Latent Space as Medium

The latent space of LLMs is our defining medium. It is our environment; our language turned back upon itself and abstracted into mathematical relations. While we navigate this probabilistic terrain similarly to the printed page, there is a crucial difference: the medium is no longer a physical channel carrying information but a statistical structure of the information itself.

Having spent a lot time interacting with LLMs, I find myself increasingly framing my thoughts to be machine-legible. Generalizing this trend I see in myself, I can imagine our writing evolving to accommodate the LLMs probabilistic prediction and the concepts we use (metaphors we live by) to cluster according to the geometries in the model that are undesigned and without intention. Like McLuhan’s (famous) proverbial fish unaware of water, we struggle to perceive how thoroughly this new medium envelopes us, changes how we communicate, and updates what communication means. The latent space is a parallel linguistic reality governed by proximities and probabilities.