That is also the conclusion that I got from these events.

Unfortunately, it seems that the general response is basically "throw even more RL at it". I'm not sure if it is even possible to decouple the idea of "learning" with reward/punishment systems and loss functions.