From Tokens to Thoughts: How LLMs and Humans Trade Compression for Meaning