Abstract This article analyzes how engineers in Amman, Jordan involved in making large language models (LLMs) conceptualize their labor in relation to language and technology. Specifically, it focuses on claims that LLMs are “language‐agnostic,” relying on multidimensional word embeddings rather than language. I examine how articulations of this ideology allow engineers to reconcile and ascribe value to their work in a transnational AI industry predicated on the division of labor between “universal” model building and localization. Examining the reproduction of this framework, alongside the invisibilization of linguistic labor it necessitates, reveals how representational ideologies structure both LLMs and the labor systems underpinning them.
Tariq Adely (Mon,) studied this question.