Six weights all start at zero. Click Train one step to see why that fails: each neuron gets the same update, so all six act as one.
Every weight starts at zero. Seems tidy enough.
