Ridiculously terrible article—lots of unsupportable assertions without any evidence. He doesn’t seem to have any knowledge of the actual constraint space on circuit design—thermodynamic, area, latency, etc.
He uses a 200hz firing rate, when neurons actually fire at < 1hz on average. He claims the cerebellum has more compute power than the cortex, which is pretty ridiculous—given that the cortex has far more synapses and more volume, and the fact that the brain is reasonably efficient. He doesn’t understand that most of the energy usage is in wire dissipation, not switching. His estimates are thus off by many orders of magnitude. The article is not worth reading.
Ridiculously terrible article—lots of unsupportable assertions without any evidence. He doesn’t seem to have any knowledge of the actual constraint space on circuit design—thermodynamic, area, latency, etc.
See my longer reply comment here.
He uses a 200hz firing rate, when neurons actually fire at < 1hz on average. He claims the cerebellum has more compute power than the cortex, which is pretty ridiculous—given that the cortex has far more synapses and more volume, and the fact that the brain is reasonably efficient. He doesn’t understand that most of the energy usage is in wire dissipation, not switching. His estimates are thus off by many orders of magnitude. The article is not worth reading.