The arrival of Large Language Models (LLMs) offers an interesting junction in the evolution of writing. While previous technological innovations (printing, internet) mainly increased the availability of text, LLMs also facilitate the generation of text. This is a fundamental difference.
In this blog post I speculate how writing when considered as a way of thinking1 will interact with LLMs.
Writing as a way of thinking
An often overlooked aspect of writing is that it is a way of thinking. It helps us refine, structure, and expand our thoughts. Writing about a topic is thinking about a topic.
Thinking through writing though is an inherently demanding process that requires focus, persistence, and stretches of uninterrupted time. When writing, we jungle many things mentally - the phrasing, the meaning, the flow, of our current sentence, and how all these tie together with the paragraph, the section, and our whole text. This process is rather challenging and requires in-depth knowledge of the text already written.
LLMs offer an easy way out of this demanding process2. This may sound useful, but can greatly hinder our thinking and idea generation processes. Initially it does so by introducing large (or small) swathes of text not generated by us. This weakens our grasp of the current text and hinders our ability balance flow and meaning throughout the whole writing. Furthermore, utilizing often this easy way out can sap our persistence, an essential aspect of good writing.
On the other hand, using LLMs properly - in a way that avoids the two aforementioned pitfalls - can greatly enhance our writing.
I see two high level recommendations that could help in that direction. First, create a setup that allows for long stretches of uninterrupted writing and resist the temptation to use LLMs no matter how blocked you feel. This way you both create ample space for your thoughts to expand and train your persistence muscles. Finally, use LLMs more in the initial phases of writing - ideation, brainstorming - and less as you progress. This way you allow for the creative input of LLMs but then give enough space for your own thinking to emerge while you also a good mental representation of your text.
-
Writing can be viewed through many lenses - as means of communication, as an art form, as a way of thinking, and as a symbolic representation of our thoughts. Each perspective would reveal a different interplay with LLMs, a topic I’ll likely explore in a future blog post. ↩︎
-
By helping for example improve the phrasing of an underdeveloped sentence, providing ideas for a paragraph, or ideas for the whole structure of our text. ↩︎