Large language models may look impossibly complex, but many of their hidden behaviors can be studied using familiar machine learning...