Writing needs a cosmology

A passage from Rebecca Solnit’s Memoir:

I encountered my fellow San Franciscan Diane de Prima’s work only later, which declared: you cannot write a single line without a cosmology.
Writing is often treated as a project of making things, one piece at a time, but you write from who you are and what you care about and what true voice is yours and from leaving all the false voices and wrong notes behind. So underneath the task of writing a particular piece is the general one of making a self who can make the work you are meant to. (Solnit 2020)

It is this creation of a self that text generation via LLMs would miss out on. There’s several layers to this.

For one and most obviously, on the level of pure content: writing via an LLM means you, the writer, are not yourself properly transformed by what you are writing about, both in terms of working through the material, but also in a very embodied way. There’s already studies out there that link handwritten text to higher comprehension of the material covered, as opposed to typed out notes. From what I gather this seems to be a byproduct of the increased effort of physical handwriting. It is a slower process involving more muscles than simple typing.

Mind you, I don’t think that typed writing is therefore inferior in some way. It’s a question of being aware of when to use what. So for increasing comprehension (taking notes), handwriting might in some cases be more useful. In other situations typing can be more useful, e.g. if the goal would be to produce a lot of text in short time, for example as a method to wear out the inner censor, or to make breakthroughs through speed. There’s a note somewhere by Nietzsche how he feels his thinking took on a new form and force after he started using a typewriter.

Now of course LLM-based text-generation misses out on this very physical process (be it writing by hand or typing) altogether, or rather it reduces it to the simple act of typing a prompt.

Another thing text generation via LLMs misses out on is the creation of a habit, the habit of sitting down and writing at regular intervals. This entails the creation of a time to write. You’re not carving out the time necessary for you to be able to write.

You’re also not working through deadlocks, through feelings of being stuck. There’s no overcoming of obstacles and being transformed as a person in the course of this. Naturally text-generation via LLMs always comes with it’s own frustrations (fighting against em-dashes and poor style and the machine not emitting what you’d like it to), but this is just a frustration with the inherent limits of this technology. This is fundamentally different from the frustrations a writer would face and work through. Nobody was every transformed by kicking his dishwasher.

We could list more ways in which LLM text generation doesn’t transform the prompter, but I think you get the general idea.

Sources

• Solnit, R. (2020). Recollections of My Non-Existence