Model Weights from Ulises Carrión’s The New Art of Making Books

I was inspired to create a new artwork based on and in dialogue with Ulises Carrión’s 1975 manifesto, The New Art of Making Books.

The context for this work is Dimitry Tetin’s Publication course, which I’m (somehow) taking in support of an MBA at Texas State University. Dimitry introduced the Carrión text as a foundation for contemporary artist publishing in general and the Publication course in particular. He also prompted the class with an assignment to consider encyclopedias and related forms such as catalogues, indexes, and sets.

I internalized the manifesto as calling for the fusion of content and its form or expression. These statements, in particular, resonated with meaning:

A book “exists as an autonomous and self-sufficient form, including perhaps a text that emphasizes that form, a text that is an organic part of that form.

“To make a book is to actualize its ideal space-time sequence by means of the creation of a parallel sequence of signs, be it linguistic or other.”

Reading the manifesto triggered a memory of the YouTube series Neural Network in a Spreadsheet. The videos make several concepts fundamental to LLM technology accessible using humble Google Sheets. What better way to demystify AI than to lay its operations out visually? It struck me that Concepts Illustrated, the videos’ creators, had achieved something like Carrión’s fusion of content and form.

I was inspired to use The New Art of Making Books as the basis for neural network model weights. The resulting weights are a set that expresses the text using a “parallel sequence of signs.” The weights naturally take the form of a spreadsheet, which I have published via Google Sheets, ala Concepts Illustrated. The figures are unintelligible to humans. And I do not intend for their use within a working model (though someone is welcome to try). Instead, I generated these weights to symbolize how we may grapple with and express Carrión’s ideas using new language.

Access the spreadsheet via this link.

I used Anthropic’s Claude generative AI model inside the Cursor integrated development environment to help me develop a program to create model weights from a supplied text. I’ve published this program, along with the prompts I used, to a GitHub repository. I intend for this ephemera of the weights’ production to provide context that may support understanding of the project.

This blog post is a network node that connects the Google Sheet and the ephemera.

Posted February 2025

Made under the ☀ in Austin, Texas.
WordPress hosting by WP Engine, thanks y’all!

© 2022 Spellerberg Associates LLC