Tag
This paper investigates whether language models encode a structured, human-interpretable consciousness spectrum in their embedding spaces, showing that sentences form a navigable manifold from lower to higher states, with implications for model guidance and alignment.