Raw LLM Responses

I did not intend to oversimplify. What I posted previously skipped an important layer. It needs to be realized that neural nets at work here are not like the neural networks in your biological brain. We learn by attempting something and a pathway for that attempt is established and carries a 'weight'. Then at the 'result-end', feedback either reinforces the 'weight' or not. If the attempt was successful, the weight of the pathways become more permanent and will be used again for similar attempts. So pathways grow and interconnect in ways ... and that's 'learning'. If your brain doesn't have enough 'pathways' it grows them, and the network becomes unimaginable in size and complexity. Though fashioning an artificial neural network has been done with components, capacity is low and it is impractical to 'grow' that capacity in hardware. So 'neural networks' are 'simulations' of neural networks that indeed employ some specially purposed integrated circuits. Software characterizes the network and determines HOW it learns, and PEOPLE write that software. Beyond that, it's still memory, data bases and Interconnectivity.