“ We rely on what’ s called a VLM, the vision language model”