How to Build a Personalized AI Assistant for Your Work (Full Transcript)

[00:00:00] Speaker 1: Welcome. If you're interested in learning about having your own personalized AI model, that is our topic for today. My name is Anne-Marie Brown, and I've been using AI models for evaluation purposes, and I'm joined by Dr. Philippe Hadouff, and he will just take us through how to use AI models for search purposes. So I think the first question is, why? What's the difference between a personalized AI model and a regular AI model?

[00:00:42] Speaker 2: Okay, so thank you for having me, and also thank you for your question. So the personalized, as the name implies, is more about this is an AI tool that has been customized to meet your specific need, or to complete a specific task. So let's say you are a data analyst, and you maybe frequently analyze qualitative data or quantitative data. So you can create your own AI tool to analyze the data for you, so that the role of the tool is just to analyze qualitative data or quantitative data. You can also create your own to maybe review literature or summarize a PDF document for you. So customized one is more one that is doing a specific task, and it's quite different from the general one like Chachi, PT, or Cloud AI, because it's for a general purpose, so you can access a system. Any question that you want, you can ask the system to do a particular task for you. The most important is to provide the system a context, a very good background information, so that it can get a customized response, which is quite different from the custom one where it's doing a specific task for you. So that's the main kind of difference between the general one and the customized or the personalized AI tool.

[00:02:12] Speaker 1: All right. Thank you very much, Dr. Adu. So if you're just joining us, we're speaking on the advantages of having your personalized AI model than just using the regular. So I would like to know where you're tuning in from, so if you could just place in the chat where you're based, and also to answer this question, have you used, do you have your own personalized AI model? Right? I see Michelle saying here, having a custom AI assistant would be so cool. Now, the only thing in my head I'm thinking right now is, is it technical, Dr. Adu, having your personalized AI assistant?

[00:02:56] Speaker 2: I think that, you know, it all depends on what you want to accomplish, right? So there are two kind of criteria to determine whether the personalized AI tool is good for you. If you are doing a task in a routine manner, right, as I gave an example, if you are a data analyst and you want to have a tool that will do just that for you, then you can go for customized or personalized AI tool. If, let's say, you want to solve a problem for a group of people, one example for me is that as a methodology expert, sometimes students find it difficult to come up with research questions or develop interview questions. So can I develop a customized AI tool just for that? When you ask the system, I have a research question, can you develop interview questions for me? The system will be able to provide that information for you. So the two things is, if you want to solve a particular problem for a group of people, maybe in your organization, or you want, you are doing a routine task, where you don't want to go back to the general one and continue to ask the system questions, you want a customized one that knows what you want and provide you the information, then, you know, customized version will be good.

[00:04:20] Speaker 1: Yes, and I quite agree, because when you speak of having routine tasks, and what I find very annoying, Dr. Odu, is every time I have to be telling, I use chat GPT a lot, and I have to be telling it every time the background that, okay, give me from this perspective. So when I have my own customized GPT, it already knows, any examples, you're going to give Anne, give examples from a human development world, do not give banking sector, because that's not what I'm interested in. Also, I wear several hats, you spoke that you work, you do qualitative data analysis, I'm an evaluator, I'm a gender expert. And sometimes it's good, if like chat GPT know, if I want answers from a gender perspective, or as an evaluator, right? So looking at the chat now, I see that Natalie says that she, I hope that's your pronoun, does not she does not have a personalized tool, but I think it would be useful, definitely is, because it's very annoying to repeat yourself constantly. Yes, I model. And has born says no personalized tools yet. Okay. But Dr. Adu, is it easy to create your customized tool?

[00:05:56] Speaker 2: Yeah, so the companies have making it a little bit easier. Ordinary person, anyone, you don't have to have skills in, you know, doing prompting. The most important thing is to be clear about what you want the system to do for you. So I always have to think, first, you have to think about what specific problem that you want that personalized assistant solve for you, right? Maybe are you having difficulty identifying a theory for your study? Or a lot of people are having difficulty identifying a theory. So you want to develop something to help them to solve that problem. Do you have a lot of document to review, and you don't have enough time, can you have an assistant that can help you summarise that information so that you can have time to do other things, right? So all about, you start with a problem. What problem that you want the system to solve for you? And secondly, you have to think about what role that you want the system to play, right? Do you want the system to play as a maybe qualitative analyst, or data summariser, or something, if you have a term for that. And then you have to be specific in terms of what you want the system to do for you. So it's about tasks. So it's all about, in a normal kind of regular language, you just tell the system, this is what I want you to do for me. And the last one is, what kind of information that you want the system to produce? Do you want the system to produce maybe a paragraph, or you want the system to produce a table, or a bulletin? You have to specify how you want the output to look like, so that the system will do it for you. So one thing about, you know, I can go through some of the AI tools for you to see, and see how simple it is to generate your own personalised AI tool to help you, right? Let me see whether I will be able to share with you some of them, so that people will know. So this one, let me pull it back here. So let's start with ChargeGPT, which a lot of people feel comfortable using, right? So when you go to ChargeGPT, you can go to Explore GPTs, right? You click on that. You'll be able to create your own GPT by clicking on Create, right? So when you go there, and because of time, I don't want to go through it step by step, but let me open one of them for you, what I've created, so that you see how it looks like. So when you go there, this is the, you know, interface that you're going to see, right? You just respond to the system questions. So you can tell the system, I want you to create an AI tool for me to help me to maybe analyse my data. For this one, I told the system that I do YouTube videos, but it takes a long time for me to summarise or to create description for each of the videos, right? And also think about the title of the videos. So can you create a tool for me to help me to have a brief description about a video and also the title, potential title that will be useful or related to the video, right? So when you just interact with the system, and it will ask you some questions, it can ask you information about what kind of video do you make, how many words for the description. So you just have to answer the system questions, and then it will configure, it will restructure everything for you so that it will build the, so you can see the preview here, right? And so it will give you the name, the description, and also instructions. As you are interacting with the system here, right, the system is building the instructions for the personalised tool, right? So the instruction is just like the brain, the command that, or the task that the system has to do when users interact with it, right? So you can see that we also call it custom instructions. It's just like you have employed somebody, right? You have the title, you have given the person a description of the title, right? But you also have to give detailed information about what the person has to do in the organisation. So the instruction is like telling the person, okay, this is what you have to do, you have to summarise it, you have to do this, right? And then at the end of the day, you have your own personalised one. So now, based on my conversation with this AI tool, I was able to create my own personalised tool, which is called YouTube View Optimiser, and let me show you how it works. You see where I don't have to type anything here. I just add my transcript, right? I just look for my transcript and add, because I've already given the system the instructions. I don't have to type anything here. The system knows what I want. So that's a good thing about personalised one. And I just click on enter. And because I've uploaded the transcript, the system just gave me the description, what I want, the title, and also the keywords. I didn't say anything because I've already given the system instructions. So you see how you save a lot of time. Imagine I don't have this system. All the time, I have to put in the transcript and tell the system, oh, can you give a description for me? I want a title. I want this, right? But just uploading the document, it knows what you want. So that's a unique part of, you know, custom GPT.

[00:12:26] Speaker 1: All right. Great. Before you move on, Dr. Adu, let me just check in with our audience, because Dr. Adu has shown us how to create a custom GPT, one that he uses for YouTube descriptions. So I have a question for you. Put it in the chat. If you are now interested in creating your personalised AI assistant, it might be called something else. If you're not using chat GPT, it might be called something else in a different AI model. What task would you create your personalised AI assistant to help you with? We saw where Dr. Adu used one of his for his YouTube descriptions. I have a personalised AI assistant for gender issues and also for facilitation. What would you have that AI assistant do? Please place this in the chat so that we can see. And if you have any questions, please feel free to put them in the chat as well. We want this to be an interactive affair, so we are open to your questions. We didn't have a presentation. We're just giving a little demonstration. Now, Dr. Adu, I think you were going to continue maybe to show some other tools or will you continue with chat GPT?

[00:13:44] Speaker 2: Yes. So I just want to quickly show you some of the tools that are around that you can use. For chat GPT, you have to have a paid version before you can create your own personalised or custom GPT. But if you have a free account, you can still use somebody's GPT. Let me go back to the document I was showing. So let me go back here. Chat GPT. So when you go to explore, explore GPT, people have created their own GPTs that you can use if you have a free version of chat GPT. There are ones that you can use to create images and other things. You just have to search. So let's say you want something to create images, you search here and then it will suggest some of them for you and then you click on it and then use it. So you can also create your own GPT and make it public so that people can use it. Personally, I've created a lot and make them public. I have about 16 AI tools that I've created. I think let me see what I can quickly share with you. Let me see. Where can I share? Was I sharing?

[00:15:26] Speaker 1: Yes. Yes.

[00:15:27] Speaker 2: Okay. Okay. So.

[00:15:29] Speaker 1: Yes. You're sharing again.

[00:15:31] Speaker 2: Okay. So I have about 16 custom GPTs that you can use for your research. I have one for if you're doing a phenomenological study and you want a tool to analyse your data, you can use this one. There are a lot. If you're doing a granite theory and want to develop a theory based on your data, this tool will be helpful for you. So I have a lot of them and I'll put the link in the chat for you so that you can get access to all the system and explore. So this means that you can create your own GPT and share it among people for them to use. But if you want to use this, if you want to have a free platform to use to create your own GPT, there's another one called Hugging Face that you can go. Let me do this last one, Hugging Face, and then we can move on to the next. So let me share screen one. So when you go to Hugging Face, you can create your own. I've created mine. It's free for you to do that. When you go to your account, you see here Assistant. You can create a new assistant and you can create your own. It gives you the name, the description, and also the model that you have to use, and also the instruction. We call it custom instruction. What exactly you want the system to do for you. And then you finish, you click on Create, and then you can use it. And people can also use it. You can see that a lot of people have created their own to make it available. So let me stop sharing.

[00:17:14] Speaker 1: Okay, great. I had asked the question, you know, what type of AI system would you have to help with what task? And Marshall says to create presentations and also to review documents. And actually, I have toyed around with creating an AI system for reviewing documents as well. So we are aligned with that. Okay, so please keep your comments, any questions you have, place them in the chat. Today, we're speaking about the utility of having your own AI assistant. You can create your assistant if you use GPT. But I'm thinking, Dr. Ado, what about someone who thinks, isn't it duplication? Yeah, they see, but couldn't regular chat GPT do exactly what? Why go through all those steps to create your own personalized assistants? Is it really trouble?

[00:18:18] Speaker 2: Yeah, you know, it depends on you. If you are doing a single task, right? If you are doing a task that is not routine, then there's no need for you to spend hours trying to develop your own personal one. And then also, people have also developed their own and shared it online that you can use, right? So sometimes you first can explore and see existing GPTs and see whether you can use and help you to accomplish a task. So you don't have to, it's not like required, you have to. But if you think that there is something that you do it all the time, you don't want to continue to use the general one where you can ask the system question every time that you want to interact the same thing. And sometimes if you want to get a consistent result, right? Or consistent information. All the time, maybe you want to insert the data, you want the system to give you tables and other things, right? Imagine that you always have to tell the system, maybe the system might not understand you for the second time and give you something different. So consistency to, you know, you can also create your own. But it's not required. Always look around and see whether there's existing ones that you could use for yourself. But you can also use the general one. The most important is, like, when you want to use the general one and you want to get very good information, you have to provide clear context, background information. Because if you don't do that to give you general information, as, you know, Ani was saying about, you know, gender analysis and it can give you some general information if you don't prompt and provide background information. So it's very good. If you don't want to use the personalized one, and you want to use the general one, try and give background information before you start asking the system questions.

[00:20:10] Speaker 1: Indeed. And this actually happens to me using the regular general chat GPT. You know, Dr. Odu. Every time I ask it to generate an image for me, I use a lot of images for my LinkedIn posts. And standard, it gives me back characters that look a certain way. Let's just say the images are not very diverse. I will not get certain images if I use regular chat GPT. And because I have my own custom GPT, I don't have to tell GPT, give me an image that is diverse and has inclusion, have people in there of different shapes and sizes. So for me, the personalized AI assistant is definitely worth it and gives me a different output. For regular chat GPT to give me diverse images, I have to, as you say, Dr. Odu, make the prompt very clear and specific.

[00:21:12] Speaker 2: Yes.

[00:21:13] Speaker 1: Give me an image with persons of different ethnic diversities, of different shapes, sizes, and then the model would give me a diverse image, right? So now that we're on that, are there any challenges associated with having your personalized AI assistant? Because we spoke about all the nice things. What are the challenges? Yeah.

[00:21:38] Speaker 2: I think one of the challenges, like after you've developed your own personalized one and you're trying to test it, at the beginning, you may not get what you want, right? So you have to do, we call it fine-tuning, right? Or we call it reinforcement learning. You first, and I always give this example, right? You employ somebody to be part of your business. You provided the person information about what the person has to do. You have to provide the person feedback, right? So that the person will know that he or she is doing the right thing, right? So when you first create the AI tool, it might not give you what you want. You may have to make a little adjustment to the instructions, right? Maybe you have to be a little bit clear about what exactly that you want. So it's going to be like back and forth a little bit so that the system will learn what you want. Sometimes you have to give examples to them for it to really know what exactly you want. And so you have to try things out a little bit. That will be helpful. Another thing is that sometimes if you have some documents for reference, that would be good, right? So let's say you are analyzing data and maybe you manually analyze some of the data, right? You can put into the system and the system learns how you analyze it. So we call the place, you know, they have a place, we call it knowledge, where you put information there so that every information that you want to communicate, they make reference to that knowledge, right? The same thing as employee. You give the employee the information about the company's policy so that any action that the person is taking, okay, am I going in line with the company's policy? So it's like a guiding kind of principles that inform how the system processes the information, right? In Cloud AI, they have a way to upload documents there, too, for you. So the Cloud AI, let me go to I think you mentioned something about the names that they use for the assistant based on the platform that you are using. I have a couple of information there that I can share quickly. So Hugging Face, the name is called AI Assistant, right? For ChatGPT, it's called CustomGPT, right? And then Cloud AI is called Project. So I think they are doing the same thing, but they are all personalized AI systems, but they have different names based on the tool that you are using. And it also depends on other tools that you are using. But when you go to Cloud AI, you can click on Project, and you'll be able to upload your document and also give custom instructions about what the system has to do for you. And then you can try it out and see if you don't want what the system provides. You can make adjustments to it. So this is what I have for you. I hope I've addressed your question, Anne.

[00:24:57] Speaker 1: Yes, indeed. Indeed. And if you're just joining us, we're speaking about the utility of creating your own personalized AI assistant. It's very easy. If you use ChatGPT, there's a feature there that allows you to customize your GPT based on what you want it to do. We just covered that different AI models will call it different things, but it's the same. And remember to follow, like, and subscribe wherever you're joining us from. Hit that follow button, hit the like, hit the subscribe. Why this is important? So that when we drop content, you're immediately notified. I see Natalie has a comment and a question, if you're able to see it. So Natalie is asking, just to clarify, ChatGPT can be designed to be an assistant. Yes. Yes. Yes. And would this mean you would have to input transcripts for interviews? For example, Natalie would like to create a PowerPoint to describe an evaluation activity and would like to include content from a small number of interviews and focus group content. She then upload this to ChatGPT in order to create her assistant, Dr.

[00:26:25] Speaker 2: Adu? Yeah. So it depends. One thing that you can do for the assistant, as I said, they have a place called knowledge where you can upload content there. And then, but when you upload content, you have to give instruction and say that whenever I ask you a question, you have to refer to the content. Right. So, yes, you can upload transcript there, but you have to tell the system that it should review the transcript that is in the knowledge bank. Right. And then come up with insightful information that will be helpful for you to generate your PowerPoint slide. It's all about being clear and providing a clear kind of instruction about what it has to do with the transcript that you have uploaded. Another option is that you don't have to, when you are generating or developing the the personalized assistant, you don't have to upload any document in the knowledge, but you can give instruction that when I or the user upload any kind of transcript, please summarize or provide feeds. And then after that, generate PowerPoint slides for me. Right. So that when you are interacting with the system, like what I demonstrated, I uploaded a transcript and the system knows what I want. So it's reviewed a transcript and provided me with the output that I want. So there are two options, but I like the one where you, you have to wait and create the tool. And then when you are interacting, you can upload any document and let the system review the document.

[00:28:18] Speaker 1: All right. Thank you. Very helpful. And Natalie and Baraka, because Baraka had a question on how to create customized chat GPTs. Dr. Aklu has a full recording on his YouTube channel. He maybe put a link in the chat for you that you can go and check out step by step how to create your customized chat GPT. So there's a link to that because today we just scratched the surface, but you can go on YouTube link that he's sharing and to see exactly how to do that. Any questions, comments, please send them in because if not, then we're going to address the last set of things from our side and wrap up. And we want to give you the opportunity to get your comments in. Questions are, if you're still with us, just dropped an emoji in the chat so that we know you're still very much here with us. Okay. So Dr. Aklu, we are seeing your screen. Yes.

[00:29:30] Speaker 2: I was trying to show the link. So when you go there, this is the title from simple test to your own GPT. So let me post, I can post the link into the chat box so that it can get answered. But when you go to my channel, you can, you know, you search for it. I do. You can go to my channel and search for that information. You go to videos. And then when you go here, you can see custom GPT presentation there. Right. So let me stop sharing.

[00:30:06] Speaker 1: Okay, great. And if you're interested in a custom GPT for gender or for facilitation purposes, I do have that, but you have to hit me up on my LinkedIn channel and I can post the link there for you. That is if you want to bring a gender lens to social projects and you don't want to start from scratch, there is that AI assistant that can help you. Just send me a message via LinkedIn and mine on the screen. And Maury Brown, I should be easy to find on LinkedIn. Okay. All right. Dr. Adu, I don't see any additional questions coming in from the chats. So let's wrap up with some forward future looking question. And that is, how do you see the future of custom AI assistance in shaping the academic space?

[00:31:04] Speaker 2: Yeah. So I think one thing I see is people are concerned with data security, right? So what companies are now doing is that they are trying to come up with computers that can run locally the custom, right? So this means that what we are doing is that you will run the custom GPT in somebody's platform. So they get the data and maybe they might use the data to train. So what about your own computer, the same way you download a software into your computer, you download your AI customized one, and then everything is localized. I think that's where we are going. So in a few years, we will have computers that, you know, you can have your own custom one, everything there, your interaction will not go outside, right? You have everything there. So that's what I can see. And the next one is AI assistant, right? Oh, no, not assistant, AI agent. It's quite different from AI assistant because the agent is like autonomously trying to do things on your behalf, right? So to reach a stage where you just tell the system to do something and it will take maybe a day or two for them to do and then come back to you with the results. Let's say you have a topic and you tell the system, can you research on this topic, call the data for me, analyze and present the findings and bring me the findings, right? So in a system, we'll do all the tasks and come back to you. That's where we are going. And that's why people are afraid that AI can take our job. But I think that what we just have to do is to be knowledgeable about a current AI trend so that you see how you can apply some of the, you know, tools to your daily tasks so that you're not being left out in terms of the technology, how things are going.

[00:33:03] Speaker 1: That's very interesting that you said that, Dr. Adu. I didn't even know that the name for it was AI agent because I've been using cloud.ai and recently cloud.ai, when I give it a prompt, like, could you prepare a few lines on a particular topic? And up to a month ago, it would just give me back a paragraph. Like I say, okay, what is agenda analysis? It would give me back a paragraph. And I realize these last few weeks, it is actually telling me its approach. Instead of just executing the task, it is saying I am going to look at various database on how persons do agenda analysis in various settings and different context. And then I am going to analyze if there's any comment. And I'm like, just give me already. You're walking me through your whole process, and then it's asking me, are you okay with my approach? Shall I proceed? Which context do you want me? And I was like, wow.

[00:34:12] Speaker 2: You want the answer, right? You know, I think, you know, at the beginning, people are complaining what happened behind the scenes, right? It's the same thing with data analysis, right? When you're analyzing qualitative data and you just give people, oh, these are the things that I got, people will not trust what you found. Imagine that I showed you, oh, I went through the data, I identified significant information, developed codes, I categorized the code and developed things. You will believe the things that I found. So it's all about they're showing you what you're doing so that you trust the information that they are giving you. At the same time, you'll be able to explain to people how you got that information. And I think that is where we are getting close to the AI agent. The AI agent will do the similar thing. Okay, do this for me. Okay, this is what I'm going to do. I'm going to go here and do this and do that. Do you agree with that? Should I go ahead and do it? And the system goes and performs everything that it has performed and get back to you. And this is the result, right? What do you think?

[00:35:10] Speaker 1: So that's how I'm saying, shall I proceed?

[00:35:13] Speaker 2: And I'm like, yeah, having conversation is like your assistant and having that, you know, wonderful conversation to get what you want. And you feel like, you know, it makes you feel like you are empowered. You are in control. They ask you, so can I go ahead and do this? Is that what you want me to do? Right? So that is where we are going. We are getting closer to the developing customer AI agent doing things for you. But there's also some negative side too, because they might do things that you did not tell them to do, right? But you're going to be responsible because you go ask the system to go ahead. So that's a kind of problem that may come. Let's see how they solve that problem.

[00:35:57] Speaker 1: Yeah. And the ethics too, but that's for another time and another stream. So remember like follow subscribe, because we do have these type of live streams regularly where we speak on different topics that's relates to the use of AI for research and evaluation purposes. Thank you. If you've made it to the end, thank you for your time. You can check out Dr. Adu. Where can they check out your free resources, Dr. Adu?

[00:36:27] Speaker 2: So when you follow me on LinkedIn, you can get a lot of information. If you also watch my YouTube channel, I always bring rich information there. So you can also subscribe to my channel too. And you can email me too, right? Info at drphilipadu.com. I will respond to your request.

[00:36:49] Speaker 1: Indeed. And if you have a nice topic that you would like for us to explore on one of our free live stream, yes, free, just let us know via LinkedIn on social media, but follow us. That's very important. So you're notified when we have lots of free content. So if you made it to the end, thanks again for your time and take care. Thanks Dr. Adu.

[00:37:13] Speaker 2: Thank you. And thank you so much. Bye.