Instructional Intimacies: Co-writing with a Large-Language Model

October 21, 2024 Summer 2024

For ten weeks this summer, Instructional Intimacies: Rules and Parameters for Actions That Would Constitute a Piece convened each Sunday to orbit the question: “How has the internet informed your language?” As we gathered online each week to exchange words and ideas, we stayed with the materiality of language, looking at how it has been formed by (and informed) different kinds of artistic mediation, from capital-C conceptual art in the 1960s to large language models (LLMs) in the present.

Could a through-line be drawn between Yoko Ono’s “instructions” like the pieces collected in Grapefruit, and the instructional language used to prompt generative AI? Is technological change the only force that pressures language to shape itself a certain way? As we attempted to cross-pollinate between Conceptual art and AI, we foregrounded reflective questions such as: How has your origin, heritage, lineage, and/or ethnicity shaped your language? How has it shaped your imagination? What do you need from it and what do you not need? Keeping in mind that social cybernetics forge and form our ways of using language just as profoundly as developments in what is often narrowly construed as “technology.”

Exploring these considerations by doing, we made prompts, prospectuses, and physical meanderings. As the mid-point assignment for the class, everyone was invited to complete an “art crawl” — a sprawling field trip through their city to look at artworks away from keyboard (AFK) and apply them to the questions we were chewing on in class. Crawls included visits to museums, galleries, experimental music venues, and a hair salon, spanning cities from New York to Quito, Nairobi to Berlin.

In the spirit of co-creating with AI by way of the prompt engineering skills that Monalisa shared with us in class, I put some of the skills I learned in Instructional Intimacies to work and attempted to get GPT-4 to formulate the art crawl assignment in the form of a Yoko Ono “instruction.” (Here’s the full prompt I eventually landed on, if you’re curious. Linked externally, because it’s long, as we learned successful prompts tend to be. Prompt engineering is hard work!)

After “scraping” and cleaning up the text from a PDF of Grapefruit, writing an ask in the form of a multi-role narrative that would “SHOW/TELL/TRICK” the model, and refining, refining, refining, here is what came out:

ART CRAWL PIECE

Discover two galleries, city's pulse.

Connect with others, shared journey.

Three hours, quiet immersion.

Hydrate, nourish, maintain energy.

Engage with art, acquire knowledge,

Return home, artistic development.

2024 Summer

A sense of deep satisfaction — giddiness, even — came over me when I finally got an output that resembled Ono’s “instructions.” At the same time, throughout the process of tinkering with the prompt, I repeatedly thought: this is ridiculous; it’d be easier and more fun to emulate Ono’s work without the aid of AI, challenging myself to enact this kind of formal transformation of language with my own creative faculties alone. GPT-4 felt less like a labor-saving technology than a new system of power relations deputizing me into submission. And yet, I was seduced by the dance of co-writing with it anyway.

Reflecting on all this, I’d like to end with one lesson that resonated across the ten weeks of Instructional Intimacies. Language, itself a technology, is never not channeled through webs of power. By attending to the workings of these webs, we can learn about the particularity of our positions, the limits of our knowledge, and the ways things could be said and structured otherwise. From there, we open vast space for transformative creative expression.