To clear out the confusion surrounding large language models, we need a new framework to think about LLMs, argue researchers at the University of Texas at Austin and Massachusetts Institute of Technology (MIT).