The Creative Process #29 - Tiny Gifts

My usual process for writing a newsletter is to accumulate a few notes first. I’ll look back on my social media posts, cringing; dig through the last few things in my exocortex notes’ Inbox; and try to find a through-line. I often look at my previous post, half to see where I got up to, and half to remember what number issue we’re up to and whether I do “The Creative Process (number) (dash) (title)” or “The Creative Process (dash) (number) (title)”. It’s a very refined process, as you can see.

Last month was so long ago. The first half of November was a work conference, which was invigorating, intimidating and a real slog to get through with sleep deprivation. I cannot be sure I didn’t momentarily pass out in a talk or two, despite my best efforts to pinch myself and mainline caffeine.

I remember one morning being particularly grim when the US Election results came out. It had all the energy of someone saying, “You know my crazy ex that burned down my apartment? Well good news, we hooked up again and she’s moved back in.” You’d ask why, but you know why, and you wish you didn’t know why.

October was particularly intense for interactive fiction with IF Comp and all. I opted out of Ectocomp, the next event on the IF calendar, and somewhere in November I could feel all the words drain out of me. My other interests module loaded up and all I wanted to do was write optimized Rust code, and daydream of writing compilers and AI agent simulations.

Near the end of the three years that was November, I bounced into making some scaffolding for the third project in my future lineup, Puzzle University. Normally I have been good focussing on my top priority project, but my world and The World was not really in a state for me to grind out some more of my cyberpunk Anne Shirley of Green Gables Cables. Puzzle University is still a long way away from completion — puzzle creation is very hard, even if you don’t wrap it in lots of story like I plan to.

Seeking Blue Skies

As with millions of people this month, I secured a profile on Bluesky, the I-can’t-believe-it’s-not-Twitter. It’s old Twitter. Classic Twitter, now with reduced fascism. Like a lot of people it captured old Twitter so well that it engendered a feeling of nostalgia. At this point, though, I’m happy to lay out a deck chair and enjoy it, but I’m not building anything on those sandy foundations. I’ve already been through Geocities, AOL, IRC, LiveJournal, Facebook, Google Plus, Twitter… The Bluesky will turn grey eventually. There’s no countervailing pressure not to.

Speaking of Twitter… groan, I mean, X… I disabled my account there. I really only kept it open out of nostalgia and that weird feeling of ownership that “I had that decent handle for a decade”. While my feed wasn’t as gross as I’ve heard other people’s were, the only reason I kept it open was to periodically check in to see what outrageous insanity Jonathan Blow and Casey Muratori were vomiting, and that was a bad mental habit to have. So I disabled my account. If Elon Musk wants me back, he can pay me.

Sure, immediately setting up an account on Bluesky seems like giving up crack to pivot to healthy heroin, but so far it’s been okay. The Starter Packs are genuinely a good idea, both to seed your feed with interesting people and block whole curated classes of horrible ones.

When you sign up they ask for your interests, which rarely exactly capture mine. They had “Writers” to which I thought, “Oh, they mean writing. Yes I like writing.” No, they meant writers. Writers talking about writers talking about writing. “It took me thirty years to write my novel, and so can you!”

And like, yep, I get it. I love libraries and local indie bookstores too. But that single check was the biggest mistake I’ve made so far on BlueSky.

I’m staying on Mastodon because I like the people there. Those people also exist on BlueSky, but it feels a bit more direct and punk on Mastodon. I spent one night writing a Mastodon-to-BlueSky post mirrorer, which worked, but I’m always too scared to automate my actual social media with my own code. Though the ease of use for both platforms reignited my interest in writing social media bots. Mark that down as Project #23 on the list.

The Long Context

My interests in AI continue, but are annealing somewhat. I’ve backed off on my Midjourney subscription as I just don’t create that many images a month. Definitely not since deciding to use human art for this newsletter. I still use ChatGPT and Claude a lot. I can feel but not precisely articulate the edges to capability that I notice with LLMs. I keep threatening to write a post on “Why leaning on LLMs to create infinite interactive fiction won’t work”. I feel like I’m very capable and cultured when it comes to AI.

Which is why it was a little shock this month to hear from good friend of the newsletter, Thom. I thoroughly enjoyed discussing tech (and really any topic) with him when we used to hang out. He moved to Melbourne where all the cool kids are, but we still catch up periodically. Late in November he hit me with a post: “You Exist In The Long Context”.

I had probably mentally bookmarked the post as it flew past in my usual information sources, but if Thom shares a thing, you damn well read it.

It’s a fascinating demonstration and discussion about the use of long-context LLMs. It has an interactive fiction game that you can play where you solve a crime. An LLM with the context of an entire fiction novel runs the game. It uses the novel to help fill out details of style and events. It’s almost the ideal IF game of what so many forum dwellers claim they’ll do with AI, except this dude did it. Sorta.

My screed-to-be on AI and IF took a ding from this post, but might still remain. As an experience, the game is interesting — write anything and the game responds sensibly. But I’ve played AI Dungeon and countless follow-ups. Stylistically it’s cool that it imbues the game with the author’s particular world and aesthetic. But as a game it missed the decades of game design that the community invented, learned and built foundations on. The game doesn’t do hints well. It’s not fair not unfair, it’s in an icky middle. You could maybe fix this with more direction from the prompt, but I don’t know. There’s a subtle web of considerations that these models don’t — and I believe can’t — take in.

I’ve yet to experiment with long-context models myself. The advertising usually is along the lines of “imagine if it could keep the entirety of the Game of Thrones novels in its context window?” While it gives the LLM a huge amount of content and context to reference, I think that’s all it gives the LLM.

In discussions, Thom and I agreed that this all seemed interesting, but the kicker just doesn’t seem to be there. When I turned my personal notes into a RAG and asked questions, the results were… okay. It doesn’t know how to synthesise my mental world from my notes. If I had amnesia, neither could I, to be fair. But what is the killer app for these things? As far as I can see in society, the main use for these new AI is writing roughly templated text for when people can’t be bothered to.

Maybe my opinion will change again next month.

Random acts of creative kindness

So what do we do when we find the world has gone insane? When innumerable potential troubles loom ahead, threatening to either sweep out the foundations or just evaporate like a fart?

I thought a lot about this in November, as were many others. I saw this quote:

“Do not accept brutality and cruelty as normal even if it is sanctioned. Protect the vulnerable and encourage the afraid. If you are brave, stand up for others. If you cannot be brave — and it is often hard to be brave — be kind.” — Sarah Kendzior

I’m in a relatively safe position in the world, along many axes. It’s important to see that as a form of power and project good things into the world.

Coincidentally I was listening to the podcast Topic Lords, the only place on the Internet where you can hear topics discussed. The episode entitled “Voronoi Cookies” had as guests a guy called Ben Wilson who I was unaware of, and Andrew Plotkin, aka zarf, IF legend and general good guy programmer and creative. While there was a lot of interesting stuff in the episode, I was really caught by an idea from Ben: he was learning to make crossword puzzles and would often print them out and leave them in public places for people to hopefully play and enjoy. Or people who would write a crossword or puzzle to bring along to a party, themed on that party. Jim Stormdancer, the host, mentioned creating Pico-8 games instead of Christmas cards for his friends and family.

I love this idea. In a previous newsletter I mentioned my revelation that I no longer wanted to chase the idea of being a big name in whatever creative field, I just wanted to create cool things for cool people. Namely, my friends and family. Building tighter communities and friendships is important, for the storms ahead and in general.

While I really don’t think I have the time to create a lot of individual Pico-8 games, or crosswords or anything like that, I want to try. I’ve kicked off my annual now-tradition of making a Blender 3d scene for Christmas as a virtual Christmas card of sorts.

I did have the idea of running a Secret Santa on the IF Forums for people to instead share tiny bits of creativity, like a vignette or a sketch or a recipe, instead of physical presents. Alas, I realised that the organization involved was beyond me at the moment (I’m organizing a work conference and it’s all my time already). Maybe next year.

I saw Carl Muckenhoupt on Mastodon declare that he was not going to do Advent of Code this year. Instead he was going to do Advent of his main project, and wrote out 25 modest daily goals for it. I think this is a fantastic idea.

I might try to use the time I’d devote to Advent of Code, and create cool things for cool people. Hopefully I can share some at the end of the month.

Have a productive, creative, kind December.