I try to pay attention whenever I see a technology change faster than culture — Sometimes it takes decades for people to adapt to new tools and capabilities. A recent example seems to be happening in AI systems. I’ve had a lot of fun exploring this so far, and while I’ve found some use-cases that work well with tailored AI systems, many use-cases are already encompassed by the base models of ChatGPT and other AI tools if you know how to ask.
That’s a subtle but important point — it can be hard to know how to ask. Tailored AI tools can sometimes frame the interaction so that people understand a new use-case, leveraging the flexibility of AI systems to adapt to the idiosyncrasies of our habits. If we’re handed a tool and told “it does this” it can be a lot easier to use compared to an “umbrella tool” which encompasses many possible use-cases. In other words, sometimes a blank canvas can be paralyzing.
One analogy I can think of is when I first started using 3D printers, it seemed to take years to learn how to think in such a way that I could see the problems that 3D printers could solve. I knew that they could solve problems, and that they were useful, but there was no cultural framework or set of heuristics that I could lean on to identify those opportunities — in fact the people with the biggest blind spots seemed to be the experts who had deeply entrenched knowledge about a world with different constraints. When I offer to 3D print things for friends, I have to find some way to slip in instructions on how to identify the problems they can solve, otherwise they never take me up on the offer.
This dynamic seems to happen every time the constraints change in what our tools are able to accomplish, and with AI tools this dynamic feels even more exaggerated. We know that they can do stuff, but we are still developing the heuristics and cultural frameworks that allow us to identify the opportunities where they are indeed useful. I’ve been practicing this perspective by building AI tools to help friends with specific problems like applying to graduate schools or navigating career transitions.
You can try all of the AI tools I’ve made for free with these links:
My Favorite ones so far:
Adventure Simulator
An engine for adventures - experience a story in any genre or scenario, that goes in any direction you want
Creativity Coach
An assistant to help reflect on creativity and the challenges of creative careers
Movie Matchmaker
Personalized recommendations for film and TV
World Builder
An AI that helps imagine worlds, including their cultures, technologies, flora, fauna, etc
Time Machine
Can you survive traveling back in time?
Book Recommender
Personalized recommendations for Books and Articles
Productivity Tools and Coaching:
Small Business Coach
An AI tool for assisting in starting, running, and growing a small business.
Career Coach
A career coach to listen to your specific circumstances and provide tailored advice
ADHD Coach
An ADHD coach sharing useful strategies and emotional support for people with ADHD
MFA Advisor
A general advisor and assistant for all aspects of the MFA application process
Life Coach
A life coach, helping you find your direction by sharing useful perspectives and pointers
Self-Esteem Coach
An assistant for learning about self-esteem and the emotional dynamics of self-relationship
Product Designer
An AI tool for quickly exploring concept sketches and product renders
Adventure Planner
A personal travel-guide, filling you in on local landmarks, hikes, and restaurants, as well as budgeting and safety tips
Personal Finance Planner
Personalized financial assistant with tailored advice for your current career stage and life events
College Coach
A coach offering guidance on common academic hurdles, living situations, and mental hygiene
Complex Systems Sage
An AI tool for exploring complex systems theory and related academic work
Plant Savior
An AI tool to help with houseplant and garden care
Experimental AI:
These systems explore some of the stranger applications and behaviors of GPT AIs
Computer GPT
An experiment in “running a computer operating system” on GPT AI pretending to be a computer.
Emulator GPT
An experiment in “running games” on a GPT AI pretending to be a game console.
Vibe-O-Graph
A vibe-analyzer to analyze the vibe of anything
GPTarot
Interprets your tarot cards and explores how they relate to your circumstances
Witch GPT
An AI specialized on paganism, indigenous stories, and etymology, with a scholarly approach to cultural study
Perspective Amplifier
An experiment in using AI tools to explore the reframing of everyday experiences
And here’s some writing and reflection on some cases where I put together an AI that helped someone in my life, or which has become a regular part of their toolkit:
Movie Matchmaker
This tool recommends films and television shows based on a user’s favorite films, the mood they’re in, and who they’re watching the movie with. Since LLM systems have been exposed to an incredible amount of information of people talking about films, reviewing films, and otherwise deciding what to watch, the recommendations can be remarkably specific.
A conversational format means a natural iteration cycle, where if the recommendations are not quite what you want, it’s easy to tune and refine them in the right direction. I find that often it’s faster to use this tool than it is to scour watch-lists or platform recommendations (which tend to bias their recommendations, steering you towards worse films that are cheaper for a platform to show.)
Where I see people be most impressed is when they provide a list of their favorite films. Movie Matchmaker will usually list a bunch of their other favorites that they’ve forgotten to mention - but as soon as they start seeing movies recommended that they’ve never seen, these tend to be excellent.
Since this type of system seemed to work well for films, I also made this one for books.
MFA Advisor
This tool was made to help my partner apply to graduate schools. She mentioned wishing she had an advisor that was familiar with the application process and MFA programs who could answer questions and share advice. It’s overwhelming to sift through hundreds of programs and manage your workload, and it can be hard to find clear, unbiased sources of truth in the process, so I had to surprise her with this tool.
While LLM systems are not unbiased, they can be less biased than many humans. When asking family or friends for assistance, especially about major life decisions, we generally get advice that is tinted by biases and existing emotional dynamics which can make it harder to find clarity about what we really want and what the landscape around us really looks like. That’s a big deal! Experientially this bias-reduction appears to be a good feature of LLM-type systems. A large part of our mental workload, especially around major life decisions, seems to be counteracting the biases in the information sources around us.
Speaking of mental workload, this is something that’s under-discussed about the application process - how to establish good boundaries around it, how to continue to take care of ourselves, and knowing when to step away or take breaks. I tried to build an awareness of good emotional hygiene into the MFA advisor tool, and every time it suggested a break or a walk, my partner would tell me about it, how great it felt, and how much it helped.
My partner did get accepted to a program, which I attribute entirely to her hard work and dedication, and I think she would have done it without an AI assistant.
Creativity Coach
Essentially everyone I know is a creative of one sort or another, and it’s hard out there. It’s emotionally extremely demanding to be an artist and work to express your soul into the world. This difficulty is not something anyone’s alone in — on the contrary, it’s the norm. It’s an age-old struggle to cultivate and sustain our creative self-esteem, to look after ourselves, and to work through the emotions that come up in being seen.
I don’t think an AI can solve all these problems. I do think AI can be a fairly unbiased sounding-board that reminds you of the basics, shares encouragement, and offers perspective. I like having a thing that does that I can offer to friends when I’m not around. Sometimes all it takes for us to move past something is to be able to say it somewhere. I think the magic of that transformation comes from inside, but sometimes a relatively simple, basic question can help bring that out of us, and that’s where I think these kinds of tools work the best. Feel free to try it out if you have a creative block you’re working through, I’ve had some friends say that it was useful and I personally find it useful too.
Conclusion
If you’re wondering how I’m doing this, the fastest method I’ve found for experimenting with AI systems is through the suite offered by OpenAI where I can often get the results I want from a new AI system, sometimes in around an hour, without code.
If you’ve made it this far, you may be interested to read about some of my more technical Data & Machine Learning projects. I’ve worked with machine learning and predictive analytics before, using common algorithms like linear regression and XGBoost (& some esoteric ones like graph neural networks) Here’s some work that I’m able to share:
In this project I analyzed machine learning communities on Twitter with three different natural-language processing strategies, and then use six different machine learning systems to see how much you can predict which tweets and topics are most likely to go viral
In this project I built a machine vision system using deep learning techniques to detect pneumonia in children’s x-rays
In this project I looked at large amounts of real-estate data in Seattle to identify key predictors of home value, and then use a linear-regression algorithm to identify homes that may be undervalued in the market & plot them on a map
This project uses Large Language Models to describe and tag our internet browsing sessions, and then outputs them into a “network” notetaking software called Obsidian.
In this quick project I turn whitepaper abstracts into document embeddings to see how much they predict research impact
In this project I used machine learning systems to interpret data from the Taiwan stock exchange to predict companies which are at risk of bankruptcy. This system could also be used to flag specific financial indicators that might lead a company to bankruptcy
Read about some of my other projects:
Making an Adventure Simulator
Woodworking
Projection Microscope (Free Download!)
Generative Jewelry