Artificial intelligence, particularly large language models (LLMs), has often been described in simplistic terms such as “just predicting words,” “a fancy autocomplete,” “a tool, nothing more.” But ...