Sing O Muse: a restricted language model for GenAI in classics teaching

Abstract This paper explores the implications for classics teaching of our experiments at Sing O Muse, an OpenAI workspace created to emulate the voices of ancient writers. To address difficulties noted by earlier researchers in digital-classics pedagogy, we decoupled from the prevailing use of Large Language Models (LLMs) in favour of a Restricted Language Model (RLM). After explaining how we used this model to reconstruct scenes from the lost epic, Cypria , we explore other uses of RLMs for learning about the ancient world, such as using Aristotle’s Rhetoric to help first-generation college applicants outline admissions essays. We caution, however, that creating RLMs of reliable fidelity requires subject-knowledge more characteristic of graduate than of undergraduate or secondary students. We find, further, that optimal RLM-use involves a feel for the seam between automation and artistry, presuming a more-than-casual erudition. We conclude that RLMs in and beyond the classroom are most promising as tools not to reduce but rather to enhance labour. In this respect, for users who invest the energies, one transformational use of RLMs may be reenactment of the performance-within-a-tradition characteristic of ancient literature itself.