Nothing wrong with combining genetic programing with neural networks, there many ways to do it, for example, you can have genetically evolved network that produ

Nothing wrong with combining genetic programing with neural networks, there many ways to do it, for example, you can have genetically evolved network that produces actual network(s) (weights, locations, connections), to me biggest is issue is with escaping local minima as efficient/performant as other methods (does not beat deep RL methods), in simple sense it stops evolving and it is really hard to figure out why.