-
@ Satosha
2024-11-30 01:49:03The earliest public model of chatGPT (version 3.0) used around 12000 mathematical dimensions to internalize the context of around 50,000 words of text. It did so by crawling through the data from internet, digital books in public domain and code repositories. This in itself is a breakthrough but consider the later versions that learnt music and art.
It begs the question : how many dimensions must a human mind have - it may be trained in text, music, art and thousands of other skills - agriculture, metallurgy, transportation to state the obvious.
And what does it take to create a consistent three-dimensional visual experience ?
Though it is hard to compare a biological neural network such as human brain with a software neural network such as chatGPT, yet it is common acceptance that a synapse in the biological terms ( roughly equal to number of dendrites per neurons) is equivalent to a "parameter" in the computing paradigm of software neural networks.
The chatGPT version 4 is estimated to have 100 Trillion parameters whereas a single human brain has around 700 trillion synapses (assuming around 100 billion neurons, each having 7000 dendrites). It's amazing that each one of us carry seven times the natural intelligence than the biggest artificial neural network as of 2024.
Even more amazing is the fact that we run such vast neural network on just 2000 calories per day - almost equal to a 100 watts light bulb, whereas chatGPT consumes as much energy as 26000 US homes. This efficiency is key to the idea of higher general intelligence. For example Elephants or Whales have bigger brains than us, but our advantage is we survive on much less due to our smaller size. No wonder there are only seven or eight chatGPT like models, while there are eight billion of us!
Another way to think about it is that AI needs to be 700 billion times more efficient to out do humans. Even then it may be hard because we already have millions of years of early start advantage. In essence, AI is no threat to humans. It is a cool utility that we should embrace to improve our conscious experience. As with any technology, there is always a chance of misuse - you can kill someone with a kitchen knife. Investigating such misuse (when reported) is the responsibility of law and order, but it doesn't mean we start regulating the industry. Regulation must always be a reactive posture - not proactive.
There is another argument given to scare the people - that humans fight with each other whereas all AIs may be singularly focused. Nothing can be farther from the truth because as soon as AI's number start growing, there will be a fight for resources among competing models. No computing system is free from competition just the way all biological systems must compete to evolve.