“Mandeep Singh predicts that Large Language Models will eventually be distilled to a point where they can run on-device rather than in data centers.”