Conloquia absentium. Le intelligenze generative tra prassi didattica e filologia del possibile

The article reports a systematic experimentation in the field of artificial intelligence applied to Classical studies through the creation of chatbots that impersonate ancient authors. The author has developed six customised GPTs (DemodoCHAT, ComiCHAT, CiceCHAT, CHATullus, VirGPT, SeneCHAT), each fine-tuned on the complete corpora of Homer, Plautus, Cicero, Catullus, Virgil and Seneca. The methodology combines comparative tests with author-specific prompts — compatible with each writer’s cultural ecosystem — and deliberately anachronistic prompts, in order to assess stylistic and content adaptation capabilities. The analysis reveals differentiated performance: lexical and syntactic reproduction is excellent, while metrical handling remains problematic. A distinction emerges between the ‘genetics’ of an LLM (its pre-existing training data) and its ‘education’ (the bespoke fine-tuning), which explains the different performances across different literary genres. The study highlights innovative prospects for the teaching of Latin and Greek (dialogic learning, exercises in style) and for philological research, such as gap-filling, authorship attribution and conjectural reconstructions. The chatbots also display metacognitive awareness, acknowledging their own limits and opening unprecedented scenarios for what the author calls a ‘philology of the possible’. Relevant issues concern the ethical alignment of contemporary models when applied to historical figures with distant value systems, a point that calls for careful calibration in rigorous scholarly applications.