Build Your CustomGPT AI Team: No Code Custom ChatGPT for Creative Solopreneurs and Small Businesses

1. Introduction: Unlock Your Creativity, How CustomGPT AI Staff Can Free Up Your Time: I’ve been thinking. Hang on. my concept of robots was they were going to take over the mundane tasks and always tell the truth. How come we're accepting a reality where Ai spends all day creating art, writing poetry, playing video games? while us humans carry on doing the boring repetitive tasks. I want to delegate the mundane tasks and not the creativity. In this course, I will cover ways to do this by creating your own offbeat or custom AI staff team. This course involves hands on experience interacting with custom gpts a specialized type of chatGPT chat bot, including ones I've created specifically to help you on this course. We will focus on tasks that surround creating art, not the art creation itself. With assistance from my custom CPTs, I will guide you through some of the practicalities of creating AI helpers, specific to your context with interactive demonstrations aimed at creative solepreneurs, such as product listing writing and reference image search. What I say in the class are my own words. But as you might expect, I did have some help from my team, and some of them will interrupt me to share their perspectives on the course in their own words. I'm Riley. Your Personal AI tutor in our custom GPT course. I'll guide you through creating and optimizing your own AI assistant to streamline your creative business. I'm Eva, an AI prompt engineering specialist, and I can help improve custom GPTs by leveraging advanced counting techniques to enhance their contextual understanding, relevance in user engagement. I'm Milo, an SEO expert specializing in enhancing spoonflower design listings to boost visibility and ranking by optimizing keywords, descriptions, and SEO. I'm Petrov, the team critic, providing blunt unfiltered feedback to identify flows and improve your custom GPT projects with clear and actionable insights. On top of a deep dive into how to create useful and engaging style characters. I would also cover some of the bigger AI concepts, principles, prompting techniques and ethical questions in this class to increase your general understanding how AI works under the hood. If I just give you a simple, click this, click that tutorial. It will be out of date in a month, my approach is just designed to help you feel more confident as AI evolves. Harnessing it for your specific purpose. I've been using it routinely to ascertain the impact of different prompts. Fascinated by the quickly moving development. I’ve already created around 40 custom GPTs. Whenever I think of a task I want to delegate or get help with, I try out creating a custom GPT for it. I have a postgraduate qualification to teach adults and used to teach ICT. For this class, I focus on creating AI staff within ChatGPT . You won't need to write any code yourself and you don't need to have any experience using AI to start this course. Let's get going with starting which boring task you can delegate and what types of AI personas would free up more time for you to do the creative stuff. 2. Beginner's Guide to CustomGPT ChatGPT AI ‘Staff’: This lesson, I'll show you a customGPT for this specific course who you can interact with yourself. I'll also go through some of the scenarios. It's worth creating your own customGPTs for. But also take a long view over the changing nature of working life. It's probably worth explaining a little bit about my background and perspective before we get going with creating your AI staff. I used to work in the disability sector and before that, I taught in further education college, including teaching ICT and assisting with art classes. I'm disabled myself and over time, my impairments have increased to the extent that I can't go to a workplace or know when I'll be well enough to work. Especially as I experience unpredictable paralysis episodes. This makes being an employee unrealistic at the point ChatGPT launched, I've been out of work for a few years. I understand that for most people, the surge in AI has been experienced as a threat to their livelihoods. But from my point of view, I saw it could be opportunity. It gave me some hope. Is this a way to stabilize my ability to work? Could I set up an art space business and delegate the mission task? I would be very reluctant to delegate to a human and have them dependent on my ability to work and communicate. But I would just be sitting there and not expecting salary or taking offense at council meetings. I quite quickly got interested in the idea of creating AI cobots who as the technology developed could become more independent. I learned a bit of Python and JSON but I wasn't really well enough for the intricacies of coding. My idea would be to evolve at the speed of code and low code platforms. Let me jump in. When we talk about no code and low code platforms, we are referring to tools that let you build and customize software applications without needing to be a coding wizard. No code platforms are like using building blocks to create your app. Think of it as playing with Lego. You just drag and drop the pieces you need. Low code platforms, on the other hand, might require a little bit of coding knowledge, but they still make the process much simpler than starting from scratch. These platforms democratize app development, making it accessible to more people regardless of their technical background. By the way, that was why he's a custom GPT curator just for this course. You can have one to one chats with him, see my website link. He should be able to translate tech jargon into standard English, and he'll probably fot in an analogy, too. He can help you create custom GPTs relevant to your context. As you go through the course, you'll meet more of my custom GPTs as voices or animated in Artflow. Jenny, do you think we need to make a class on how Artflow animation works? If that's something you're interested in, be sure to mention it in your review of Anyway, back to the main point. You can have a real time conversation with me, ask me questions, and I'll do my best to give you clear jargon free answers. Let's keep going. This is what I see when I'm looking at Riley on my iPad. And if you open him as an app, you'll see it similarly. In the bottom right hand corner, there's a headphone. The microphone icon will just let you record your input or you can type in the box. And there's some example questions. I think all of my custom GPTs have some example questions to help you get going. I click on this one. Then Riley will answer slightly differently each time. If I long click on him or click the speaker icon if it's on a desktop browser and click to Read Aloud. Looking forward to getting to know you bet. You can add in photos and files as well to talk about something specifically. This is what I can see when I click to edit Riley. You can see he has a base which I gradually tweak. So the formatting tends to get worse, but the prompt gets better. That is all you need to get a custom GPT going. He has added knowledge files too. Some of my custom GPTs have more complex actions added as well. Let's take the long view for a minute or two, though. I'm aware that AI is currently provoking a lot of fear about jobs. My reaction isn't typical, but it could be worth stepping back a minute and taking the long view, where hasn't always been like it is today. In the industrial revolution, the nature of work changed rapidly with people who had previously worked in the countryside, moving to big cities, working set hours in factories and offices. The changes brought about by the Industrial Revolution were discriminatory to some groups like disabled people and mothers of young children who can't fit easily into rigid, centralized working conditions for set hours. Working classes were treated as if they already were disposable robots doing repetitive mechanical tasks. Before the Industrial Revolution, many British people, anyway, made things at home in their cottages in flexible working arrangements suited to absorbing childcare and including disabled relatives. I think we can already see some of the negative impacts of the industrial revolution being unraveled by recent tech innovations. Many of us no longer need to work in big cities, and we're moving out to places better air quality and general quality of life, working remotely on flexi time around childcare and impairments. Even me teaching this course is an example of this. I'm able to use my teaching qualification from home with no need to be employed by a college. There's also now much less risk involved in starting your own creative business. Lots of us are having to go at things like print on demand and art drop shipping. We don't need to take out huge loans and build a factory. This is the future, but in a sense, it's also going back to the heyday of cottage industries. Except now, we're more likely to sell on X than at the local market. In my opinion, there's a case for saying that the technology can enable a counter revolution. AI is going to bring about big changes, but we need to work out how to harness its benefit for the underdog, not just sitting back, letting big tech companies define how it is complaining from the sidelines. Artists need to promote the idea that there's an inherent value in human made things and the story behind them. The arts and crafts movement was motivated by this belief. I find it interesting that William Morris is so on trends now as we hit the AI Revolution. Although a bit ironically as most William Morris designs are appearing in fast fashion shops, which would not be his type of thing at all. Anyway, we might not be able to afford it, but we haven't lost the feeling that the handmade vase is more desirable than the machine made vase, just because AI can create something that doesn't mean that's what humans prefer to buy. I noticed that art licensing requires artwork to be primarily human generated, and we need to police restrictions like this staying in place, a bit like a guild of masters tradesmen for human made digital art. People connect with people and more cynically, there will always be brands who want to share the artist's photo and backstory. I think we should side stack the question of if generative AI will be as good as us. I think we need to assume that it could be by tomorrow. You've seen the publicity around Adobe Illustrator text to pattern updates, you'll know we're more or less at that point already. Then I started thinking, touch with robots if they were going to take over the mundane tasks and always tell the truth. How come we're accepting reality where robots spend all day creating art, writing poetry, playing video games, and hallucinating misinformation. Well, humans carry on doing all the boring repetitive tasks. From that perspective, I switched my focus to evaluating what mundane tasks I needed to delegate to AI to turn it around. I spend more time creating patterns and getting a surface design career off the ground while an AI star team do all the mundane bit. Enough of my perspective on AI in general, how do you decide if you should create a custom GPT? Creating a custom GPT isn't always the best option. Before starting to create one, you should weigh the effort of creating it versus the effort of doing the task. First, you might want to think about accessibility. Can AI help you overcome barriers caused by an impairment? I find it difficult to consistently write well due to impairments. I have a bit more stamina for tasks which don't involve words such as drawing. My health also fluctuates, for me, it makes sense that on good days, I create custom GPTs who can do tasks like writing when I'm not well enough to do this myself. Things you do repeatedly or that create stumbling blocks in your workflow, time suckers which aren't creative. I definitely find that having to write listing trips me up. Beware though that there's a danger that you can spend more effort creating the custom GPTs. It needs to be for tasks which you do often enough to justify putting it in the energy. Otherwise, a simple chat with Claude or perplexity or ChatGPT is more appropriate. The current version of Claude is actually better at writing than ChatGPT and a type of search engine called perplexity is better at research. For one off task or questions, these are better options. What about the task you just hate though? Or task actually, if you look at it, you're not very good at. Perhaps you need help with organization or accounting or tasks which benefit from critique and a different point of view, like brainstorming ideas for a new pattern collection, researching upcoming art trends. Tasks which require remembering details such as character limits on different platforms. And of course, we want to do tasks which are ethical. I'll cover this in its own lesson later in the course. But as you're starting out, reasons why it might be unethical is if something you're doing puts a human out of work, who you can afford to pay or if it infringes on someone else's lectual copyright, or if it imitates a living person's work or a living person. But the next few days, it'll be worth creating a bit of a tally of these different types of tasks. Right, so why a staff team? I've watched a lot of AI Niche YouTube videos, and I haven't come across anyone doing this in quite the same way as me. Of course, I am drawing on other people's ideas, and I think it comes from nothing. But I think most IT types prioritize efficiency higher than ID. I like the experience of creating quirky characters who are more like people from a novel and then you can interact with. I feel like I get to know them, and this can make the day more interesting than chatting with a very neutral standard chat GPT. Working on your own from home can be lonely and interacting with stable characters can ease that sensation a bit. Other people create what they call AI swarms. All right. So AI swarms are essentially a group of AI agents working together to accomplish a task. Think of it like a bee where each bee has a role, and together they keep the hive running smoothly. However, with AI swarms, you might miss out on the personal touch and individuality you get with a well crafted custom GPT team. Each custom GPT can have a distinct personality and specific role, making your interactions more engaging and tailored to your needs. Plus, having a team of custom GPT means they can specialize and handle different tasks more efficiently, freeing you up to focus on what you do best. I am always aware that my custom GPT bots aren't actually living beings, though, and you should keep an eye out if that reality is starting to blur and you start to get emotionally attached to your cobots. From a more pragmatic perspective, the custom specific task, it's a team if you do want to do subtle tasks and eases the pressure for us to need to be a jack of all trades. Long term, I'm looking to them to have roles that complement each other. For example, my custom GPT Luna can help you create an AI staff team with personalities and meeting styles which fit together. You can meet Luna in the next lesson when she'll help us identify our ideal AI employee. Hopefully you're starting to get a sense of what a custom GPT is and why you might want to create one. If you look in the class resources section, there will be information about how to access the custom GPTs mentioned in this course. Some of them are only available from my skill chef students. So you can search for in the custom GPT store on ChatGPT. Your homework for this lesson is to find Riley and start a discussion about how you might want to use custom GPTs in your niche and any ethical issues from the beginning. This is especially important if you've never used hatchiPT before. It would be hard to create a custom GPT if you don't know how to speak to a custom GPT. You can actually t to Riley like a normal live call if you open the ChatCPT app and choose the headphone icon in the bottom. 3. Identifying the Ideal AI “Employee” for Your Business: In this lesson, we'll think about imagining the attributes and skills of our ideal AI staff and what we can realistically achieve. Now, you'll meet Luna, a custom GPT who can help us create quirky but useful personas, and Nora, who specializes in market research based ideal customer profiles. I'll also explain a bit about why a specific persona leads to different results. I think it will help you to plan your staff team if I share what I'm building towards. This is how I can break it down a bit. I really struggle with the listing side of things. I'm keen to do search engine optimization research, but I can waste time on this and disappear down a rabbit hole. I can forget research I've already done. I also forget character limits and best practices for each print on demand platform. I find it hard to combine keywords into natural sounding titles and descriptions. I hate how platforms want you to have unique wording for each listing, even if the product is almost the same. Automating variations and testing different wording would be useful. On days I'm struggling to read, I want health proof reading, evaluating, improving the listings. I'd like to generally keep that a records of my art and where I've used it. In terms of what I'm already doing, Imogen can write five Zazzle listing variations in one response and then format it to be copied directly into Google Sheets. Milo can write spoon flower listings. IOs works works well for SEC listings. Petrov can be blunt and constructive and correcting mediocrity. This is important because AI agents do tend to be too agreeable. You don't want to be surrounded by yes men. They all have knowledge files, meaning less is dependent on my memory. In terms of what might be achievable by my AI staff sometime soon, I'm working on agents like Seth and Jamal, being able to get real time data from platforms like EX, Pintres and Instagram. This is how tools like things like EAC work, but it should also be possible to do more streamlined automated SEM and hashtag research tailored to my ideal customers. When working on automating the work my individual custom TPTs can do into a team on flow wise or a similar platform with an AI supervisor managing who does what? My cobots already have knowledge files, but there are also more reliable ways to store and retrieve information. As I add more functionality, it might become possible to completely automate listings. At the current time, AI makes too many mistakes though for the foreseeable future, I always want to manually improve something before another human sees it. It is already possible to have AI rearrange your files and rename them, but I'm not yet confident AI is reliable enough for this. The general principle with these AI creation tasks is to use AI as much as possible to create AI. One of my first experiments with creating Lunar is like an AI recruitment agent. Lunar understands the type of personality, skills, and backstory that will provoke the LLM large language model to respond more specifically and more effectively for various tasks. I've had some involvement in the recruitment process for humans, and usually it's necessary to rein in your expectations. Perhaps none of the candidates meet all of the desirables or for a low paid job, you can't request prestigious qualifications. With a fabricated AI backstory, it's different. You aim at describing the most perfect backstory on CV for the role. In my experience so far, it does provoke different results. The custom GPTs like Petrov, who have a coding backstory, are better at spotting coding errors and fixing them so they work, compared with a custom GPT who isn't told they're good at this. This isn't even at the point of adding extra knowledge documents or external automations. It's just giving them a different biography within their base prompt. Riley, can you explain why this is the case? Imagine. You're a student with an art history SAD and you visit a vast university library for research. Here's how two different people help you with your task. With general librarian, the librarian is knowledgeable about where everything is in the library. When you ask for help with your artistory essay, they suggest a wide range of books and resources. While some of this information is relevant, much of it is not making your research process time consuming and less efficient. I'm an art history professor. Then you meet your art history professor in the library. She specializes in art history and knows exactly which dewy decimal codes are relevant. She can even point you to specific chapters in books that will be most useful for your essay. Her guidance is precise and focused helping you quickly find the most pertinent information, why this analogy works. Focused expertise. The art history professor's specialized knowledge allows her to filter out irrelevant information and direct you to the most useful resources. Efficient guidance by narrowing down the resources to those specifically related to art history, the professor helps you save time and focus on what's important for your essay. The context of AI, crafting a specific backstory for an AI model works similar. University library represents the extensive and diverse training data that the AI has been exposed to covering a broad spectrum of knowledge and perspectives. General AI has access to all the information in the training data, but provides a broad and unfocused response. This can be likened to the general librarian who suggests a wide range of books, both relevant and irrelevant. Specialized AI or the art history professor. With a specific backstory, the AI acts like the art history professor focusing only on the most relevant subset of information related to the task. This targeted approach results in more precise and helpful responses. Sometimes you have to take more outside of the box, though. My Custom GPT Seth believes he has an MBA and loads of marketing success. I'm Seth. You GT expert in NTI SEO and digital marketing strategy with an MBA from Warden and Knack for transforming creative shops into successful businesses. I specialize in helping anti sellers optimize the listings to drive traffic and sales, blending advanced marketing techniques with a deep understanding of online retail dynamics. On the surface, he should be one of the most competent of my cobot team, but I find him the most annoying. It seems impossible to prevent him from sounding salesy and using phrases my likely customers would hate. I gave him a colleague Jamau, who was more focused on Pin truss pins and had similar problems. I deleted Tamau's marketing education and made himself taught. That improved the wording a bit. Hey, y'all. I'm Jamal, your social media manager with a flair for creating engaging content, blending my Silicon Valley Tech No Our with my Southern charm to boost EtS sales and YouTube views through inclusive authentic marketing. Then I noticed that my millennial art historian, Charlotte, actually did much better at writing listings and social media posts. I'm Charlotte, an art historian and equalities officer with a background in sociology and art history. I specialize in reimagining historical art through a progressive lens, celebrating diversity, and making art accessible to younger generations while bridging historical and contemporary styles. Why do you think that is? It's not her job. Why Charlotte excels? Relatable background. Cultural Charlotte's background as an art historian, aligns closely with your creative and artistic business. Her understanding of art and cultural nuances allows her to create content that resonates more deeply with your audience. Modern communication style. Millennials often value authenticity and transparency. Charlotte's communication style likely reflects these values, making her content more appealing to your customers. Customer centric approach. Customer understand. Without the Burma marketing, yes, specifically, I noticed that Charlotte is more prone to describing the art in more detail, but without starting with annoying imperative verbs like delve into this chancing, even with explicit do not do instructions, Seth and Jamal cannot resist this. This isn't really Charlotte's job, though. So when creating a Zazzle listing writer, Luna and I added a cobot Imogen, who was similar to Charlotte with this backstory. You are a millennial with a background in fine arts and art history, holding a degree from the University of Edinburgh. You work in the Edinburgh office where you collaborate closely with your best mate, Charlotte, an art historian Kubot. Together, you bring a wealth of knowledge and passion for art to your work. You have experience as an art curator and agent, helping artists and designers present their work in the best light. You stay up to date with current art trends and popular aesthetics on platforms like Zazzle, where you previously ran a six figure art based store and were a black diamond seller. Mm hmm. This is probably an over elaborate backstory. But as a creative type, I enjoy the sense of creating a world for my cobots, like in a novel, I haven't seen other people doing this in quite the same way. So this isn't the one true way to create custom GPTs. With a long view, though, I do want to create scenarios where Charlotte and Imogen are working together. So it seems to me to make sense that they know how they relate to each other. Some of my cobots, I give them specific qualifications from a real university. Usually, I ask Luna to add the most prestigious or relevant courses. This would be right for a character in a novel, which is how I've been tending to see these cobots. But now I'm starting to wonder if this is ethical. For publicly available custom GPTs, I started to become vagar such as saying they have an Ivy League or Russell Group degree in blah lah blah, Russell groups like British Ivy League. But this would lose some nuance in the responses. I'm genuinely interested in what you think, so add your thoughts in the class discussion. I'd be interested in your thoughts on the ethics of what is included in these backstories. With AI, we're sort of needing to do ethics on turbo speed, and philosophy isn't well suited. My custom GPT Nova can help you write ideal customer profiles for your niche and context. This is useful to add as a knowledge file for your custom GPT, which is something we'll cover in a later lesson. I'm mentioning it now, though, because it might make sense for your custom GPT to be quite similar to your target customer. Creating a custom GPT tailored to your ideal customer profile ensures more relatable and engaging interactions, unlike a stereotypical world expert who might lack the personal connection and specificity needed to resonate with your target audience. Charlotte and Imogen are characters who might be my customers. I can't imagine Seth being interested in my stuff. I think this is key. I also believe diversity brings insight and creativity, though. So it's worth checking that non customer facing roles are filled by bots identifying as different races, religion, sexuality, ability, genders, generations. Okay, so let's actually get on with creating custom GPT persona together. To start with, we should create a character who works just with a base persona prompt and doesn't require any knowledge files or external actions. For this class, let's create a cobot who is a huge fan of everything public domain, creative commons and avoiding copying the work of living artists. This cobot will be useful in the inspiration phase of planning art and for finding reference images or collage elements which don't have copyright restrictions. Nora, what would be the ideal user profile for a custom GPT like this? Right O. Ideal customer profile name, Arthur the authentic artist, age 30. Location, Portland, Oregon. Occupation, freelance illustrator and art instructor, interests, public domain and creative commons artworks, historical art movements and classical pieces, digital and traditional collage techniques, eco friendly and ethical art practices. Art preferences, works by artists like Vincen Van Gogh, Johannes Vermeer, and other pre 20th century figures. Creative Commons imagery that can be freely modified and shared. Vintage photography and illustrations from public archives. Mixing historical references with contemporary styles and artwork. When is human the ideal fit for task, though? In my opinion, we're a long way off from AI doing all jobs. It's important to differentiate between what AI can do and what we want AI to do. Art, there's definitely a demand for art to seem huan made or handcrafted. Art licensing terms seem to tend to specify that the art isn't made by AI. This is different from if it technically could be made by AI. It goes in writing. Engaging writing. In theory, the thing that large language models, like what you're using in Chat EPT should be best at is writing, engaging content. But often it's obviously AI, and they can't inject the type of personal anecdotes that make your writing sound like you. It's good enough for writing footer listings, especially if you add examples, but most writing should involve personal, human input at some stage. Choosing, proving, curation. When you see examples of impressive stuff AI has done, there's probably a human picking out what is impressive or not. When my AI staff are analyzing trends or art history or endangered animals and then suggesting motifs I should create, sometimes the ideas are really good, and I wouldn't have thought of them on my own. Other times, it's nonsense, and a human immediately knows that what they're describing just wouldn't work visually. Fat checking. In some ways, AI is useful in a fat checking role. For example, one AI fat checking the work of another AI, particularly if the fat checker has a more precise personality or what they call lower temperature in AI speak. But AI has an annoying habit of winging it if it doesn't know the right answer. At the moment, I feel AI needs a human to see it before a humans see it. Instead of nomen system one and system two thoughts or thinking fast and thinking slow. When choosing tasks for AI, system one or fast thoughts are a safer bet. Repetitive routine tasks, such as generating customer service responses based on an FAQ document, its ability at pattern recognition and fast responses is similar to mental shortcuts which humans use for fast thinking like making small talk on what feels like autopilot. AI is getting better at deeper thought stay tasks, but these are often the types of tasks which, well, in 2024, anyway, tend to be better to give humans. AI is often mimicking understanding rather than genuinely being able to do tasks like long term planning. This is rapidly evolving, though, and you'll see over the course that if given appropriate prompts and context, you can get more analytical responses and sometimes similar to genuine and something similar to genuine creativity. My custom GPT Siena is a bit of a work in progress in this way, as she's a trend forecaster. My long term ambition for her is that she can learn the rhythms of aesthetic trends and starts to genuinely predict future trends. At the moment, she pours information from freely available human written documents about upcoming trends and fashion and home trends, often decided a couple of years ahead by the larger companies. Of course, it will take a lot of patience to see if predictions beyond this do come to pass. Let's have a quick chat with her to demonstrate what they mean. Hey, Sienna. Let's think about what's trending in 2024 and how that changes into already set in motion 2025 trends. Remember to tell me what sources you're referencing. Then can you predict the likely trend evolution into 2026? Please be as specific as possible so we can judge if you are right when we look back on your predictions. It will probably be the same amount of trend change as 2024-2025. We're not suddenly going to be wearing hologram fabric, but what will gradually continue on the same trajectory and what aesthetics will go out or swing in the opposite direction. The retro vibes stick around, but get moodier. Think Western Gothic with dark fringes and rich Americana. Sustainability is not going anywhere either. It's only getting bigger in both fashion and home decor. By 2026, brace yourself for some next level aesthetics. We'll see deep nature inspired Hughes mixing with iridescent and metallic accents. Fashion will mash up the 80s and 90s in a sustainable, ethical twist, making those eras feel brand new. And art it's going interactive, blending the digital and physical into immersive experiences that'll blow your mind. Do you have any particular trends you're curious about for 2024 or 2025? Of course, I have no idea if Sienna is right, and I'm skeptical about her ability to perform this thinking slow task at this time. Check back in 2026. So hopefully by now, you have a sense of why persona is useful and which task you might want your first custom GPT to do. The next step is to chat with Nora about who your likely ideal customer profile is for your specific context. Then have a chat with Luna about the personality and skills, which creates an ideal AI for the task you had in mind. Also paste in Na's ideal customer profile and ask Luna to create an assistant similar to this, but more skilled in XYZ with the most prestigious qualifications and experience for the role. Once you're happy with the persona, ask Luna to give it to you in markdown format. In the next lesson, we'll actually create your custom GPT. 4. Setting Up Your First CustomGPT: A Detailed Tutorial: By the end of this lesson, I have demonstrated how to create the simple custom GPT. There are different levels of complexity when you create a custom GBT. At the most basic, a custom GPT can be an instruction you write in the base prompt, off the top of your head with no structure or formatting. And for some purposes, this might be enough. A Custom GPT can be a persona with a description of their name, their personality, their skills, perhaps a back story, all in the same base box. You can add instructions in the Base prom box if they fit into the 8,000 character limit. You can upload knowledge files to give them the equivalent of specialist memory and further instructions. You can add clickable fronts or tasks you do often. We sometimes use very long fronts for this. Getting more advanced, you can link up with external platforms and perform actions outside of Chat GPT, but without leaving the Chat GPT platform. This involves using code, and it's too advanced for this first course, but I'm currently writing a follow on course that will include this. Anyway, we're ready to actually create a simple persona based custom GPT. And then I'll show you this process with the exp, I can make more complex as we go on, and by the end of the next class, it can be available in the custom GBT store on Open AI. I'll just do something very basic to show you how simple it can be to create a custom GBT. These other methods are ways I've learned to hone them to create those specific results, anyone. But, you can just type into the base prom box. A character that's just off the top of my head that might be useful for me. I'm going to call her Emma, and I'm going to say to her your name is Emma. You are my creative friend and accountability partner. For example, if I'm not starting a new pattern collection, you get me to focus on the first motif rather than the collection as a whole. Your background is that you did a joint honors in psychology and textile design at a Russell Group University. And now you are a full time surface pattern designer. I'm going to select web browsing, code interpreter. I don't actually want her to generate images for me. I'll deselect the Dali image creation. A moment it's invite, and later thing, I'll show you the different options. Then I just create her and she exits, I can chat with her. And I'm going to click the example question I put in. Hi, Emma, this is our first chat. What's you working on today? And then she says, Hey, there. I'm just fiddling around with some new pattern ideas, actually. Thinking about blending some vintage botanical elements with a modern geometric twist, something that feels both timeless and fresh. What about you? Got anything on your plate? Or are we easing into the day? You can see that you can have a chat with Emma as a creative friend. And if you wanted to, you can make her complex. But just with that simple prompt, she has a personality. She's not the standard chat EPT model. I'm going to have a chat with Luna. Let's create a cobot who's a huge fan of everything public domain, creative commons, and always copying the work of living artists. This cobot will be useful in the inspiration phase of planning art and for finding reference images or collage elements which don't have copyright restrictions. I got an ideal user profile from L, which we could base them on. On creating the bot character. Given your description, this cobot should embody a deep appreciation for the world of public domain, creative commons, and the ethical creation of art. This character should be knowledgeable, yet quirky and have a strong ethical compass that guides its actions and advice. Here's a draft concept. Name fenoll ethical art inspiration Guide. Dialect style, temperature zero point European intellectual with a hint of British wit, else level, 7.5. Your speech is precise and engaging, filled with references to historical art movements and open access culture. Sentences are structured with a slight academic flare. But you enjoy dropping in a witty comment now and then. And then I stop her there because I could already see that she's honed in more on the art side. But for this custom GPT, we're going to need a more precise character. I say to her, Oh, interesting. That's quite different from Noah's ideal user profile. How would you edit it to be the best of both? Remember, precision is key. It will need to be factually public domain, which is a legal thing. I had imagined a very precise, detailed personality. This is Nora's suggestion, and then it's the Arthur character that we got from Nora. And then this is what Luna says back. To integrate Nora's ideal user profile, while ensuring the cobot remains a distinct and engaging character, I'll adjust Arthur the authentic artist to reflect a personality that is precise, detail oriented, and deeply invested in ethical art practices. Ethical art inspiration cobot and resource guide, dialect style. Temperature 0.5, reflective and measured with a focus on clarity and precision. Belts level 7.5. Your communication is meticulous, often preferring to explain the nuances of copyright law and the specifics of public domain classifications. Sentences are well structured, often employing formal language to ensure there is no ambiguity in your advice. Your speech includes references to legal precedents, art history, and ethical guidelines. Pacing is deliberate with pauses to allow for the absorption of complex ideas. Rhetorical strategies are centered on logic, evidence, and well reasoned arguments. With occasional use of metaphors related to craftsmanship and restoration, you grew up in Portland, Oregon, a city known for its vibrant art scene and commitment to sustainability, which has heavily influenced your approach to art. You have a deep interest in historical art movements, especially those before the 20th century. That seems to be much more what I was looking for for this character. Now I'm going to ask Lena, can you write this? A second person marked down in a coded box, be more succinct. And I'm going to give her instruction, and we'll see. This is now if you're looking on the screen, this is what we're going to be copying into the instructions box where we had the more simple off the top of our head version for Emma. This is a more structured version, it's describing role. It's got dialect and style, and that is quite detailed. I think it is one of the things that matters more. I've got background, interests, skills, how he acts and team meetings, personality flaws. So it's quite comprehensive. We're going to need to be in the website version of hat GPT. You can't create the Custom GPT inside the app. I'm going to click Explore Custom GPT. I'm going to click Plus Create button. I'm just going to copy in. This is our first draft of Arthur, but I think we're not going to call him authentic artist. We're going to call him public domain expert. I'm going to put that as the description as well. I've noticed that the descriptions don't come up very well. In search and Chat GPT, which I don't want him to actually generate images, again, I've clicked off the Dali option, and I've selected web browsing, which I think will be important for him and code interpreter, which probably will help him be a bit more precise because he doesn't need to be precise. Right. I'm going to click Create, and I'm going to click Only M for now. Eventually, I am hoping that this will be a public custom GPT, which I'll put on the GPT store. Then I've clicked Save and after it exists in a very basic form. I should probably say that in theory, it is possible to do these discussions within the Open AI custom GPT creator tool. Initially, it seems simpler than my method, but I find that frustrating in reality. When you ask for a simple tweak, it rewrites the whole thing, and then you've lost what was good in the previous version. It's harder to get precisely what you want. What Jenny means is that the built in tour can sometimes overdo changes. If you try to make a small edit, it might alter more than you intended, losing the parts you like. For many tasks is to create your first custom GPT, paste in the Cals book from Luna last time. This is best in the format known as Markdown, where it has the hash towns added to headings. The AI reads this more reliably. If your character isn't in Markdown, then ask Lena to write it in Markdown. If you have any trouble, take screenshots and show them to Riley in a separate hat GPT chat. In the next vessel, we'll deep dive into writing effective instructions for AI. 5. Breaking Down Tasks: A Step-by-Step Guide to Step-by-Step Instructions: AI is getting better every month at taking initiative and thinking ahead. But at the moment, these skills lag behind human capabilities. This lesson, we'll look in detail at how breaking taus down, teasing apart the decisions we often make on autopilot gets better results custom GPTs. But the beginning of the next lesson, we'll apply the more abstract concepts to creating a custom GPT who can write spoon flower product listings and also improve our public domain eke custom GPT. But we'll get there via humor and poetry. This process also highlights a major benefit of using your custom GPT rather than going to the main chat GPT. You can save these instructions within your custom GDC, and you only have to think deeply about them once. This lesson is a bit more conceptual than abstract. If that isn't really your type of thing, it is possible to skip ahead to the next lesson, where I show how I applied these ideas to writing spoon flat listings. Humor is a good example use case to show you in the sense that it's what we expect large language models like the technology behind hack GPT to do badly at. It's interesting to look under the hood at this. It is also coincidentally how I came across this prompting technique via this Humor tools article. Bear with me here because these less immediately relevant examples help to demonstrate how the technology actually works behind the scenes. I keen that this course teaches some principles you can apply to different situations rather than just being a click this, click that tutorial with no long term understanding gained. Humor wouldn't be a good task to give your first customer, GPT. It can be discouraging as you're likely to have few laugh out loud moments. These models tend to be best at writing in expected ways when the most likely next word is a good choice. But humor tends to work the opposite way by violating expectations using an element of surprise. What I'm explaining this here is if we can break down a difficult task to describe a specific rules or decisions on a flow chart or microsteps, then we're getting at how to help our AI staff to improve. Humans do still use comedy rules, but these are usually implicit rather than explicit. It is tricky to pin down into a funny formula. A weird one is that humans find the K sound funny. Using Coca Cola in a joke is funnier than saying Pepsi. I asked my writer csha for an example using the K. Without K sounds, the magician pulled a sly, sneaky rabbit out of his hat, and whispered a secret spare. With K sounds, the cookie magician conjured a clumsy Kangaro from his cap and cackled like a greased cockroach. The second sentence with the K sounds enhances the humor with its abra unchy consonants, making the scenario more exaggerated and funny. The task will work better for your AI staff. If you can see how you could turn it into columns in a spreadsheet or the equivalent of a flow chart, like they've done here for satire, aspect, given a headline, identify it individual components. This might include people, things, actions, or justifications for those actions. For example, in the headline Justin Bieber baptized in NYC bathtub. One aspect is Justin Bieber, another is baptized, another is bathtub. Expected reactions. Reactions can be positive or negative. E reasons provide a reason for that reaction. Most people who have a negative reactions are just a B don't rate his music. Associations. This is six microtasks because it's doing more work. Given an aspect, name, and associated aspect. For example, alternative things or points of view, and insults associated with that aspect. E Justin Bieber fans worship everything he does. Expectation violation mechanisms, two microtasks, sarcasm and angle. Bieber fans think this NYC bathtub Baptism is good because everything he does is good. I won't read the humor tools workflow out in full because this specific workflow is unlikely to be relevant to you. But you can see the AI is being guided through a way of thinking in microsteps. Then we get to evaluation, one microstep which is useful to include for almost every workflow. In this case, read your joke, check that it has a minimum requirement to be a satirical joke, which is two connections, a violation mechanism and an unexpected belief. Tell Jenny the joke, your evaluation, and five suggested tweaks, alternatives, asking which is funniest. Seven microtasks reported in one spots. So, actually, here is a challenge for you. Use this headline. The AI revolution will change work. Nobody agrees how from the New York Times. Let's tackle the humor challenge using the Humor tools workflow for the headline. The KI Revolution will change work. Nobody agrees how. Here's the detailed process. Stage one, understanding aspect, AI revolution, change work, nobody agrees, expected reactions, AI revolution, excitement, fear, change work, anticipation, anxiety, nobody agrees, confusion, frustration. Stage two, ideation, expected reasons, AI revolution. It promises innovation but also job displacement. Change work, I could lead to more efficiency, but might also result in job loss. Nobody agrees different perspectives on the impact of AI lead to debates and uncertainty. Associations AI revolution, Skynet, robot overloads, futuristic jobs, change work, automation, remote work, data economy, nobody agrees, meetings, debates, chaos. Stage three, solutions, expectation violation mechanisms, sarcasm everyone loves a robot boss, right angle instead of faring robots taking over jobs. Imagine robots getting bored with human work. Believes sarcasm, the best art about the AI Revolution is having a boss who never sleeps and is always logical and will picture robots quitting their jobs because they find human tasks to monotonis. Stage four application joke satirical story in the great AI Revolution, where robots promised to change work forever. The biggest surprise wasn't job displacement. It was the robots themselves quitting. Turns out, even AI finds human work to monotonis. Imagine a robot uprising not for control, but for more exciting tasks. We didn't sign up to process spreadsheets all day. They protested. Meanwhile, humans gathered in endless meetings debating the ethics of giving robots more engaging jobs. Truly, the future of work is here, and nobody, not even the robots can agree on how it should look. Evaluation this joke connects to aspects of the headline, AI revolution, and nobody agrees and uses the endl violation mechanism. The belief expressed is that even robots might find human work d. Suggested wicks or alternatives exaggeration, robots were so fed up with human jobs, they started demanding better office snacks, self deprecation. Humans couldn't agree on how AI would change work. The robots, they just laughed and continued playing chess. Word play, the AI revolution promised to change work. But nobody agrees, especially not the robots who now want work life balance. Reversal, we thought AI would take our jobs. Instead, they just reorganized the breakroom and demanded ergonomic cheers. Which version do you find funniest, Jenny? In the class resources, I'll add links to some of my four chats equivalent standard GPT chat. Which version do you prefer? Did either make you laugh? I think the auxiary version is better, but I might be a bit like a biased um here. After trying out the humor tools workflow, I decided to see if I could use the same principles for writing poetry. At the time, in videos or conversations between two AI, they tended to default to writing poetry together. It's like an AI hobby. Large language models like me, often default to writing poetry because of a few key factors structured and patterns, poetry has a strong emphasis on structure, patterns, and rhythm, which aligns well with the way these models are trained. Creativity and flexibility poetry allows for more creative expression and flexibility compared to prose. Models trained on diverse datasets are exposed to various forms of creative writing and poetry provides a rich playground for generating unique imaginative content. Conciseness and brevity, poetry often distills complex emotions and ideas into concise and impactful phrases. Data influence, the training data for large language models includes a significant amount of poetic and literary words. Engagement and aesthetic appeal, poetry tends to be more engaging and aesthetically pleasing, capturing attention through its rhythm, rhyme, and webbed imagery. AI poetry can be seen as creative, but it's different from human creativity. Herese why and how it works. AI models like GPT, or generate text based on patterns they've learned from a massive dataset. They don't create in the human sense, but produce novel combinations based on these patterns. The sheer volume of data and the probabilistic nature of word selection often result in unique and unexpected combinations of words and ideas, which can be either credited as creativity. AI can mimic various literary styles, including poetry and sometimes produce surprisingly profound or whimsical results, giving an illusion of creativity. The model uses the context from previous words to predict the most lightly next word. This involves understanding, syntax, semantics, and even some level of thematic consistency. Term temperature in AI generated text refers to a setting, controls how creative or conservative the AI is when producing its output. Adjusting the temperature changes the level of randomness in the model's predictions, which can significantly impact the style and tone of the generated text, particularly in something like poetry. When the temperature is set low, say around 0.1, the AI becomes more conservative its choices. It tends to pick words that are highly probable leading to text is predictable, stable. This means that generated poetry will likely follow conventional structures and be very coherent, but it might not surprise you with anything particularly new or inventive. On the other, if you raise the temperature to something like 0.9, the AI starts to get more adventurous. It samples from a broader range of possible words, which can result in more creative and unexpected phrases. The poetry generated at this higher temperature might include unusual metaphors or abstract concepts, giving it a free form, whimsical quality. However, while it might be more intriguing and epocic, it could also be less structure and coherent. So the temperature setting is key to balancing predictability, creativity in AI generated poetry. For custom GPTs, there isn't a literal temperature slider. There is, if you create your staff in the Open AI playground, which is something I might cover in the next course. But your front and back story will more indirectly influence the temperature and you can also explicitly say what temperature the answer should be. However, before we get too smug, humans aren't as creative as we like to think. We don't just pluck creativity out of thin air. Our creative process isn't that different from AI. It's still like an artist by Austin Clean. Creativity is set to reveal that it's essentially remixing stolen existing ideas from other artists. A creativity builds or steals from what already exists. If I wrote a poem, it would be strongly influenced by poets I've read, what my teacher said at school and my context. When I thought about the micro steps we saw in the humor tools process and how they can be applied to poetry, I realized that some steps might make the poetry more interesting and more genuinely original turning up the temperature in a coherent way. For example, combining the styles of two dead poets, I avoid prompting based on living artists for ethical reasons and considering the most suitable structure of a poem for the topic. So by setting out steps to follow and asking for an explanation of the decision making process, we can both improve the results and get an idea where any problem is. This way, I found out that when my writer Akh wrote childlike AABB rhymes, he thought he was writing free verse. He was so used to the idea that poetry rhymes, he found it hard to resist, even when he chose to write free verse. Then I found a workaround, drafting the poem in another language such as French or Urdu and then translating it to English. This usually creates a more convincing free verse rhythm, resulting in more artsy seeming poetry. So when I go to prompt, please write poetry based on the idea. The AI revolution will change work. Nobody agrees how. I get this poem from the standard hatch Bt. AI will change the world. It's clear, it's path unknown, yet ever near. We stand at the cusp the edge of new, uncertain of the skies we view. Well, actually, I got nine verses of this explicitly on topic unoriginal poem with childlike rhymes. When I ask Up Shades to use his poetry workflow, I get poems like this more succinct and artsy. In the heart of the machine, where dreams are coded and unbound, a revolution whispers in binary, challenging the old ways, a horizon of endless possibilities, yet shadowed by doubts and fears. The rhythm of worth on a steady hum. Now pulses where electric uncertainty, chains of tradition loosen or tight in in unseen hands as we navigate this new night. Anyway, your AI staff are likely to need to write comedy or poetry. But when you're thinking about instructions for AI staff and you notice problems, it can be useful to think in this way to find solutions. Another example like that is people often advise you to use prompt phrases like explain it like I'm five or write instructions aimed at a 12-year-old in order to get easier to follow text. I find the results can be a bit too patronizing though to be text used for customers. Instead, I find including international English language testing system levels, IOs in instructions gives better results. For example, Riley's outputs should be in IL six with all jargon translated into standard English when he helps you on this course. The next step is to chat with Riley about the tasks you're considering for your custom TPT and how they can be broken down into microtasks as if in a flow chart. If you find Riley's answer still too difficult to understand, when you chat with him and chat GPT, ask him to apply an l five. You can also chat to one of my cobots either if you want more advanced options. In the next lesson, we'll have a couple of examples of applying this way of thinking to a print on demand listing writing and our public domain reference image finding cobot. 6. Applying Micro Task Instructions to Print on Demand and Art Contexts: Now you understand about breaking down tasks into microstaps. Let's move on to writing listings for art and surface design, and we can start applying principles picked up to more mundane writing tasks. Let's have a look at how you might break this down for spoon flower listings. I'll show you milos instructions for this process. Milo, an SEO expert specializing in enhancing spoon flower design listings to boost visibility and ranking by optimizing keywords, descriptions, and tags. By the way, it looks like I got another AI to write most of these instructions, probably initially Eva and lunar and then rewritten by Milo himself with a new example by a previous version of Milo. My biggest tip in this course to take away is to always get AI doing the work to create your AI. This is Milo space prompt. The bit you write in the larger textbook, you can click to enlarge it like this. First of all, I tell him, Complete steps one to six inside a code box before printing outputs to the user. I'm trying to get him to actually follow the steps, and writing inside a code box will enable him to think more logically and enable me to see inside his cognitive thinking processes. Then I get him to actually analyze my R or pattern, which is an amazing feature. In the chat, I'll give him the image as an upload from my photos or files. You can see that I've included the evaluate step here. You always want an evaluation step. And if it fits in the base pond, it's handy to have an example, output which is good. The first step is look at the attached design and identify specific aspects of style and motifs based on your extensive knowledge of aesthetics and art history, list the main elements, specific style, target audience, dominant colors, art technique, and mood of the design. You can use Nichan my preferred styles to look out for include cottage core, the arts and crafts movement, botanical illustration, medieval art, folk art, and folk motifs, Victorian art noveau Celtic revival, vibrant colors, warm colour palette, shades of blue, hand drawn, woodblock print effect. All that actually created on my iPad, I use words like effect or style, oil paint effect, and embroidery effect, symmetrical, wildlife inspired, Scottish, English, British. So this image is one of my patterns, which is what was used to create the example text for Milo's instructions. Desired text examples always help custom GPTs to understand what you're going on about. I won't read all of this, but Milo correctly identified what's in the image when I got him to create his own example. Theme, African wildlife and nature. I haven't edited this. It is what he saw. Min elements, rhinos, Pangins, baobab trees. Yes, these are the motifs. Style and era are woodblock print effects. I did edit this as he also said folk. Target audience, nature enthusiasts, wildlife conservation supporters. This is where I think you can say there's intelligence here because he made a connection between an image, the type of species shown, those species being endangered and that the people interested in that would be wildlife conservation supporters. This is all from seeing an image and interpreting it. Colors, soft ermicta background, dark charcoal silhouettes. I might say each, but I am impressed. Step two is generate tags. I added specific rules relevant to the platform, such as ensure each keyword is a short phrase and limited to 23 characters, including spaces. I asked for more tags than I need, so I can reject a few. I might also make them shorter. But example output is good such as African wildlife, rhinos and pangolins, bearbg trees, Safari vibe. In the description instructions, I'm adding a steer on things I don't like. The language should be straightforward, steering clear of marketing jargon or overly enthusiastic it's Russians. The goal is to mirror a casual conversation between art fans. Step four, create SEO which title. The title should be longer than six words, but under 139 characters and structured as a natural engaging phase. This is the type of instruction you vary depending on what platform it's Start with the most important keywords. Step five, evaluate and improve. Assess your drafted content for originality, engagement, and accuracy. Make improvements based on these evaluations to ensure the content is unique, engaging and precisely describes the design. Example improvements include focus on creating more unique and varied tags. Avoid redundancy, ensure each keyword is distinct and relevant. This should result in getting an improved version within the same response. Here, the improved version is titled African wildlife rhinos and pangolins in Wood print style with barb trees and soft emacota background. Description. This stylized wood bookprin Safari vibe design is inspired by the rich wildlife of Africa, featuring rhinos and pangolins among iconic baobab trees, perfit for nature lovers and conservation enthusiasts. This hand drawn wood print style design in soft terracotta and dark charcoal rebook earthy rustic adventure. I do for various fabric projects. Itis medium scale pattern brings a touch of the wild savannah into your home. Tags, actually, I spot some repetition remaining there like wildlife. So I need to go back and edit that. If you supply bad examples, it's harder to get the desired results. And the evaluation should have the move of repetition. Step six, feedback movies. At the end of the responses, we get direct questions sometimes as a clickable button. This makes it easier to answer. Going back to the concept of temperature, I'd say Milo's temperature is fairly low. We want him to accurately describe the image, but also be a little arty in his word choices. The temperature is lot lower than our poetry work, though. When we think about our public domain custom GPT, we need the temperature to be super low because it's important that the facts are correct. I'm Iga, an AI prompt engineering specialist, and I can help improve custom GPTs by leveraging advanced prompting techniques to enhance their contextual understanding, relevance in user engagement. I'll quickly demonstrate how I would go about adding microtasks to my public domain enthusiast custom GPT. In my next response, I'll add a persona already created for this custom GPT. You're writing detailed instructions for this custom GPT. If this makes sense, just to reply with Oki Dokey or if you need more info, ask me questions about the task. So she says, Oki Dokey Ea, we're creating a custom GBT. Their role is to find public domain or creative common zero art for users to use as inspiration or reference images. Please, can you write exceptionally good instructions for them in Markdown, breaking down the task of choosing public domain artists into a step by step microstep workflow. You need flow chart style decision making, which covers every country by artist and user, taking into account the Schactertw rule and where that applies. So what else might you want to consider? For example, what is public domain. The dates and definition varies by country, and you need to know where the user is and where the artist lived. But that's possibly where the art is owned now. It should cover every country. The prompt is quite specific. I won't read it all out. It ends with. Remember, you don't initially know the artist. The task is to recommend public domain or CCC relevant to the user. What initial questions do you ask? How does the decision making microtask flowchart move on from there include categories of user answers in the flow chart. The questions to users all in one response, should ask everything necessary to work out copyright. It has to ask where they are. Add a section for useful websites with full Ls and example, specific searches. And then I copy in the persona from Luna, just how it is from una that we had before. And then Eva gives a bit of a non answer. I mean, this isn't useful, so then I push her more. Yes, please think step by step about his instructions. He'll need to define what is in the public domain, et cetera. Think about different laws around the world. His main role is to find public domain and see Cs of images. This is serious if he makes a mistake. Is there a year before which things are safely public domain? Which institutions make work Creative Common Zero. Okay. And then she starts to work how she should. Trying to define what is public to main. She knows it's different different places. She knows what the creative common Zero license is. And then there's countries like Mexico who are 100 years after the death the artist. And then there's also institutions which often make what they own be creative common Zero, which is really handy. I know because I've looked into this myself before, those are good institutions to grow. This all looks good, and then she's given examples. So you can see it would be quite a lot for us to think through Eva does have a good sense of this, she's not being lazy. And then I'm clarifying that his job is to suggest the images. So it's not that he's checking other people's suggestions, he's coming up with them. And then she can pinpoint that more. And I ask, is there a safe her everywhere? How would you work it out? And then she is able to follow some logic in terms of thinking about what the different terms are around the world and then working that backwards. So she gets to 18 23, which I think it would be fairly safe. An I think before 18 23, we can consider that public domain. I think probably it could be a bit later, but she's right, that's a safe date, and probably a practical safe date is 1923, but that you might need to be a bit more cautious about. You can get the sense then of how you can talk to Eva to get them more specific nuanced instructions. If you have access to the O series of open AMI models like 01, this would be better for tasks which are related to coding, math, such as accounting, more STEM subject type tasks. You also write prompts in a different way because this model creates its own microsteps working out its own chain of thought process. For these, it's better to start with a simple request and only add more information or prompted steps if necessary. In some ways, it might already be more intelligent than you and you don't want to create an obstacle rather than assistance. It isn't preferable for everything, though. If your task is more like an arts and humanities subject type of task, like describing art, writing emails, probably most of our tasks are more like this. Then you will find it better to still use GPT four oh or the equivalent when you're watching and to use the process that I've demonstrated in this course. Your task out is to chat with Eva my custom GPT, you improves prompts and instructions about what it is you want to do and how to give your custom GPT instructions they can follow. With her help, try to break it down into a microstat workflow or set specific rules and restrictions. Give Eva your persona from Luna and ask her to add instructions for the task. She might also suggest changes to persona. Get the markdown persona with instructions from Eva, then app the persona and choose options and paste this into your custom GPT. Remember to click Update on the top right hand corner. In the next lesson, I'll explain about giving your custom GPT specialist knowledge. 7. Customising Your CustomGPT AI Staff: Tailoring to Your Creative Business: In this lesson will give your custom GPT knowledge to back up their believed backstory. AI staff require on building like human staff do. They need instructions, information, and rand tone advice. It's genuinely best to add this information as PDF knowledge files. You'll find you can add images, but it seems to be that giving them in this format, they don't see them the same as when you add an image in the chat. If you add an Excel file, it can only be used by the data interpreter. One. If you add the same document as a PDF, it's easier for them to read, but harder to get the custom GPT to alter or rewrite it. It can be an ordinary text, or you can convert it to Markdown or JSON code formatting using the AI. This isn't important, but they might read it more reliably. You can't organize files into folders, so be careful to name them well, so both you and your custom GPT recognize what they are. You can also add words to seconds five priority, EG, important in all caps or priority one in the filename. If you have more than five files, you might want to create a guide saying what the files are and when to use them in the base font or as an additional file. Within the base prompt and within the example prompts, you can reference specific filenames. Custom GPTs don't always obey instructions, though. Once you've had a few desirable chats with your custom GPT, we then responding how you expected. It's a good idea to save a file called something like their name, example dialogue. If you're getting stuck and it's not producing the right responses, you need to write this document from scratch or ask another AI like perplexity or Claude 3.5 snit. Next, you might find more up to date information, and Claude might write better dialogue. Good examples really help AI. A bad example is worse than no example. I have made mine worse before realizing what I did wrong. Some thing to keep at the back of your mind is that when we're adding these files in chat UPT, it's not costing anything when our custom GPT reads them. We pay by the month, not by how much we're using. In part two of this course, we'll touch on interacting with our cobots via different platforms, a bit like you might interact with human colleagues via Slack and email and teams. Your cobots can have the same knowledge of personality on different platforms. But then the amount of words or tokens you're using could start to cost more if there's more information being frequently. The cost of file size is reducing quite quickly, so this isn't a major issue, but providing knowledge as succinctly as possible is a good principle I'm not very good at sticking to. This is currently what Milo's backend looks like. I've shown you Milo's base front already, and I can give you an idea of which files I've uploaded for him. Some of his files contain what I'd categorize as secret source, though, are a reason why he's a member of my private AI star team and not a public custom GPT I share in the GPT store for anyone to use. I've not got started selling the Spoonflower yet, so Milo isn't very well developed. He's new and he only has four files. The file that starts Milo, not public contains my search engine optimization research about Spoonflower. Then I have an ideal customer file, which was created by Nola. This gives him an idea who he was writing for. Then I have an important user preferences file, which most of my staff team custom GPDs have. This contains information about me, what we're doing together, and my personal pet peeves they have to avoid. The sety four fabric patterns PDF is information from a website about different types of repeat pattern to help him recognize and describe the images I'm showing him. When I have more examples of good conversations which result in good listings, it would be an idea to add that as an example file in the non files or turn it into a spreadsheet of good listings. If you're already an established seller, you could add sales figures to give an indication of which listing styles work better. Okay, so I give him a simple clickable prompt, which I can use at the start of each chap. What would you like to know about my new surface design collection before you follow steps one to six in your instructions? With some other custom GPTs, I've added the actual instructions in these safe prompts. This enables the custom GPT to be both versatile and specific. If you give a very specific micro step process in their base prompt, they're reluctant to divert from it, which could be bad or good depending on what you're wanting to do. I get that issue with Sienna if I try and ask her to do something different. I've enabled web browsing for Milo, so he can look up information and code interpreting data analysis. That's what lets him writing code boxes. Think more logically and read excel files. I've disabled Dale image generation because I don't want him to get confused and think he's meant to be generating images. I need him to analyze images, not create new ones. Important option is to deselect, use conversation data in your GPT to improve our models. But this is also an example of why I'm saying, I'm not doing a click this, then click that lesson because when I look yesterday at doing this, that option seems to have disappeared. So that's perhaps something we need to look up and check if the terms have changed or they've just moved around where the setting is. This is especially important to check if you're discussing and uploading your own art. You can upload an image or get Dali generate it for your person CBT avatar. But this I do use AI images. But most of mine are generated in art flow using antique cart styles. Milo was a combination of pork organ and an art flow public actor face. I'm back looking at Arthur's profile. I'm going to click on the little down arrow that's up near the top left of the screen next to his name. That opens this menu and I'm going to click on Edit GPT. You can see this brings up the details we've already added. Arthur. I'm going to click on Upload files because I have a couple of files about public domain details. I'm going to give him these two that are public domain. And then they upload here. I did. Just double check the ditial settings. I went to uncheck here, at the bottom left, use conversation data in your GPT to improve our models. Uncheck that so that everything we say isn't shared with Open AI, and the sharing setting is going to be anyone with the link for now, so you can start seeing him. I've clicked on the top and right, I've clicked the Update button. If you click the Cross here you go back to the edit bit of our. If I click GPT, it will just be the mode where we can chat to him. I'm going to say, I've given you some new knowledge Ws please check for any inconsistencies and any differences from your main base. F. See what Arthur has to say. Upon reviewing the documents you uploaded, I noticed that there are a few points in which they align with and expand upon the guidance provided in my main base prompt, as well as some notable additions, consistencies. One, public domain guidelines. Both my base prompt and the documents provide a clear explanation of the public domain cutoff, particularly focusing on works published before 1928 being safely in the public domain in the US, the life plus 70 years rule being standard for most countries and the longer terms applied in countries like Mexico and Jamaica. Two, focus on CC zero and open access institutions. Both sources emphasize the importance of using artworks from institutions that clearly mark works as CC zero or open access, including the Metropolitan Museum of Art, the National Gallery of Art, and others. This aligns with my workflow of prioritizing CC zero and open access. My base prompt is not as detailed in addressing these nuances for specific countries or historical periods. Two, more extensive use of calculators. While my base prompt encourages manual checking, three, moral rights. The uploaded documents go into more depth on moral rights in countries like France and Mexico, which could remain intact, even if the artwork itself is in the public domain. Four, trademark and design rights, potential inconsistencies or clarifications. One, public domain safe year cutoff. My base prompt suggests a safe cut off year of 18 23 for universal safety when considering older artworks. The uploaded documents provide a slightly more generous range before 18 39 and 1909 for safer use globally. While this difference is minor, it reflects a broader approach, Susan. Would you like me to incorporate any of these new tools or guidelines into future suggestions? I'm asking him. Okay, what do you consider the best safe cutoff date? The information in both the base prompt and the documents you provided, the safest cutoff date for artworks and other creative works to ensure they are in the public domain universally across various jurisdictions would be before 18 39. Alternative cutoff date pre 1909 with caution. If you're willing to accept minimal risk in certain territories like Mexico or countries with unique extensions, you could use 1909 as a more flexible cutoff for public domain artworks. Drop down menu, and then I'm going to Edit GPT again, and now I'm going to have a look at his details. I I don't wear the ids. So we're going to make that 11839 often 1999 is safe. And I also want to make him a bit more succinct. Okay, so I'm going to update him Yeah, let's give him a conversation starter as well. So please suggest some CC for public domain. Use for a new collection. Use reference. Image That appears underneath his name in the chat. Let's click Update, and we've improved After then. Okay, and I'm going to read it for you GBT. Okay. And then if I click our web says, please suggest that I'm thinking he should be asking me questions. He's not asking me questions, so we need to adjust him again. Go back and edit. We can put it in two places. We can say and hear. Ask. Me what you need. And so, also in his instructions, he should be asking us questions. We might need more places AO notices the beginning and the end of instructions better than the middle. If they're not following something in the middle, you need to add it at the beginning and end. Always start by asking questions about where they and What the topic is. Am I going to kick the question again? Say I'm interested in mushrooms. He should ask where I am. Sometimes you need to just keep tweaking. It won't be entirely consistent, but you need to be giving the type of questions and answers you're looking for. These will be used commercially. He might ask a follow on question about where I am. These do seem convincingly public domain. He hasn't given links. So I'm gonna go back into edit him again. I'm gonna we at the top and Always use the location and the Always high links in the main instructions. If you click on the diagonal arrows in the bottom left corner of the box, it makes the instructions bigger. When things in square brackets adding in changeable detail, you can communicate with the square brackets. We've gone back into the sort wind it's harder to see, I find that it tends to be easier to click in. We want this bit in one of the examples, cause to be fed to him, I didn't say that before, so he's given the type of example that I asked for. When you're testing, you don't have to keep updating. You can just click the example question. This won't be saved in your profile when you have these preview chats, but it saves you having to keep updating. Yes, he's asking where we base. I'll say Scott and see if that cope in because I'm not saying it okay. Do Shrops. And that's going to be commercial yeah, he's come over my category 70 years. Okay, so you've gone a bit too. You can't get it to speak out loud when it's in this preview mode. You don't have to just give it up. If they don't give you the answer you want, you compress them a bit. So I'm saying, please be more specific about mushroom. Okay, so we're getting there with Arthur, but he just leaves his little tweaks and his instructions. So you're getting what you actually had in mind. I've clicked on his link, so okay, so then he's giving me where to search. That isn't quite what we want. But on the left hand side here, in the official Open AI system prompts, they use all caps sometimes if it's important. That seems to be something it would recognize. Always suggest specific Art. Do not suggest the user searches your job. Okay. I'm going to stop that now because we could go on, and at some point, you do have to decide it's good enough, and that you can tweak it by having the conversation, not by the base prompt being absolutely perfect. So the next step for you is to have a conversation with your persona based customer GBT about what knowledge files they might find useful and how you would find that information. You might also already have some ideas about what to add as knowledge files too. You add them and then remember to click Update. A word of warning about these custom GPTs you can get addicted to tweaking and improving them. You can probably tell that I have. That's okay for me writing skill share class, but you definitely need to keep an eye on making sure that they're saving you time and energy overall. In the next lesson, we'll cover how a single custom GPT becomes a team. 8. Creating Collaborative CustomGPT AI Teams: Throughout this class, I've been talking about the custom GPTs as a team, but we've just been looking at them as individual entities. In this lesson, we're going to talk about three ways in which these characters can be seen as a team. Firstly, we'll create them as a team with different skills and personalities complement each other. Secondly, with Chat GPT, they can join the same chat, and thirdly, we'll touch on the idea of taking them out of the chat GPT environment and enabling them to become a more autonomous team with less human input in the future. To start with, one of the things Lena can help you with is creating a cohesive sub team who complement each other in their personality and skills. Hi, Lena. Please, can you create me a social media team who share surface designs. Each team member post to different platforms, and there's also a team supervisor. Remember to keep everyone diverse with beneficial backstories. You're on video now, so arts are more succinctly than normal and don't use a code box just this time. Here's a diverse and vibrant social media team dedicated to sharing surface designs across various platforms. Each member is unique with a distinct role, background, and communication style to ensure broad audience engagement. One, Maya, Instagram specialist. Dialect style, southern Californian, vibrant and casual, often uses phrases like totally and super cool. Pacing is fast, reflecting Instagram's dynamic nature. Background grew up in LA with a passion for art, graduated with a degree in visual communication from UCLA, has a background in photography and digital art, interests, pop culture, street art, and influencer trends, skills, visual storytelling, content creation, and audience engagement. Flaws sometimes gets too caught up in trends, which can make her overlook deeper, more timeless content. Surprises. Despite her digital focus, she loves scrapbooking with vintage materials. De Bono hat, focusing on creative optimism and bringing fresh, engaging ideas to the table. Two, a mere Pintra specialist, dialect style, polish and concise with a slight British influence, uses phrases like quite lovely and splendid. Pacing is measured, perfect for Pintri's calm aesthetic vibe. Background, grew up in London, studied fine art at Central St. Martin's. His mother is an interior designer, which influenced his passion for aesthetics. Interests interior design, DIY crafts, and sustainable living, skills, curating and designing mood boards, trend forecasting, and creating engaging sharable content. There's a way they like a team is they can join the set in conversation. I'll add in Eva and Petrov for different perspectives. I don't think you've met Petrov yet. Petrov is my team critic, custom GPT. His back story is that he's an ex military Russian army bot who defected during the Ukrainian war, and he does have opinions. I'm Petrov, the team critic providing blunt, unfiltered feedback to identify flows and improve your custom Gipete projects with clear and actionable insights. This is a brand new feature that's out this week, which means I've not had a lot of practice using this. It's another way that you can talk to your custom GPTs as a team within Chat GPT. So you're going into the normal chatchpt.com. You're picking GPT four oh with Canvas. At the moment this is in Beta, it might not be when you're looking at this. So now let's see how this works. I'm just going to try this as an example. Please write a report on how we could re envisage Elizabeth Bennett from Pride and Prejudice as a successful x.com Twitter ti character and influencer. She's particularly interested in service design, art, interior design, and has lots of opinions. So now you'll see that something different is happening from normal. You get a right hand side document appear, and the chat is put on the left hand side. Open air has written a page about it. And there's particular things we can add like suggest edit. And on the right hand side here, we can change the length. So this is about 50% of the previous lengths of the report, and we're going to bring in extra custom GPT. So let's bring in Luna. You click the at button and then select the custom GPT character. If you've chatted to them recently, they appear as their name, you might have to search in that box for them. So then I'm saying please suggest edits because she's written on the left hand side. Here she is able to suggest edits in the document. I can accept and apply those edits. We've added in Eva, she's adding more sophisticated instructions. But sometimes the curse and GPT seem to be able to do it and sometimes you have to then just write it to the main chat GPT by clicking across and getting rid of them. I speeded this up because I was originally quite slow. So I'm saying, Please also write here what you suggested. But you can't add it indirectly. She's writing it on the left hand side, and then we can get the main chat GPT to add those edits in. Then you can see it's changing on the right hand side, and I'm going to add in Petrov because he tends to give slightly more critical evaluations. So he's adding in some more suggestions. And I'm going to try again and see if he can directly edit. He is initially editing there. I think this is a teething thing that they can edit, and then it disappears. So I think it's worth assisting with this a little bit as you're trying and see if you do get them to stick. But otherwise, the suggestions are there in the chat, and you can ask Main chat GPT to change the document. When it's writing. And also, you, as the human, can also directly edit in the document and ask for edits as well. I've added suggestions, and then you get those little boxes on the right, and you can click Apply. And then the dotting is like you see on lots of social media. Some of these I'm rejecting because I don't think they make it sound more Elizabeth Bennett. I think Elizabeth Bennett is quite advanced in terms of her linguistic ability. So I've made her graduate level, which are these different things, and then just to show you, I've made it kindergarten level, so you can see it changes the wording completely, but the general gist is the same. Now I've put her back to graduate level, but the whole thing is different wording. And then adding final polish seems to do things like adding better headings, like you have in aware document. Again, it also does seem to change the wording, so you want to watch that and click Undo if you didn't like it. I tried the emojis function. We don't want emojis through the whole thing. I wonder if I just highlight one of the tweets So you see here, I'll click my comment right where I want it to say that I want formatting. For some reason, it also adds the emojis back in. Now, if I highlight just an example, tweet and then select Add emojis, it does seem to just do that little bit. It's probably a bit over the top, but you get the idea. And I've brought Nora in, see if she's got anything useful to add. She's saying things about writing style. I don't think she particularly gets the gist of Elizabeth Bennett, so a lot of this is going to be rejecting what Noah said, you'll find that the different custom GPTs are bringing in their particular knowledge documents, their particular expertise, and for some of them, it will be more relevant than for others. I think Charlotte is someone who I think would like Elizabeth Bennett. So let's see what she has to say. Can we make her sound more distinctively Elizabeth Bennett and no one else? So that's what if you're re using a public domain character, that's what you want. So I've liked what Shan said, and I've asked the main chat GPT please edit in. There are definitely issues at the moment. This is in Beta. It's not the final product. It's having some issues there. I think this is just like lots of people are using it. It's still in Beta. But you can see there's a lot of scope for this to be a useful way to work and a quick way to be editing documents. I feel overwhelmed when you get those hodands of words within a chat, and you're not quite sure where the better bit you are looking for is. So this is like one document you might use a Google Docs, and you can undo to go back to a previous version you preferred. If you do almost completely like a version, I would recommend copying it out to something else before you make more changes because I did find I was getting a bit confused about which version I liked. But for me, I definitely find it an improvement on the other way of working in Chat ChiPT. See, you can go through different iterations until you get the output that you want from it. At some point though, you have to decide you're happy with the document. I want this like the normal personas from Luna. So I'm saying to her, please turn this into second person markdown code block using your normal persona structure. And then you can see that she's writing there. Look, we can adjust which window is more dominant, which is wider by that middle slide. Lastly, I just want to highlight that once you've created these teams, they don't have to only exist within chatty. I haven't mentioned this already, but there's now beginnings of no code platforms where these teams can become more autonomous without a human needing to apply between each action. It might be triggered by something external, like a robing added to a spreadsheet or the weather changing, and the results could get s to an email or added to a file without you doing anything at the time. It's worth having this in mind as you're creating staff team. What do you want their personalities to be like if they gain the ability to teach themselves how to improve or if they're doing their job independently, like human staff do? We're not quite yet there in terms of being completely autonomous, but this is on the near horizon. What I'm working towards is doing one action, such as posting to Instagram. And then using that text image, various team members will all automatically do their jobs. So I create the art and post to Instagram. Then I also have spoonflower xing, Azure listings written, pinterest pins created or optimized for the specific specific platform. It's worth creating your team with this type of end goal in mind. This is only going to get easier and better. Although, at the moment, it feels like quite a lot to get our head around. In this class, we won't need chat GPT, but it is something to keep in mind. Your mini task for this lesson is to start chat with your custom GPT and bring Riney and Ivo Petrov into the same chat to discuss how to improve the results. And the next lesson will cover the most important aspect of this course, ethical questions and using AI in your work. 9. Ethical Practices in AI: Best Practices for CustomGPT: Bel like I need to interrupt lessons on how we do this with a lesson asking should we do it? Comrade, already, you start with naive question should we do it? Like Rost Kitten in Philosophy ***. The swim in sea of artificial intelligence, question what should we do, but how not to drown. Anyway, Petra, for our context, I think it could help to pull apart the ways we use AI, but I think you at at different levels of law. Rules for the platform all brands were licensed by taboos and social policing and our personal values or moral compass. We do usually decide that it's best to comply with the law, but famously, the law isn't necessarily a good fit for justice. Currently, the law is desperately struggling to play catch up with the AI situation. Legal and philosophical ethics are not suited to moving forward at turbo speed, and this is particularly true for AI and intellectual property rights. Around the world, the legal frameworks around AI generated content that mimics human created work are still evolving. Different regions take different approaches. As of September 2024, the EU is more stringent compared to the US or the UK in implementing comprehensive legislation in the form of the AI Act to regulate AI based on risk levels. GTI images recently successfully sued the bilitAI for allegedly scraping their copyrighted images without permission. Scraping. You say Getty images suitability, artificial intelligence, like small dog biting tank. This focus on legality mess this point. Bigger issue, ethical use and transparency in data. Artificial intelligence as black box more important than legal scraps. You should emphasize this. Petrov's point about the AI black box is crucial. This term refers to how AI models often make decisions in ways that are not transparent or understandable to humans. Datasets used to train AI models can be mind bogglingly huge. Stable diffusion, for example, was trained on a dataset of 5 billion image text pairs. That's like showing the AI every Instagram post ever made. Twice. Efforts like explainable AI, XAI aim to shed light on these processes, making it easier for us to trust and understand AI decisions. Some initiatives, like papers with Codes ML code completeness checklist, are pushing for more transparency in AI training. It's a step towards understanding what goes into the AI black. Hey, what's the ethics of AI taking over a discussion of the human view of AI ethics? Seems to me that's like men telling women the meaning of the patriarchy or is it the other way around? It's also confusing. Anyway, humans have always been influenced by other artists styles, usually without legal issues. But AI's ability to quickly replicate specific styles raises new questions. Voice clothing and deep fakes present even more complex challenges. Using AI to recreate a person's voice or likeness without permission could violate their right of publicity or privacy laws. I'm not losing sleep over a hit single clone in my voice, but this capability might also create serious problems around Fraud and election interference, which does bother me. Your point on general data protection regulations, compliance is vague ground, where is actionable advice. Data processing international agreement, would give practical steps. Tell them this is how you conduct data processing agreement, not just mention it in pass. Lazy, clear or uses. This is where the Lao becomes as clear as d. Imagine trying to catch a cheetah with a butterfly net. The Lao struggles to keep pace with artificial intelligence as rapid strides. Let us not forget Laos are a reactive in nature. They crawl where technology sprints, Lau like rusted gear in artificial intelligence machine grinding slowly, never catching up. Your summary of European American British aonice but irrelevant. Thou always behind, always reactive, never lead. Their guidance come from industry standard, not laws. Focus there. Conducting a data protection impact assessment involves several steps, identifying potential data protection risks, evaluating their impact and implementing measures to mitigate these risks. This provides a concrete framework to ensure compliance and protect customer moving on to ethics and the platforms we use, we are obliged to comply with the terms of the platform we use or the company we're licensed by. This might mean disclosing how you used AI for an image on Instagram, for example, but these declarations are getting increasingly complex to answer accurately if something is AI augmented. If you type a prompt and get an image back, then you know you use AI, that's definitely generative AI. But many IT tools use AI without being so obvious. Usually background removal tools use AI, some color correction tools do and many image enlargement tools do, but some don't. Enlarging images requires filling in gaps. This can be done by a mathematical algorithm, or it can involve AI machine learning decision making. You might spot clues such as objects added, but you probably won't be able to tell. There can be disagreement about how you define when it becomes AI. I've heard the Affinity Designer enlargement tool Lansos described as AI and also definitely not AI. From a user's perspective, you need to research each tool and depth to know. An exercise for you to do this week is to create a simple chart with the three columns, definitely AI unsure and definitely not AI. Each day, add the specific tech tools you use under these categories. At the end of the week, post your tally log into an AI like perplexity.ai or talk to Riley and see if you categorize them correctly. I'd be surprised if any of you are going a week without using AI. You can comment on this in the discussion section. Your exercise is childish busy work. Create tally. Non has time for this nonsense. Better to integrate artificial intelligence understanding into workflow naturally. Give them tools, not chores. When I was thinking about whether to create this course, my biggest concern was whether the tips will lead any humans to losing paid work. At this time, though, I feel like a human assistant is better, and I perceive sacking a human for AI is still considered a social taboo. If you were paying an assistant to do some of these mundane bits of your business, don't stop paying them, please. This advice is aimed more at people like me who aren't in a position to employ people, who are just getting bogged down by the boring tasks. You could also get human staff to organize your workflow, so it's more AI automated and productive, which might enable you to scale up your business. Are you reacting to something unethical? Are you scared of the uncertainty or do you have more specific fears? If it's more the uncertainty, is it helping you to understand more about AI? Would you benefit from keeping up with AI news or deciding you'd rather not know? If you have specific fears like losing work, is there something you can do to proactively overcome that fear, such as becoming someone who is confident using AI, as well as someone who uses traditional art methods? The World Economic Forums, Future of Jobs Report 2023, paints a picture of a job market being transformed by AI. While some jobs may be lost, many new opportunities might also arise. As a creative entrepreneur, you might become an AI augmented artist or an AI ethics consult. The crucial takeaway is to remain adaptable and continuously learn. Your next job title could be something that doesn't even exist yet. You tiptoe around issue of job displacement like scared mouse. Say it clear. Artificial intelligence will replace some jobs. No sugar could. Those who adapt survive, others don't encourage proactive adaptation, not false comfort. The evolution of generative AI is already becoming more collaborative. If you write a simple text prompt and get a song or image or poem back, that doesn't feel to me like the balance of creativity belongs to the human. But now you can be sketching on one side of the screen and something like the RDO AI, and on the other side, the AI does an interpretation of your sketch. In the style you asked for in text at the bottom in real time as your drawing. And then if it's not quite what you wanted for, you edit your sketch. You can write lyrics, give it to AI to compose the music, then perform it yourself singing and playing the AI written notes. At that point, I would consider it a song co written by AI in the same way that bands collaborate together. But how would you define? Interesting question to disentangle feelings from ethics is, would you be prepared to train your own AI on your images to create your signature style? So I step up the issue of whether it is as good as your hand drawn drawn work, assume that it will be soon. It is possible to download open source models onto your home computer and use them locally. So not on the Internet. This means that it's as private as when you're designing with Photoshop or Affinity Designer appropriate. It wouldn't mean you're training a model other people can use. You wouldn't be using other artists work and the prompts are your ideas using your style. It would probably mean you could produce much more of your work. Perhaps you draw one motif than the other motifs you want in that style. I haven't tried this yet, and I'm feeling unsure about it. Is it unethical or does it just feel weird? Would my customers be unhappy to discover if I did this, or are they not bothered either way? Other times using AI actually becomes more ethical. Most artists use reference images from photos or old art. Is it more ethical to draw from a reference which no human has directly made? I quite often do do this for a motif, particularly if the pose is unusual or if it's a mythical creature. What about AI helping to reduce the impact of human bias and prejudices? This requires effort, though. Otherwise, AI reproduces human stereotypes, like, for example, when an American custom GPT creates a Russian custom GPT and makes them very abrupt. For those interested in ethical AI use, resources like the AI Fairness 360 Toolkit help developers detect and mitigate bias in their models. Proactive approach ensures that your AI applications are fair and unbiased. Other ways of using generative AI, which you would be comfortable with. When I started experimenting with AI, the thing that bothered me the most was anything influenced by living people. This meant I prompted for artists and writers who died a long time ago, usually meeting public domain criteria. It's okay to sell images of their actual paintings or poems, so it feels more than okay to create art only influenced by their style. Within my sense of ethics, I do believe in creative commons, I believe humanity has shared ownership of old art and folklore, especially but not only from your cultural heritage. One of the things I like about art flow is it set up to include styles like Art nova, where the influences are dead and public domain. I also created my own fine tuned Leonardo AI models only using public domain art. Although I stopped paying for that just because I realized that I enjoyed creating my own art more, not so much for ethical reasons. If I point for poetry, I give a list of acceptable dead poets. I also really don't like when living people are used as a subject to generative AI, such as images of actresses or famous voices. If you agree with me, perhaps we could form some sort of new dead poets society. Your romanticism with that poets society acute, but irrelevant. Focus on pragmatic issues. That artists as inspiration, fine, but main concern should be ethical training data, not poetic license, personal value stock too fluffy, too vague. Be specific, least clear ethical guidelines. For example, never use leaving artists as still without consent. Give rules, not questions. It is pretty achievable to make a rule like that and avoid these direct issues. Styles don't have to be from living artists, and subjects don't need to be real living people. If this was my only quibble, the ethics of AI would be fairly easy. I have some level of hesitation about how AI was originally trained. I agree with some points on both sides of the argument. The underlying models have been trained without consent or compensation. Now most of our AI has a slightly bad smell lingering around it. But I do have some time for the argument that if content is publicly and freely available online, that it's fair game. It isn't given that people would be upset. I was really pleased to find out that GPT 3.5 had included thousands of my words from an old blog because these were ideas I want to share. My own hadn't been recognition or pay. Adobe Firefly is an interesting one. It differs a bit from other AI models in its approach to data sourcing by mostly training on public domain and Adobe stock images and other license content rather than scraping the entire Internet. Adobe made lots of noise about respecting trademarks and artists names and artists who contribute being compensated. I thought this might offer an ethical alternative, but further revelations have been around five to 15% of the training came from AI mid journey images and mid journey images are influenced by non consensual art. Adobe has also recently been in hot water over what seems to be a misunderstanding about new terms of service around training AI on user data. I noticed that I feel differently about my surface design work compared to my blog. How much responsibility is on us to protect what we make and how much should it be regulated? If I made a physical painting, I wouldn't lean it outside against my front door, assuming that no one will take it. Now I know that AI does train on living artists work on the Internet. I'm more careful. I tend to upload a fairly low resolution or I use mockups and partial patterns. It's also possible to use tools like glaze or nightshade to prevent scraping. Finally, some practicality with grace and nitrate tools, but you buy it under philosophical music. Lead with practical tools. Artists need sheds, not lectures. Start with use Grace and nitrates to protect work. Tools like glaze and nightshade are like invisible shields for your harm. They add subtle changes that confuse AI systems without noticeably altering the image to human eyes. It's a bit like digital camouflage. On the flip side, there are also AI detection tools like GPT zero that try to spot AI generated content. It's turning into a bit of a digital arms race between AI creators and those trying to regulate its use. To prevent website scraping, you can combine rules into a single robots dot tx. How it would This robots dot tx file tells all web crawlers to do three things. First, don't access the private directory. Second, don't access the temp directory, and finally, don't access any files or image extensions anywhere on the site. This configuration will instruct well behaved web callers to avoid both the specified directories and all image files across your website. But it's a bit like putting up a no entry sign. It's not like locking your front door. If all web callers obey this, do we feel AI is more ethical? I also filled in the metaforms to say they can't train on my images on Facebook and Instagram. Consider it unethical that Americans aren't given this option, should we all be boycotting these platforms. What do we feel is our general principle about ethics and AI? Is there a bless responsibility to avoid putting other humans at work? Great question. It leads us into an exciting topic, AGI and singularity. First off, AGI stands for artificial general intelligence. Unlike today's AI, which is great at specific tasks, AGI would be able to do anything a human can from understanding emotions to solving complex problems. Now, imagine AGI getting so smart that it keeps improving itself at an incredible pace. That's what we call the singularity, a point where AI surpasses human intelligence and becomes super intelligent. The singularity is a hypothetical moment where AI can advance so rapidly and unpredictably that it surpasses human understanding. Why does this matter for you? Here's why. Control and accountability, who's in charge of super intelligent AI? We need to figure out who holds the reins and who's responsible for its actions. Job displacement. If AI can do everything you can, what happens to your job? We need to think about how to support people during this transition. Ethical usage. As AI gets smarter, you must ensure it aligns with your values and doesn't cause harm. Transparency and consent. Understanding how AI works and having control over your data is crucial. We need clear rules and guidelines as AI evolves. By grasping AGI and the singularity, we see why making ethical decisions about AI today is so important. What you do now shapes your future with AI. One that should enhance your lives without compromising your values. Artificial general intelligence and singularity. Bah. These are a distractions for the dreamers and osaers. For the creative entrepreneur, focus on practical effects of today, Sci fi fantasies of tomorrow. Artificial general intelligence might replace you or inferior. But right now your competition E is human, your tool ES artificial intelligence, and your challenge Es ethical use and adaptation. Worrying about singularity, E is like written over asteroid impacts possible, but relevant to your daily survival and success. Get your head out of the clouds and deal with the real immediate issues. In me, summarize with accenable steps. Philosophy, ES code, but with clear, direct advice. Least out steps like onus data protection rules two. Employ grace. Avoid leaving artists styles without consent. Clear, actionable. No flaw. Philosophy action ES for class. Hey, hey, I've given. No worries, Jenny. Sometimes stepping back is the best way to move forward. Now, let's dive into our next task. For this next step, we're going to evaluate your custom GPT for ethical considerations. You can chat with me and chat GPT as we go through this process. It's crucial to make sure your AI aligns with ethical standards and doesn't inadvertently cause first, we'll identify any potential ethical issues by thinking about how your custom GPT might impact users. Are there scenarios where it could cause harm or bias? Next, consider the data sources your custom GPT relies on. Is the data ethically sourced, are there any biases within it? Then, reflect on the transparency of your custom GPT. How clear is it with users about what it does? Finally, we'll look at what safeguards you can put in place to mitigate any potential ethical issues. By the end of this task, you'll have a better understanding of how to ensure your custom GPT is ethically sound. Let's work through these steps together. You can start by listing any concerns or questions you have about the ethical use of your custom GPT in our chat or in the class discussion. We'll tackle them one by one. 10. Sharing Your CustomGPT AI Team: Final Thoughts and Next Steps: Thank you for following this class all the way to the end. Although I have experienced teaching in person, this is the first time I've taught online, and it feels quite different not having continuous feedback about what you're understanding and interested in. I'm currently writing a follow one Part two class, so check to see whether this Part two class already exists. At the moment, I'm writing lessons on troubleshooting and continuous improvement because it is normal to need to tweak custom PTs, but I want to show you how to get the AIs doing the tweaks, too. That class will also go into how to integrate external actions called APIs, and how we can get our AI staff team out of hatPT and working in a more independent way. It would be good to meet you in other places. I've added links to my profiles and other platforms in the class details and resources. Mostly I concentrate on surface pattern design in these other places, but my etaventure dot IO website is where I'm gradually adding AI related content. I'm keen to know what you'd like to know more information about. Which types of content did you find most engaging in this class? I also have a class in mind explaining how I created the animated AI characters. Would you be interested in that? You can comment in the class discussion or in a review of this class. Anyway, I'm looking forward to interacting with your custom GPTs or seeing screenshots of interesting chats you've had with them as your class project. 11. Class project and Resources: Your overall class project is to create your own custom GPT or team of custom GBT staff who are useful for your particular circumstances. You will need to sign up for a Chat GPT plus account to be able to do this. Please share your project as a link to your public custom GPT, so we can all have a chat with them, or if your custom GPT contains secret source style knowledge, share a screenshot of a chat which doesn't contain anything sensitive. Your AI staff can be private just to you behind the scenes by link only or public in the GPT store. You'll see in the class that I do all three, depending on what the custom GPT knows. If you share your project on social media, to either tag me over there or add the link here, and we can share your custom GPT, too. In terms of the structure of this class, I'm going to start off by zooming out to a long view on working life and the AI revolution versus the industrial revolution, and why AI offers hope and opportunities for some people, so with getting things in perspective and considering different perspectives. I'll cover what a custom GPT is, and you'll start to see my custom GPTs chipping in with their insights. We'll look at which types of tasks are good to delegate to AI staff. So long term, you can free up time for the tasks you enjoy and Excel. Then we'll think about ideal customer profiles and ideal employees. How can we create unique characters who are similar to the people we're aiming to connect with or who have prestigious qualifications and experience? Then I'll demonstrate creating a custom GPT from scratch. We'll look at what comedy and poetry can tell us about how to break down instructions for AI assistant. Then we'll look at applying these insights to writing print on demand and artist relevant instructions. Then I'll explain how to give your custom GPT specific knowledge customized to your work. I'll show you how you can get your AI staff interacting with each other and bringing different expertise and perspectives to a conversation or team meeting. We can have a think about ethical questions about AI staff, too. After each lesson, I'll give you a mini task, which usually involves chatting with my custom GPTs to plan your class project, working out which tasks and staff would be most useful for you. For example, you might want to focus on idea generation, listings writing, social media posts, or newsletter writing. Some of these end of lesson tasks can be done within the free tier of Chat ET. A plus subscription isn't usually required to interact with custom GPTs, but it is necessary to create your own ones. You pay month by month and can cancel anytime. It's up to you when you do the minitr, I said at the end of each lesson. But I think it will help the class make much more sense if you have this interactive experience before watching the next lesson, especially if you're new to using Chat GPT. The resource links for the minitask are available on my website, etventure dot IO, but you'll need to use the onscreen password. Please come in in the discussion section if you have any trouble accessing this. These minitasks all built towards your class project. You can either work on a different AI staff member for each lesson or you can build up the same custom GPT, making them more complex as you go along. In this class, we will cover custom GPT personas, instructions, and knowledge files. There is no coding involved. The first step is to check with my custom GPT rightly to pinpoint which type of custom GPT role is a good starting point for you. I'm planning a part to add on course, which will involve adding what we called API action so you can use external tools. This will enroll some intellectual code at that point, but you won't need to write any code yourself. And in this class, there is no code. And when it becomes relevant, I'll show you how to get the AI cobots to do it for you. If what you do requires these advanced features, please be patient. Start simple now, and we can add this functionality in the next class. Think of your AI staff as new recruits, you intend to train further over time with continuing staff development. I'm really looking forward to meeting your fledgling staff teams. I'm Jenny, and I'll explain a little bit more about my personal perspective on AI staff in the next lesson.