The simplified approach makes it easier to see how neural networks produce the outputs they do. A tweak to the way artificial neurons work in neural networks could make AIs easier to decipher.
Modern neural networks, with billions of parameters, are so overparameterized that they can "overfit" even random, ...