Distant Poesis: Stochastic Prompting as Literary Creation

Abstract This paper introduces the concept of “Distant Poesis” as a theoretical framework for understanding a novel form of literary creation that operates at the intersection of programming, randomness, and language models. Drawing on Franco Moretti’s notion of “distant reading,” I propose that Distant Poesis involves writing programs that generate stochastic values (such as random word pairs or stylistic variables) which are then inserted into prompt templates for language models. This creates a new form of literary creation – one that operates at a distance from the text itself. By examining two specific implementations – lexical pairing and stylistic variable sampling – this study demonstrates how the meta-creative act of designing both sources of stochasticity and prompt templates constitutes a form of poesis in itself. Through analysis of poems generated via these methods, I explore how this distant approach generates both coherent individual works and diverse textual ecosystems, challenging traditional notions of authorship and creativity while expanding the conceptual territory of computational poetics.