All set? Okay. So welcome, everybody. We're so glad you all were able to join us for using generative AI tools for research and writing. So some of you may have joined us before. Lauren and I have done this presentation a couple of times, but because the world of AI changes so quickly, we have to change this presentation pretty much every time we do it. So we are very glad if you have joined us again. My name is Maria Barefoot. I am a librarian. I am the online learning librarian here at UD. So I'm going to provide some library perspective for the research part of this. Lauren, you want to introduce yourself? Yeah, and I'm Lauren Kelly. I'm also with the University of Delaware. I'm an instructional designer. And my department is called Academic Technology Services. And so I love tag teaming with Maria all the time for this kind of research and writing. aspects of tools that you can use. So I will be covering the free versions of co-pilot and also Gemini. Okay. Laura, I'm going to let, do you want to talk about the overview or you want me to? Sure. So what we always aim to do is to infuse a little bit of the importance of AI literacy in these workshops. But we also want to get down and dirty with the tools themselves because sometimes there things that we pick up on that other people don't in everyday life. So if your staff, if you are a teacher, a researcher, if you're a student in some other aspect of a hat that you wear and the work that you do, what we aim to do is across these tools is to, from the librarian perspective, Maria provides what these tools can do and why they're good and what are the best uses of them. And then I am also providing that same lens for co-pilot and Gemini. And I'll be very honest full disclosure with as the free tools some things are pretty bare bones so there's some tips and tricks that we are going to offer that we feel like are really the best these tools today so this one are you got this one yeah so we're going to do this in two parts the first part we're going to focus on a couple tools that are specifically designed for research and those the main focus will be on perplexity and illicit um i am going to mention a third tool during that part that is called Consensus, which we have looked at before, but there are a whole lot of tools out there that are AI fueled for research. So I'm only briefly going to touch on consensus today. And then the worksheet is going to include more about illicit. And then Lauren is going to talk about co-pilot and Gemini and some of the new developments that were able to use with co-pilot and Gemini and some tips and tricks for that as well. we will be sharing that worksheet link whenever it's time to use it. So don't panic right away. We're going to be plugging links into the chat the whole way through the presentation. Absolutely. But if you like to get on stuff a little bit early, it gives you a sneak peek if you have to set up monitors because we know people might be doing stuff from their phone or doing it from at home. So it'll give you an opportunity to kind of get set up the best way that makes sense for you for the session. So out the gate, just like we are teachers, we always like to set expectations for responsible and ethical AI. So as we kind of joked about as people were coming in, these tools are extremely dynamic. We haven't really seen anything quite like this that changes on a dime. I guarantee you something has changed within the time that we get ready to present right now. But the tools are very dynamic. So I think you just have to pack your patients and just know that probably the best thing is if there are tools that you really, really they do like, just work to kind of follow those tools and keep up with them as best you can. LinkedIn is a really good resource for following people that they do this day in and day out. I don't think I'll have them get a lot of sleep, but they keep up with all the tools. So that's a really good way to do it and follow people that have blogs or podcasts on these tools, because that's a really good way to try to not be as overwhelmed. Having conversations about plagiarism and academic integrity in any work that you do, whether you're in the workplace, whether you're doing research, whether you are teaching is extremely important. Some of the things that I know, Marian, I notice in our work being in academia and being in higher ed institutions is that there are not a lot of conversations in the classroom. There are a lot of students that are confused about what it should look like and, well, is this plagiarism if I do this and I use this tool or what if I did this? Is this, you know, my being dishonest in some way? So it's really helpful to just lay the groundwork and have those conversations about what it means to you. Because I really think we're at a really interesting time where everybody has very different ideas of what that looks like based on generative AI. And it depends on what type of work they do, what experiences they've had. Like if you look at artists, for example. So that's a really good example. The other three bullets are they have everything to do with those conversations, right, about generative AI, what's responsible, what's ethical, what really makes sense, disclosing it. You'll notice that within our documents. We do disclose where we've used an AI tool because that's the right thing to do, right? We didn't do this all by ourselves. So that type of disclosure is very important. If you're teaching, if you're doing research, if you're in the workplace, I do recommend disclosing. It is important to let people know, hey, Jim and I help me out with this or co-pilot help me write this, just so that people are aware. And of course, bringing into practice the same things you're paraphrasing, you're citing appropriately. And Maria has a really wonderful resource that we'll talk about in just a second that helps with that. So it's just really important to establish those guidelines. Go ahead, Maria. Okay. Oh, you know, let's do this and then we'll go to the library. I forgot that this slide was first. But this is a really quick one. These are resources that are specific to the University of Delaware, so we know we have some folks on here today that may not be affiliated with the University of Delaware. So there are asterisked by particular ones that I believe will probably not be public facing for you because they're within our ticketing system. But the very first two are very much aimed toward the teaching and learning process and practice. And there are considerations and it gets into policy language and a plethora of resources that you might find helpful. Okay, Maria. Okay. And another tool, especially for the faculty members that are on this call with us, is the library has developed an AI tutorial. This tutorial is geared towards anyone who is new to AI and wants to understand how they work, some of the ethical implications. We cover a little bit about citation in the tutorial. So you can plug this link into your classes. It's great for students to complete. There's like a little quiz at the end of it. So if you were hoping to give them a little better understanding of AI, you can have them turn in their quiz results as a verification that they completed this tutorial. And of course, if you have any questions about using it, you can always contact me or use the Ask the Library chat and somebody will direct you to me. So I'm going to throw this in this. the chat, this link. Oh, you already did it. Thank you, Lauren. Okay. I'm losing track of like which one of us is doing what. And let's move to the next slide. And the next one's me, Anne Maria. So when we first started doing these sessions, we thought, where can we put all this stuff that we're learning about? Like, we need a space. So we both really love Padlet. Do we have any Padlet fans? You can raise your hand if you're a Padlet fan. But we felt like Padlet was a a really great place to kind of house everything and categorize it. So we can put this in the chat as well and make this link available. But there is a QR code. You know, if you're floating around and, you know, you don't feel like accessing this from your computer, you can use your phone and you can pull it up on your phone and bookmark it if you'd like to. So it is a little overwhelming if you're new to generative AI. But we tried to categorize where if you're looking for particular things, you could find them. So it's the gamut. It's AI literacy, it's AI policy, it's AI-specific tools, if you want for images or audio or for presentations or video. If you like to track AI tools, there's something there as well. So there's a little bit of something for everything with AI for this particular resource. All right, Maria, you can move on. And then I want to just preface our space before we get. get started. We are sharing a lot of resources and we are going to be doing all of this, of course, in this remote space and virtually. So when you're navigating screens, you're trying to take notes, you're trying to actually do things. It can be a lot. So it's a safe space where we have a good number of folks, but you can feel free to unmute and speak up if you feel comfortable enough, right, to ask questions. You can also use the chat that way as well. If you need to see something again, just say it. Just let us know, hey, I don't know how you you got there. Can you please, I got lost somewhere. I got stuck. Please let me know how to get there. I also want to mention that this is being recorded. So we will share the resources along with the recording probably a little later this week. If I could turn it around really, really quickly with the editing. So we'll make sure that we do that. A lot of people have to leave a little bit earlier. But we do have more time on the back end. This session actually will conclude at 115. So we do have extra time here. So we can slow down. You can choose to not do any of the activities if you don't want to and you just want to watch and learn with us and just listen. So feel free to learn the way that you learn best. Okay. Go ahead, Maria. Okay. So with that said, we are going to get into perplexity and elicit. First, I'm going to give you a little overview of perplexity. Then I'm going to give you a little overview of elicit. And then we're going to go into the worksheet and you can choose which activity is most interesting to you. So we don't expect you all to do all the activities we have in this worksheet. Just pick one that you find interesting or compelling for some, in some way, shape, or form. Whoops, I clicked the wrong direction. Okay. So before I get into perplexity, I want to talk a little bit about understanding the different places that we come from when we start a research project. So I think the majority of the people on this call are faculty and staff. And so a lot of us have been researching for a very long time. As a librarian that works a lot with freshman students and students that are new to the college experience, learning the idea behind collecting information and then synthesizing that information can sometimes be a pretty challenging skill. And one of the biggest problems is whenever a lot of students come with the perception that when we do research, we ask a question and get an answer. And it's like a one -to-one relationship. So when they're formulating questions, sometimes we search those questions as if there is a single answer to those questions. And so when we're developing that more complex way of thinking, actually, I think some of these AI tools can help along that route. And one of them that I'm going to show you is called deep research. It's a new tool from perplexity. And one thing that I like about it is that it kind of highlights the thinking process behind asking questions. Another piece of this is that a lot of times students ask questions in very popular language. Like in the language we would use if we were communicating about academic topics in a magazine or a newspaper, something for a layperson to read. But when we search for academic literature, there's often a whole different vocabulary that goes along with that topic. And so another thing that AI can be useful for is helping steer students towards academic language, developing the terminology of their topic. And then whenever we talk about understanding the information landscape, The other reason that I recommend a tool like perplexity or consensus or elicit, instead of something like co -pilot or Gemini, when you're on the research side of things, is because they will always steer you to a source. Where did this information come from? Not all AI tools have that capability. And so understanding, what am I looking at? Who was the publisher? What was the purpose of creating this information? all of that falls into like the knowledge that students have to build as they become competent researchers. So it's not that the AI tool is like completely killing that thought process. If you're using the right AI tool, it can actually support that learning that goes into like tracing and evaluating and understanding where your information came from. So with that in mind, I want to show you guys how perplexity works a little bit before I give you time to play with it. So this is the tool. It's perplexity.a.I. If you've never been to perplexity before, it will ask you to log in. So you can see that I am logged in right down here. You can go ahead and set up a free login. You can either use your Google. If you're a UD person, you can sign in with Google. or you can set up a different, you know, you can use a personal account too because it's free, it's totally up to you. I actually have a personal account and an account through work so that I can kind of compare how they function differently. So it's up to you is all I'm saying here. When you get into perplexity, this is your search bar where you're going to build your prompt. And I want to point out a few of these tools. First of all, it's defaults to auto. Auto is the free version. It's what everybody in the world has access to. Everybody also has access to about three pro searches per day. I will show you the difference between an auto search and a pro search. The biggest difference is that the pro search gives you more sources. Where this one is going to give you like four sources, I think. This will give you like 10, 11, depending on what you ask. And then the one that I'm also going to show you is this deep research. So deep research is actually a really cool. It's really cool because it shows you not only the results, like it gives you the resulting summary with links, but it also shows you how it got to that summary. So I'm going to show you that in a couple minutes. Before we do that, this is the other piece that you really need to know about to use perplexity effectively. It defaults to searching the web. That's like everything, like doing a Google search. So, yes, you will get some academic information with just the web turned on, but most of the information that you're going to get is what we would consider popular information. Most of it comes from organizational websites, blogs, that kind of thing. If you turn on academic, you'll see that it adds like a little hat, like kind of the Google Scholar hat, to your search. So it's not that you're doing one or the other. it's that you're searching both the open web and academic sources. If you want your students only searching academic sources, you have to turn off that little web one. And then you can see that we're only searching this. As far as I've been able to tell, the academic search only searches semantic scholar. I have never seen other things. Like it almost always links me out to semantic scholar. So that's where it's getting its content from when you say search academic sources. It's just like going to semantic scholar. Okay. So I want to show you some of, let's see, this was the deep research one. And I can tell that this result was one of my deep research results because at the top it says deep research. And if I click on the little down arrow, you'll see that it actually walks me through the entire research process that it did behind the scenes. So the first thing I asked it is the cruise industry becoming more popular? I chose this question. This is actually a question that I got through the library chat at one point because it is such a, it's such a great example of a question that is phrased in popular terminology, but actually does have like academic underpinning specifically in the business literature. So if I look at this, it tells me, okay, I'm going to have to gather data, trends, and reports. I'm going to focus on 2025 because you asked, like, for recent information. And then it tells me what it searched for, which as a librarian, I was like, oh, my God, this is wonderful. Transparency. There's transparency here, and I got very excited. So it says, I searched for cruise industry becoming more popular, cruise industry growth trends. Like, you can see that it's using more than just one search string to find information. Because this was an academic search whenever I turned it on, it's trying to find information in academic sources. And you can tell that the AI tool actually really struggled with the phrasing of this popular question in an academic setting, which, is a very similar struggle that many of our students have when they try to search a library database, but they're using popular language. It doesn't always work out well. We're kind of mimicking that same issue here. It keeps telling me that return to no results. So I can't validate this information, but I'm going to refine my query and look for things like cruise passenger growth. And so it keeps searching and it keeps telling me I can't find anything. I can't find anything. And let's see if I can find the one where it finally found something. Okay, so now it has decided, this is like mimicking the choice process that a lot of students go through. So it couldn't find anything in the academic semantic scholar stuff. And so it says, all right, well, I'm going to go to a specific site that the AI thinks, apparently, thinks is at least trustworthy enough to go into an academic search. And it says cruise industry news.com. So it goes to this specific website and then it looks for and it says this site is an authoritative source for industry specific information. And by narrowing the search to this domain, I aim to bypass the lack of results from broader queries. That means I couldn't find anything in semantic scholar. So I'm going to go to the open web. And so it does that. And if Eventually, it finds some stuff from this cruise industry news, and it gives me a little summary. It does eventually talk about structuring this report, and if I want to look deeper into my sources, I can click on this and see what it gave me. So some of this stuff does come from semantic scholar. It eventually found some scholarly information, but it really had to go through a lot of steps before it made it to academic information. Now, if we compare that, this was a search that I did in the, it's the exact same question. The same query is the cruise industry becoming more popular. But instead of telling it to search academic sources, I just did an auto search. This was not deep research. And it gave me five different searches or five different sources to work with. And so I got something from this go walkabout .com. I got something from the week, something for J.P. Morgan, Expedia, Forbes, and then Reddit is in here as well. So this is very similar to getting a Google search result summarized for you, like an AI summary. Except this one actually links you to where the information came from. So students would be able to then evaluate that piece of information, the same. way that they would if they clicked on it from Google search. So that's one of the big reasons why I recommend this tool because it allows you to link to sources. Let's, I don't want to, I think I showed you all that I want to show you about perplexity. Let me check in with the chat here. So we did have one. Okay. Justin had a great scenario. He said one of my librarians at the university told me that students are using perplexity and getting hallucinated sources. I suspect that the students got the sources from chat GPT or other tools and just misreporting that they came from perplexity. How accurate do you find sources to be? So what I did for paid versions, there are other models that you can access, but perplexity is pulling from chat GPT. It is pulling. And so here's the other thing. It depends on how your students asked the question. So I've never had perplexity link me to a fake source. I like like when perplexity provides links it's always linked to like a live website somewhere um if I ask perplexity give me a list of sources on this topic then it absolutely hallucinates made up sources so it's like if they are asking the question and then checking the links for the most part it seems to give real sources but if they ask for a list of sources essentially they're asking the AI to write up some sources for them. And then it mix and matches titles. Sometimes it makes up authors. So yes, that is that is always possible. Does that answer the question, whoever asked that one? Yeah, that's really informative. Thank you. Okay. All right. Awesome. So then I'm going to switch over to elicit and consensus. Let me go back to my slides here because I think there was some other things that before I, before I send you off to all to practice. So that, all that perplexity stuff, a lot of that I consider to kind of be like an undergraduate level research tool because it like, it helps researchers take the step from all I'm doing is reporting other people's ideas to I'm starting to ask questions. I'm starting to understand like I have to ask follow ups. I have to get more deep. also because perplexity has a tendency to give so much popular information, I kind of see it as like a stepping stone. Now, the next two tools, consensus and elicit, I see those as tools that are geared towards more advanced researchers, researchers that know the difference between different types of academic literature, like systematic reviews versus a standard literature review, or somebody who knows the difference between a randomized controlled trial and a literature review, like that kind of stuff. Because that is what these things are designed to show. They actually have more advanced tools for pulling data out of research papers, like look at the specific outcomes, look at the specific populations that we're working with. So where perplexity is like a summary tool, these are like, let's compare academic papers kind of tools. Consensitizing has this little consensus meter thing, which is what I have a picture to show you. So this is what the tool looks like when you get there. You can turn pro on or off. It does only give you like three or four pro searches per day. So you might get like an error message if you're on the free version. But I want to quickly show you the way that you ask questions makes a big difference in these tools. So this is a question that a researcher from my discipline might ask, somebody who's an educator, like, are physics first curriculum models effective for college readiness? You notice the consensus meter did not tell me yes or no. Like it didn't tell me that yes, definitely, or no, not at all. It just kind of gave me an overall summary. And then it gives me some links that I can check out, some actual articles I can go to. On the other side of things, if you're asking like, a question that might come from the health sciences or that is more qualitative in I'm sorry, more quantitative in nature, then the consensus meter might tell you that there is actually an answer. There are 85% of the sources said yes, CBT can treat fear, possibly it was 10%, and there were a couple papers that refuted that idea. And then we can take a look at them and see which ones say what. so I kind of see this tool as like somebody who is really starting to become a more advanced researcher that wants to know what does the literature as a group what what consensus have we come to and then finally I just want to point out illicit um elicit is the third option for looking for seeking information before you get to the point that you're going to synthesize your information when we're still looking for stuff elicit is another option. You can see that certain things are labeled pro. So it has like a lot of a lot of options, but they're not all available to you in the free version, which is what Lauren was talking about earlier in the session. However, I do have a couple examples from a list to show you. And I try to always get an example from both a quantitative type of question and a qualitative type of question. So this question was inspired by a faculty member who was working on feelings of belonging in, I think they were actually looking at black men, but this one, I changed the question a little bit. So I went with bilingualism and feelings of belonging. And I asked the question, how does being raised bilingual affect children's feelings of belonging? And you can see that for free, it will give me like four papers to choose to look at. because it will link me right out to the article if I want to read that. Like if I want to go find this article, it will link me out to Semantic Scholar or it will link me to the publisher website using the DOI. That's what those two links are for. You go back here. The other thing that I love about semantic Scholar is that in addition to the summary, and it says bilingual education can promote a sense of school belonging for emergent students. Like, it gives me this overall summary, but then I can add in a specific comparison that I want to make. And as somebody who loves things that are laid out in spreadsheets, this tool is like my favorite. I love it. I love having a spreadsheet with everything as easily comparable. So if I wanted to look at, if I wanted to compare the main findings of each one of I click on that and it automatically adds in, oh, I added the outcome, sorry, before I was playing around with this before we started. I already added a column about the outcomes. So this one is looking at the sense of belonging. This one is looking at the, it says that the abstract doesn't mention any specific outcomes. And then I added a third column that summarizes. is the major findings. So parents positioning of bilingualism as a positive resource. This is some of the main findings from each paper. So it's a very quick way to like compare, is there a lot of data that supports this idea that I have in my head? I did actually play around with this, which is like where it says add a new step down here. I asked it to show. me, let's see, add a new step. I asked it to show me a list of concepts, which is what you're looking at up here. So essentially what that's, if you are working with early career researchers, you know, this is a good way to help develop that academic language that we were just talking about can sometimes be an issue for new researchers. That's shifting from popular to academic language, this kind of thing can actually help with that. Now, the downside is you can only get two extra columns on the free version. If you want to do more columns, then you have to use the pro paid version. That's why they keep telling me to upgrade in different places. Same thing down here. You can't actually download your Excel file unless you upgrade, which is very sad to me, at least. So I'm going to check in with the chat here. Lauren, is there anything? Oh, it's been popping. I've been answering the question. I can see things. There's been a lot of rave about perplexity and wanting to know pricing and deep research. Of course, you got everybody pumped on deep research. But within the worksheet, Maria put a little bit of information about perplexity that free users, it's a limited number. But I think someone said they couldn't find the exact number. So I think they're being vague on purpose. Yes. They are absolutely. They do not have. I scoured the internet for what is the number? How many searches am I allowed for free? And I found like a vague post that says around five. So like I don't know how they're calculating that, but around five was the answer. Yeah, Rachel mentioned, and I have seen this before, when you jump on perplexity, it'll do this thing. We're like, hey, do you want this free trial? So for UD folks, we do not have a license for perplexity. Now, you can on your own do that free trial if you wanted to do that, that would be your prerogative if you wanted to do that, but we do not have a license for it. But much of what Maria is showing you, you can do for free and you get a decent number of searches per day for certain things. So you could do some damage. You can do some good research damage with that, I would think. Absolutely. Absolutely. I agree. So Lauren, I'm going to show them, did you already plug the workshop link or the worksheet link into the chat? Okay. So this is the worksheet that Lauren linked to while I was talking. The top of it. it is a little comparison chart of things that you can do with each tool. So you can see we have consensus or consensus elicit, perplexity are all on here. And we've kind of tried to indicate the stuff that you need to pay in order to use. So Lauren's going to cover some of the Gemini and co-pilot stuff next. But I think let's give you guys like five minutes to play around with one of these tools and then we'll come back together. So on page, uh, Let's see. There's activity one is labeled for perplexity. And you can do that one if you work a lot with undergraduate students. This one's geared towards imagining what it's like for your student to learn how to use these tools. Elicit, imagine that you're either a newly minted researcher or maybe a graduate student. So you might want to play with these two. I'm going to stop my screen share. And we will be active in the chat. So if you guys have questions, please feel free to throw them in there. or unmute and we will answer questions that way. Feel free to like throw in the chat what you're discovering and what you're finding. We always find that intriguing when people find really cool things. So please feel free to do that. So we'll give you, you think, in five minutes, Maria? Yeah, let's do, because I want you to have time to go over. So let's do five minutes and then little discussion and then we'll go into co-pilot and So did anybody have any trouble doing a forced copy? I just want to make sure everybody's okay. If anybody had trouble, please feel free to unmute or you can drop in the chat. Hey, I'm stuck. This didn't work for me. Help or something to let us know. But feel free, like Maria said, try illicit if you want to, try perplexity if you want to. These are here so you don't have to do the thinking. Let's take like one or two more. minutes and then we'll chat about what you discovered. Okay. So at this point, I'm kind of curious as to what kind of choices you all made and what you found out about using any of these research tools to find literature. Would you, which one did you try? Would you recommend it? Would you recommend it to students? Would you recommend it to colleagues? What do you all think? Feel free to the chat or to unmute. Maroon mentioned that you like the deep research more than the regular web version. I could see how this might be helpful for grant writing. How about anybody in here that teaches students? Would you recommend this to students? Would you ban your students from using it? What do you think? Hi, Maria. Can I jump in a bit? Actually, at USC Prize, we are working on the kind of the three, cross for writing using AI. So we propose students using illicit and with several tools like consensus and notebook or spice space, but like we not make it like break it into here. That's what they need to do, but we make it more like research flows. For example, the first one, if you want to find the idea for your topic, you can use maybe publicities or chat CBD or stump. So that you can have the idea of like what it could be about. And then in the second step, it will be literature review. Like first you will be like gathering information using consensus as spiced by eclosite or and yeah. And then you can do the managing reference using Zotero. So like we try to make it like the flow. So that's why they will be like go by step by step. But I really love the Google doubt that you're working on. So I'm not sure if I can use it because like, We made, but we haven't go into detail, like, what's that you need to do? We just, like, make it by flow. Like, here is the, the first state, here the tool that we, like, suppose, like, propose you to try and use. But we haven't do, like, any, like, detail, suggest, like, activity that you did. And I think it's super helpful. And if we, we incorporate it into our teaching, it will be really helpful. So I wonder if I can, like, steal some of your ideas and using it. Yeah. Thank you. Yeah. Yes. The answer is absolutely. And Lauren and I designed this in a way that, like, one of the things she's going to show you is how to, like, export and move things out of AI and into your other research tools like Zotero or, like, right into your Word document and all that stuff. So, yes, please use it. Please use the worksheet. It's, I think we put a Creative Commons license on there, right, Lauren? So feel free to use. There was a question in the chat about safety. results. If you are logged in to any of these tools, it will save your results from session to session. If you do not log in, then it will not save your results. If you want to, I'm going to do a real quick share, Lauren, so I can show them the share button. And then I'm going to hand it over to you. If you are using these tools and you have a result like this that you want to share with somebody. You can get a link to the results or you can use the share button. You can make it private to yourself or you can share it with people using the link. Or I guess you can put it on social media, which I've never done before. But yeah. So you can get a shareable link for most of these things. Even Elicit has a shareable link right here. And it's just like setting permissions on a Google document. There was someone that mentioned about the use of data. They do use your data. Although they say, hey, we're not going to use this to train with models. All the more reason to just say, just be careful what you're putting in the models, make sure that what you want to be in the models, you actually want to be in the model. So I would say no sensitive information. So that's critical. You can turn that off in perplexity. You can tell if you go into your settings, there is a way to say do not do not use my questions to train AI so probably should have shown you guys that but you go into your settings and then it's just under I'm so sorry should have been more prepared um under yeah someone's looking for it our scott said oh can you show that setting yeah the pool five just make sure you show your scream Maria if you're showing if you're just going to yeah okay here it is I found it I'm going to share my screen and show you how I just wanted to find the right thing before I started screen sharing. So if you go right next to your name, you click on the little setting button and then scroll down to where it says AI data retention. AI data retention allows perplexity to use your searches to improve AI models. Turn this setting off if you want to exclude your data from this process. And I have mine turned off. So this is what it looks like turned on. it'll be green. This is what it looks like turned off. Very cool. Good. I forgot to that feature. Okay. All right. Any other questions? We're going to jump into part two. Go ahead. I'm so sorry, Lauren, if I took up to the question. Oh, no, no, no. That's okay. This is great. We have a lot of really great questions. Okay. We always try to anticipate what you all will ask. So this is really great. So, all right. So we're going to move into part two. Can everybody see the screen? Okay, thumbs up. See any thumbs up? All right. Okay, very good. All right, I want to take that as yes. Okay, so there are three things. If you run out of time, it's okay because you have your forced copy of the worksheet. So just keep all that in mind. So we're going to go into something that's called customized instructions. So for those of you that if you've been following other AI tools like chat GPT or cloud, these have the capability to in their settings somewhere. And sometimes within individual chats with like writing style, for example, it will allow you to customize instruction so it remembers you or it remembers your style or remembers how you want to receive the generated outputs. So I'm going to show you a way that you can do that for Gemini and for co-pilot. So that's one tip and trick that we'll share. The other thing are prompts for assistants. I'm a sucker for an assistant. I do use generative AI tools as my assistant to work with me collaboratively to, for example, develop workshop materials to help me with a really sticky response when I'm, you know, responding to right now we're playing a conference. So if I'm trying to kind of cord a speaker and I'm not sure what to say, I will use AI models to help me craft that language to make sure I send the right messages, for example. So assistants are excellent. So there'll be three different assistants that I'll share with you today that you can test on your own so we'll have time to actually play you can pick which one you want and then something that's called prompt decorators that I just learned about more recently that actually give you as a user a little bit more control over the generated output it can kind of help with since we started talking about hallucinations a little bit it kind of guides and steers the AI model a little bit more it's you still not going to 100% control it but it gives you a little bit more where you could be the front seat driver, so to speak, with that. Okay, so before we dive into it, I want to share just a couple things. So one thing that I have learned in my workflow is that I move across models. What tends to happen is models are super inconsistent, like Maria and I talked about. They're very dynamic. So sometimes they'll go down and you're in the middle of your conversation and in a flow with the model and all of a sudden you need to shift gears and go to plan B and use a different model. So always have a few different models on tap that you are accustomed to. Now at the University of Delaware, our recommended tools are co-pilot and Gemini. So those are the ones that I recommend. So toggle between Gemini and toggle between co-pilot. Funny story right now, co-pilot, I can't find my chats anywhere. And it's not just me. Maria can't see hers. Another colleague of mine, she mentioned it. And I'm like, where did the chats go? So I'm not really sure where the chats are going right now. If you work with chats and you have a conversation, I just don't know where it's going right now. So it's always good to have more than one. What I have here that you all can kind of, you know, take a look at and you can decide what works for you is there's the ability to mix and match. So for example, based on the strengths of the model, depending on what tasks you choose to do, you can decide which model generates the best output based on the task that you want over another one. And you can play between all of them. So here's just a really great example of that. Now, in terms of Gemini, I hands down think that Gemini is better for reasoning. So Gemini like Perplexity also has a deep research. I don't know who came up with that first. I think Google did. But if you're following the tools, you know that they play off each other and It's really hard to know what the roadmap is overall with a lot of these tools because they're just in competition with one another, unfortunately, which is terrible for us as the end users because we never know what's going to happen with these tools. But for the most part, this is what I gather that are the best strengths for Google. And then I've played across the models for asking the models like, well, what are the best strengths? What are the best uses for them? So these tend to be what they are. So you'll notice between on the next slide, co-pilot and Gemini, they're very good at doing a lot of the same tasks, right? So any type of drafting papers, brainstorming, outlining, summarizing, giving you key points, providing you with insights, they're good at that. They can, for the most part, do that. For certain things, one is better than the other. I tend to think a lot of times Gemini has a little bit more insight and goes a little bit more in depth, hence the reasoning aspect of Gemini being better. Gemini is also very good with analyzation. They call themselves the multimodal AI model. So there's also that. So you can upload audio files, image files. There's all types of media coding files that you can upload into Google Gemini. And you can do that part for free. And then, of course, when you're getting into research and looking at reasoning for that, it's really great for that. Now, what's very cool, and Maria helped me discover this, is that there is a new feature that Gemini has that's called this double -check feature. And so when you are talking with the model, you're putting in your written prompts, there is the ability afterwards at the very end of all of the conversation. There are these three dots, and the three dots will give you this menu, and then the double-check response option is there. Now, I feel like this is inconsistent. I feel like it disappeared a couple times that we were preparing for this presentation, but the last time I checked, it was still there. Now, it can be inconsistent with it. Now, what it's supposed to do is double check and it creates this kind of legend of highlights, so to speak. But for the most part, what ends up happening is that you can click on this drop-down arrow within the highlights, and then it's going to give you those references. So very similar to like perplexity where you could off to the right, you could see all of what your sources are. Well, these are kind of embedded in the conversation with Google, okay, for Gemini. In addition to that, what's good to know is that these will disappear. So it's only, you're only able to see it as long as you're using it. Like if you jump out of that chat and then come back to the chat, it's gone. You'd have to initiate it again, at least for right now. you'd have to initiate it again. If you were to take this and copy it and paste it into a Google Doc or try to save it in some way, it gets lost. It's not going to stay there. So just things for you to know. Maria, am I okay? I see things popping up in the chat. Yes, just confirming kind of what you're already saying that it's not the easiest user interface. That kind of stuff. I didn't see any questions. Okay, perfect. Now, on the next slide here, this is actually. what it will do, which this is cool. Maria, when you and I were practicing with it the other day, I couldn't find this, but it was there last night when I was playing with it. So again, at the very end of the chat, once you've initiated for this double-check feature, then all of a sudden this hyperlink appears that says, understand the results. So if you click on that, then a window pops up, and it creates this legend of what does the color coding mean, right? So it should still, like if I'm using dark mode, but if you're using like a, a regular mode and it's white, you should still see it as like kind of a rose color and like a green. You should still see that same. So it should be accessible. You should be able to see those those same colors. So apparently with the green, this is similarity across what it's generated in the output, right? So maybe there are other sources that have, that it's compiled and it's summarized that, not necessarily direct quotes or anything, but that it's immersed in this resource that it's giving you, and that's why it's highlighting it green. Now, if it's red, then that means that, well, it's different and it's not necessarily grabbing it from a resource, but if it has a resource, it will make that link available. And then if it doesn't highlight it, it wasn't really able to find enough information based on the information that's there. So whether it's good or bad. I think it's a nice feature, but again, when we had that conversation about hallucinations, just keep in mind with that non -highlighted text, that could be hallucinated, right? Because if it doesn't know where it got it from, you just don't know. Okay? Any questions? You can feel free to unmute, chime in, or anything like that. Please feel free. All right. So let's talk about copilot. So very similar. You see that top bullet. It can draft essays and papers. It can summarize, outline, brainstorm insights also. It is helpful for all of those things. Generating outputs. If you built a class a long time ago and maybe you didn't have student learning outcomes, it's really great. You could put a summary of what you covered, for example, or a description, and it'll completely generate learning outcomes for you. So for things like that, it's great in pulling out and abstracting information, giving you key points and summarizing. It's a wonderful tool that way. But again, I do think Gemini is a lot better with reasoning. And more often than that has given me a lot more rich of a generated output than co-pilot has. So a very cool feature that Co -Pilot has that Gemini does not have is it has a prompt gallery. So this is a great starting point. Very early on when Gemini was booming, like really, really booming, it was everybody talked about prompt engineering. It was so important for you to be able to prompt. So it's not as important. You do still need to know how to ask the right questions to Maria's point in her section about researching and using these tools. But it's a lot easier. You have a lot more guidance and a lot more help. And there are other resources in that worksheet that you will have available to you at the very end where there are other prompt repositories or guides that you can use too. But I thought this was a really nice feature and it continues to grow. So initially when Maria and I found this and discovered this from this task area here, there were only maybe like four or five ways to filter. So now there's probably like 15. So it's continuing to grow. So you can filter by the app, which just means if it's going to be like a word document, a PowerPoint file, an Excel spreadsheet. I'm not sure how helpful that is, but it's there. But I think the most important things are going to be the task in the job type. This co-pilot agent popped up last night. I don't really know much about that and how it's helpful just yet. But the job type and the tasks are great because this will break down by like human resources or particular, you know, department. So I think this is extremely helpful for staff members, right? And then the tasks are as well. So it's like create, learn, things like that. So very, very helpful. And there are two ways to get this. So when you're actually in co-pilot, there is an area that will actually, say, view your prompts. And it's right, you can see here right above the window where you're actually talking to copilot. pilot. And then when you start doing that at the end, it'll say, do you want to see all the prompts? And so that's another way to get it. So that'll bring you to, if you go back, to this full page here. That'll bring you to this gallery. And again, these resources are also on the worksheet that you have too. Okay. So I'm going to stop there for a second. Do I need to check in? Nope. Okay. So I would say let's get ready to dive into the worksheet. Maria, can you do me a favor? Put the link to the worksheet in one more time in case anybody missed it. And I'm actually going to toggle over, oops, I'm going to toggle over to the worksheet. And I'm going to do this with you. So, all right, so I have the worksheet here at the left, and then I have Gemini. I'm going to start with Gemini first. So, all right, so let me give you a little context. So as I mentioned before, customized instructions are big with other models like Claude or like ChatGPT, where those were the very first things that they came out with. You're like, oh, my God, cool. I can say I'm an instructional designer. You know, I use universal design learning. The things I do every day are this, this, and this, where you can totally tell the AI model who you are and what you do as your profession to really aid you and the tasks that you do throughout your work during the day. Well, Gemini doesn't have that. Copilot doesn't really have that. So I, full disclosure, I used Claude to help me build the customized instructions to be able to put into, say, a Gemini and say a co-pilot. So if you take a look at your sheet here, this is, I just used a role for staff, right? For example, so I'm an administrative assistant who manages and I'm just basically saying, here are the things that I do throughout the day. Now, the key here for the prompt is what I want for the customized. instructions is for me to be able to take this and bring this to any conversation that I have. If I know I'm doing pretty repetitive tasks and I'm doing the same things maybe all the time, right, and for the model to be able to hold my context and who I am and how I want it to generate my output. So that's the key. So if you'll notice at the very end of this prompt, I have said here, and I'll make this just a little bit bigger, I have said here, okay, when given the task, you'll remember my role and these customized instructions when given the task by me, you will always help me save time by doing this, this, and this and this, right? And then your task here is to just remember the instructions each time I ask for tasks. So that's the way I'm tasking it, so that in this very first prompt, there is nothing the model necessarily needs to do. Now, depending on the model, it might summarize to let me know it's understood what I've said, or it might just say, okay, well, let me know when you're ready for your next task. So, for example, if I click over here to Gemini now, I've saved it here, and I'm going to go ahead and I'm going to run it. Okay? And so Gemini is like, okay, understood. Now, what I will tell you is the models are inconsistent. When I did this last night, it's summarized for me. And I said, so I understand this is what you want me to do, yada, yada, and it hit all of my points. And then it said, let me know when you're ready. Now, this time, it's generating it a little bit differently. All right? So I'm going to go over to co-pilot now, and I have my same chat here. here, okay? And I'm going to go ahead and let me just make sure, yep, it's all here. I'm going to go ahead and click that and it's thinking. And same deal. So very similar in this vein, this time. Now, both of the models understand me and we're using those customized instructions. Now, this is an export. I'm going to minimize just a little bit so I can grab it. Okay. So I'm going to now take the rest of this prompt. And I'm going to go ahead and copy it. and now I'm going to put it into each model. And then I'll run them. Okay, and we'll flip back in just a second. Okay. So, okay. So pretty much what I said here is that I want to do this project. I would like this outline of it, right, because it's this white paper. So I just kind of want to get an idea of how to write this. So Jim and I really goes into quite a bit of detail. So it started the outline. It's letting me know the executive summary should be a page, that they're key talking points. And it remembered that I actually want to infuse and bake in data into this white paper in this outline as well. So it has given me those perspectives also for the introduction, one page. And it's following this same consistency across it all here for the third bullet. It's saying here one to two pages. So it's giving me really great guidance on how to develop an outline and in turn to write this paper, right? Now I'm going to switch over to co-pilot. And so co-pilot, a little bit more generic, right? Let me just, let me zoom in just a little bit. I'm co-pilot here, right? So co-pilot, it's just keeping it quick and dirty. It's executive summary, key points, right? It's a lot less information for me to follow. it's not really giving me suggestions based on how long it should be and so forth. So I would say Jim and I wins here because it's giving me a lot more detail. Okay. Any questions on this particular activity? And you can feel free to try this one on your own as well. I wanted to just start one to give you to give you a feel for it. So you could very easily, you now have language that you can use. So I would argue that, for example, if your role is very administrative in nature or if you're a backdrop now. And so you could flesh in part of your job description, right? Or, you know, work that you do every day that you have outlined, you can plug and play that information in here and create your own custom instructions as well. So, Marie, am I okay? It's really quiet. Well, I mean, we can give them, it's 105. We can give them five minutes to practice and then wrap up if you want. Let's move on because that one off at that time went by really fast. Okay. So let's go by fast. So let's go to activity four. I want to show you one assistant. Okay. So I'm actually going to start a new chat for Gemini. And I am going to, let me see. I think I want to do a writing tutor. So the way this works, let me pull this out a little bit more. So I gave you kind of a description of three different assistants, a research assistant, a study partner. We can all use a study partner no matter what type of work we do, a writing tutor, same deal, right? So the writing tutor, just keep in mind there's probably something you're going to need to upload, right, if it's an essay or something like that, or maybe it's a white paper, et cetera, right? So there are fillable prompts that are here, right? So I didn't bold everything, but where the brackets are, if you wanted to do it on your own and do your own thing, you can plug and play. If you don't feel like doing that and you just want to test the prompts, I have completed prompts already here for you. And you can pick one. So if you want, the sound of my voice, you pick one. You'll pick your model. It's right up here. So either co-pilot or Gemini is here. It'll take you right to it. You can sign in. I'm going to go ahead and do, because this one is a really quick one. So I'm going to actually do the quiz one right here. So I'm going to copy that. And I'm actually going to plug that into Gemini. And then I'm going to enter and plug it into co-pilot. So I'm going to let Gemini work. And then I'm going to go ahead and do a new chat or copilot and do the same thing. Okay. So I'm going to go back to Jim and I real quick. And I can see here are my questions. Now, I'm not a subject matter expert, so I probably would need a subject matter expert. and definitely we need to verify all of these. But given the scenario, if you're using this, it's most likely you're going to be the subject matter expert in your area and you can verify if the information's right. Okay, so this is Gemini here and what Gemini rendered, and then here is co-pilot and what co -pilot rendered. So again, it looks like some of the same types of questions. There are all multiple choice. If I wanted, I could change the question type as well. Okay. So that's an example there. So I'll give you about a minute to play around. And do I need to look at the chat, Maria? Are we okay? See a lot of... There are some questions about the co-pilot agents are a custom feature, but you have to have a 365 co -pilot license. Correct. Correct. Do you want to clarify anything with that, Lauren? So really nothing to clarify. At University of Delaware, we are using free tools. So this is kind of an alternative to that. To my knowledge, there's no way to do those for free. So the customized instructions that you find here, at least an example or a use case of how you can use them, is here. So that's something that you could bring with you to the chats. And you can have that and just customize it each time, but keep your role or persona, you know, the function and you're giving it context for the model because it does need all of those things. Also, there's a question about do we do we need to customize our prompts for depending on which like whenever AI first came out it really made a difference whether you were in Claude or chat GPT or do you think that you need to customize specific to the one you're using or do most of them work effectively for the same kind of prompt? So I tend to have a habit. So for example, if I were using Claude, Cloud already has my context. So at this point, I don't necessarily always need to customize. But what I will do is I will keep those prompts if it loses the context. Because as these models change, like Claude is a good example. I used to use Claude all the time, but Clyde has gotten a little bit stingy. It doesn't give you as many free chats. Like it'll break you out. Like it'll say do shorter prompts and it'll keep making you want to upgrade. Like it'll keep trying to sell you on the upgrade. So I could do a longer prompt and customize, but then somehow with the algorithm that might give me less prompts before it's going to say, okay, the next time you can access Cloud is going to be at 4 a.m. Yep. It's great. So it's really, it's less about customizing your prompt now and more about just knowing how many freebies do I have left. I will also say that perplexity, like the ones that I showed you, those don't do fantastic with command type prompting. Like all of, as a researcher, as a librarian, most of my prompts are question-based. And when you move into the, like, write this with a specific voice using a specific tone, like, perplexity is not fantastic at that. Like, that's a better, better for one, like either Gemini, co-pilot, something like that, that's designed for the writing piece. Exactly. Now, if I have Cloud fans out there, Claude already has built in, you can design your own writing styles. And you can actually, within a single prompt, you can go back and forth to say, I want this formal. Oh, now I want this in my learning design lens. Now I want this. And you can build those, tweak those, continue to edit them. Now you can do that same thing just more manually with Gemini and with co-pilot. So I've kind of given you a hint that you can do that to steer the model the way you want it to go. So the more context and the more structure that you provide, the better you're going to have, the more effective your generated output will be. And that brings me to prompts decorators, which is the next piece. So prompt decorators that I learned about, those are the special and specific instructions that you give AI models so that it's more logical, it's more organized, and you help steer that a little bit. It can still hallucinate. I don't think we're ever going to get rid of that. I mean, think about it. In conversations when we're talking, do we not hallucinate a little bit sometimes? Like, we're not, you know, sometimes we might steer a little bit. We might embellish a little bit. So models do that same thing. So if we circle back to this guide here and you can play with these on your own too. So the key here is that plus plus, plus whatever the thing is, right? So in the first example, I'm done reasoning. So remember I said Gemini is really good with reasoning, right? So if you're using, co-pilot, let's just say, and maybe it's not giving you a really good reasoning. It's not helping explain why it came up with a certain response, because usually it's not going to just do that on its own. It's going to be a, you ask it something and it's going to give it to you. Every now and again, you get a golden nugget. We were like, oh, I didn't ask for that, but it gave me that, right? Like Gemini gave me that this should be one page. This should be two pages. I didn't ask for that, but it gave me that. So when you're talking about these decorators, you're asking it specifically to fact check or give you sources or to critique it or to refine it based on a certain number of iterations. And so that's what these are examples of. So what you're seeing here are examples example prompts. So like here is one that mixes reasoning and step by step. Here is another one that I'm mixing in the number of iterations for something. And then here is like the raw thing here. Right. So this is what that one looks like. So the end just means you put your number in. And then mixing, you know, because to me it makes sense to say, okay, fact check it, but then also cite my sources, right? So I can create a little hybrid combo of that. So you can feel free to test these out and see, like I like this one here. So I used another A model to help me discover this one. So, you know, drinking eight glasses of water daily is necessary for health. And you can test that out. You can fact check it. You can cite sources. So these are great activities if you're teaching to get students to learn how to use. the models and to learn how, you know, what's going to be effective and how they do their research. So these are nice little ways to do those pieces. So I'm going to stop for a second to see if they're questions because we're about at our time. And I know we did a lot. We shared a whole lot. But I also want to say that we are still here. Our information is at the very end of all of this. So a couple different things. And I can actually share this stuff in the chat as well too. So we do have some other workshops that are coming up. And I realize I put December in there. Let's see. So I'm going to put these in the chat. And I'm also going to share that these are upcoming ones. So if you want to join us again, you can. You can feel free to do so. So I'll put these in the chat. That also gives you access to our site where the rest of our workshops will be for the remainder of the spring semester. In addition to that, we love an opportunity for you to tell us how we're doing, and if you would like to see other things and share feedback, so I'm also going to put that in the chat and get those administrative things over with. But our contact information is within this slide, or within this slide deck, so you can always find us and ask us questions, because we've started to do this and people reach out from all different institutions, and we're all in this together. So please know that, and I'm actually going to stop my share, if I can see everybody. Just know that you're not by yourself in this. You do have us as resources. You have each other as resources. So let's just keep learning from each other. All right. Thanks, everybody. I hope you had a good time, and I hope we all hear from you. And I'm free to stick around a few more minutes if people have questions. Thanks for joining us from Oregon. Oh, nice. Excellent. Very cool. Yay. Thanks, Kathy. Sorry, I have questions, Lawrence and Maria. Africa, nice. We had someone from South Africa. Yes, go ahead. I'm listening. We are listening. Thank you so much for hosting this seminar. I love it. I would love to choice, like, if I have another open seminar. And I wonder if you will organize something, like go further, like if using AI for data analysis in research, just like the advanced part of this one. Yeah. Yeah. So we can put our emails in the chat, if that's okay, also. And it's in the slide deck. So that's me. Yeah. So, yeah, we're definitely. And that's me. Thank you. Let us know. Yeah. Just let us know. Feel free. Thank you so much. Thank you. All right. Good job, Lauren. Good job, Maria. We did it. We needed the extra 15 minutes. We definitely. We did. I feel bad. I think I took up a little more time than I should have. I'm so sorry. We had a lot of questions. So it was good. I'm going to stop recording too. Okay.
Using Generative AI Tools for Optimizing Writing & Research
From Lauren Kelley February 25, 2025
22 plays
22
0 comments
0
You unliked the media.
In this session, discover how generative AI tools such as Perplexity, Elicit, Consensus, Gemini, and Copilot can streamline literature reviews, enhance writing, and support research processes. This hands-on session will guide faculty and researchers in leveraging AI for more effective and comprehensive searches. Prerequisite: Participants will need a smart device to explore AI tools during the session.
Session materials:
Session materials:
Presentation Slides Link
Forced Copy link to worksheet
Padlet for resources
Chart with comparisons:
- Tags
- Appears In
Link to Media Page
Loading
Add a comment