Appreciate it Markus :) That's a good question, I'm not sure that information is anywhere. Because the parameters are what's learnable, that's the number that matters the most.

To make an analogy, the brain has aprox 10^11 neurons and 10^15 synapses. That's 4 orders of magnitude more connections than units. Therefore (and this is just to make a very rough approximation), because the synapses are the equivalent to parameters, GPT-3 would have around 20 million neurons.

Yet, there are important differences. For instance, the number of synapses per neuron in the brain could be much larger than in GPT-3.

Cheers :)

Writing at the intersection of AI, philosophy, and the cognitive sciences | | Get my articles for free: