 So to start off the day properly, I would like to do the land acknowledgments. I'd like to acknowledge that UBC is hosting this symposium and is located on the traditional ancestral and unceded territory of the Muscovite people. As we're meeting here virtually today, I'd like to acknowledge that we here in the lower BC mainland are also often on the unceded territories of the Squamish, Slewa Tooth, and other Coast Salish peoples. You may be joining us from many different areas and I'd like to take a moment to appreciate and consider and give respect to the lands on which we are situated. I appreciate the land where I am as it provides me with many opportunities. And when I acknowledge being on the territory of the Muscovite people, it's rooted in an understanding that I, as a resident of Vancouver and a member of UBC, am privileged to be learning and working on territory that is not my own. A roadmap for today in the bottom right is our symposium website. Will, if you'd be able to maybe drop that into the chat so people have that, the slides for the sessions today will be posted there. We have something that I'm still quite a starstruck that we managed to pull off for that, or that that we, well, we have Sven Torter, who is the CEO of Jubal and the founder of H5P is going to be coming today talking to us about H5P in the age of AI. Afterwards, you're going to be hearing me rambling on for another hour, hour and a half and creating an effective feedback before we take a look at how we can do that within H5P. And then 1130 to 1250 will be community chat, very much like some of that we had yesterday. And then we're going to have a closing plenary, which is looking at a panel of faculty here at UBC from various faculties showing how they use H5P in their classrooms in new and inventive ways. So without further ado, I'd like to introduce the CEO of Jubal and creator of H5P, Sven Torter. I was doing some research on my introduction today and was blown away when we look at just how extensive H5P is being used. There are over 200,000 installations and over 200 million users worldwide. UBC is just another one of the 433 universities that is using H5P and this spans 42 countries. So when we think about educational impact, this is one of the tools that has had just an astounding impact on how we are able to provide content to students that make use of best practices of things like active learning, feedback, flexibility and creating multiple choice questions, giving us flexibility and being able to branch. So it's a tool that allows us to do our job better. Now with generative AI coming on the scene, this has opened up several other areas of possibility and I would like to now welcome with immense pleasure Sven Torter to talk about H5P in the age of AI. Sven Torter, thank you very much for coming. Thanks a lot Simon, thank you for the introduction and thank you for inviting me and most of all, thank you for hosting this event. It helps spread the word about H5P and also helps people be better H5P authors and H5P contributors. So I'm going to share my screen now and we'll be talking about generative AI and H5P for the next hour. I'll also share a link in the chat to my slides and the things I'll go through today so that people can follow along and they can also go and revisit later. I'm trying to make this very practical and useful for those who listen in. So the vision of H5P is to empower everyone to create generative use interactive content. We've had the same vision from the start and it's very important for us that we are many people together like here who can join in on the vision and do things like this and spread the word. We don't have any marketing budget or marketing people. We had one for a few months but it didn't really work out for us. It's a community movement. So it's us here on the virtual table who are driving H5P forward and these types of events is the heart and soul of H5P. And then what happens when we meet generative AI? We can certainly create content a lot faster with AI as an assistant. We're going to look at some very concrete examples of that and you can take with you a lot of recipes from this session on how to use generative AI to build H5P content. It makes it much easier to build a rich content repository. So we have the H5P hub which is in alpha stage now. It's possible to use it from Moodle, for instance, but it's not fully launched yet. Generative AI will make it much easier for us to populate the hub with content that is built with AI and built by humans as well. So quality assured by humans but the process is sped up by AI. And I think that is important. I think some of the AI solutions that exist for H5P including one I'll show you that H5P core team has built are among the very best AI solutions in the world but AI is not at all able to make perfect results so needs to be a good mix of human and machine. Machine can only speed us up. It doesn't do the workforce. So what are the results when we have better AI assistance for learning content? I think one of the big dreams within online education is more adaptive content, having one content item, being able to serve both the strongest students and those who need more explanations, serve things in a visual way, in an experimental way depending on how the students prefer to learn or how the teacher prefer to teach. But having one content that can do it all is too expensive today. With AI we can achieve those super smart content items that can deliver a customized learning experience for each individual or each scenario. We can cover more with interactive content. So I don't know exactly how it is in Canada. I know in Norway that for both public projects and also private publishers, etc. they are not able to cover the entire subjects with interactive content. Typically maybe the biggest courses but most of the time you either have to pick between very simple interactive coverage or high-quality coverage for just a small percentage of the subject. You are not able to cover it as well as you want with AI. We can do that. So what we hope to get this is that we can have much better content, much faster with AI and the end result is better learning. There are lots of research papers as you probably know about, how much better interactive video is than normal video and if AI can help us with that, it is tremendously powerful. So it can help us assist and create drafts. We will see that today. We will have the AI create some drafts for us. It can handle mundane tasks like putting definitions on the right places in an interactive video. We will see that also today. AI can speed up ideation We will see examples of how AI can come up with ideas for us on how to build higher-order learning activities and AI can also be used to improve language, etc. But super important that authors remain in control. I won't talk a lot about this. I think most of you are familiar with these things already so it will be very practical for the rest of the session. This is the roadmap for the H5P Core Team 2024 when it comes to AI. So we are building and maintaining prompt recipes that people can just use for free. They are just below. We will go through them today. We've had some of them earlier, but for this session we've built a complete set of recipes for all the things you can easily build with AI and H5P. We have recipes now that we'll have a look at. You guys have links to the page so you can use all these recipes to build content yourself. We will be maintaining them. That's actually important because if you use GPT, for instance, it keeps changing. The recipes that work today might stop working in a month because GPT is interpreting things differently. We'll be keeping up with GPT and trying to make sure that these recipes always work. Then we have something we call Smart Import. We've built for H5P.com. It's an example. We won't spend a lot of time on that, but just use it as an example for something you can build with H5P and AI. There are similar projects around the world that also uses H5P because H5P lends itself extremely well to AI since we have a very well-defined data structure. You just have to tell the AI the data structure of H5P and then the AI is able to build H5P content. Like, interact the books very easily. There are several other projects that also does something similar to Smart Import. In Smart Import this year, we are going to support more languages. Today it's just English, but French and Spanish will be added. So I think French is good news for Canada. We'll support that. We'll be supporting more video providers this year. We support only YouTube now, I think, but adding support for Panopto and Vimeo. We're going to give authors more control. I'll demo it today and you'll see that you don't have that much control today. You give it a resource like a video and then you get to choose which concepts in the video is important for you, but nothing more. So in future versions you can, for instance, tell it what are the learning objectives. And the moment I know the learning objectives, it can give a much better result. So imagine you want an instructional designer to build interactive content for you. You wouldn't just send an instructional designer a video and know other instructions and hope it turns out well. You want to tell them what are the learning objectives. So that's coming in SmartImport 2.0. We are making some UI improvements. There are lots of universities using SmartImport. There are like three, four more every week starting to use it. So we've seen some things in SmartImport, the initial version that doesn't scale well and we're improving that. And we're also completely re-implementing how we integrate with the large language models. And we have people from universities testing SmartImport 1.0 with the prototype of SmartImport 2.0 and they are very impressed with how much more precise it is still than the first version. And we're also adding support for more content types. So single-trade set, for instance. Also the trace will be supported in SmartImport 2.0. So that's a little intro. And then we'll jump into more practical things. Some examples of how you can use GPT and HIP together or you can use GPT to build HIP content. So we'll start out with the single-trade set. And as you can see in this document, I'll give you a little overview first. We have several tips. So how to use it to build single-trade set with a prompt example here. And we have bolded some of the things in the prompt that we think users are likely to want to change. So the subject. Obviously put in whatever you want here. How many questions? How many alternatives? Things you typically might want to change in your own prompt. And then we show you an example of what it looks like in the altering tool and paste it in and also what the end result is. And similar here we have summary. We have drag the words. And we have three more content types. All of these are content types where you can just have GPT create something for you and you paste it directly into HIP and you get ready-made content. So multiple-choice quiz is very powerful. We'll show an example of that. Fill in the blanks, mark the words. And we've also today added our section here with more specific prompts. So here you'll learn how to make, for instance include learning objectives or age or stuff like that to get a more fine-tuned result from GPT. And the last thing we'll show you is how smart import works. We'll take it from where we can skip single-trade set. We can go summary. So here we want to create a summary about chat GPT. So I'll take the prompt here from the recipe page and I'll go into chat GPT and I'll paste in the prompt. And now chat GPT will start generating content for us so that we can get interactive summary very quickly. This is for GPT. We optimized it for GPT. You'll be able to use the same prompts in other competing tools as well. It might be that you have to do minor adjustments to get the format right, but the idea is the same. We've tested some of it in part. I think it was and worked quite well. So here we have GPT created interactive summary for us and actually cockpit code. And I'll go to hp.com but you can obviously use WordPress or anything for this. And I'll go interactive summary. And the trick here is GPT quiz. The trick here in summary is to go to textual mode because if not I would have to paste one line at a time into the visual editor. It would be very cumbersome. If you go textual, you can paste it all in like this and then you have built interactive summary. So it's super powerful. So this is an interactive summary built with GPT and obviously I could go in and I could just change the subject or I could tell GPT this is for three-year-olds or this is for 10-year-olds. This is for people who study a PhD. This is the culture, et cetera, and it would give me exactly what I want. So it is actually, it can seem a little bit cumbersome that you have to go between the tools but the big advantage is that you have full control and we're all becoming prompt engineers these days. So just having these recipes on how to get the right format from chat GPT is very useful and then you can have it build really nice things for you in HPP. Those not familiar with interactive summary, it's a super nice way to summarize, for instance, what happened in the video, what happened in the lesson and instead of just giving the students a summary, you make them build it themselves and they'll remember it a lot better. So in total, we have two questions if I may ask at the moment curtain into this. The first one asks, would students get the correct answers when they type these questions created by chat GPT as far as I can understand, when they take those questions and feed them back into chat GPT, would they get the correct answers? So is there a way that they can sort of game? Yeah. I will do this for exams now. Yeah. So for the format of assessment, this is really cool, is there a textual node feature on the WordPress H5P plugin? Yeah, it is exactly the same. Wonderful. Thank you very much. Those are the two questions that popped up. All the recipes we go through now, it can be used on WordPress or Moodle or any version of H5P. So that was interactive summary and then we'll go to the next one. Drag the words. Maybe we take... Yeah, we can take drag the words. So I'll do the same here. I copy the prompt. Go to chat GPT. That's a new conversation with my little AI assistant. And what you should note here is that we've asked it also to provide feedback. So here we say, include three paragraphs and three to seven drop-offs per paragraph. Add explanations. Here is the format you need to follow and here we tell chat GPT how it is supposed to add feedback based on the user's answer. And that is important because often as authors we know that adding feedback per alternative adds a lot of learning. But it's also time-consuming, very time-consuming, so we don't always have time for that. When we have an AI assistant, we can actually make it do it for us and we just have to review. So now I got the drag the words created by chat GPT. I'll go add content again and just drag the words. And I actually don't remember what this was about. Northern Lights. A Northern Lights quiz and we paste in what we got from GPT in this text field here. Save. And now we've done a lot of work in a very short amount of time. And we'll want to review this. So here we have the Northern Lights where a natural light is playing in the Earth's sky. The room is very popular for Northern Lights. So hotel rooms in rooms these days are terribly expensive because of all the tourists. These displays can vary in maybe color and complexity, I don't know. Metacost by SolarWind and I'll just answer some random things here. And now we'll see what is right and wrong and we'll also get the feedback section at the bottom here. So we answered in winter months we answered oxygen and it will tell us incorrect oxygen particles are responsible for green and red colors. So here GPT has also added a feedback section for us which is very powerful. Maybe not super good examples of feedback here but I'll show you a task later where I actually think the feedback is contributing significantly to the learning. So that was track the words. I won't show you all the examples so you can go through some of these on your own. But this one is really good. Multiple choice question set I will see even better feedback. So I'll copy the prompt here as well. New tab and we run it. So these are questions about the sun and here we also see in the prompt that we've told it for each distractor add explanation about why the distractor is not a correct alternative. And we also tell it what format to use. So this is the one where we'll see quite good explanations for the incorrect alternatives. IRON for instance here we see it says IRON while present in the sun is not a major component and then it explains what the sun is actually made up of. So that is very useful feedback if you choose an incorrect alternative and adds value to the multiple choice quiz. So we'll copy it and we'll go to add content. We go quiz question set and what is important here the title sun quiz and textual mode again. Question set is also that type where you have lots of fields to fill in but you can go textual and you can just paste in everything that you got from chat GPT and then we save. And now we have this nice little quiz as I remember IRON for instance was one that had a nice explanation. It's not a major component the sun is primarily composed of hydrogen and helium. So what happens if you go carbon? Carbon is a very element that is much less abundant in the sun compared to hydrogen and helium. So here you have really good learning. What is the process that powers the sun? Is it the photosynthesis for instance? Photosynthesis is a process used by plants and other organisms to convert light energy into chemical energy not the process that occurs in the sun. So I think this is a good example of GPT adding value and doing a lot of work for us. So if we take fission the process of splitting atomic nuclei is not what powers the sun the sun's energy comes from fusion which combines so that is also I think a good explanation in my opinion. So yeah I think using these tools and GPT can help save a lot of time. We'll do one last example before we start looking at smart import filling the blanks and marketing words are quite straightforward I think for filling the blanks we can do it quickly because here we have GPT add tips for us which is also something that we might not have time for ourselves adding tips to the blanks So here we have it give us a text about behavioral psychology and add some blanks where the blanks are the most important words or concepts and we also ask it to add tips for each blank and that makes it a little easier to answer and also more clear what we're looking for So give it a few more seconds and we should have the fill in the blank and we can see what it looks like fill in the blanks behavioral psychology and we paste in here and we see what we get So here you see all these little icons typically when I create the fill in the blank I don't have time for it but with AI we can have it add these tips the tip is known for operant conditioning for instance here we have learning by observing others as a tip another important concept is learning by observing others another word for that So also a nice example of the AI both doing the basic work for us and maybe adding some extra value that we don't have time for ourselves Okay, and then we'll go for a more advanced So in this tip 5 section more specific prompts our idea is what you can add to the prompt to ensure that you get what you are after So add information about your target audience add the learning objectives add what style you want the content to be do you want it to be fun or sophisticated and you can also ask GPT to pretend that they are something So we have a couple of examples we have a funny Mr Bean quiz example prompt I won't do that now, not so academic but it does show some variation we have also a more specific London quiz example prompt So here we ask it to we can copy it in and then we'll look at it where GPT works So here we tell it that first we just tell it to create a quiz about London then we can get anything obviously then we tell it that the target audience are Spanish master degree students in English under fourth year which will tell it to have quite sophisticated questions and then we tell it that the learning objectives are to learn about the most significant historical events in the middle age for the city of London So with the learning objectives as well GPT knows much better type of London questions it should ask and hopefully it will be much more happy with the outcome than if we just told it to create a London quiz I think it just have one alternative left to right there we go so we can copy the code now and we can see what we got quiz again London quiz we use textual mode again and we paste it in and here we see what went in 1066 had a profound impact on London leading to significant changes in its governance, culture and language so if we go black death for instance we've also now told it to idle explanations so the black death occurred in the mid 14th century and although it had a significant impact on London's population did not directly lead to changes in governance, culture and language as the Norman conquest did so we got a good explanation for why it was wrong so by giving it information about the target audience and what we want to learn we get both better questions better distractors and also I think better explanations on the incorrect questions, incorrect alternatives Sven Toref I may there's a question over here do you have any examples for math and statistics I don't think we have that in there now I don't know how well GPT handles formulas and not ML or math check but I think it would do really well if there are no formulas I haven't tried with formulas I don't know how well it handles it okay so that is the GPT and we'll also spend a little bit of time on smart import the smart import example and what I wanted to do now is for you to choose topics that we can use AI to create content for so I will put this resource into the chat and you guys can go in and type in what subjects you want us to want us to use the AI on so this can be this can be Pythagoras it can be butter it can be cooking it can be dancing I don't know any type of subjects so just click on that link and then you can fill in a form and tell us what you're interested in ballroom dancing it was the first one we'll give it a little bit of time and see what the most popular subject is poetry Victoria good ideas let's play some music definitely in Canada with hockey in the lead here so far so maybe we should try hockey I don't know much about hockey but we can do a hockey and we can do something else it's easier to go like this learning Spanish and hockey if we take those two or maybe this one the concept of monsters in cinema it looks very interesting so we'll take that one and hockey and we'll have smart import build some content for us we'll see what YouTube have about the concept of monsters in cinema the problem with modern monsters not quite the same how to create a movie monster okay let's go for that one I'll just make sure that the video has it seems to be a serious video so I'll copy it and I'll go to H4b.com smart import and I'll just paste in the URL to the video about monsters and then for hockey I suspect if I just search for hockey I'll get some entertainment stuff for hockey matches so how to play hockey maybe ice hockey rules and instructions that should be possible for the AI so we'll try that copy that URL as well and we feed it to smart import so now we have those two working on transcriptions the processor has four steps provide content we provided YouTube links but we could have provided we could have uploaded videos we could have uploaded doc files PDFs etc the next step is to review the transcription third step we find concepts in the video and the user gets to review if they want those concepts or not to be covered and the last step is to choose what to build you want interactive books interactive videos etc so this is this should be a nice example of how you can use AI and not just use it by hand as we've looked at so far but here we have actually built a tool combining HVP and GPT to do more sophisticated work for us more quickly the first one is ready with a transcription and we'll just go for it start the next step and now the hockey one is ready looks okay as well and typically the transcription is very precise but if we have a mix of languages if we had some Spanish words and some English the Spanish word could be misspelled because it might try to spell it as it would in English it doesn't understand that it's a mix and then it goes a bit wrong we also have an overview here of what works very well with smart import and what doesn't and it will be the same with other implementations as well it is GPT limitations so if it's very conceptual like biology, history, psychology it has certain rules or concepts it works very good if it is more procedural how to do this, how to do that it doesn't work that well I've tried it with cooking recipes it actually works quite well with that too not as good as conceptual content and currently it doesn't read mathematical equations it doesn't see what you see in the video so if it's a teacher in the video who constantly refers to us we can see here etc it won't work super well but for other content it should be good so now we have the hockey concepts face of game clock, goal hockey, ice drink, major penalty minor penalty all of these seem to make sense if we had some strange concepts here we would be able to remove them but we can just take all of them and for the movie monster angry color and as you see the definition of color isn't the general definition of color it is from the video in the context of the video of movie monsters detail is the same they also seem to make sense and then we can choose what to build to interact the book higher order questions, I promise to see an example of that we can also go crosswords for that one we'll have it generate that hockey content and then for the monsters let's do some conceptual the conceptual dialogue cards maybe and a quiz and an accordion and now we can go and look at our hockey content so here we had crossword first I don't think we have time to solve it together but we can just to show solution so here we see central hockey words are being put into a crossword stick, penalty, overtime etc and then we have then we got the movie monster stuff in here then we have the higher order questions so here are ideas for project based learning about ice hockey, how to play ice hockey so one idea is create a strategy for a team to effectively defend against power place in ice hockey create a game plan for a team to successfully execute breakaways and score goals in ice hockey so these are ideas for teachers on what they can do in class for instance if they are running flipped classroom they can consider doing these classroom activities so helps ideation and then the last thing we had it build about ice hockey was to interact the book so it looks like this we have a picture from youtube front page, title etc and the video player was a bit slow but here we have it so we have 8 pages of interactive book created for us first page is the interactive video with the reader friendly version of the text below with paragraphs and if we play the video how to play ice hockey we see all the concepts, all the difficult words have automatically been explained for the first time the word is mentioned or the concept an explanation is also popping up in the video that's a goal when a player successfully shoots the puck into the opposing teams net so you see lots of concepts that there was ice rink for ice hockey it's a large flat area made of ice typically indoors where the game is played and in the middle here we have a quiz where did that go so here we have auto created some questions about the video if a player is illegally impeded from behind when in possession that was a long question what happens when a penalty is called a goalie in ice hockey I don't know maybe penalty box no you don't put the goalie in the penalty box I guess that would be harsh how long is each interval in ice hockey game I actually know very little about ice hockey but maybe 30 minutes 20 minutes okay it's probably a quite intense game so lots of questions there lots of questions at the end here as well so depending on how long the video is there will be more question points seems like we have some trouble with youtube what are the questions at the end and then on the following pages we have a glossary so all the concepts with explanations we have concept cards so what is ice hockey what is ice rink what is game clock what is goal what is penalty and then we have explanations for it so trigger the students to think about what penalty is we have a quiz fill in the blank questions here true false more fill in and also multiple choice and then we have a drag the words so here we match concepts with correct definitions we have flash cards that are not conceptual but more about what happened in the video 12 cards have been built for us what is offside in ice hockey another attacking player enters the defending zone ahead of the puck and then summary and as I said initially this resource won't be perfect it has been built in just a few minutes if we were going to build this by hand with all the questions in the video all the definitions all these eight pages it would take many hours probably built in a few minutes but we can't just publish this asses it's a good first draft we would have to go through it and maybe remove some stuff we don't want add stuff and correct stuff maybe we're not happy with all the distractors for instance and we want to adjust them so maybe we spend an hour reviewing and improving and then we can publish but we still save a lot of time so when we ask our customers in surveys how much faster are you they say typically four times faster when they count in the review time they are four times faster at building content using AI in this way and I think we in the future this is only going to get better one or two questions a number of people saying how fantastic a small input is if someone created something like an mp4 file of a video that's on their computer not hosted on any sort of cloud service are they able to upload that into smart import or would they have to upload it onto something like YouTube and then yeah no you can you can go here with choose file and upload a video and that will work work very well and we will store it and serve it from AWS fantastic is there a smart import only available within h5p.com yep and then there is presumably instructors can upload course readings to prepare no stakes none of the checks are there copyright issues or are there no copyright issues at stake for a smart import I think smart import is less copyright problematic than many other things the AI that the things you upload is the root source it is the truth so only use that so as long as you have the copyright to what you upload you should be safer than if you for instance have AI generated content for you based on nothing but if you don't have the copyright to what you upload or you don't have the copyright to make alteration of it then you are in trouble you need to have the right license to to the original material to the video or to the document that you upload thank you can just quickly show the other things we built the monster stuff so here we have the stand-alone flashcards built we could have had a interactive book about monsters built of course as well and we have a quiz about monsters referencing nature rather than other moving monsters is crucially in designing a unique and dynamic creature maybe so here we have a large monster quiz and we also got an interactive book about monsters with the interactive video and and the same type of content that we got for hockey so both of them created in a few minutes there I don't think we have time to dive more deeply into it we just have a couple of minutes left thank you very much thanks a lot for the time and thanks for hosting again see you guys