>>11895792if it's new, it's not been trained.
On top of that, the network's model was probably based for some particular problem. Adding another node at a later date, it'll probably just tend itself to 0 and let the rest of the network do its job the same. This is actually a problem that occurs when setting up some neural networks too, hence why understanding the problem is vastly more important than just throwing nodes and horsepower at it.
tldr if the rest of the network is trained, the new node is gonna do fuck all
