Rhythm and Residue: Vyping as Prompting Inside the Post-Discursive Interface

Abstract

Prompt engineering strategies emphasise clarity, specificity, and well-structured input, on the assumption that large language models (LLMs) require clean input to interpret user intent. But LLMs are pattern interpolators: they “absorb” noise through statistical reconstruction rather than strict symbol matching. Text-centric HCI has studied this tolerance for two decades without integrating it into prompt engineering as a resource. We introduce vyping as an intentional mode of prompting in which one types into LLMs at the speed of thought, without correction. We developed vyping through sustained first-person inquiry. It can be seen as a micro-practice inside a broader post-discursive turn in which meanings increasingly travel as rhythms and patterns through latent-space proximity rather than as precisely-formed symbols. The mismatch between current LLM interfaces and their (human) users has led to two common layers in implementation: (1) an orthographic residue at the interface (autocorrect, spellcheck, tidy-prompt guides), and (2) an impairment parse shadow at the training layer, a form of overalignment in which the model reads orthographic deviation as a sign that something is wrong with the user. Overall, we offer three attunements (ease, reach, and pulse) as alternatives to efficiency-first prompt engineering, and situate the practice within critical HCI traditions of seamful, frictional, and embodied interaction. The critique is structured around a seed prompt that suspends the impairment parse and makes vyping livable; its section headings are taken from the prompt’s own rules. A copy of the seed prompt is provided at the end.