Cabbage Sweeter than Cake? Analysing the Potential of Large Language Models for Learning Conceptual Spaces