Robin <mixent...@aussiebroadband.com.au> wrote:
> As pointed out near the beginning of this thread, while current processors > don't come near the number of neurons a human > has, they more than make up for it in speed. I do not think so. The total number of neurons dictates how much complexity the neural network can deal with. To take an extreme example, a worm brain has 302 neurons. Even if they could operate a million times faster than a computer circuit, they still only give you 302 data bits to work with. (Assuming you cannot overlay them from a peripheral mass storage device like a hard disk.) There is no way you could achieve intelligence with that. The human brain has 86 billion neurons, all operating simultaneously. In other words, complete parallel processing with 86 billion "processors" operating simultaneously. ChatGPT tells us she has 175 billion parameters in Version 3. I assume each parameter is roughly equivalent to a neuron. I assume they are run in a massive parallel process (MPP). I doubt all 175 billion can be evaluated simultaneously. It is not as MPP as the human brain. I do not know if they are all on-line simultaneously, or if they are overlaid from mass storage. Even if they are overlaid, they are so much faster than a human neuron, that would be more than equivalent to having them all on-line. So anyway, that is roughly twice as many parameters as a human brain, and these parameters can be evaluated faster than a human brain. Maybe not that much faster if they are not fully MPP, or they need to be overlaid from mass storage. It is definitely bigger than a human brain in total data access. But not orders of magnitude bigger. The limiting factor is not speed so much as data capacity. If you want a computer roughly as intelligent as a person, I guess one with 175 billion parameters could achieve that if it were programmed correctly. But, if you want a super-brain capable of taking over the world or fooling the human race into destroying itself, I suppose you need something much bigger than 175 billion parameters. I wouldn't know how much bigger, but I am guessing a million times bigger would be sufficient. They are millions of times faster. But, as I said, a worm brain with only 302 neurons might be a trillion times faster than ChatGPT's ANN, but it would still only be capable of rudimentary processing. The total number of data bits is a limiting factor, along with the speed.