Stuck: The Deep Inertias of Large Language Models and How to Digress

Master project

Due to ideological, infrastructural and commercial entanglements, Large Language Models (LLMs) like ChatGPT suffer from deep inertias resulting from problematic histories, which they obediently reflect and enact. Moreover, LLMs make it more difficult to engage with such issues because they act as engines of ahistoricity, muddling situational specificities with a generalised use of language, while they get lodged into an ever-expanding loop. The exhibited work invites digression—meaning “to veer away from the appointed course”—from this loop. It follows the operational procedures of LLMs, but in a different way, reorientating the goals of LLMs to predict the next word in a sentence towards a form of divination directed at a specific person. The work thus enables visitors to transform the generalized clichés of LLMs into language that helps them navigate and negotiate their realities.

Fact Box