That’s a good point. What humans see as cognition in LLMs is often just good associative memory, or (generously) intuition. A model that can echo “I am aware” is not necessarily, but in larger LLMs more of the functional implications of awareness begin to apply.