This is the insight missing from everyone comparing LLM parameter counts to human neurons or synapses. The human model gets a new version every day, and the digital one costs $5B of energy and a year to do the same.