Webinar Slides

Prompt Library for Educators

Prompt Framework: 5 “S” Model

Oral Exam Prompts

Summarize a Text Prompt

Lesson Planning Prompts Collection

Classroom Jeopardy Prompt

Apples to Apples Prompt

Amanda Bickerstaff

Amanda is the Founder and CEO of AI for Education. A former high school science teacher and EdTech executive with over 20 years of experience in the education sector, she has a deep understanding of the challenges and opportunities that AI can offer. She is a frequent consultant, speaker, and writer on the topic of AI in education, leading workshops and professional learning across both K12 and Higher Ed. Amanda is committed to helping schools and teachers maximize their potential through the ethical and equitable adoption of AI.

Well, welcome, everybody. I am Amanda. I am the CEO and co-founder of AI for education.

Amanda Bickerstaff: This is part of our AI launch series of webinars really excited to have you here today as always, we want you to get involved and thank you so much for everyone that's already starting to chat and put their where they're from in in the chat. But we want this to be an opportunity for you to learn from each other as well. One of my favorite parts about this webinar series, that they're now on 15 or 16 of lost count is to have the community that's been built around what we do. So

Amanda Bickerstaff: please take the time to, to chat, to ask questions, to have share resources. You're a learning community together and a global one at that. So please take that time also prompt library and prompt engineering. I want you to try this out on your own. This is meant to be as interactive as possible. I may not be able to see you, but I know that this is an opportunity for us to learn together. As I said, this is part of

Amanda Bickerstaff: part of our AI launch cut series, and so we're doing prompting zoom today next week doing top mistakes educators make with AI, which will be kind of common mistakes that we see through the work that we do. Both Harry Pickens and I and then how to combat that.

Amanda Bickerstaff: And then we're gonna be doing a session in November on evidence of learning which is a really nice counterpoint to the technology aspect, and we'll be adding more. In fact, please everyone, if you can do our our final feedback form and give us more ideas about what kind of webinars you'd like to see. That would be really great. And so usually, we kind of start where we're talking. We're kind of getting right into it. But we're gonna take a little bit of a step back and do something

Amanda Bickerstaff: technical. So if you were with me last week, you know that I don't like to spend more than 5 to 10 min in the professional development on the technical aspect. But I think this is really important, because what we're talking about today is like prompt engineering. And prompt engineering is a brand new field. And it's not one. It's both what we experience through chatbots. But also what's happening behind the scenes. So if you have a if you're using a chatbot directly, you are prompting that chatbot, if you're using something like a mess

Amanda Bickerstaff: magic school, or you're using. Use Jasper AI to do copywriting, or you're using all these different applications. Prompt engineering is happening in the background. So we're gonna start with a little bit of AI vocabulary. And so when we think about what we have here is that we've got a couple of things we should think about. So traditional AI or classical AI is gonna be really about more of this kind of pattern recognition. So

Amanda Bickerstaff: approach where machine learning is when a computer learns from data and then makes decisions or predictions based on the pattern identifies. And so if you think about this, you might have been using tools like predictive text or you have. We've had adaptive testing, or there have been other types of tools like writing tools or grammarly that we're able to learn for lots of data and then apply patterns. And so it's pretty much like very, very structured their algorithms

that were leading that

Amanda Bickerstaff: which is more traditional, classical. AI. And then you add some pieces about that, like natural language processing, which is how machines are gonna be able to read, understand, and generate human language

Amanda Bickerstaff: so that ability to say type, type, dive, and they can understand that I am writing words and that word part form sentences, and that there is the ability to create meaning from those words. And so that's a really important component of how these AI models work. And then the next piece is when we talk about large language models. So large language model is essentially what ChatGPT is

Amanda Bickerstaff: okay. And so these are these big models that are trained on a whole host of information like the Internet. And so GPT. 3.5 was trained on to text only. And GPT 4 was trained on text and images. And so they're able to like, take all this huge amount of data, and they generate human like text based on their training. So like, how they were trained in the sense of whether they ingested how humans were part of it.

Amanda Bickerstaff: They actually rate, the output, the parameters that are built in like all little rules in which they like it follows. And then the way that you prompt it. So we're gonna talk a lot about that today.

Amanda Bickerstaff: Then what we see here is the large language models like ChatGPT are considered to be generative. AI, which is that we? And we talk a lot about that which is that idea that this creates new content. So it's not quite machine learning anymore where it's very about about prediction and patterns like it's much more about the ability to create new content, like music or images or blogs or students essays which we hear a lot about is that it's able to do that through its training.

Amanda Bickerstaff: And it's something that we're seeing more and more interesting ways in which this is set to transform the way that we teach the way that students learn the way that you know technology works in general, the other 2 pieces I wanna talk about before we move on is the idea of temperature. So if you, if we all right now, collectively, there are 121 people now on this webinar all into ChatGPT. Right now, and entered one prompt like.

Amanda Bickerstaff: what are you, or what's the weather or sailors left on tour. What's gonna happen is the responses are all gonna be either slightly different or very different. And that's because the the model is actually designed to have temperature. Or it's a parameter that creates the creativity, the randomness. And so it means that you could ask the same thing multiple times, even the same context window, that same window. And you're gonna get different outputs.

And the reason why that's important is that when we think about prompt engineering, even if you have the best prom

Amanda Bickerstaff: that worked one day, it may not work in that same day. It may not work, you know, at all the next day, because it's going to be designed to have that creativity. And you can see that if you ever use bing, it actually allows you to pick a creative mode where it really goes very, very high temperature or a like boring mode. Essentially, I don't think it's called boring, but a mode in which it's not very creative. They could call it precise, that has a low temperature.

Amanda Bickerstaff: And then the final thing is, if you've been here before, you know that hallucinations are a big part of of models, and there's not any tool that's been based on these generative AI models are going to have hostinations in which the model generates. Something that isn't accurate, but can be very, very convincing inaccuracies. And I see that like at this point, you know where this is a little bit more technical than we usually do, but I think that it would be

Amanda Bickerstaff: like. This is the idea that we're coming together, and we hope you watch this again if you need to. And we're gonna present more information. But I think it's important for us to contextualize this information because it's something in which a prompt is actually really important to understand that you are a computer scientist

Amanda Bickerstaff: for the first time, you're a computer scientist, even if you've had no training. Because that's what you're doing is you're actually creating with language

Amanda Bickerstaff: technology and outputs. And so we're gonna look at how that works. And just a little bit more technical, and then we'll get to the more practical piece. But the idea of how this works is that every time you type into ChatGPT, or bard, or bing, or you're going to see is that you're going to create these tokens. And so every sentence like this sentence here. You know, ha! Has multiple tokens. And so that's where pretty much. It's not every word, but for every a thousand

words, is about 700 or sorry. Every 750 tokens.

Amanda Bickerstaff: Oh, man! Every 750 words. They're about a thousand tokens, because sometimes what you're gonna see is that they break up, the misspelled words will have actually smaller tokens, and then whole words will have their own token. And so the reason why this is important is that not every model can adjust, you know, like.

Amanda Bickerstaff: like enormous amounts of of tokens. But it's also how probability works. Cause. Remember, these are not things. I'm gonna put this on a pillow. But these tools are not thinking tools. They are predicting and computing tools. And the way that they work is that you take. You know, words like we just saw and you. But they become tokens and they get turned into numbers, which is how computers work is. They work with numbers, and then based on that, they go into the large language model, and then it predicts

Amanda Bickerstaff: the next word. So in this case, if I put in this like, please complete the sentence. This guy is

Amanda Bickerstaff: the way this would work is that that would now lead to

Amanda Bickerstaff: probability of what the next word is. And so in this case the next word is most likely blue, and it's going to give you that as an output. And this is how these tools work. And it's able to do this

Amanda Bickerstaff: at really like amazing skills. But it's doing it word by word. And so it's why, sometimes it could even make up words and so it could do things that are pretty crazy at times as well, because the probabilities can be off, or it could be a little bit confusing. And especially if you're not thinking about your prompting techniques, it can lead to things that are not going to have a very good output.

Amanda Bickerstaff: So now, like, why is this? We're gonna jump, jump into the hows and why's and what's of this? And so this is, gonna get us to the need of actually getting started and thinking about prompt engineering. And so the first thing that we're gonna look at is the why.

Amanda Bickerstaff: So the way you prompt a large language model affects its output. So if it's like all these words are becoming tokens and becoming well, you know, they're gonna go and they're gonna pee come a text array, and then it's gonna go there and it's gonna become a you know, probability that's really complicated. So I don't want us to necessarily get stuck on that like these. These models can be very complicated. But the idea of this is to really inherently kind of

Amanda Bickerstaff: situate you all, and thinking that it actually matters what words you use and matters what order to use this in? It matters in which, like what like, how you're actually formatting the request, and even how what you do after you request. And so it's just really important that that is why you can't just type in, you know, a sentence and expect a great output like you actually have to put some time and thought into this.

Amanda Bickerstaff: So that's the why. But what if it means that it actually is something in which you can become a good prompter and like to become a good prompter, means that you follow some, some like actual prompting techniques which we're gonna talk about today, including something called priming. You can provide the right examples using the right phrasing. And even what you do after that initial prompt is all gonna be part of getting to the best output possible.

Amanda Bickerstaff: which is something that we all want. We all want to use these models to give us something really meaningful for our practice or just for our students. So this is why it's so important to learn these skills. And then the final piece is like, now let's go and do it. So we have. There are tons of prompt frameworks. If if Vera is here. She has something that is great which is called craft. We have the 5 s. There are all kinds of different models.

Amanda Bickerstaff: but really, what we're gonna do is talk about a a way to get into the habit of asking good questions and creating good prompts. So that's the starting place. But because we remember talking about temperature is that they're going to be creativity built into the responses. So it means you're gonna need to experiment. And that means that you need to be patient, and you know and resilient in the experiment. And and we saw that I think it was one of our very first webinars, where I was

so like our rubric, prompt live.

Amanda Bickerstaff: and we had that prompt mess up 4 different times in 4 different ways, which is a good example of like even, you know, live on a webinar. These things don't always work as expected, and so it's going to rely on that experimentation and that patience and so these are the ways in which we're gonna do this. So we're gonna think about like how we set up a prompt. We're gonna show you a framework. And then we're going to actually think about how you work with this

Amanda Bickerstaff: together today. And so we have this 5 S framework that we have. And Dan's gonna drop this in the channel. We have 2 versions, one for teachers and one for students. And I'm gonna walk this through in terms of what this framework and how we designed it. And so the idea of this is that there are 5 steps you can take.

Amanda Bickerstaff: The first is that you want to always set the scene. And so it's really important to think about what you want the chatbot to respond as and so, if it is, the training is the entire Internet, it means that if you ask a very open, ended question, it doesn't necessarily know where, in that like how to respond and what you want it to respond, as so we call that priming. But setting the scene. So if you're writing

Amanda Bickerstaff: a stem lesson plan. If you're doing an English lesson plan. If you want to be a historical figure, you want to make sure that you give the chatbot that context.

Amanda Bickerstaff: So what role or expertise, or the environment it should use to guide its output? That's a really important first step. So what is this like you wanted to be taking on like Isaac Newton, or you wanted to take on the guys of Shakespeare an expert, or whatever this may be, you always want to start there.

Amanda Bickerstaff: The next is, you want to be as specific as possible. So you want to create this very. So that's a lot of specificity and clearly abiding the tap

Amanda Bickerstaff: and providing details or ex examples on what you would like included. So if you're doing a lesson plan, we we don't wanna just say, create a lesson plan. But we wanna say, use a 5 d model. We wanna say the length we wanna say it has to include these 5 things. If you're asking it to create an objective, you can even say frame it as students will be able to or another version that's appropriate to your context. But the more specific you can be.

Amanda Bickerstaff: and providing examples is gonna help with that output being more appropriate to what you want.

Amanda Bickerstaff: The next piece is simplifying your language. If you're spending a whole a whole host of time on doing this work and creating the very best prompt you're you're going to actually miss out on where the magic happens which is really using your expertise and continuing to prompt after the initial prompt, to to refine and get to the output that you want. And so we really want you to be conversational, simplified, but not actually create something that has, like a whole bunch of jargon, but something that remember this is about probability. So

Amanda Bickerstaff: so the more probable you're, you know, like the way it's framed, the easier it is to understand, the easier it is to give you something back that's going to be of use to you.

Amanda Bickerstaff: The next piece is structuring the output. So make sure it's like, if you want it to be in bullets, if you want it to be in a chart, if you want it to include specific components. That's gonna be really important for this as well. And so we we love like, you'll see. We'll look at a couple of examples where we show like actual. This is what we want included, and giving it very specific details about what structured format should be in.

Amanda Bickerstaff: And then the final piece is like, I said, this is where the magic happens. There's 2 pieces of magic. One is that you are the expert, and that you're going to be using these chat bots to get to something that may that matters for you. Especially when you're using it, one on one. But the other piece is that your your ability to keep prompting and refining and sharing feedback. Let's say that this isn't quite right, or I want this to change is really gonna be a great way to get to something that's gonna be of a lot more merit for you. And so we want you to think about this as we get started.

Amanda Bickerstaff: And I'm actually gonna call on a special guest which is really exciting. So Kelly is our our prompt kind of expert now. And so, Kelly, I want to introduce you. So Kelly and I met oh, man Kelly, like 3 or 4 months ago. Where Kelly is a transitioning teacher, and was looking to get into Ed Tech. We met at a networking event in New York City, shut up so New York City, New York, at Tech Meetup

Amanda Bickerstaff: and we had this conversation, and at this early stage was like, Hey, Kelly, like, I'd love you to come on our prompt libraries. I think I really like like really think has merit. But it's really hard for me to keep up with. So why don't you introduce yourself, Kelly, and talk about how what you thought so. It's been pretty technical, and I think this is more technical than we usually get. But, like, how did how did you get to a point from where I first talked to you about this.

Callie Pinkas AI for Education: where where you have no experience of AI to like, where you now are, building prompts pretty much every week for sure. Yeah, so exactly like you said, I'm a former educator. My background's in early childhood education. We met in at an edtech meetup. And I was very excited just to get going with anything. And yes, I had never used ChatGPT before, or any form of large language, model or any form of generative. AI

Callie Pinkas AI for Education: I knew what ChatGPT was, and that was the extent of it. So when you first were talking about all of this I was like, this is not for me. But you were so passionate about it, and so it made me excited about it. And now making the prompts is definitely the most exciting part of my work with you and AI for education.

Callie Pinkas AI for Education: yeah, I mean, do you want me to walk through the process of what I do. And you know, when I'm thinking about the prompt or yeah, definitely, yeah, no, that's good. And so how long does it take you to do? A prompt that's like, so we'll talk. Look at our prompt library, but we go pretty intense. We have like a prompt that is one that you can, that you can use one that's an example that you can plug in, and then some examples of how to remix for yourself. So how long does it take you to do that? Usually

Callie Pinkas AI for Education: probably about 30 min. I will say that sometimes I spend a lot of time sort of thinking about. You know the theme that I want to go with. Because I think that that's the most fun part. So like I said, my background's early childhood, and I've only taught Pre. K. But that's my time to sort of think. Maybe I want to be, you know, an Ap. Us. History teacher.

Callie Pinkas AI for Education: or in tenth grade, or, you know, a fifth grade French teacher or a seventh grade health teacher. And so, you know, I have fun with that. And so I'll really take some time thinking about that so maybe other people could do it in less time. Or but you know, maybe they're having less fun with it.

Callie Pinkas AI for Education: And then, you know, from there thinking about exactly. So sometimes I do have to do research. Your background is in science. So you probably have to do less research if you're writing a prompt about a ninth grade bio teacher. But I don't remember my bio. You know as well as you might. And so sometimes I literally have to research. You know what is Mitosis. If I'm going to write a prompt about, you know, creating an assessment based around.

Callie Pinkas AI for Education: you know, students learning mitosis. But again, once I've done all of that, then I think about the wording, we have a specific framework. You know that that we've developed priming. The chatbot.

Callie Pinkas AI for Education: you are an expert teacher. Specifically, you know the lens that we want the chat to think through and the wording, as you've said, really matters the order of the wording. There was one that I did recently that was actually through an early childhood lens

Callie Pinkas AI for Education: where we were talking about a kindergarten class learning, patterning and I wanted ChatGPT to produce patterning in the real world, but the way that I had worded real world. It was giving me examples that were.

Callie Pinkas AI for Education: you know, ways that I could use real materials in the classroom, which would be great for a teacher if I were still in the classroom. That would be great. I could use candy, or, you know, animal figurines, but I wanted to literally have it give me ways that I could point out in the real world patterns. So then I had to change the wording, still using real world, and then it eventually gave me, you know.

Callie Pinkas AI for Education: animal stripes stripes on, you know, houses or fence posts. So you do have to have patience, as you said, and be able to sort of, you know. Now I could see exactly. I was like, oh, it's this, you know, wording of real world. But maybe when I started that might have taken

Callie Pinkas AI for Education: some time. But with practice it's only been a couple of months, and you know, so you really can learn it very quickly.

Amanda Bickerstaff: It's great. And I think just to kind of call out a couple of things that you know when we, when we build these, we're doing like the most right guys like we're doing the most because we want you to be able to use it. And we want it to be really usable. And I think that this is where it gets really interesting because it takes us 30 min, or maybe a little bit more to kind of come up with the idea and to try it and to try it on multiple chatbots.

Amanda Bickerstaff: But for you like like, it's gonna be a lot easier to take this, and like either start with our prompt library and remix it, or to use the framework to do that. And I think that that's where the power really comes in here is that it is something in which, like, you know, we're giving you like, even with just a couple of changes, it can really, radically change the outputs. And that's why we spend so much time on this. But you getting into that habit or teaching your students about

Amanda Bickerstaff: it. It should just make you really much more efficient.

Callie Pinkas AI for Education: So what we're gonna do is Kelly, can you bring up your screen? And we're gonna look at an example that we did recently. This is actually so. It was quite funny. Kelly and I were talking before this. And we first started, I think, when you came up we had like 30 prompts, and now we have 75, and Kelly was like you. We have them all. What could we possibly create? And but we've we've created a whole. I looked at them. And I was like, Okay, you have rubrics. You have assessments. So you know II feel like you. You did it.

Callie Pinkas AI for Education: And then the more we talked I was like, Oh, wait! Reference letters! Oh, wait this wait, parent teacher, conferences.

Callie Pinkas AI for Education: the part of a lot of applications

Amanda Bickerstaff: exactly. And we like, I think, every once in a while I just like dump like 10 ideas on call like people that I talked to, and the things that I see. We bring in. And, Kelly, there's a question from Carl. Do you use AI to help with the prompting process. Reverse. I do. I didn't at first, and then the more that I got familiar with it and got comfortable with it. Now I do, actually, and you'll see. So in the one that I was going to pull up. That's a great question. So

Callie Pinkas AI for Education: can I answer your question while going through this one? So this was one that I did for an oral exam. And this was close to my heart because I was a French student. And I did have oral exams. So you know.

Callie Pinkas AI for Education: really put me back in the day? So this was the question. Can everybody see this? Yeah, you are an expert t-shirt. So that's the priming skilled in designing and administering student assessments. Again, part of the priming. Create a 6 question. So when we have this on the prompt library, these are blank parts that you can insert in. And this is part of the testing, so that you know when you do this for yourself, it's all. All the kinks are worked out. So it's really easy for you

Callie Pinkas AI for Education: blah blah blah ap French, they're studying Candide by Voltaire.

Callie Pinkas AI for Education: So then it spat out this immediately. The first problem that I noticed was when I took advanced French the questions were not in English, and you know, fair point, I didn't specify that, so I went back. I read through the questions. We're fine, but I went back and then respecified. The exam should be in French.

Callie Pinkas AI for Education: but just in case, you know.

Callie Pinkas AI for Education: provide an English translation.

Callie Pinkas AI for Education: I went back. It was great.

Callie Pinkas AI for Education: Everything's in French. But then, like you? Asked Carl, was it

Callie Pinkas AI for Education: so then I went back because we do have a section make the prompt work for you. So then I went back and used ChatGPT, what do you think? How long do you think it'll take the students to answer this question. And when I first started doing this, II was really doing everything like.

Callie Pinkas AI for Education: what do I think that people will want, you know, when they're using this. But honestly, if you're interacting with the chatbot.

Callie Pinkas AI for Education: you know, why not use it to help you? And honestly, II found that it's really been helpful. So you know, I went back. Here's the time, you know. I now I can't remember if I used that as part of the make the prompt work for you. Maybe, I said, you know, think about consider the time that you know the students may

Callie Pinkas AI for Education: may take for each section. How can I help my students prepare for the exam. So this was definitely one of the, you know. Make this prompt work for you. Think about how you can help the students and ask the chatbot how it can help you. Because these were great, read and discuss the text, blah blah, blah! A lot of these you probably would have already done for yourself. You know. Practice discussions great. Obviously, it's an oral exam you're going to be, you know, speaking to do that. But maybe

Callie Pinkas AI for Education: some of these, like a mock oral exam or real life connections. And then also, another thing is okay. If that struck your fancy, real life connections. You can go back and say, Okay, regarding number 9, can you, you know, elaborate on that? Give me some examples of real life connections. You know some of these might not be applicable, or might be some things that you've already thought of.

Callie Pinkas AI for Education: That's fine. The idea is that you know Amanda talks a lot about having it be a thought partner. It's not going to solve

Callie Pinkas AI for Education: everything. And it's not going to, you know, always have unique ideas. It's just to sort of get you thinking, and and, you know, be additive in that sense. And then this was an interesting thought. As well. Can you help with pronunciation. II wasn't sure

Callie Pinkas AI for Education: if it would, and it did. A phonetic pronunciation which I thought was kind of cool this could day, you know it was pretty good, actually, it. It wasn't exactly realistic, you know, realistic that you would use this I think that someone teaching at that level would probably be able to pronounce this, and this wouldn't really be, you know.

Callie Pinkas AI for Education: the the best way of using this, but it's good to know that that it was able to do that, and pretty accurately, I thought, and then just some other. What might other things be to consider if I were doing this? Yeah. So I think that's a great question. And I definitely do.

Amanda Bickerstaff: Yeah. So Janet actually had a great question about how confident are you in the translation to other languages. And we talk about this a lot. This is not ever sorry I get it coming back. This is never where you want to end ever. And so like. This is a great starting place, though, that you use if you are a expert like in the sense of, you know you are building this out, but you need help getting started. Then this is where you can

Amanda Bickerstaff: at least get the. You may not have read candy that closely in a while, or you're wanting to get like an extra set of eyes. So I think it's definitely something in which what we want to do is like, have it be something in which it's always going to be up like a a brainstorming partner, a place to get started, but especially with languages. It does a pretty good job of basic like, very kind of more common languages, but it's going to do

Amanda Bickerstaff: worse as the language becomes less common. But again, that's pretty much for everything, though, with ChatGPT and other tools, is that you're going to see it. Remember, it's a predictive in a like a probability engine. And so the less

Amanda Bickerstaff: and knows about that through the training it has the worse the outputs are going to be. So this is just a good example. The reason why I like this one is that I think before we started we didn't. We never would have thought that we would get to something. This close like this complicated multiple languages structure output with 1, 3 sentence prompt. And so we wanted to show this one cause. It's a good example of how. The more that you do this and the more expertise that you use.

Amanda Bickerstaff: you build, the more the more kind of interesting, unique novel applications you can use and so in in terms of like Harry's question, Hi, Gary, he's on our webinar next week. And so the idea that, like, you know, we using we don't actually use ChatGPT to build the prompt and I know that some people will do reverse prompting where it'll ask you to do that reverse prompting. We don't also don't use that's something like prompt, perfect

Amanda Bickerstaff: and so, in a sense of that's where it's giving you like the perfect prompt, because those prompts tend to be very, very large, and they often tend to work better with the non free version. So we try to create something that's understandable that gives you like, 80 and that's going to be something that's unique, based on our approach.

Amanda Bickerstaff: because it is something. If you, Kelly, do you want to change over to the prompt library? We've created this tool in 2 ways. It's not only for you to go and be able to have a starting place. But it's also androgical or pedagogical tool cause the way it's structured, like you can clearly tell our educators to the way it's structured is it gives you? I gives you a thing that you can change if you want to go to, maybe unit or lesson objectives.

Amanda Bickerstaff: and then what we have is we have the prompt. This is a very simple one that can be done. We've got the example. That's the scaffold.

Amanda Bickerstaff: And then we've got to make the prompt work for you, which is the creativity. And so the reason why this is created this way is that it is designed to be about building as much AI literacy as it is about the perfect output cause. We know that there are no output that's ever going to be perfect the first time, because your context and expertise is absolutely key in that piece. And that's why we like this ability to do this work. And so it's gonna answer a couple of more questions before we go on.

Amanda Bickerstaff: We so chats to be compared to bard, I think, for something like this. ChatGPT is gonna be significantly better. We use quad for some things like we use quad for more communication. We use, Claude, that does better on larger context windows. We have a bigger piece of text. You want to parse. We have one. Kelly, do you wanna bring up the version, the prompt that we did that you need to actually upload a piece of content.

Callie Pinkas AI for Education: Yes.

Amanda Bickerstaff: while we're doing that. And then so in the case of just one more question from hope so definitely ability to to reasonably predict time.

Amanda Bickerstaff: even though Kelly asked it, that didn't get into the final prompt because it does a pretty poor job. It actually can't. It's again, it's predicting. So we just like it was more of like kind of seeing the thing. So we definitely do not

Amanda Bickerstaff: We don't do that work. Like if you wanna come off share what we look. That's okay. Kelly don't. Wanna don't wanna give anybody like I get. I get Carsick a bit. No, you're perfect. And so but this idea, though, that like while we're yeah. So while we're doing this, we're actually kind of navigating the edges, so to speak, hope so. I always think it's really funny, because, you know, some of our

Amanda Bickerstaff: prompts will be like supposed to be 5 to 10 min, and then it'll give you something that's like a full lesson. So again, it requires your expertise. And that's why it really can get to the point where, if you keep doing in that refining piece, it's gonna get better and better and so everything that happens after that first prompt is part of this process. So I'm gonna actually, while we're doing this.

Amanda Bickerstaff: yeah, to Aaron's point like the timing. I always love it, you know. We have a 90 min, you know, 4 90 min lessons, and it's gonna be like 3 h in some schools and and 30 min in other schools and other spaces. So that's actually something that that messes up quite a bit, but would definitely not be a used case for prompt engineering that we would use all that much.

Amanda Bickerstaff: So I'm just gonna open back up the the kind of resetting this because we also have a version of this for students. It does the same thing. But I just wanna kind of call out. And like, we're just gonna do a bit of repetition by learning is that when Kelly showed you how she creates it.

Amanda Bickerstaff: he's always going to prime always. We do not have a prompt in our library that doesn't have priming, that is, that it's cutting the pie, so to speak, and making it a smaller amount of of, you know, a smaller amount of of knowledge, so to speak, of training that that's going to navigate. You got the specificity which we saw that the more specific that you are, the better the output would be.

Amanda Bickerstaff: Kelly. It's not a very particularly challenging piece of of language. It's very simple, and it uses keywords like oral exam. And it's using a a very kind of common technique that's going to be in a lot of the training data structuring the output to have both English and and French as an example, and then providing that feedback and keeping doing it is really gonna be where you're gonna see the most value. So I'm gonna I'm gonna actually go to our, he did. Okay, yeah, do you wanna go back and do that

Callie Pinkas AI for Education: share screen.

Amanda Bickerstaff: So so, Claude, is anything so, Claude, so we only build for free versions. So there are witness, not Mega prompting. This is not which are really great. These are not like crazy, crazy, complicated prompts that ask you questions. But what we wanted to do is make sure this is equity based, and this is accessible by everyone. So the only tool right now. Well, not the only tool, but the most commonly used tool right now, where you could upload a document and have it

Amanda Bickerstaff: should be part of the response is going to be Claude. And so we we have this example. You wanna come back up and show this to summarize text example. And you could cut and paste text in this happens a lot. And so like, if you wanted to cut and paste an article, or a Youtube Transcript, or something on those lines. You could do that with chatGPT or Bing or barred. But we wanted to do is show an example. So let's go to Claude. And then

Amanda Bickerstaff: we can like.

Amanda Bickerstaff: pull up. Yeah, yeah, let me do it. Okay, so we're gonna do. We're gonna go to Claude.

Amanda Bickerstaff: I'm gonna go here.

Amanda Bickerstaff: gonna come to. Can you see my prompt library? So I'm gonna go to this is summarized text, which is gonna be in. I think it's in our lesson planning. It's also in the students, one as well.

Amanda Bickerstaff: I'm just gonna do summarize.

Amanda Bickerstaff: We have a lot of prime snow.

Amanda Bickerstaff: We have 75 prompts to get a little bit big guys. But here, let's do summarize.

Amanda Bickerstaff: Yep, there we go. And what we're gonna do is I'm gonna pull this prompt. Remember, this is the example that you can change for your own piece. Here are the examples how you make to. We actually have identified Claude or ChatGPT 4 cause there's only 2 that you can do that with. I'm gonna go to Claude.

Amanda Bickerstaff: If you're not use cloud. Claude is a tool that's been designed by anthropic, and it's it's human. First approach. But what you can see is right there you can add files, and you can add a series of files. If I put this in your expert student with expertise and summarizing and pulling out important sections of a test. So it's gonna be, I'm gonna go. And I'm gonna find a you know. Gonna go and do

Amanda Bickerstaff: text. Webinar, shoot blog on

Amanda Bickerstaff: building a website. So I'm going to go to this. And I'm going to say how to build a website, a beginner's guide. And so I'm gonna

Amanda Bickerstaff: I'm gonna take this. And I could do something where I there are plugins where I can do this, and I'll go, hey, like

Amanda Bickerstaff: you know. Take this text. But what I'm gonna do now is I'm gonna cut and paste this into a document. Sorry. That's gonna give everyone whiplash

Amanda Bickerstaff: but I'm just gonna do this here. I'm gonna put it into a document. And you may always remember that we're these are important to create you know, to think about

Amanda Bickerstaff: copyright and those types of things. We don't want it to necessarily be used and publish this. But if you wanted to help this and summarize this text. So I'm gonna do is I'm gonna

Amanda Bickerstaff: save it as a export to a Pdf which it works. It works really well. Pdfs

Amanda Bickerstaff: which is nice, and I'm gonna put this in, export it. And then. Now I'm gonna go back to Claude. What I'm going to do is I'm going to upload this document. You can see my slides for today should have did it.

Amanda Bickerstaff: It didn't download. Hold on a second, everyone. So let me just come off sharing. But while I'm doing this like, what I what I'm doing is I'm gonna take this text. But hopefully, it actually

Amanda Bickerstaff: hold on. Just wanna make sure export to Pdf, and then next. technology guys super fun replace. Okay? So it should be there. Now.

Amanda Bickerstaff: one moment I'm gonna share my screen again

Amanda Bickerstaff: and gonna go here. There we go entitled to, and I'm gonna upload and says uploading it. And so now, what it's done is that's building websites. You wanted to read the text. You wanted to summarize the text and identify the main idea steps and key vocabulary. So this is a quick and dirty version of this. But what we're gonna see now is that this is a step by step

Amanda Bickerstaff: it's got covers choosing a blogging platform and domain name, selecting a template, adding important pages like about and contact. It gives you main ideas from the text.

Amanda Bickerstaff: So pick a specific niche, choose a template that fits your bog style, etc., the key steps, which are gonna be similar to that and key vocabulary. And we could use this actually have it defined. The vocabulary

Amanda Bickerstaff: as we keep prompting. And so this is an example of like a very structured prompt that gave us week. It gave us back it, summarize the document. We can double check it, but that's something to see. It's it's then identified the main ideas and stuff's in key vocabulary. And so it. And then it didn't bullet points. And so this is something that's been structured. So this is an example of like using another free version. And

Amanda Bickerstaff: so Rachel Richelle says, like Tldr, use that a lot. But in the case like you know, this is something in which it does a good job of of creating a space which is not always easy to do, which is kind of summarizing. Let's say that you have English language learners, or you have students that need extra scaffolding that you wanna be able to kind of ha have an opportunity to have something a little bit simpler. This is a great example of doing that I know that Claude isn't available all over the world.

Amanda Bickerstaff: So what we can do is we can take this text and upload it in chunks potentially with other tools. You can also get around it with you know. So Poe uses Quad and others. You can use play lab or other tools to create your own. But it definitely is something that

Amanda Bickerstaff: it's a good space. And I know this is about equity as well. So we, I wish we could make this publicly available. But you're going to see that there are. They're also plugins and websites that you can do this work. But this is a great example of even like a Youtube video that you want to create a transcript for as to see that Vincent just said on a similar, he does a transcript with his recorded classes. And then Cloud helps and make a mind map for his students to reflect back on. And that's gonna get easier and easier with with multimodal

Amanda Bickerstaff: that's coming out. So that's really cool. Thank you, Vincent, for sharing that. So I'm gonna go. And we're gonna kind of wrap up this section with some really interesting new things. So I'm gonna actually, I know that Cali and I you know, we we try to keep it as simple as possible if it's replicable as possible. But there's some cool things that are coming.

Amanda Bickerstaff: They're not the most reliable, because the models actually change quite a bit over the course of like the usage of this. But there are a couple of areas in which you can get to my documents. Well, that's like I'm off share before you guys see my Doc, nothing in there but like but I wanna make sure that like we're pulling this up. So let me pull back the

Amanda Bickerstaff: there we go. Here is our slides. There are some kind of interesting

Amanda Bickerstaff: new techniques. And so that you might not be aware of these. But there's some techniques of actually using some phrasing and some words in your prompting that academic papers has found to be successful in improving the output.

Amanda Bickerstaff: So things like, be concise, is actually a lot of because this is really expensive. So remember, prompt engineering is you creating with technology through your prompting.

Amanda Bickerstaff: But also the tools that you're using are prompting in the background. And so it can get very expensive. So if you have a really long, pro prompt, and a lot of like a really long set of words. It costs a lot more money, those tokens, Eric, you actually have, like you, pay per token, so to speak, and so the more concise and output is, the better it is for the like the cost of it. So this is something in which you can create something that is supposed to have higher value, or because

Amanda Bickerstaff: better output. There's a new one that's been shown is that? Take a deep breath.

Amanda Bickerstaff: It's another example. So actually asking it to slow down has been and take a deep breath, is has shown to have an impact on the quality of output. And so I've tried this a bit with our with our prompts. I haven't seen enough of it yet to to add it. But this is something I think, about a lot, and I used in my own kind of prompting

Amanda Bickerstaff: the next one is to ask. Ask it to ask you questions. So Harry and a couple of other people are like, Well, how do you? You do like reverse prompting? The Carl asked as well. So you could do something like this where you say I'm gonna I'm writing a lesson on you know. Seventh grade recycling.

Amanda Bickerstaff: What do you need to know, to help write that lesson. So ask me questions, and then it will provide you the questions that it needs. And then you say, okay, take all those answers and then create that lesson.

Amanda Bickerstaff: So that's something you can also do to try. Another really interesting one. And we'll like these are all like, you know, to publish, which is really fascinating. Is this idea of this is important for my job. So like it actually like saying that this is important, like having an emotional kind of part of this. It says, like, this is really important for me. This is something that's, you know, gonna get a make or break my job. I could get a raise like all these things. But making like an

Amanda Bickerstaff: emotional plea, actually can create a better quality of output. And then the last one is, explain your answer. So actually having it explain the answer that it gave you also can do supposed to help with the quality of output. So these are kind of other prompting techniques like we believe and like, we're very comfortable in this

Amanda Bickerstaff: where you're, you know, doing this. But if you wanna kind of try and please like this could be a really fun lesson for your teachers. If you're working in a Pd or your students, or you're like, okay, we're gonna try these different piece and can clear and contrast. Does it really like help to ask, take a deep breath. And how funny does that like? What does that mean? And like having like kind of this engagement piece, where you can see, like how this works, knowing that, like these models change quite a bit, and they don't always work very as expected. But they can be pretty fun.

Amanda Bickerstaff: And so the last thing I'm gonna show before we go on is before we kind of go to questions is something that also can really help with prompting. And so we want you to use our prompt library. If it's a starting place. If you have great prompts.

Amanda Bickerstaff: send it to us, we'll we'll work with you to create one. We have plenty of prompts on the website. We have been supported by other people. But if you go to chatGPT, there's a new. There's a thing that's not super new, but it's relatively new in the sense of a way to get better outcomes. So you know, we have ChatGPT and I have both versions. Kelly only has 3.5 right now, cause we really want this again to be accessible. But if I go down here

Amanda Bickerstaff: and I go to my settings. There are a couple of things that I can do which is my I can do. Custom instructions I could do settings if you ever, if you're ever using something for sensitive pieces, you could clear all chats you can create, you know. You can make sure that you can turn off chat history and training. I suggest that if you're not gonna be like coming back all the time, or you wanna use this in a private mate way. But then, also, there is the opportunity to set a custom instruction

Amanda Bickerstaff: and so customize your interactions. Chat to Youtube by providing specific details and guidelines for your chats.

Amanda Bickerstaff: Whenever you edit your custom instructions will take effect in all new chats you create existing chats won't be updated. Okay, so this is where you can actually put in. I'm a high school biology teacher. Oh, okay, sorry. Oh, dang it again! Oh, man, I look out like Aaron yelled at me. Amanda again. You can't actually yell at me. But luckily Kelly is here to yell at me, which is good. I love that that's so funny.

Amanda Bickerstaff: okay, can everyone see this now? Okay.

Callie Pinkas AI for Education: yes. So this is a custom instruction. So funny.

Amanda Bickerstaff: You know what guys technology? It's easy, you know, just webinars with, you know, couple of 100 people in it. So it totally works out everything perfectly. But I'm gonna go back here. So at the bottom, you're gonna see custom instructions. And then what you can do is this is where you can enter. It's only gonna effect that comes after. But I'm gonna I'm a high school biology teacher in the Bronx

Amanda Bickerstaff: that uses a 5 E model common core

Amanda Bickerstaff: State standards and and Gss. Standards.

Amanda Bickerstaff: and what I can do is now I could even put in here like I use this

Amanda Bickerstaff: output or structure. or my lessons where I can cut and paste it in. I can put in so much information about me. And then what that means is that the way we prime is suddenly. Now, if I'm using this primarily for teaching or for instructional design or for students. What it will do is it will cater its output to get better instructions. So it's like always priming it for you. And so in this case, like like you can say and like. And if you want to go from like

Amanda Bickerstaff: teacher world, so I wanna be the best travel planner I'm gonna be. Whenever, however, you use this tool or financial model, or you can do that as well, and you can also do how you would like it to respond? How formal or casual, how long or short should the answers be? How do you want to be addressed so like like Amanda? Or it could be something where like, I don't know, you can be a superhero

Amanda Bickerstaff: to be a superhero, and then, should we have an opinions or remain neutral, so like what you're doing. And so in this case, if you're using this in a classroom, what you can do is that this is gonna be something that you could actually create and prime before you go in where you don't want it to be formal. You want it to be very sure of its answers. You want it to remain neutral, you want it to avoid certain types of pieces. You wanted to introduce you as Miss Bicker staff, or if you taught in the Br, the Bronx, miss, and never had the last name and 3 years of teaching there.

Amanda Bickerstaff: So the idea here, though, is that you can actually go in and only gonna affect your chats going from that moment. But this is an example of more kind of advanced prompting that you can do. It's readily available, but we often miss it, cause we're so much is happening. And so this is just an important way to start thinking about how you can continue to do this work and that you are. You're going to be in a position in which you can make this work even more for you, because the only

Amanda Bickerstaff: ultimate goal here. All of this, like everything that we do at Ia for education, the ultimate goal is to make your lives easier as teachers, to create spaces which you can personalize, learning that you can support your students, that you can build AI literacy. But like, really like, the idea of this is, how do you make this work for you? In a way that is meaningful and tactical. So what I'm gonna do is I'm gonna kinda come back and all these slides and everything that we have

Amanda Bickerstaff: on the website. Dan's gonna actually put in the feedback form for today. Kelly's if we couldn't figure out the QR. Code. But you can have Kelly's career code as well for Linkedin connect with us there. But the ultimate goal again is that when we think about prompt engineering. It is you creating. You are a computer scientist.

Amanda Bickerstaff: Kelly, you are computer scientist. And we have actually had conversations. Or Kelly has not believed. She's a computer scientist. And I have to say, you know more about creating these prompts than most people. And this is the opportunity to really create spaces in which you are actually able to create something with technology in ways that have never been possible. And so everything that you think about is really going back, and it sounds silly, but like just something as simple as this piece.

Amanda Bickerstaff: and then using our prompt library or your own expertise will really help you get to a better output.

Amanda Bickerstaff: So I'm gonna stop there. And for those of you that have to go, I know we're we try to get you guys out at the 45 min mark or a couple of minutes behind, and I know we have people from all over the world. So please like, go chat you know. Go to sleep, or whatever you need to do, but if you wanna stay in chat for a little bit we'll stay on and answer a couple of questions.

Amanda Bickerstaff: if you have any. If you don't, we really appreciate you. You guys are the reason why we do this like we want you to probably find value and thank you for sticking with me with a little bit more technical at the beginning. It's definitely more technical than we usually get but yeah, if you have any questions at all, please let us know but maybe, Kelly, I'll just ask, like, what is the favour like when you think about the prompts that we created. What do you think is the one that's your favorite?

Callie Pinkas AI for Education: I think the apples to apples one which I hate to say is another French one which honestly has nothing to do with it, being French but it was just so different, I think, you know, like we said when

Callie Pinkas AI for Education: we started, and you were showing me the ones that you already had. They were very straightforward, and there's nothing wrong with that. Those are the ones that's like the bread and butter, the ones that are, you know, clicked on the most. But I had never imagined or conceptualized, that that could be something that

Callie Pinkas AI for Education: you know, that those are the kind of things that as educators, the fun things that you want to do with your students because you want them engaged, and you want them learning the material and remembering it. But I would have never thought that a Chatbot could help you

Callie Pinkas AI for Education: create something like that. And it really did. If anybody, you know, knows that game. It did a really great job. And same with the jeopardy one which was not a language, one that one was either a history or a science one. The science ones are really easy. To conceptualize as well. But yeah, same thing. Really fun framework for a game. Yeah.

Amanda Bickerstaff: yeah, I think this is where, like with multimodal chat duty coming out like the apples apples, game you could actually like have it build an image to go with the text.

Amanda Bickerstaff: and like even have the text. So Dolly 3 is coming out, which is gonna do a better job of creating and refining, and that you can include text and I think that that's where you take this. So we're definitely gonna be doing like a pass, especially when multi mob comes out and it's more widely available. But it is gonna be so cool to see how much more you can do. But I think this all this ultimate goal of exploration and experimentation and patience is where you can start to push and pull these systems

Amanda Bickerstaff: into ways in which they could help you, and that you didn't think are possible. And I think that this is where I get really excited, and that, you know, we're where we're doing this work, that there is this opportunity to keep

Amanda Bickerstaff: trying new things. And and you know there are so many great like applications out there. But what I love about our prompt library and the work that we're doing is that this is you. You get to be the you are the actor, you are the driver, you are the expert. And then what you're doing is you're using the AI to support your context, your unique needs cause. I love it so much in terms of so like Carl saying, like you're doing a Pd. Session for university staff.

Amanda Bickerstaff: And I think it's so great to like you can then say, like, it's gonna be like, only for our contacts. It's gonna be an Eli version, or like an English version. It's gonna be a research version. It's gonna be an adjunct version. And that's where it gets really, really interesting in terms of

Amanda Bickerstaff: how it's going to be something that's really powerful for you. And it's so. It looks like we are the prompt one. It's so funny apples apples. We'll get that to use in a second amber. But if you just go to our website and it's gonna just be prompt library collection to apples apples, but we'll get that out there. But I think this is where we get to the the idea really fast that this is something that there's so much power. And your expertise.

like your expertise, is the power here.

Callie Pinkas AI for Education: I think it'd be really cool, actually, cause that, you know, that is the fun part like creating the examples. But then we remove everything and test it so that it's easy for people to access. And I actually think it would be really cool if people wrote in with the ways that they use it. If they do so, people use the apple's apples, or if they use the jeopardy, you know, for however they use it, whatever they you know, whatever subject or topic you know, that'd be really cool cause that would also give us a sense of

think of how people are using it. Or,

Callie Pinkas AI for Education: yeah, that'd be helpful and very interesting. Yeah, absolutely. I'm gonna throw this one into this guy. There we go! Hold on!

Amanda Bickerstaff: Here you go. So maybe what we'll do is like we'll actually do a piece where we'll put something out there in which

Amanda Bickerstaff: we can have like some some work together where we can look at actually tapping and like like brainstorming like these different ways. Because I think this is where we get so excited. About this work. Is that alright? You know, it's so funny. I actually just put the wrong one in. But anyway, we'll send you the apples apples tomorrow. But I wanna say, like we're coming up on time. And so first of all, I just wanna say thank you to Kelly.

Amanda Bickerstaff: who was very nervous before she came on today, and I keep putting her in situations that she's a perfect example of that like believe and growth, mindset and yourself, because this is something where I don't think that this is on her Bingo card for this year and working with AI for education, helping build prompts. But thank you to Kelly for joining and sharing your wisdom, and we'll continue to create great prompts. Secondly, I wanna thank everyone for joining. We have so many great people, and thank you for

patience as we continue to work. And again, thank you so much for your patience that little bit of technical at the beginning, and I just wanna say, you know, have a good night or morning or middle of the night. Please go to bed if it's a little night, and we'll be posting this tomorrow.

Amanda Bickerstaff: following up with an email so you can share with your colleagues and rewatch, if you'd like, and always connect with us. We really care about you all, and hope that this is helpful for your practice. But thanks to everyone here, thanks to Dan, who helps, and thanks to Cali for being such a good support.

Amanda Bickerstaff: thanks everyone.