TinyTim: A Family of Language Models for Divergent Generation

Abstract: This work introduces TinyTim, a family of LLMs fine-tuned on James Joyce's `Finnegans Wake'. Through quantitative evaluation against baseline models, we demonstrate that TinyTim V1 produces a statistically distinct generative profile characterized by high lexical diversity and low semantic coherence. These findings are interpreted through theories of creativity and complex problem-solving, arguing that such specialized models can function as divergent knowledge sources within more extensive creative architectures, powering automated discovery mechanisms in diverse settings.