and we'd look at what this might mean for teachers, because there are plenty of snack oil merchants who are looking to make a fast buck off the back of this. And I promise I'm not one of them. Well, now you you have no skin in the game. And you know and you know the the the the charlatans aren't
there. They're not teachers. And so I'm not commenting on any educators here in in this operating in this field, but rather you can see the promoted tweets on Twitter and stuff like that. There are words like, you know, you know, I'm sure people have seen them and, you know, so like reskin and chat, GTV trying. I don't know. Am I am I offending any teachers? No, no, you're fine. I'd say it's 80% of my Twitter feed now is just these promoted accounts that are telling me that I could earn thousands of
dollars in side hustles if I just learn how to do these five things in ChatGPT or Dolly or Mid Journey. And yeah, it's it's a pain for my Twitter feed, Let's put it that way. So I thought we'd explore it because there must be some value, because there's a whole lot of money being invested in this and around the world. But it makes sense to start with definitions of what we're talking about. What is artificial intelligence? OK, so AI Kind of. For me it's this.
The study of using data in ridiculously large quantities to attempt to program intelligent machines that mimic human like behaviour. That's my the definition that I'm kind of going for here. Obviously the purpose of it I guess is to improve how quickly, accurately and successfully it can complete tasks that we may ask, you know, humans to do
right now. We're kind of seeing the beginnings of that sort of journey and attempt to mixed success, I think is is fair to say. I mean, when I think of artificial intelligence, I think of Was it Deep Blue? The chess? Playing a robot? Yeah. That that taught itself essentially through running through millions and millions of different games. Hard to play chess to the point where it was unbeatable.
Was that was it deep blue? I mean, I know they they put deep in front of, put deep into everything. The reason it's called Deep Blue is that particular sort of so AI is like a bit of like a good like umbrella term for lots of other kind of other techniques that are AI, but there is definitely a sort of like hierarchy to them. So AI is sort of the the holding concept, the overall sort of umbrella and then you kind of have things like you know machine learning which is a method of making predictions
with that doesn't sort of use predetermined human coded rules and that kind of analysis, vast volumes of datas to find patterns then kind of construct a model. So that's kind of your large language model esque type of ideas. You also then have as you kind of mentioned there the this kind of deep and that's kind of like this deep learning. And what they try to do with deep learning is that again it's a subset of AI where they are
literally trying to kind of feed and create this kind of neural network that you know effectively learns by itself. And so that's where chess element comes from. So anything that kind of has deep blue, deep mind, anything like that is kind of using this kind of deep learning kind of process that uses artificial neural networks to try and get it to comprehend and understand and attempt to reason.
But not always very well what they should and shouldn't be doing at various points which wise that they can analyse all that data and make those predictions because that's all they are. And that's a kind of really important thing to remember when we're talking about AI, particularly sort of generative AI, is that it's just a massive prediction machine. And that has massive ramifications for how you should treat the kind of information
that it, you know, the outputs that it then provides you because they're not necessarily searching the Internet for accurate information. Everything that it creates is new in the sense that it hasn't necessarily been written and copyrighted previously. So. But it can sound really convincing in the terms of what you want it to give, which can make it seem like, Oh well, you must have picked this up from the Internet or you know, I've
asked you for paragraph three, Chapter 4 and War and War and Peace. And I want you to summarize that. And then it kind of gives a really what you would assume you know, jurotish well, not necessarily accurate plausible idea what that actually could be, where it's an actual fact. You know, it sounds like it could be summarizing it, but the reality is, is that it's just made something completely up. Yeah. I mean, I think that's a really important distinction there. Yeah, because I was almost conflating machine learning with
artificial intelligence. That's not quite the case, no. Machine learning is a a subset of AI. It's a type of AI, but it's not the same thing. And machine learning has been around for, yeah, ages on our phone. You know, machine learning is the kind of stuff that like, I don't know if you you're probably, you know, far much, far, much more of a life than I do. But I like making sure that if I take photos of people that my phone like, tells me who those
people are. So if I wanted to find all the photos that I've had of you that are in my phone, I could just search for Kieran Mackel and it would kind of show me all the photos that I that it thinks is you. Now you kind of start training that model, then eventually it kind of gets to learn who you are. And so, you know, every now and then it just pops up saying, hey, here's, you know, 5 new photos of of you and Kieran Mackel. That's kind of like machine learning. That's it kind of learning from that data that you kind of feed it to kind of get to know you and say the phone, iPhones, I'm sure Google are doing the same
as well. You know that kind of feature has been there for for ages. So, so yeah, machine learning, nothing kind of new, no, it's not something that's just brand new to market. Such the the newish side of it is this idea if it's been these generative AI tools. This idea of you can prompt it with various things and then it kind of gives you an output which could either be a text output or an image output depending on which AI tool you are choosing to use. My phone does that sometimes.
It does it for me, or it feels like it's done it for me. I've actually been training mine to identify me as someone else in case I ever need to be extracted, and then they'll have my records. No Who's this guy? Is artificial intelligence likely to change the way the teachers work? I know this will be a surprising answer or not, but I actually don't think it's going to necessarily fundamentally change the way that we work. I think it's just going to support teachers and how they do
the work that they currently need to do. I don't think we're a whole myriad of reasons that you know, we're going to automate a lot of the things that we would like to automate from a data GDPR perspective, from a cost perspective. I don't think it's kind of with anywhere near that yet. But what I do think it will do is kind of, you know, be the sort of digital teaching
assistant to the teacher, I can't say teaching assistant, otherwise we might get confused with what we mean. So it's definitely assistant to the teacher. I kind of think the way that it kind of the best way that's going to kind of help is if you think about the Pareto principle, so you know 20% of the efforts deliver 80% of the results. Like the hardest place to start is obviously that first like 20% of that, 20% like when you're there with like a blank sheet of paper because you have to do a fill in a form or you have to sort of write a risk assessment
for a trip or you know, plan a lesson and you're just there with that kind of blank sheet of paper for ideas. You know that first 20% of that 20% is really hard. And I kind of think that what generative AI can kind of really help us with is to kind of provide that launching point to kind of help us with that first 20% of the 20% that delivers the 80% of the output, if that makes sense, which is the hardest part. But just by giving us that kind
of launchpad, get these ideas flowing, I think it's going to be really, you know, quite useful in that regard to teachers because they say it kind of gives them that that first could be that first step or, you know, just, you know, a couple of ideas that could, you know, be what they need to then, you know, get into a state of flow so that they can kind of finish off whatever task and activity that they need to do. So yeah, I don't think it's going to massively change what teachers do, but I think it will change sort of how teachers are
going to do it, particularly I think with things like and again depends on budgets and data protection and things like that. But Google services they're starting to integrate Bard which is Google's large language model. Microsoft it has Co pilots which is their large language model
and all these and this kind of assistant we want to a better word can talk to your office apps. So Co pilot can talk to outlook, word, PowerPoint, Onenotes whatever it may be. But the same for the Google equivalents. And So what what would really do is kind of really support teachers bringing all the information that currently sits separately within those
individual applications and bring them together and being able to kind of compile that information together and you know, produce some sort of output. So you might have for example teacher planning a let's stick with planning a school trip. So she might want to ask Co pilot, I want you to bring up or kind of show me everything that I've done so far to support the school trip to the Tower of
London. And so the e-mail that she sent to the to SLT asking for the trip would come up automatically. The, the Word document that she put together for the letter would come up the PowerPoint presentation that she's done for the parents to explain the purpose of the trip and you know how they can support and all that stuff will come up. And obviously then you can kind of, you know, combine all of that data that's in those particular documents, you know to potentially then create
something, you know quite powerful. And so I do feel that side of it, that sort of digital admin assistant, you know, that could actually probably, you know, again not change the way teachers work, but change how teachers need to do the work that they need to do. However, as I mentioned previously, there's the whole thing about, well, how much this does is this going to cost schools extra. And obviously as we're talking about, you know, young children here, you know, you need to give
Co pilots access to effectively your Microsoft account. And so if you are a complete Microsoft school, that is all the data that you may hold on children and whatever else within those different applications. Those then need to go off to a Microsoft server or you know Co pilot to do the things that it needs to do. And obviously you know there are massive implications and if anything you know does go awry. But that is definitely I think
one way that I think will save teachers time from that kind of admin side of it and that kind of aggregation of all this data that teachers accumulate or maybe it's just me and I just accumulate lots of emails about different things. And you know my filing system of organising things has never been as I would like it to be. I'm I'm no Adam Smith with his notion and things like that. But you know, I think it could help teachers with that regard and I think that's not the
reason teachers come into school. So anything that helps them to that side of it, which means that they can focus more on their teaching. I don't want to go as far as saying that there's probably like a moral reason there that schools should be looking into this because by going down that Ave. they're opening up more kind of ethical issues. But I definitely think exploring such avenues, particularly in this current climate, is something that, you know, schools, large trusts, the DfE,
should, should be considering. I think it sounds really interesting. I would love that as a teacher to have something that would help me structure my personal organization. A lot of the a lot of the early conversations about artificial intelligence seems to allude to the outsourcing of thinking. And obviously, the main theme of the podcast is that thinking is the most important thing teachers can do. Are you saying that there's no
way that artificial intelligence will ever take the thinking art of teaching? You know and and you know, Hey, Bard, I want a Lesson plan on, I don't know, rock and Romans. Is that that day ever going to come? Yeah, that day will come. But I don't necessarily classify that as the AI doing the the thinking for you. As I say, you know, I would be amazed if anyone takes that, you
know, puts that into a generative AI tool and, you know, doesn't then do anything with it. As I say, you know, it starts off the thinking process. It gives you some ideas to to think about but you know fundamentally you know you then have to I think with that information again the 1st port of port of call and it kind of comes back to something we'll discuss later on is you know you have to check for the accuracy if you're using a large language
model to do this. You know you have to check for the accuracy. So if you. So my preferred way of doing it would be that you have, you know, quite a tight curriculum in the terms that you're quite explicit as to what want students to know. So let's say keeping on the theme of rocking Romans. So you know, Caesar tries to invade in 55, he tried again in 54 and he fails because I think the weather was particularly
bad, something like that in one of the invasions. And the the Celtic tribes were able to kind of have a bit of a ceasefire between themselves to come together to defend the Romans. Let's say that's what you wanted out of that first lesson. If you told the AI that you know, these are the parameters of my lesson. This is what I want kids to know. I want them to know that, you know, Julius Caesar invaded 55, then 54 and then it failed. One of the reasons why the 54 invasion failed or two reasons why the 54 invasion failed was
because the weather was against him. So he had to invade later than he originally planned. And the Celtic tribes of Britain at the time, you know, came together to, you know, stand against him, make me a Lesson plan from that is going to be far more beneficial, far more effective, far more accurate because you, the teacher are kind of, you know, narrowing those parameters for which this predictive model is going to be working from.
So the output that it gives you is going to be far more accurate to what you need versus something like hi ChatGPT. I'm teaching a lesson on the Romans. It's for year three. You know, hit the button, go, because you know that's just going to deliver all sorts of weird and wonderful things with potential inaccuracies. So the more thinking that you do actually before you ask what you want, the more useful the that output is going to be in sort of
a big. And that's kind of partly the reason why I've gone down this kind of like this prompt Ave. because I kind of really want to make people aware that this AI tools and technology is can be really effective. But it really is sort of that intersection between teacher expertise and your knowledge of how to provide a really useful prompt. It's not an outsourcing of
teacher expertise. Far from it. Now will we ever get there though? I mean, you know, who knows? But right now we can't. We shouldn't. I I was listening to a podcast and it was from a totally different field with the Talmud AI and they were saying that between now and April next year there's going to be a massive explosion. You know, if we think about what? The explosion that happened last October when the world became
aware of this large language and model, and then the 12 things that have happened in 12 months since then, But I think this next, what, six months will be massive. So who knows what will be possible. But I really like the idea that it's going to take a lot of expertise in two domains now to actually get the most from it. And you can't just type in. I'm teaching fractions. Off you go. You know, there's my fractions plan, you know that. That gives me some reassurance. We're recording this on Monday the 30th of October 2023.
You know, this is probably going to be out of date before even. I don't know when you're planning on putting this one out there, but I would go likely before where this gets put out to to listeners and elements of what I've said is probably going to be out of date already, which is just goes to show how fast moving you know this field really is. Yeah, absolutely. You mentioned prompts, so you're alluding to the newsletter that you have started to release.
So myself, past participant on the show, Pats McRae, who I'm sure people will know from all his concise books, which made me think that perhaps they were mostly written by AI. I'm not sure yet. I haven't had the guts to to ask him yet whether he was actually just using large language models to write most of those, but I'm sure I'll ask him at some point. We kind of had this idea of putting this out there to help
teachers navigate the the murky waters of these generative AI tools that are out there. Obviously we alluded to earlier the kind of the kind of Twitter bots that are appearing here there and everywhere and other kind of various other you know AI tools that are marketed towards people using and seeing if we can provide a different Ave. of support which I say it's
predominantly through this idea of of prompting which is where you most of these generative AI tools like ChatGPT and Dolly, Mid Journey, whatever they may be predominantly require on the text inputs. And that's called your prompt. And effectively the this entire sort of field has arisen recently called prompt engineering which is effectively
I'm really starting to not like the term, but I'll use it because I can't think of another one. But it's effectively the science of how do we create effective prompts to make sure that the output that we then get from this generative AI model is what we envision that prompt to be. So as I said the these are effectively prediction machines. So if we are really vague, you know, you say our one Lesson plan on the Romans, you might well get a Lesson plan back that
includes things like this is how you should support different learning styles, for example, which evidently, you know, we know is not a good use of our time, but clearly within the the trained data sets it has information about learning styles. And so because we were quite vague with our prompt, we, you know, it gave us back something. So the more we can kind of find the underlying structures and play around with those underlying structures of what effective prompting is, the higher quality, the output
effectively and hopefully the more useful than it then is for teachers to to use in upcoming lessons or admin for example. And so yes, one of the way that we started this process is through our our weekly newsletter, Teacher Prompts, which is there to kind of upskill teachers not just in sort of prompts, although it we try to make sure that it does do that. It's kind of there just to give dominantly, you know, educators, you know, up to date news in the
AI world that sort of where possible sort of directly influences education. So for example, at 2 bits of news that will probably make it to tomorrow's newsletter is that you know the government's just recently handed over Oak 2 million to explore AI use there. I've had a bit of a play with some of these tools already and you know they need work. But I'm sure you know the more this is in front of teachers,
the more useful it's going to be. And you can now upload if you purchase ChatGPT Plus to access the the GPT 4 model, which is you know, has more parameters, is trained on more data, more accurate, you can sort of now drop PDFs and other kind of files into it and it can then you know, summarize and you can ask questions and interrogate it
about that particular file that you've dropped in, et cetera, et cetera. So and then we go in about you know what effective, you know, some effective prompts that will be useful for for teachers. Yeah. So every week and I like that you always have you know in this in this I don't know edition in this this one episode you you will find and then you get like a couple of links and you've got and this image yeah always try throwing a generative AI image there just to they're quite good.
I mean, you know, we had some good fun making some of the other day, didn't we, in our little little group chat of the weird and wonderful things that you could you could do with that. I really enjoyed the the UP sequel down. But it was funny because you immediately, you could immediately get that sense that this was UP If this was, you know, it was all themed on the idea of UP. And, you know, I'm not going to say what's, you know, holding me back from creating high quality
educational resources has been my poor graphic design ability. But I'm far more inclined to give it a go now that I know I can just prompt Dolly to give me a few, you know, useful images to use. So yeah, it's going to be useful. Yeah, I I mean, it does take some prompting because I gave it some lazy prompts and got some nonsense back. My best one were The The Beatles, but they're a band called The Beatles, but they're actually Beatles. And then dishwasher has fallen.
Dishwasher has fallen. It was a great one. They will put these when the episode goes out. We'll find them and people. Can have link? Yeah, good guy. Yeah. I think it's season what we're using. We're using Bards image generator, so it's called. We were using Bing. Bing, not Bard. Yeah, Bard is Google Bing, Bing is. Microsoft Yeah, that's the search engine that one uses, exactly. So technology and education have
a very interesting relationship. Sometimes millions of pounds can be spent in ways that were perhaps ill advised. What are the chances that this AI revolution is smoke and mirrors like those other aspects of technology that have proven not to be as effective in AI? Don't know. Changing our practice or our ways of working, I mean there's always the chances. Now you know we've seen it with, you know, interactive whiteboards were not the solution that we were sold. They would be 1 to one devices
and iPads certainly haven't been the thing that they've done. The reason I feel like this is different one because you don't necessarily need almost there's a continuum here. You know if you were to go as I mentioned you know the whole virtual admin assistant via Bard or Co pilot and that's something that your school purchased as part of you know, the Microsoft license, then yes, obviously there is a likely to be an
additional cost to that. But you can get in there, you know, without having to without price being a barrier. You know ChatGPT 3.5 is free to use. Bing is free to use as well. All you have to do is just make accounts for these things and I feel as well you know the one to one iPads or you know one to one devices, the interactive whiteboards, all of that stuff that was very much student
facing uses of technology, this sort of generative AI stuff. Yes you might put resources that you've created through this in front of the pupils but it is very much all happening sort of you know in the background backstage as you were. You know it's the planning processes. It's the that you know that deep thinking that we're doing before we get in front of the classroom. And because of that element I think it will be different I
think because I mean the the irony is you know you know I'm I'm I'm no fan of you know Soonak as you know. But I think we all laughed a couple of years ago when he said he was going to plough a bit of money into AI and we all sort of can you know laughed at him. But you know I think you know credit where credit is due. You know, he was potentially right in that, you know, this is something that is going to
really potentially, you know, support teachers. And I think, you know, with their putting their money where their mouth is to a degree by giving you know, Oak Nationals some funding to kind of explore this. And Oak National are putting all their tools out there, which is good and interesting. So, you know, teachers are going to, you know, see this. I'm sure there are teachers who still don't know how to access these things and that these
things exist. Absolutely. But those teachers that know about interactive whiteboards, those teachers that know about large language models and generative AII, bet they've played more with generative AI and what that can do for them than they ever did about, you know, let's have a look and see what what smart can do. You know, they found infinite Clover and thought great, this is my job done and I don't need, don't need anything else. So yeah, this feels kind of different because price isn't
necessarily a barrier to access. There are three versions that you can use. And because it's sort of, it's not happening in the classroom, it's happening in that kind of planning stage. You know, I think that there is more scope for this to be successful than previous elements of technology. But again, all it would take is 1 massive data breach, then the whole, you know, House of Cards to come crumbling down somehow. So yeah, got to be careful with
it. I think it has more chance than previous tech endeavours, but as I say, it's certainly not infallible by any means. If it's used in the way that you describe, it could be used. I think there's a fairly good chance that it yeah, it could be. There is I mean I think I mentioned this, I can't remember which episode it was. There is a quite a large. I think it's a charity organization. That's involved in the education of other people and their investment in AI has been trying to get the the computer or the
program to give feedback on misconceptions and things like that there. It feels like that's the kind of thing. I mean, you would need a whole lot of information. To make an accurate prediction about what the misconception was or why the error had formed, Do you reckon there's, you know, is that indicative of how people in, you know Silicon Valley at the minute maybe a little bit
further beyond, are are funneling their investment in education and AII or do you think it is as as a sort of an assistant to the regional manager? That's a good question and what I haven't really thought about. I would say the former purely on no data or any you know, smart hypothesis behind it other than you know in Silicon Valley, you know data. Data is king. There's that running joke where if you don't pay for a service,
you know you are you are the you are the, you are the product, Which is why I always pay for, you know, pay for Apple and try to avoid Google wherever possible. That's my biases coming in there. And so the more data you have, the more valuable data is, the more it can be used to target at specific things. And you know, we've got no idea how advertisers are somehow going to find their mitts into all of this data and do stuff with it to try to target more ads at you in a more kind of
specific fashion. But it's evidently, you know, in the past, it's paid off having as much data about as many people as you can. So you can sort of generate, you know, rough profiles on who these people are, what they're likely to be into, who they're likely to vote for and you know all that sort of real semi conspiracy Cambridge Analytica sort of stuff that you can kind of get into I. Mean we don't have to legal back in to navigate this storm that
is potentially brewing here. Perhaps we'll say yes, I I see what you mean now because yeah. And lots of tech geniuses try to get involved in education in different ways to varying degrees of of success. So it'll be interesting to see how they go. I mean, what's really clear? Is that over the last 12 months, you've spent a lot of time exploring this field and getting to know this field because the, you know, the, the quality of your answers today absolutely
blown my mind. Well, maybe maybe it speaks to how you little I've properly engaged with this, but you know, I've learned a lot about how this actually might be. You know, my skepticism is being eroded gradually and with every response. And so it's really clear you spend a lot of time thinking about this. But how we can get the most of it? What are three things all teachers need to know about you think? Yeah, these are the three sort of products. Not, maybe not even products,
but types of product that you should be spending your time reading about nigh to get ahead of the curve or even just to catch up. OK, I'll do 3, then a bonus one. So for the teacher, if you were to go sort of 1, one tool to rule them all sort of thing, one place to go. So there are must be hundreds of start-ups starting up sort of every every day. But if I think you know you
really can't go, it's it feels so weird saying this, but you really can't go wrong with Microsoft Bing as your kind of best free at the point of use generative AI tool for text. And images. All we need is a Microsoft account and you get you know effectively, effectively, you know, unlimited conversations with this AI chat bot that you can use to create lesson plans, create quizzes, think of explanations, get thinking about analogies, generates interesting
images that you might want to use in lessons. For example, the reason why chat Why Bing is particularly effective and useful is because Microsoft have plowed a lot of money into Open AI and so Bing is hooked into the ChatGPT for back end. So it's using the most up to date, certainly the most popular large language model to generate
text and images because it uses the recently updated Dolly 3 model to help with images as well. If you usually go through Open AI themselves to get access to ChatGPT for I think it's something like £20 a month I think. So, you know, you can save yourself a lot of money straight away there by just doing it all through all through Bing. They're not quite the same. They're also going to tweak and style it quite differently. But you know effectively, and this is kind of an important
point as well, kind of buried within this one, is that on any AI tool that comes up or that you may come across, have a look and see what it's large language model it's using to draw its data from. Because if it's using the same back end, you can probably get the same answer using a free version. It may not look as pretty, but what you actually ask for the content is I would say 9095%
going to be the same. So if you haven't used any tools yet, you don't know where to start or you've perhaps used ChatGPT 3.5, the free version, I would definitely say it's worth having a go at looking at Microsoft Bing because as I say, it's free and has the most up to date models that it draws upon. So that's definitely going to be high up there linked to this. So these are all nicely linked together.
The kind of second thing is that there are two classes of sort of AI tools sort of out there right now. These are names that I've kind of given them. They're not ubiquitous by any sense, the imagination, but there's a sort of a closed AI tool. What I mean by that is that the prompt is effectively invisible from you and all you will be asked to do is input some information. So the stuff that Oak had put
out recently, that's a good example of a closed AI tool because all you do is ask for what year, group, subjects, any particular topic that you want which you can input yourself, you press enter and sort of you know there you go. It kind of brings you back something and there are you know some advantages to that. One is that it's, you know, quick and easy. The the user doesn't have to learn how to do effective prompting.
And as I say, it's quick and easy. But there are sort of some downsides to that in that it doesn't sort of offer a serve and return type interface. So what you're given is what you get. You can't necessarily go back and tell the AIO, I don't really like question 4. Can you please change that to something else or I really wanted you to include the Norman started building castles quite quickly when they invaded in 1066. You've missed that. Can you please put that in your explanation? So you can't kind of do that. So it kind of really kind of
limits you. It gets you there quickly, but sort of really limits you as to what you can do. The other type which is the type of tool that Bing is that ChatGPT is that Dolly is is what I kind of call open tools in that you need to then provide that prompt so you're in complete control as to what that prompt should be. And again some advantages disadvantages are is that well
you're not limited to the type of requests. So loads of these tools the way they sort of they'll usually say oh you know we have a lesson planner tool, a quiz creator tool and you can have that for free. But if you want the assembly creator tool and you want the school improvement plan tool, then you know it's, you know, 699 a month, you know extra whatever it is, if they're using ChatGPT 4, you know you don't need to worry about any of that because you can control the prompt yourself. So you can just ask for those things because all of those,
what those types of applications and tools are doing are just effectively keeping that prompt hidden from you. If you have good prompting engineering, you know you'll be able to get, you know, probably 80% of what they give back to you. So that's why it's good because you can prompt for anything that you want. You're only limited by what you yourself can think of. You have complete control of what that input is. A big bug bear of mine is that and this goes on to my next thing that teachers should know. But because most of these things
are trained on American data and sort of American and they have, you know, biases built in built into them, a lot of the output is going to be American English. So like my favorite, the ending of a prompt is always say the output must be in British English. You know, small tiny thing, but it's going to, you know, save you a bit of time. Can't do that with the other ones unless behind the scenes they've also coded for that. It must be in British English, but you will never know that until you've got a prompt.
But put your information in and got some data routes, it does allow for that kind of serve and return. So if I put a prompt in and it comes back with something that I don't like, I can quickly say, hey, I really liked this bit, this bit, this bit, but I'd really appreciate it if you included this. So you can kind of then sort of, you know, narrow down what you want and then really narrow your focus. It develops for the teacher expertise in prompting. So they actually understand the underlying features and structures of what you know and their mental models of what this technology is and how it's going
to work best for them, something that doesn't necessarily happen with these closed tools. And let's say going back to what we said, you know, is this going to get rid of our thinking, It's definitely not because the user in these open tools can use their own expertise to their subject phase, the curriculum to guide these large language models to the output that kind of best suits their context. And so don't necessarily get that again with that closed sort of tools.
Now obviously there are some disadvantage to this one as well. There is a big learning gap. You have to kind of learn how to prompt and what effective prompts are and that takes time. And so obviously that's then going to perhaps impede how quickly you get started with this. If you don't feel successful straight away in generating what you want, you might feel a bit disheartened and not motivated to give it a go. But I think the initial effort is worth the outcome by using.
And again, all biased here. I have a newsletter that kind of tries to teach you how to do this open style use of AI tools. But I think in the long run, you know, it's better for teachers to use that than some of these closed tools. And the kind of the third thing, again, we've kind of talked about this before, but it's really important, is that it's not always right. So you may think that you've delivered something, got something back that's, you know,
perfect. I think I saw something recently on on Twitter about someone who wanted some descriptions of, I think it was like gothic castles or something like that. And she wanted it from from Dracula. So she wrote, I mean again, a prompt of, you know, to try and get that information back and ChatGPT, whatever she used, brought back an excellent description of a Gothic castle
in relative style that I as a as a novice in Gothic literature would not be able to say, well, that's definitely not, you know, Bram Stoker. But you know, one of the first comments was, hey, you know, that's actually like not the description that's used in Dracula, right. And you know, that was a surprise to her. So it's never right because it's not connected to the Internet necessarily. It is just a prediction device. It will be really, you know, it's well, they always say like
as long as you say anything with enough confidence, like you'll they'll anyone will believe anything. Like it is literally like that Incarnate. It's so good at making you think that what it's producing is the right thing and it's trained on data. Human created texts and humans have biases that will have implications for the output of that data. And so you really kind of have to check and make sure that one,
it's accurate. And second of all, that humans, natural cognitive biases aren't necessarily drawn out of that output as well because you could get yourself in a bit of trouble I think if you don't check for that one as well. And you do end up producing text that's may provide a certain interpretation that you know you believe something that you or you've provided some data or some information you know to
pupils that isn't necessarily correct because it is. You know the data was biased, so it's produced a biased output. And the kind of the bonus one obviously is that teachers should definitely sign up to Teacher Prompts newsletter, because that will definitely tell you everything that you need to know. I'm currently working on a beginner's guide to AI and like PEPS did with evidence, Snacks, you know, if you refer. I think it's going to literally
be like if you prefer two people to this, you know you'll get the guide. If you refer six people you'll get the the basic image guide as well. So that's that's my half term job, so I'm hoping that will be done by definitely not by tomorrow but hopefully next week Once everyone's given at the OK that will be available there as well for people to use. I love the. I love the. Yeah, the incentivized sort of thing. Yeah. And I remember speaking about sort of just watching him with
his chapter from his upcoming book and stuff like that there. So that I think that's a really interesting way to. Encourage engagement. And I'm looking forward. I mean, I'm excited about the potential for artificial intelligence. Just off the back of this, what about an Irish conversation? So, you know, whenever we finish up, I'm going away and I'm going to try and read as much about engineer prompt engineering as I possibly can. You know what? What do you do for a job? Well, I'm a prompt engineer, you know, they say it's going to be a very highly valuable profession, since I think that
is probably, you know, the jobs of tomorrow kind of brigade going a little bit too far, but it's definitely an interesting, interesting Ave. Yeah. I mean, if there's one thing I've learned since starting my own business recently is that if there's anything that could make your life easier, someone's got that corner of the market sewn up already. So if you want to save yourself some some money. And get exactly what you need. And then I think the prompt engineer's going to be really, really important for getting the most of them from from
artificial intelligence. What's your bold prediction? For the next 12 months, the outputs from these models are only going to be as good as the data that the model uses to learn from and Oak National, because all of their resources are open source. Effectively, large language models can comb everything that Oak has will create.
And so that means that there is going to be, you know, models that are trained specifically on education data and specifically education data of the the 2014 National Curriculum for England. And so as working in education in England, if you have models that are trained specifically on that data, that's going to be, you know, particularly useful for teachers and that's where whatever Oak do next could be really quite interesting over
the next 12 months. So I think bold prediction there is to kind of keep an eye out on what Oak are doing because they're going to produce something I think that could be really useful and valuable to teachers. I hope they open it up. So it's not just a kind of a closed tool, but having it specifically trained on, having it specifically trained on education today that's completely relevant to, you know, teachers in England is going to be quite a powerful thing, but they're all going to open that data up to other models as well.
So you know, it may not just be Oak that has access to all of that. The biggest weakness that I've found in generative image creation right now is that it is spectacularly poor educational diagrams. So if I want were to ask it to create a show me a bar model of 4 * 6 or, you know, here's the here's a word problem, show me a bar model that goes with that. Can't do it.
So I'm kind of confident in the next 12 months. Well, I feel not confident. I don't have no insider knowledge, but I feel like it's an easy win for them to like if you nail the educational diagram, you're going to make a lot of teachers very happy. If I can just say, you know, consider the split attention effect and create a simple diagram of the water cycle aimed for pupils in, you know, year four who are in the, you know, studying the English curriculum. And I might say, you know, it must include, you know,
evaporation, condensation runoff, whatever else you want it to include. Enter. And then it produces a useful graphic that takes into account the split attention effect. So it's not key in the corner. It's, you know, labeled where it needs to be. And I don't have to worry about the copyright of that. That's going to be really powerful for teachers. Same way for, you know, you and your math journey. I'm sure if you could say, you know, I want you to show them this model using algebra tiles or you know, cuisinare rods, whatever it might be and you could get generative AI to do
that, you know, that'd be really powerful. I've told Johnny, you know, he needs to get in that market really, really quickly, make it happen because no one's doing it right now. And kind of the big the next kind of prediction is that I do kind of think you know I think more teachers are going to just keep on using it and find interesting ways to you know make it work and make it make it
work for them make it useful and hopefully cuts down a little bit on their on their teacher workload. And if I can play but a but a small part in that for what is for me, you know hours work on a weekend to put that newsletter together, Be a happy man, yeah. Because at the minute, obviously we made those movie posters. We know that it currently struggles with thin words. So there are some hurdles to overcome before diagrams and models are a realistic prospect.
But yeah, but imagine being able to sort of put in the prompt and then you would have that representation ready for you. I think it would open up a lot of doors. So fingers crossed. I mean obviously. No more marking or sort of considering how they can get the most from it. So it'd be interesting to see what happens with them because they've got a massive data set on which to train and stuff. I don't know your your reaction seems to think that they haven't started exploring this and have we just discussed that they should explore this or I thought, no they're doing they're doing some interesting
stuff. No more marking. Absolutely. Yeah. Daisy is a is an intellectual powerhouse and she's very good at getting rid of the the rub, finding the bits of gold within all the all the nonsense that are out that's out there right now. And they do some, Yeah, they do. They do good stuff. Cool. Well, like I say, it's an absolute pleasure and I think I've learned a lot. I am excited about the next 12 months, really looking forward to seeing what happens and definitely going to read more. I'm going to sort of pick your
brainer offline for where you going for your information first time because the minute I'm getting it from from guys here financially. Incentivized for this to be a success, You know, so they're all very much in Palo Alto and and the surrounding suburbs and and have a, you know, have a good skin in the game. You know, they're not in education fields, but you know, they're incentivized to make sure that technology continues to develop and stuff. And so it'd be interesting to find out, you know, because I doubt research papers exist, you know, because it's happening in
real time, so to speak. You know be surprised if you go on to if you go on to our our litmaps. There's a there's an area there's a thing that I check every now and then about ChatGPT and engineer prompting and stuff like that that I kind of put together. Ethan, I think it's Malik. He's he's a pretty sensible AI engineer prompter researcher chap. He's worth. He's worth a follow on Twitter if you don't. He's got a sub stack too, which
is worth a with a subscription too. And he's a tech guru. No, he's a teacher. He's a lecturer at university in America. So he's not in the tech. He's a he's a one of us, not, not a tech guy one. Of US Independent Adjudicator, Independent Adjudicator, absolutely. Nice, excellent. Well, I mean to be continued I think, yeah, see another 12 months.
Well, it could be some sort of robot hosting the show by then. Well, fingers crossed, fingers crossed. Yeah, that's the other, Yeah, we haven't explored the the vocalization stuff yet, have we? No, that's for another time, because I want to see how we can work that in this episode never really happened. It was all just an AI voice. I mean, it's it struggles with my voice. It's got your itch. It's your voice down pat. Yeah, it does it pretty well, doesn't it? Yeah, but it thinks I'm American. Well, welcome to thinking deeply about primary education.
That doesn't sound like me at all. No, that doesn't. To be fair. All the set to do then is Seth. Thank you very much for joining me. Thank you so much for having me, Karen. Appreciate it. And everyone at home. Until next time, thanks for listening.
Learn more about The Story of Maths and ACEL Pro - www.alta-education.com/tsom-overview and...
Learn more about The Story of Maths and ACEL Pro - www.alta-education.com/tsom-overview and...
Learn more about The Story of Maths and ACEL Pro - www.alta-education.com/tsom-overview and...
Learn more about The Story of Maths and ACEL Pro - www.alta-education.com/tsom-overview and...
Learn more about The Story of Maths and ACEL Pro - www.alta-education.com/tsom-overview and...