Discover more from Mars College
The Simulated Times — a newsletter from the future powered by AI
During Mars College 2023, we created a newsletter from the future, powered by AI writing tools and the art of collaboration.
This past year at Mars College, I initiated a collaborative literary art project called The Simulated Times.
The project — which utilized AI writing tools such as GPT 3.5 — was originally intended as a platform for us to explore new approaches to creative writing, but is taking on a life of its own as part of our Mars College literary brand, Nomicon Press, which we seek to expand in the coming term (join us!)
The weekly newsletter, which claims to present news from the year 2050, allowed us to playfully speculate about our rapidly-changing technologic world and what it might be like 27 years into the future, particularly for decentralized alternative societies living in a remote desert bioregion. However, the project also allowed us to develop new opportunities to write collaboratively — with other humans, as well as together with AIs.
Each Friday during the Mars College 2023 term, I held an editorial board meeting where we would brainstorm together and pitch story ideas for the next issue. In the following days, my pals and I would generate draft articles using LLM algorithms (Large Language Models — the more technical term for AI writing tools) to help us fill in the details. As editor-in-chief, I would then copy-edit the drafts (using my human brain as well as a second round of AI tools) and lay them out for printing.
The newsletter included some recurring columns, such as a review of local cults and their rituals by Fakepope ("Papal Skeptics"), an advice column by Daphne ("Dear Agony"), and a finance column by Chatsubo.Z ("Investing in Mars"). Drmbt also contributed an amusing opinion column ("Human Voices") that parodies The Onion's "American Voices", using AI-remixed headshots to represent everyday people.
We also used The Simulated Times to practice world-building for some of our individual projects. In a pair of articles, which I penned in collaboration with Vanessa, we tried to imagine some of the origin backstory of her Little Martians sci-fi project, first running an exposé on some mysterious nearby server farm ("The Human Memorial"), revealing that the corporation has been operating a metaverse for the uploaded minds of rich tech entrepreneurs. In the next week’s issue of The Simulated Times, we followed this up by publishing a Letter-To-The-Editor, setting the record straight about the tech firm's shady history, describing data theft and alleging their ties to a pornography ring.
The project also gave us an opportunity to amplify our local memes and in-jokes at Mars College. For instance, the advertisement for "Black Bullet Tooth Paste" emerged after an amusing show-and-tell presentation by Electric Sam about his progress during the 2023 semester trying to develop a DIY bio-gasification apparatus (attached to the back of his converted shortbus, "The Magic Tool Bus"); in a memorable moment during a weekly “Thunder Talk” event at Mars, Sam had proudly presented to us a fully-carbonized charcoal turd (from his own personal waste stream) that was "so clean you could eat it", spurring my imagination for a personalized-medicine company that collects your poop samples and generates customized nanoparticles from the resulting biochar.
Another article, composed by Vishakha, lampooned the predominance of vegetarian food at Mars College by presenting a news item about a standoff among planetary colonists on Mars (“Rogue Martian Hides Rations — Colonists Must Eat Only Beans”).
Preceding our midterm showcase, Drmbt wryly contributed a survival guide for "Day Forty-Five" — a mythical day which marks the half-way point of the Mars College term, and we also reprinted the flyer for our midterm festivities that weekend, designed by Kirby.
On Tuesdays, during the Language sections of our “Simulation Sessions” coursework, we held classes led by Jmill at Mars to help teach each other best practices on how to use these cutting-edge tools. Invaluably, he introduced us to the GPT plugin he’d created for the Obsidian notetaking app, demonstrating how he uses it in his everyday writing workflow.
While some people preferred to generate their article drafts by conversing with the new chatbot interface that OpenAI had just rolled out (ChatGPT, also known as GPT 3.5), many of us adopted a workflow that still used older LLMs like GPT3, which function as “text completion” algorithms. As this is actually what an LLM does — it generates a continuation of whatever text a user submits to it — we learned how to guide the AI to work creatively, within the future worlds we were imagining. By designing a custom pre-prompt that claims to be an internal memo leaked from an editorial board meeting, the algorithms were able to write a lot of wacky but coherent article drafts, situated in the “solarpunk setting” of autonomous desert-dwelling decentralized encampments resembling Slab City.
My pre-prompts mention the date of the article (helping bias the LLM to include futuristic details), include a comment from the editor as well as the article title, and typically contain the first few lines of the draft copy which I then have the algorithm complete.
This is a more realistic approach to using LLMs for creative writing, since the models are trained from actual documents scraped from the internet. My pre-prompt method also allows me to preface the article with certain editorial directives, such as a quick note from the editor asking for more details.
I find that the chatbot interfaces of more recent LLMs tend to generate text that is a bit dry and robotic, lacking the idiosyncrasies that help make good prose enjoyable to read; this is because the hidden pre-prompts of companies like OpenAI and Microsoft are designed to restrict their products to be helpful and informative and uncontroversial, with the goal of creating AI assistants as commercial products.
Nowadays, this pre-prompt technique is considered "jailbreaking" an AI, but that phrase is retrograde and misses the point of how LLMs actually work — as completion algorithms.
I learned the power of this pre-prompt technique last year at Mars College, when we created a book of wisdom ("Mulabonding") authored by five different chatbots each with very distinct personalities.
Five different artists at Mars College each composed their own pre-prompt for their own chatbot, providing not only a short characterization of the AI entity's personality and their backstory but also a short list of example questions and human-designed answers that exemplify the style and mannerisms in which the chatbot should speak.
This list of example Q&As is called a "few shot", a technique introduced to us by JMill during a 2022 lecture about language models; including this information (which remains hidden, since it never appears in the output) helps bias the way the model will traverse the layers of the LLM's neural network.
Consequently, it enabled each of our chatbots to exhibit very distinct and amusing personalities, including some very fiery (and often inappropriate) answers from the Fire Spirit chatbot, Chatsubo, which I had programmed.
I've long valued the practice of collaborativity in artmaking, and The Simulated Times project has been the deepest I've ever explored with it.
Of course, the writings in The Simulated Times have all been edited by me, and thus they share a cohesive prose style that comes from my own poetic taste — literary elements like assonance and zany neologisms can easily be recognized by friends of mine familiar with my past writings.
But the most interesting aspect of this project is that you otherwise really can't tell which parts were written by me, compared with the parts written by my human collaborators and the parts written by AIs. In most cases I can't even tell which sentences were AI-generated, since I was handed fully-formed drafts by my co-authors!
I believe something transcendent happens when two artists collaborate — a synergy arises — and the work takes on a voice of its own; we observe this commonly in music, with duets and cover versions and remixes. I believe it doesn't matter whether I am collaborating with another human or with an AI; a new texture appears that I could not have generated all on my own. This phenomenon is especially valuable for sci-fi, since the AI can help build wild ideas that we can spin off into speculations and hypotheses.
Such collaboration between multiple artists has always been desirable to me for the way it softens the spikes that the Ego can install in our works.
While I do appreciate the humanity of an experienced author peeking thru the prose and exposing a little of their spirit to the reader, for most of my life I've been too shy to be autobiographical — maybe it is because of my career as an academic research scientist, where I've painstakingly chosen my words to be factual and to avoid bias.
A lot of times, poetry (which can also exist in prose form) speaks the most brightly when it DOESN'T make direct statements; good literature often makes a reader think about context and possibility.
I enjoyed some of these contributions so much that I actually used two of them as source material in my "sound collage" performances for our Livecoding Club, remixing them further and inadvertently spawning new Mars memes (ie., "Please collaborate with me"; "Cannabis will make you happy").
I also decided to spin off one of the story ideas into a video animation, as the article (pitched by Karo) seemed a bit too mythical for the near-future setting of The Simulated Times and would be better suited to a visual presentation with a narrative voice.
The video of Karo's legend turned out so successful that several of us want to spin off more stories from The Simulated Times into videos of AI animations; using the excellent tool platform developed by Martians (Eden.Art), we expect to be doing much more narrative video production during the 2024 semester.
At the same time, I have more issues in the pipeline from the rich amount of material we generated together at Mars College this past year. I already have mined this extra material to produce a special summer edition of The Simulated Times, given out at a salon event Vanessa held in Brooklyn at the Regency gallery, displaying the ceramics and storytelling of her Little Martians project.
I am a strong believer in the power of physical art objects, over digital ones, and I believe having a physical newsletter to hand out at events or to tack up for display in public spaces helps to create an atmosphere of being present. Nevertheless, (for now) I am also making the work available online, and you can view facsimiles of all the back-issues here.
This year, for Mars College 2024, I intend to continue operating The Simulated Times. I am especially excited for more columns and advertisements, and am eager to collaborate with graphic designers who can help improve the layout and design elements of the publication. SUBMIT AN APPLICATION TO ATTEND MARS COLLEGE 2024!
Stay tuned!! Keep an eye on the Simulated Times tumblr page!