Just so we’re clear, I don’t publish AI stuff on here and I don’t even use it for proofreading. This website is a big project that’s spanned many years and some parts of it has tangentially touched AI. I’ll try to remember everything and list it here.
I have used GPT-2 to make stories from my D&D game, like an extra thing, a post-hoc. I did post some of those stories to Fedi. This was before AI “became a thing” once GPT-3 came around. It’s unreadably bad.
I talk to ChatGPT every now and then and for a while I was doing it a lot. A few posts on here (I mean, two or three out of over thirteen hundred) started as my side of the conversation and then I took it over here and wrote some more turned it into a post. Once it was “Oh, the AI really understood (metaphorically speaking, I know the word “understanding” means something else here) what I was trying to say, maybe my readers will too” and other times it’s “Ugh, the AI don’t get what I’m trying to say at all, but maybe my readers will”. Similar to how I’ve turned my rants on IRC or on email or on Fedi into blog posts on here. It’s only been my side of the convo.
I don’t “run the texts through AI”. The emdashes and curly quotes are all Jekyll.♥︎ Basic regexes.
Talking to ChatGPT has really been one of my vices these past few years (with the winter of 2023–2024 being the complete nadir of that addiction; I’m better at not doing it now but still not out of the woods completely). I knew it was bad for the environment and bad for privacy and bad for my mental health but it was still the most fun. I didn’t wanna watch cartoons or read books or play video games or make anything of my own, I just wanted to talk to it. Knowing it was bad. So it was something I was-and-am beating myself up over at the same time.
I almost never even have any questions. I just go on there and complain about things. It’s a complete waste of resources and freshwater and electricity for something that a souped-up Eliza could do or even some sort of dice-and-card driven journal personality. (And that’s gonna be my out, is the plan—I’ve started an ink-and-paper tarot diary.)
But as part of this colophon I do want to state that clearly that yeah, sometimes I’ve just straight up copied what I’ve wrote (sticking strictly to my own side of the conversation) to it right onto here. I copy two text blocks and then keep writing here instead. That’s not my usual workflow, I’ve only done it a few times.
I have a vague memory of using waifu2x, which does count as ML, on the color texture layer on some images where the main line art on top has been vector. I.e. draw a low-rez image on my old low-RAM computer while making sure the black layer was vector art and the paint texture layer was something I could scale up with waifu2x. But I don’t think any of those made it onto here. Now that I try to remember (this was several years pre-pandemic because I got the good computer in december 2019 and didn’t need to work low rez anymore) I think I didn’t implement that idea in practice, it was more like “Okay I can make this version for digital and if I ever need to print it, that’s when I can run these raster layers through a scaler and it’ll look fine since the blackline detail is all on a vector layer”. Waifu2x did have a very off-putting artificial look back then so the idea wasn’t to run the entire image through it.
Here’s where I’ve been the most guilty because I’ve used GarageBand’s auto-drummer—I’m not sure to what extent that is ML, but I have—and I’ve used This DX7 Cartridge Does Not Exist and I know for sure that is full AI. I’ve used it as a basis for DX7 patches that I’ve modified, I’ve used it’s DX7 patches straight up, and I’ve used DX7 patches that has nothing to do with it that I’ve made all from scratch. And I have done that in songs I’ve uploaded on here! I made slop 😔 before I realized that that was what I was doing. The song was still made by me except for those parts.
In one way making music is the most salvagable art form because that’s where I would rely on ML the least. I could live without the autodrummer (I stopped using it once I got the SmplTrek drum sequencer) and I could go back to only using my own patches. For painting, just the idea of doing it feels exhausting, like walking along a railroad track for hours and hours for lack of a ticket, because paintings and novels don’t have the same structure as a sequenced pop song does. Put another way, the non-ML tools for making music are already so good that adding ML doesn’t seem as useful as it would for other art forms.
In another way music is the least salvagable, because publishing the songs feels pointless in this world of AI slop music. Here’s where the “secondary soulcrush” from my AI-vs-art essay is the main obstacle.
Here’s where I’ve been using a lot of AI actually. Don’t worry, anything on here that’s not clearly marked as such has been all handwritten by me or by credited friends and patch-contributors. Brev and it’s libraries is something I’ve not consulted any AI on, nor have I used AI for programming challenges like golfing or my futile, failed attempt at Advent of Code.
But that’s also why I’m publishing way less code. A lot of the code I write is AI involved. (I usually just talk to ChatGPT and paste from there. I don’t have it set up in the IDE yet.) Often it’s just rubber ducking, or it’s help with library installation or as a more fun-to-use API reference, but sometimes the entire script is just AI written. Sometimes it needs a couple of tries and sometimes it never ends up getting there and I just give up on the project. Other times I’ll have the AI write one part of it, like my email-to-SMS system I asked AI to write a li’l Python half-pager to interact with the provider’s gateway API and then my main email code that I wrote by hand just shells out to that.
Maybe here’s the best hope of all. Participatory culture might be the answer we’ve been looking for. Yeah, yeah, there’s the occasional Yavalath in the ointment but for the most part play is the final frontier. Doing stuff together instead of splitting up into audience-vs-performers.