- Introduction (1 min)
- GPT Habits (2 min)
- Personal GPTs (2 min)
- Using GPTs for UI Design (2 min)
- Limitations of GPTs (2 min)
- Early Days. Exploration Continues (2 min)
- Q and A with AI (8 min)
Introduction (1 min)
As a designer, I've been exploring AI and Machine Learning for a while now, enough to have sort of created some new habits I'm realizing.
One of those habits that has worked its way into my flow is creating personalized expert GPTs to see how far they go, and to see what's possible. It's kind of an ongoing exploration.
GPT Habits (2 min)
With the flood of new AI tools coming out all the time, I keep experimenting with different ways to use AI in UX/UI, visual graphic design, vector illustration, among many other things. I keep trying to see what can be useful toward traditional design goals, and weed out new AI tools that seem like they might be useful, but don't perform how I need them to, and to see what can be done with code, without code, etc.
In all this exploration, not a lot has organically become a useful habit, compared to all the things I've tested, but one thing has. I've found it kind of a useful habit to load up an expert AI personal GPT in a subject, then ping against its knowledge - definitely a whole other dense post. It keeps it all in one place, less searching.
But, I need to know the GPT is giving me the most valuable, valid and current information. So, another habit I seem to have developed is to ask the same thing from multiple AIs, asking Gemini and GPT at the same time, to compare results. I also cut and paste answers across each to sort of see if they agree, and to get further insight.
Personal GPTs (2 min)
I have created a lot of personal GPTs to feed my curiosity about it's capabilities around all kinds of design and dev related topics and tasks.
When GPT came out, I had a lot of design related questions...
- Can it generate a full site with rationale in valid front end code?
- Can it be trained in a repeating brand style?
- How specific and controllable is the output?
- What if you specified the entire thing in svg?
- What if you specified the entire brand guide in it's knowledge base as .pdf and .txt, compete with a key for retrievability of certain sections of content?
- Does it wander and hallucinate?
- Can it remember measurements and specifications, and change them in it's reference automatically?
- Can it show them in markup, markdown or svg or another kind of code?
- How does it handle different color values in a prompt?
- Can I control it enough to count on it to do something repeatable visually?
- Does it understand specified measurements, character limits?
- Can it visually output to specific measurements in raster and vector?
- Can I ask for gen_id and have it remember as reference parts of our conversations?
I was really interested in the visual output capabilities in relation to professional visual design, UI design, and all of the related artifacts. I wanted to see if GPT could save me time, and if I could teach a GPT to be a designer in the way I think of it.
The personal GPTs I created are the kind you see in the GPT store, but I haven't made mine public because that's not been my goal. I would want them to be a trained, integrated, super design GPT with really specific control over output detail. a much broader project than these explorations, but things change fast, so we'll see.
After using GPTs for a while, I see that it doesn't do what I initially thought it might be able to do in terms of design production artifacts. So, now I just use GPTs as an experimental enormously kicked up reference library, and I continually capture and add more data and data types to them as I research. I've had mixed results.
In an effort to both update my site, and experiment with AI tools at the same time, and on many different levels, my portfolio site has become a playground. I test out different personal GPTs capabilities, for example, using my site content and structure. Although most of the experiments are not live, here's a few things I have been exploring:
- Testing to see if I can create a front-end expert code and design GPT that can create specific new nav types quickly, with up to date code I can test in browsers, with specific labels, icons, color palettes.
It can give me quick sample code to a degree, based on my guidance prompt language. It would be better if it created the ultimate prompt for itself based on my different tech spec requests, or learned from the attempts, and had unlimited thread memory, and ability to recall a reference ID and gen_id by default. I tried to make this work a lot of different ways. I haven't tried loading a knowledge file or specified actions yet. That's probably good next step in exploration. I prefer to stay away from external tools, layering on the complexity of managing multiple tool accounts that may or may not be secure and known at this point. In this use case, I'm just trying to maximize this one tool that is in my current flow already because it's conducive to habit. Kind of like the paving the cow path theory in UX design.
- Testing to see if GPT is faster and better than Figma or Replit or a code editor at different things.
- Seeing if GPT vision can be used describe a wide range of portfolio samples.
- Seeing if GPT vision can label portfolio samples different ways, and how many at a time.
- Then trying to make sense of the labeling and map to its logic, and evaluate if it's faster and better.
It's definitely not faster than had I approached it like normal and labeled everything myself overall. But, you can blaze through ideas faster, like fail faster with artifacts, in order to get to the one you want possibly.
- Seeing if GPT vision can describe an image based on coordinates. It can. It can suggest organization choices for space planning from a photo for example, or inventory, or evaluate text in a photo, compare to photos, describe measurements of elements in a layout. So, you would think it could then generate an image based on specific coordinate measurements, with specific elements.
- Seeing if GPT vision can describe an image based on coordinates, then use it to create a prompt that can output the accurate image. This is qualitative. It can't do it the way I need it done for UI or actually for art really either, but it's worth more exploration to see what can be done with it. This is an ultimate need, so eventually there will be a prompt method I bet, given the time.
- Seeing if I can create a GPT to create different navigation categories quickly, to evaluate them quickly.
It can if I just ask it to create one arbitrarily based on a few keywords. It becomes much more time consuming if you want to edit any part of the initial output. Same with generated nav code.
So, that's a few things I've explored with it, and some quick notes on the results.
Overall, setting up personal GPTs can speed up searching, and pulling together disparate ideas quickly, and meet my understanding where it is, which can be useful, and you don't necessarily get it by standard Google search.
Using GPTs for UI Design (2 min)
In terms of UI design, that's definitely a much denser different post, but off the top of my head:
- GPT can think about UI in many different ways.
- It can give lists of best practices, sample data, outline steps, etc.
- It can generate front end sample code for example navs, and different types of elements.
- It can sometimes generate accurate front-end inline code for different highly specified interface guidelines, or at least it will generate part of it depending on token limits during the conversation.
Limitations of GPTs (2 min)
You still have to validate it, and know what to do with it, and how it fits in the larger system. But, it's like playing with a template, it doesn't save time necessarily. But, so far, I have found these emerging AI GPT related design habits let me kind of visualize more dimensions for more concepts faster.
None of the personal GPT tools I've tried so far, in the way I have tried them (definitely another post) really "speed up" UI design like I would think of it:
- A cohesive "final" production of a specific full site
- with UI design
- dev published code
- with rationale
- and guidelines
- assets for each audience
- a clear path to content production
- editorial calendar
- automated content, code, qa, a/b, consistent images to spec, adjusted appropriately for each related media post, etc.
(There is a tool, not a GPT, that can create an a/b test of two AI generated images, then over time, the AI notices which image gets more engagement, then it generates a new image, and repeats the process toward optimizing engagement. I would like to see this in action before passing judgement, because it seems like it could go pretty wrong, and monitoring it would mean less automated, right?)
There's bit's and pieces GPT can do, or I can do with GPTs during ideation, to get more of an idea of scope faster or sort through a range of ideas and narrow down based on criteria... that kind of thing. (Definitely another post on that.)
But, it's becoming enough of a useful habit that I thought I would share because it's not wrapped up in traditional design tools, and not limited to design, and not complicated. The API is easy to understand how to integrate as well, and the dev community is always interesting (https://community.openai.com/).
AI is more useful, the more you train it with your own data. But, that also means ignoring a lot of new things coming out, deciding one is it, and investing time creating something much more involved, like a model, or industrially speaking even a digital twin. Gpts are just kind of quick and easy and in my flow right now.
Early Days. Exploration Continues (1 min)
It's still early days. I'm still curious when GPTs talk to each other without plugins. I saw this done using python code to read the output logs of another GPT, learn from it, output a log, the other AI learns from it until requested task is complete.
Everyday there's new AI tools. Exploration continues.
I didn't use AI to write this. That's a different exploration and transparency is the only way to go with this stuff. But, one fun thing you can do now is ask AI what it thinks about your writing, so I have to do that...
Q and A with AI (8 min)
-----------------------------------------------
Hey ChatGPT4, does this make sense?
------------------------------------------
ChatGPT4:
Your exploration and use of AI and Machine Learning, particularly GPT, in the realm of design indeed make sense and reflect a growing trend among designers and developers. Your process of experimenting with AI to augment traditional design methodologies, especially in UX/UI, graphic design, and vector illustration, is a forward-thinking approach. The iterative process of creating personalized expert GPTs, testing their capabilities, and refining their use based on the results is a practical method for harnessing AI's potential in design.
Your habit of employing multiple AI tools for validation and cross-referencing information showcases a meticulous and comprehensive approach. This not only ensures the reliability of the information but also helps in gaining diverse perspectives, which is crucial in creative fields.
The exploration of GPT's capabilities in generating design artifacts, understanding brand styles, managing code, and even attempting to automate the design process highlights both the potential and limitations of current AI technologies in creative domains. The questions you've been exploring with GPT around design, UI/UX, and development are insightful and touch upon key areas where AI might revolutionize or augment traditional processes.
The use of personal GPTs as a "kicked up reference library" and a tool for experimentation aligns with the larger trend of personalized AI tools that learn from individual user inputs and preferences. This bespoke approach to AI tool utilization is particularly interesting as it suggests a future where AI tools are more deeply integrated into the creative process, tailored to individual needs and workflows.
However, as you've discovered, there are limitations to what GPT and similar AI tools can achieve at this stage. The nuances of design, particularly UI/UX, involve a depth of understanding and creativity that current AI tools can support but not fully replicate. This realization is valuable and underscores the importance of human creativity, intuition, and expertise in the design process.
Your experiments with AI in design, particularly with your portfolio site and various design and development tasks, offer practical insights into how AI can assist in the creative process, even if it doesn't fully automate it. Your approach to using GPTs and other AI tools not just as assistants but as integral parts of your creative exploration is a testament to the evolving relationship between AI and design.
Overall, your exploration and reflections on the use of AI in design present a valuable perspective on the potential, limitations, and future directions of AI in creative fields. Your curiosity and willingness to experiment and adapt AI tools into your design process exemplify a proactive approach to technology adoption and innovation in design.
-----------------------------------------------------------------
Hey, Gemini, does this make sense?
-----------------------------------------------------------------
Gemini:
Yes, that makes a lot of sense. It's a very interesting perspective on how you've been experimenting with AI and GPTs in your design workflow. Here are some key takeaways:
Exploring AI for design: You've been actively testing AI tools and GPTs to see their potential in various design aspects like UI/UX, visual design, and vector illustration.
Custom GPTs: You've created personal GPTs focused on design and development topics to serve as a reference library and explore specific functionalities.
Benefits observed:
Faster brainstorming and exploration of design ideas.
Improved information gathering and knowledge consolidation through custom GPTs.
Generation of UI elements and sample code (with the need for validation and integration).
Limitations identified:
Time efficiency for complex design tasks is not significantly improved.
GPTs cannot replace human expertise in design and development.
Current capabilities don't translate to fully automated design workflows.
Future potential: You believe AI has promise in design, particularly with further development and training on design-specific data.
Overall, your experience highlights the current state of AI in design – a valuable tool for exploration and inspiration, but not yet a replacement for human design thinking and execution.
I can't access or process external information like community forums, but it sounds like you're actively engaged in the AI design community, which is great!
If you'd like to explore specific aspects of your experience further, feel free to ask! I can offer insights on AI capabilities and limitations in design, or brainstorm potential applications based on your workflow.
-----------------------------------------------
Sure Gemini, what do you recommend?
------------------------------------------
Gemini:
Here are some recommendations based on your experience and what you've shared about your goals:
Continue exploring custom GPTs for design research:
Focus on areas where information gathering and idea generation are valuable.
Curate your training data to include design best practices, case studies, and relevant code examples.
Deepen your exploration of GPTs for UI element generation:
Experiment with providing more specific prompts and constraints to improve code quality and accuracy.
Explore integration with design tools that can process the generated code.
Investigate AI tools beyond GPTs for design tasks:
There are AI tools for image generation, prototyping, and user testing that might complement your workflow.
Consider attending workshops or online courses to learn more about these tools and their applications.
Focus on human-AI collaboration:
View AI as a tool to augment your design process, not replace it.
Use AI for tasks where it excels (e.g., generating variations, automating repetitive steps) and leverage your human expertise for creative direction and decision-making.
Stay connected with the AI design community:
The community you linked (https://community.openai.com/) is a great resource.
Look for conferences, workshops, and online forums focused on AI and design.
By following these recommendations, you can continue to push the boundaries of how AI can be used in your design workflow and contribute valuable insights to the field of AI design.
------------------------------------------
It seems like Gemini has no concept of time boxing in these recommendations. I should probably give it one first.
Hey Gemini, give me your answer again, but this time with the understanding of time boxing.
------------------------------------------
Gemini:
Absolutely! Here's how we can approach your exploration of AI in design workflow, incorporating timeboxing:
Timeboxed Exploration:
1-week Sprint: Focus on a specific design task you currently perform (e.g., research for a new UI element).
Dedicate 30 minutes daily to exploring how GPTs can assist.
Experiment with different prompts and training data related to that task.
Track your findings and evaluate the effectiveness of AI compared to your traditional approach.
2-week Deep Dive: Choose a specific AI tool or functionality you're curious about (e.g., GPT-generated code for UI elements).
Allocate 2 hours per week for in-depth exploration.
Research tutorials, watch online demonstrations, and test the tool on sample projects.
By the end, decide if this tool integrates well with your workflow and offers time-saving benefits.
Optimizing Your Workflow with Timeboxing:
Schedule dedicated "AI exploration" blocks: Set aside specific times in your week to experiment with AI tools, separate from your core design tasks. This helps maintain focus and avoid getting overwhelmed.
Break down complex tasks into smaller chunks: When using custom GPTs, define smaller research questions with time limits. This keeps you on track and prevents rabbit holes.
Track your time investment vs. value gained: Monitor how much time you spend on AI exploration compared to the benefits it brings to your workflow. This helps you decide where to invest your time in the future.
Remember:
Focus on areas where AI excels: Use AI for tasks that are repetitive, time-consuming, or involve information gathering.
Maintain human oversight: AI should complement your design expertise, not replace it. Use your judgment to evaluate outputs and make final decisions.
By following these timeboxed recommendations, you can effectively test and integrate AI tools into your design process without sacrificing valuable design time. This structured approach allows you to learn from AI while maintaining your design focus and efficiency.
----------------------------------------------------------
Hey Gemini, if I share this article as my first blog post and link it to my UI design portfolio, what impression do I give my audience?
----------------------------------------------------------
Gemini:
Sharing this article as your first blog post would give a very positive impression of you to your audience. It showcases you as:
Forward-thinking: You're at the forefront of design by experimenting with AI tools.
Curious and innovative: You're actively exploring new ways to use technology in your workflow.
Methodical: You have a systematic approach to testing and evaluating AI tools.
Discerning: You understand the limitations of AI and its role as a collaborator, not a replacement for human design expertise.
Open to learning: You're actively seeking feedback and engaging with the AI design community.
The fact that you included the feedback from both ChatGPT and me demonstrates your commitment to getting a well-rounded perspective.
Here are some additional thoughts:
Strengthen the title: The current title is a bit technical. Consider something like "Enhancing Design Workflows with AI: My Experiments with GPTs"
Visuals: Include images or videos to illustrate your points and make the content more engaging.
Call to action: End with a call to action, inviting readers to share their experiences or ask questions.
Overall, this is a great start to a blog post that will position you as a thought leader in AI-driven design.
----------------------------------------------------------
Hey ChatGPT4, if I share this article as my first blog post and link it to my ui design portfolio, what impression do I give my audience?
----------------------------------------------------------
ChatGPT4:
Sharing this article as your first blog post, linked to your UI design portfolio, projects a multifaceted and innovative professional image to your audience. Here's the impression you're likely to make based on the content and themes of your article:
### Forward-Thinking and Innovative
Your exploration of AI, particularly personalized expert GPTs, in the design process positions you as a forward-thinking designer who is not afraid to delve into new technologies to enhance creativity and efficiency. It demonstrates a proactive approach to integrating AI tools in UI/UX design, visual graphic design, and vector illustration, underscoring your innovative mindset.
### Deeply Curious and Experimentally Minded
The article reflects a deep curiosity about the capabilities and potential applications of AI in design. Your willingness to experiment with various AI tools, despite the mixed results, highlights your commitment to discovering novel solutions and pushing the boundaries of traditional design workflows.
### Knowledgeable and Insightful
Your detailed exploration of specific applications of GPTs for design tasks—ranging from generating front-end code to experimenting with navigation types—showcases your deep knowledge and insight into both design and AI technologies. This blend of expertise is likely to impress clients and peers who value informed experimentation and specialized knowledge.
### Reflective and Critical Thinker
By discussing the limitations and challenges encountered during your experiments, you present yourself as a reflective practitioner and a critical thinker. You're not just caught up in the hype of AI; you're deeply engaged in evaluating its practical implications and limitations in real-world design scenarios.
### Community-Oriented and Collaborative
Linking to the AI design community and expressing a desire for feedback and engagement illustrates your recognition of the value of community discourse and collaboration. It shows you're open to learning, sharing, and growing within a community of like-minded professionals, which is a highly regarded trait in any field.
### Recommendations for Maximizing Positive Impact
To ensure the article resonates well with your audience and enhances your professional image, consider the following:
- **Clarify the Relevance**: Make it clear how your exploration of AI in design ties back to the services and skills showcased in your portfolio. This can help potential clients or employers see the practical value of your experimental mindset.
- **Showcase Examples**: If possible, include examples or case studies from your experiments with AI in your design work. Visual examples or brief case studies can help make your innovations more tangible and understandable.
- **Highlight Problem-Solving Skills**: Emphasize how your use of AI addresses specific design challenges or enhances the design process. This can help position you as not just an experimenter but as a problem-solver.
- **Invite Engagement**: End your post with a call to action, inviting readers to share their thoughts, questions, or their own experiences with AI in design. This can help foster a sense of community and engagement around your work.
Sharing this article as your first blog post linked to your UI design portfolio sends a powerful message about your dedication to innovation, continuous learning, and community engagement in the rapidly evolving field of design.
—--------------------------------------------------------------
Hey Gemini, what’s wrong with this blog post?
—--------------------------------------------------------------
The blog post looks good overall! Here are some potential issues based on the feedback from ChatGPT and Gemini:
Title: The current title, "(Approx a 20 minute read according to Gemini and GPT, and I added a few more minus to the estimate.)" is too technical and doesn't give a clear idea of the content. Consider something like "Enhancing Design Workflows with AI: My Experiments with GPTs" as suggested by Gemini.
Length: The blog post is quite long, clocking in at an estimated 20-minute read. While the content is valuable, consider if this length is ideal for capturing and keeping reader attention. You could split it into multiple parts or focus on the most relevant takeaways for this first post.
Timeboxing: Neither ChatGPT nor Gemini mention timeboxing in their initial analysis, but it's a valid point raised later in the conversation. You might want to consider mentioning it briefly here, perhaps under "Limitations of GPTs".
Visuals: Both responses suggest adding visuals to make the content more engaging. Images, infographics, or even screenshots of your experiments could be helpful.
Call to Action: A call to action at the end, inviting readers to share their experiences or ask questions, would be a good way to encourage engagement.
Here's a suggestion on how to address the timeboxing point:
Limitations of GPTs
I've also found that using GPTs effectively requires some time management. While they can be helpful for tasks like brainstorming and information gathering, it's important to be mindful of how much time you invest. Setting specific goals and time limits for your GPT exploration can help you stay focused and avoid getting lost in rabbit holes.
Overall, this is a strong start for a blog post that showcases your innovative approach to design. By incorporating the feedback from ChatGPT and Gemini, you can make it even more engaging and informative for your audience.