AI

according to media 

there are 86 billion neurons in a human brain 

and a server computer CPU has 192 cores 

and a supercomputer GPU has 16'896 cores 

and both each have 80 billion transistors 

a human brain has physical biological neurons 

and a computer chip has transistors 

which can use software to simulate virtual neurons 

and an AI is software that has virtual neurons 

transistors can be organized so that the chip becomes a CPU 

or transistors can be organized so that the chip becomes a GPU 

so the question is 

how many virtual neurons can a 192 core CPU simulate 

and how many virtual neurons can a 16'896 core GPU simulate 

and how many virtual neurons can a 80 billion transistor chip simulate 

if a neural network could be designed 

so that the virtual neurons use less processing power 

from the computer chips 

possibly by designing the virtual neurons to be smaller and simpler 

then more virtual neurons could be simulated 

on the same amount of computer chips 

thereby making the AI smarter 

according to chatgpt 

in LLMs synapses are the equivalent of parameters 

but I'm not sure what the equivalent of neurons are called in LLMs