>>6Ironically the key researchers in the(neural networks) field tend to hold the same position, its oversold as magic AI that solves everything, ignoring that at its core there is only brute-force pattern recognition built on top of matrix math and huge datasets.
Its hyperspecialized(ignoring transfer learning) and really compute intensive.
People are switching to simpler methods all the time, NNs are exceptionally hardware dependent(Nvidia is making a killing on high end GPUs for this) and slow.