> LLM's are cool but not even close to the power of the human brain
There are only 20 Billion Neurons in the Neocortex. GPT-3 has 175 Billion. GPT-4 scores 130 on IQ tests. They are absolutely in the realm of the power of human brains. They're just exposed to different stimuli and have very narrow input/output.
> There are only 20 Billion Neurons in the Neocortex. GPT-3 has 175 Billion.
No, parameters are loosely analogous to synapses (they are weights on the connections between neurons) not neurons, and the neocortex is not the whole of the brain. There are about 140 trillion synapses in the neocortex, 600 trillion in the brain. And neural connections may not explain the brain's capacity, as mutual interaction with other body systems is involved.
> GPT-4 scores 130 on IQ tests.
Methodological issues (or simply methodological unclarity) with the reports of GPT-4 test results aside, IQ tests do not directly test general intelligence, even under the assumption that g is the same thing as general intelligence. They are a proxy which works for testing that in humans, because in humans the capacities directly tested by them correlate with g. We have no idea what IQ testing results of LLMs, even conducted appropriately, mean, if anything.
> They are absolutely in the realm of the power of human brains.
No, to steal from Pulp Fiction, an LLM and a human brain “ain’t even the same f###ing thing. […] Ain’t no f###ing ballpark neither. […] It ain’t the same league. It ain’t even the same f###ing sport.”
There are only 20 Billion Neurons in the Neocortex. GPT-3 has 175 Billion. GPT-4 scores 130 on IQ tests. They are absolutely in the realm of the power of human brains. They're just exposed to different stimuli and have very narrow input/output.