At The Design Project, we're always exploring new technologies. In particular, we are focusing on understanding how designers can become more productive using Generative Pre-trained Transformers (GPTs) developed by OpenAI. Our designers Euge and Matute discussed how these AI models are not only revolutionizing the design industry by enhancing efficiency and creativity but also empowering designers with tools for better communication and idea generation. They delve into the practical applications of GPTs in everyday design work, from automating mundane tasks to conceptualizing innovative designs. The conversation also touches on the importance of adapting to evolving technologies, ensuring that human creativity remains at the forefront of design. Join us as we explore the exciting intersection of AI and design, and discover how GPTs are becoming indispensable tools in a designer's toolkit. Enjoy! TDP GPTs here.
Read
Matute: We are starting to experiment with new technologies at TDP. We always like to use the latest technology possible, both in crypto and also in AI. We are right now experimenting with GPTs, which stands for Generative Pre-trained Transformer. It's a type of artificial intelligence model developed by OpenAI. In the context of ChatGPT, GPTs refer to the different versions or iterations of this model used to power the chatbot. And also, you can train your own GPTs. The ChatGPT application utilizes these GPT models to simulate human-like conversations, answer questions, and provide information across a wide range of topics. The performance of ChatGPT depends on the underlying GPT model it's using, with newer models generally offering more sophisticated conversation abilities. And maybe you're wondering, why would I train a specific GPT with a GPT creator? Training a specific GPT model using a tool like GPT Creator, which allows for customization and fine-tuning, can be valuable for several reasons. Number one, specialized knowledge. If you need a model that has expertise in a specific domain, like law, medicine, a niche area of technology, or in our case, design, fine-tuning a GPT model with specialized data can significantly improve its performance in that domain. Also, customize responses for applications where you need the model to respond in a certain style or tone, like a company's customer service bot. Training it with data reflecting that style can yield more appropriate and consistent responses. Reducing bias and improving safety. Standard GPT models can sometimes generate biased or unsafe content because they are trained on large, unfiltered datasets from the internet. Fine-tuning with carefully curated datasets can mitigate these issues. Language and localization. If you need a model that is proficient in a less common language or dialect, or one that understands local cultural nuances, training with specific datasets can be crucial. Proprietary or unique data. If your use case involves interacting with proprietary or unique data, like an internal database or specific literary works, training a GPT model on this data can enable it to understand and generate content relevant to this unique context. Compliance and privacy. In cases where data privacy and compliance are important, like healthcare or finance, training and monitoring specific compliant datasets can ensure it operates within required guidelines. Enhancing performance on specific tasks. For tasks like answering specific types of queries, making recommendations, or generating particular types of content, a fine-tuned model can perform significantly better than a general-purpose one. And last but not least, incorporating user feedback. Over time, user feedback can be used to improve the model's responses, making it more effective and user-friendly for its intended application.
Euge: Matute, how do you think the design world is actually going to benefit from this? I think it's interesting to consider that, a lot of times, AI is perceived as a big scare for professionals in general. I believe that everyone is a bit scared about AI and how it's going to affect their job in the future. It's interesting to think about how it might affect design. At least, this is my thought, and I would love to hear yours as well. Some people, like me, think that design is not going to go away whatsoever. Human design, or someone behind the design, is just understanding a bit more about how we can collaborate with the new tools that we have, which nowadays are completely new, super frightening, and we still don't really know how to take advantage of them. Perhaps there's a lot we can take advantage of, but it's all so new. There are so many tools, some good and some really bad. So, my feeling about the future as designers is that we need to adapt to these new tools and to AI in general and new technologies, but that humans are not going to go away from design. Is this something you think, or what's your position on this future we see ahead for our profession?
Matute: Well, I think it's a new tool, and like every tool, it will erase certain tasks but will help us make another task more efficient. Some examples of how we can use AI and GPTs in design practices could be, for example, for idea generation, assisting in brainstorming and conceptualizing new designs, or with content creation, automating text generation for web and graphic design elements, erasing the famous Lorem Ipsum. Enhancing UX design by creating interactive elements and analyzing user interaction for better user experiences. Personalization, tailoring design to individual user preferences and behaviors. Automating repetitive tasks, handling routine tasks like code generation and document formatting. Serving as a research assistant, quickly gathering information and trends to inform design decisions. Also facilitating communication, improving the interaction between designers, clients, and stakeholders, providing feedback, offering initial evaluations of design concepts, promoting accessibility, aiding in the creation of more inclusive and accessible designs, and predictive analysis, forecasting future design trends. If you want, we can talk about some GPTs that we created at TDP.
Euge: Yeah, I would love that. I'd love to learn a bit more about how this process of creating GPTs has worked. If you found any new learnings on design while creating these GPTs, I think that's interesting also. In the end, you're designing a GPT, but not in the UI, UX way we're used to in our everyday work. Like how this has worked and how perhaps you think we might be helping designers in the future with these new creations of GPTs. I think it's super interesting that as designers, we have the tool to create activity. As you said, it's like, what is it called, natural language? I think that's the way you create GPTs, right? With natural language, exactly. Yeah, I think that it's super interesting that as designers, finally, we can actually program something because we're so used to working with devs that are constantly knowing how to basically direct computers and systems to do whatever they want. And it's the first time I, at least me, who doesn't have development training or developer training, can actually make a computer do what I want. So that's, I think, so interesting, not only for us designers but for anyone that wants to use a GPT to create it. So, I think it's also interesting for us to tell people how we actually created them, what we learned, and if there's any specific tips or tricks to creating a GPT that you have.
Matute: Right, well, as you said, since the OpenAI Dev Day, which was the first one ever, and the release of the GPT Creator, we can now fine-tune and create new specific GPTs without needing to know how to code. And that's, at least from my perspective, revolutionary, and it democratizes access to AI for more people. In the design project, we apply the design process in the same way as creating GPTs. The first step for us was, with a couple of designers, to do a brainstorming of all the ideas that we could use GPTs for in our day-to-day work as designers. We came up with like 16 different ideas, and after that, we had five minutes, ten minutes, it depends, to prioritize and vote on each idea. Do you understand what we think is the best option that we should build first? Right. After that, we set up an objective for our quarter, and we decided that we wanted to build six GPTs in the first quarter of 2024 and another six in the second half of the quarter of 2024. And then, after that, we started talking with HGPT because we should also use it in this process. We started to define the problem that HGPT will tackle, the solution that will be provided, the target audience for that specific GPT, the objectives of the GPT, and last but not least, the limitations of the GPT. Because we should take that into account if we want to have the best performance of the GPT. One of the very common limitations that we have in most of the GPTs was that we don't want this GPT to create images since ChatGPT can create images with DALL·E 3, but we were very focused on text interaction, text-based interaction with the GPT. We disabled this ability and focused just on text generation.
Euge: I'm sorry, just one quick thing on that. It's super interesting how, when you ask Dali or JGPG to create, I don't know, some piece of UI or anything, it still is completely lost. Like, GPTs are not yet trained, at least for now, to create real UI. I think it's interesting how, in the end, integrations and different things can work because GPT, on its own as a UI designer, is lagging a lot. But it's interesting to see how that can develop in the future and also how it can grow into actually being a tool where UI can go through AI, and we can think solely about user experience. I think it can add a lot of value to the users. That time when we can actually think so much more about the user experience rather than the look and feel, perhaps GPT will be amazing.
Matute: Right, at least for now, maybe in a couple of months it will be different. But for now, it's very limited with the understanding of a screenshot of a UI and what we should improve from that UI, and also to generate new UI. It's very good for conceptualization and maybe to create a mood board, but we still need to do the work that we do as designers. That's why we focus more on the UX side, which is more focused on the research, on the definition, also assisting for communication with stakeholders, and not that much on the generation of UI and visual design. So, once we defined the problem, the solution, the objectives of each GPT, the target audience, and the limitations of the GPT, we needed to create instructions for the GPT. Taking into account all that information, we started to build customized instructions for each GPT, saying what it has to do when interacting with the user, when looking for information on the internet, when creating an answer to give to the user, also understanding what are the limitations of the GPT and how to communicate to the users which limitations it has, and many more things that we can set up in the instructions of the GPT. And once that is done, the GPT is set up. But that is not when our work finishes as designers because we all know that we need to first test our new product before releasing it. So, we did two different kinds of testing. The first one was with alpha testers. That was the people that are actually working as designers inside of TDP. So, we tested it first with our team. And then we asked for a small group of people that are designers and also have access to Chativity Plus. And we did a little group of beta testers, and we tried with people from outside of our team to add more feedback, gather all the feedback, and make improvements inside of each ABT. And that's how we end up now releasing our first GPT open to all the design community and also for non-designers because everybody can use them. Right now, we're focusing on optimizing GPTs for designers since it's our day-by-day work and we want to not just try to sell it, but to actually use it in our process of design.
Euge: Sorry, adding a bit to that, to the idea of adding distributors for our day-to-day, I think it's interesting to let everyone know that part of our process, while ideating on the GPTs or understanding what we could create, was, "Let's build something that makes sense for us." So, the time we use is actually, in the future, a better time management and it's more efficient for our process as designers and our everyday work. So, perhaps it's also about understanding, in general, for anyone that wants to create GPTs in their day-to-day because it's fun. I think that the first GPT you create, a good experience or a good idea, is to create a GPT that works for you. So, you can test it out. You can see if it works. And actually, you start finding in that GPT the answers you need, for example, for making a new recipe, or if you want to turn into a vegetarian diet, okay, let's create a GPT that actually helps you ideate a new diet based on vegetables. So, in that way, while you're solving for yourself, you'll try to make the best GPT ever. It has the answers you need. So, that's perhaps the best way to make that first approach into the GPT world, into the creation of GPTs. I think that's a good idea or a good path to take while trying to start in this process.
Matute: Exactly. In the best-case scenario, it will be super popular, and everybody will use it. If they type any keyword inside the search bar of the GPT source, they will find it, and that's awesome. But in the worst-case scenario, you are using it every day. You are being more efficient in your work, and you can just start to use it as a team in our agency, in TDP, as a team all together.
Euge: 100%. Yeah, yeah, yeah. And it's super fun to see how it can grow with the new instructions, with new models you add, and so on. I think it's a completely new world, and it's super fascinating as a designer to start developing something, as I said before, that a computer actually responds to what you need. And it's super exciting to see how it actually works and starts getting bigger and bigger with the new knowledge that you add to the GPT.
Matute: Right, exactly. Well, maybe we can say a little bit about the activities that we created. Yeah. For now, we have seven created, and we have nine more to be released in the future. The first one is a stakeholder communication assistant for designers. We also have an MVP definer, a user persona generator, a social media caption generator, a design principles expert, an agile methodologies expert, and an accessibility expert.
Euge: Which one are you most excited about?
Matute: I think a Stakeholder Communication Assistant is crucial for designers because most of the time, or at least with the majority of designers I talk to, the struggle is understanding how to translate our concepts and ideas. It's about how we communicate between designers and people who are not designers, those more focused on the business side, or maybe with developers. Having that assistance to just type in whatever you want, saying what you did, and then translating it into easier language for non-designers, I think that's the best tool for now, yeah, 100%. And adding a bit to that concept of stakeholder communications and how GPT can work, I think that...
Euge: Something very valuable I learned about designs is that designs on their own don't go anywhere. However, designs that are good, have good communication and background, and can justify the decisions behind each of the different parts and components, are the ones that actually go far. Without clear communication, designs basically go nowhere. You need someone to communicate what the designer wants to do. I think it's super important to have designers as that spokesperson. So, this GPT about communication, I think, might go very, very far, in the sense that we all need it. It's always super complicated when people with business terms and developers with their complicated words come and talk to us. It's interesting how we can work interdisciplinary with them. But at the same time, it's super helpful to have this tool, a GPT, to help us go through our day-to-day.
Matute: Right. The idea of these GPTs is not to replace the designers but to give superpowers to the designers. I love that about GPTs.
Euge: GPTs that give superpowers.
Matute: Exactly. And so, we can have a super team of designers at TDP to tackle harder tasks because, I think, the tech industry is always evolving and making it more complex. So, it's great that we, as designers, can work on other things and focus on other types of tasks, and not just on this boring task or difficult task for designers, and to have that assistance.
Euge: It's boring, it's boring. You can say it haha. It's like, let's focus on design and leave everything else to the GPTs, right?
Matute: Exactly. So, you can focus on what you actually want to do and what you are best at doing.
Euge: Yeah, I love that.
Matute: So, to finish this interview, I think we should say why we are releasing it right now. Next week, we are heading to ETH Denver. It's one of the biggest events of Web3 in the USA and also in the whole continent of America. Since everything in Web3 is so new and early, we want to showcase to the businesses and the projects that are inside this industry that we are also working with new technologies. We are always updating ourselves with the latest news and technologies. And that's why we are releasing our first GPT just before going there. So, if we contact people who are creating new things with the latest technology, we can showcase some examples that we are also doing the same.
Euge: Yeah, and I think that it's also interesting to perhaps, with this whole talk we had about the process and everything, is that this GPT creation, we're taking it very seriously. It's not something that we just want to release and get it over with. It's something that we actually want to add value with the GPTs to designers. And it's like looking for the best way we can add value to designers. We can add value to designers and other professionals working with designers. GPTs are here to help us. And I think that it's a very interesting way for us to test out new technologies and see actually how we can keep designers updated in this world that changes every day. And I think that GPT is a super interesting way to introduce ourselves to this.
Matute: Exactly. Because I remember you started this conversation saying that maybe sometimes it's a little scary. And I think we shouldn't be scared. We should be curious, and we should always be experimenting with these new technologies and trying to apply them in our workflow instead of thinking that we will be replaced. I think we will be replaced if we are not updated in this always-evolving industry, right?
Euge: Yeah, and we work in technology in the end, so we have to keep updating. We work with technology. Everything changes every day. Figma tomorrow might not exist anymore, and something completely new might come in. Some years ago, we had Adobe, and then we have Figma as our main tool. So, let's set up tools and leave them as soon as we want. That's like the interesting part of testing everything out, seeing what works, and seeing what doesn't. That way, you can actually choose which GPT to use in an informed way and decision because things change constantly. And as designers, we have to keep updated, have to stay at the forefront of innovation, and yeah, I'd say it's the best way to design cool, new, and amazing things, I think.
Matute: Right, I hundred percent agree.
Euge: Nice. Well, GPTs are coming. From the design project side, we hope that all designers want to come and see what new technologies look like and how any designer can actually achieve it. And if you want to create your own GPT, go for it. It's amazing. It's easy. It's fun. That's all of it. And once again, do that. It makes you curious and solves your own problem. That way, you learn a lot, and then you can start applying those GPTs to others and to other problems, other needs, and other goals in general. So, have fun with GPTs and create a lot of GPTs.
Matute: Yay!
