Abstract

This work introduces TinyTim, a family of large language models fine-tuned onJames Joyce's `Finnegans Wake'. Through quantitative evaluation againstbaseline models, we demonstrate that TinyTim V1 produces a statisticallydistinct generative profile characterized by high lexical diversity and lowsemantic coherence. These findings are interpreted through theories ofcreativity and complex problem-solving, arguing that such specialized modelscan function as divergent knowledge sources within more extensive creativearchitectures, powering automated discovery mechanisms in diverse settings.

 

Quick Read (beta)