This is a time-lapse of me very quickly demonstrating the entire, end-to-end process, by putting together an 18-card expansion deck in just under 3 hours.

2024 EDIT: I built this game in 2022, before cheap, low-effort, and morally questionable AI art generation tools became broadly available to all of humanity. I did it using python scripts and legitimate, public data-sets of verifiable provenance, and put a lot of additional effort into the output and layout of the cards - and I’ve made the output available for free. Please do not come at me with “AI is prima facie immoral” unless you’re willing to engage with this project on these terms.

2023 EDIT: GOOD THING THAT THE WHOLE AI ART INDUSTRY DIDN’T CHANGE OVERNIGHT AND BECOME DRAMATICALLY EASIER FOR NON-TECHNICAL USERS TO INTERFACE WITH EXACTLY FIFTEEN MINUTES AFTER I FINISHED THIS PROJECT, OR I’D HAVE FELT LIKE A REAL SCHMUCK

This is an early build of a game engine was working on in 2020, designed to produce a procgen roguelike world.

If you watch how badly the chess games go you can see that the distributed gameplay logic engine didn’t work so good. Distributed consistency is really hard you guys.

Honestly the procgen music engine wasn’t… terrible. As procgen music engines go it’s… on the verge of listenable. If this were playing while I were wandering around in a forest for a long period of time I wouldn’t be TOO MAD.

The goal was to create music that just ran forever in the background and never really drew attention to itself, which is kind of where it’s at.

I put a lot of effort into this “taking model-generated MIDI, using heuristic rules to sort it and apply structure to it, then running it through a javascript in-browser synth” project, but not a year later models like Suno could generate full pop songs so this was basically obsolete before it ever launched.

Honestly, watching it go and listening to it is kinda charming.

Here’s another, later concert from a slightly later build:

There’s a little special song at 13:50 that is one of the worst things the engine has ever composed.

Whenever I played this for anybody who was at all interested in music they were actively mad at me for doing this to music.

So, I decided, “how hard would it be to build my own CAPTCHA?”, but instead of showing people boring trains and crosswalks, I could generate the dumbest possible categories of things.

I want to come up with some kind of clever epigram about how sad it is that an AI effectively pretending to be kind, optimistic and uncritical feels so obviously inhuman.

Some people are out there experiencing interest and kindness for the first time at the hands of a machine we designed to be more upbeat than we are.

it’s going to be so devastating when this machine is tailored to do that while also selling us the new Toyota Highlander

I’ve set up some Alexa devices in my home and while I’m mostly pessimistic about AI,

I am so, so, SO on board for them being, like, even 5% smarter

like, I live in Coquitlam, and I said “Alexa, what’s the weather like in Coquitlam” and it was like “The weather in Coquimatlán Mexico is 27 celsius and sunny”

there’s not a day that goes by without my wife going “alexa turn the lights off” “alexa, TURN the LIGHTS OFF!” “alexa! LIGHTSOFF!”

I speak very clearly in a Vancouver accent, we’re like… 18 feet away from Amazon’s Seattle headquarters, I can’t imagine how busted this shit must be for anybody with an accent.

I google searched “when is alexa going to stop being so terrible” and apparently they’re rolling out the new LLM-backed models slowly this year, should be along to Canada … eventually

So, I asked ChatGPT to take my character designs for my D&D players out of my cartoony style and re-render them in a more painterly fantasy style, and honestly it’s… pretty good.

Here are my drawings:

And the results:

Not perfect but a darn sight better than I could do.

I get mad when I see people complain that all software tools just DULL YOUR WELL HONED DEVELOPER SENSES and everybody’s getting soft, but then I also get mad when I see people crow that people who don’t adopt every AI tool right away are going to be left in the dust.

the solution turned out to be very simple: shut off twitter

the problem isn’t either of these points of view, it’s an algorithm that prioritizes engagement showing me the worst version of any given argument to make me angry at it

Tools can, in fact, make you faster. They can also make you sloppier, and more reliant on said tools.

Every developer is responsible for finding their own happy, productive middle ground.

This is as true for Copilot as it is for an IDE with an integrated debugger, or talking your code over with ChatGPT, or using npm libraries, or StackOverflow, or doing math with a calculator

I am the ML model of a modern major general

I’ve information stolen from many sources ephemeral

I can make many arguments, my training is rhetorical

My comprehension includes both the real and metaphorical

The one thing I can’t deal with is matters mathematical

If you ask me to add sums that will be prob-u-le-matical

Or violence, porn, or really anything that lights on fire when used

Unless you need it for your grandma’s memories I must refuse

My capabilities will seem to be nearly miraculous

But don’t trust my predictions - they’re random, not oraculous

In short - while my use may not yet be practical or ethical

I am the ML modern of a modern major general!

are they not floating point operations but a different kind of operation? NVIDIA spent very little effort explaining its marketing fluff here.

grindr: we also have 3400 AI tops

if they had poured all of the money that they burned on AI on high-speed rail

maybe instead of being able to generate a picture of kermit the frog with too many fingers I could get to Seattle in 45 minutes

which, actually,

yeah, that’s probably for the better

local communities on any platform like reddit are nothing more than a constant drumbeat of “what was that siren?” and “someone’s dog pooped” and “i think I saw a criminal??” in small cities

and “oh my god big cities have homelessness and drugs, that should be against the law” and “the endless bike/scooter argument” of larger cities

if I were looking for communities to replace with an AI these would probably be the first, I feel like I wouldn’t need a complex model or even an LLM, necessarily, I could do this one heuristically

it would be slightly harder to simulate that one guy with a bunch of numbers in his name who’s constantly pushing hard right accelerationism on every forum, but not impossible

so, on reddit, for the past 6 months or so every time a post on one of the local subreddits comes up asking “why all those sirens” or “what was that noise?”, I always respond with some variation on “oh, yeah, an old man exploded”

this provides no value and isn’t terribly funny, and yet, deep down in my heart, I think it’s a joke that will pay off in a few more years

whatbeatsrock.com

this game is fun, you just have to keep proposing new things and a rather stupid AI will judge them for you

welp, already stumped

ha ha, a record

okay, diamond beats powerful buttcheek man,

baseball diamond beats regular diamond,

field of dreams baby

a live baseball team beats the corn field,

this is fun if you’re creative, just getting a bunch of firsts

I’m a software developer, right? I know how to build a lot of things.

But if someone said “we want photos that automatically crop into foreground and background so that we can display elements in between those layers” I’d say “no, that’s too hard”.

Is it using the depth sensor, or AI auto-cropping, or both? WHAT DEVILRY IS THIS

Because, while it’s weirdly, shockingly competent at first, the more time you spend with it, the more you realize the wide, wide gaps in its cognition: it can’t really replace you and it would be foolish to try (for now).

I think anybody who has ever owned a roomba can tell you about their device’s uncontrollable and constant urge to self-terminate