A graphic demonstrating the relative sizes of gpt-4 and the human neocortex in terms of param count

Created with the help of Claude 3.5 Sonnet

How do "test-time" compute advancements change how we should think about this?