Paul Graham compares how we describe LLM cognition to how 16th-century astronomers used the heliocentric model for calculations without fully believing it, suggesting we'll gradually acknowledge what LLMs actually do rather than calling it thinking.