Meaning without reference in large language models

Lots of excellent references showing LLMs have rich internal models such as color spaces. Good argument meaning doesn't require "grounding" in perceptions (with citations).