Thank you for sharing this. Do you think the syntactical issues could be addressed with fine tuning or some other kind of parameter tweaking? That's frustrating hah.
I've been building hobby LED art projects with WLED (exclusively built on the ESP32 platform). It's been a blast. These little boards are so powerful and the open source community continues to amaze me.
My preferred controller platform is of the QuinLED line - comes with power distribution, voltage regulators, fat copper lines, configurable data-line resistors, and smart auxiliary hardware support all for an affordable $30-$50 per controller. (quinled.info)
<https://kno.wled.ge/> - WLED homepage and probably my favorite clever URL of all time.
This is interesting work, thank you for sharing. What hardware would you buy today for experimenting? Seems like the new gen of macbook pros are pretty powerful?
I love project ideas like this. Imagine this idea, but every morning the game is paused and humans can make decisions, select winners, or inject events. Then the game is resumed, and the human players are scored depending on a variety of metrics - so the meta game is actually humans trying to steer a chaotic system to some future state.
OK i'm kind of geeking out on this one. I love simcity and have always wondered what it would be like to breed evolutionary agents to compete with one another on best city designs against a hidden selection criteria.
It'd be kind of fun to just let this run on a raspberry pi using a local model and display the emergent world on a wall hanging display :P
Thanks for sharing.
Update: What would it take to run this locally / offline? I'm not quite sure how the cloud flare layer works. Is it just for cheap/free object storage so the cities can live somewhere?
My quick story: I built up an old 90s cyclocross bike and his website was the main reason I have this beast of a frankenbike gravel bike. I found his article "8 of 9 on 7" and it changed my life: Take a 9 gear cog, remove one, and it fits perfectly on a 7-speed cassette body.
Then I found his other article on an alternate wiring for a shimano mountain bike RD-310 7/8 speed drive train (which unlocks 9-speed ability), which thus let me use the rugged 7/8-speed derailleur for the cassette WITH shimano dura-ace indexed bar-end shifters (which use, get this, 9-speed spacing on an 8-speed index because it made their system "proprietary"). All of this works together flawlessly <3 <3 RIP sheldon brown.
I flew from central US to western asia (via moscow) and it was an interesting experience for the reasons you mentioned. I think I left early Saturday morning local time and arrived Sunday evening local time. I saw a sunrise, sunset, sunrise, sunset in 18 hours of travel time.
We all admire your absolute mastery of your own habitual reflexes and mind. For the rest of us, there is a daily battle of wits, desires, weakness, and habit.
If I could snap my fingers and break toxic habits and patterns, I would have done so decades ago :)
In my experience the LLMs work better with frameworks that have more rigid guidance. Something like Tailwind has a body of examples that work together, language to reason about the behavior needed, higher levels of abstraction (potentially), etc. This seems to be helpful.
The LLMs can certainly use raw CSS and it works well, the challenge is when you need consistent framing across many pages with mounting special cases, and the LLMs may make extrapolate small inconsistencies further. If you stick within a rigid framework, the inconsistencies should be less across a larger project (in theory, at least).
I've visited this museum and it was the highlight of my trip to the netherlands. I also wondered, for hours, about how cool it is to hook up modern hardware to these old systems. Can you imagine playing one live, similar to how an artist would play a synthesizer kit?
This technique is surprisingly powerful. Yesterday I built an experimental cellular automata classifier system based on some research papers I found and was curious about. Aside from the sheer magic of the entire build process with Cursor + GPT5-Codex, one big breakthrough was simply cloning the original repo's source code and copy/pasting the paper into a .txt file.
Now when I ask questions about design decisions, the LLM refers to the original paper and cites the decisions without googling or hallucinating.
With just these two things in my local repo, the LLM created test scripts to compare our results versus the paper and fixed bugs automatically, helped me make decisions based on the paper's findings, helped me tune parameters based on the empirical outcomes, and even discovered a critical bug in our code that was caused by our training data being random generated versus the paper's training data being a permutation over the whole solution space.
All of this work was done in one evening and I'm still blown away by it. We even ported our code to golang, parallelized it, and saw a 10x speedup in the processing. Right before heading to bed, I had the LLM spin up a novel simulator using a quirky set of tests that I invented using hypothetical sensors and data that have not yet been implemented, and it nailed it first try - using smart abstractions and not touching the original engine implementation at all. This tech is getting freaky.