For reference, the “one human brain” estimate comes from FLOPS = 86 billion neurons×1000 synapses/neuron×200 Hz = 10^16 − 10^17 FLOPS a mode of estimation that I suspect Kurzweil would admit is tendentious.
I think that this is almost certainly an overestimate, in that the 200 Hz rate I don’t think accounts for the full compute cycle to which a synapse is relevant. I think it makes more sense to either consider effective synapse compute cycles at something closer to < 10 Hz or consider the compute units at scales of groups of synapses on dendritic branches or even whole neurons. Perhaps both. Either way, you should trim at least 1 OOM off that number for that.
I think that this is almost certainly an overestimate, in that the 200 Hz rate I don’t think accounts for the full compute cycle to which a synapse is relevant. I think it makes more sense to either consider effective synapse compute cycles at something closer to < 10 Hz or consider the compute units at scales of groups of synapses on dendritic branches or even whole neurons. Perhaps both. Either way, you should trim at least 1 OOM off that number for that.