Using Theurning in Deep Learning?

In deep learning, we optimize over the imput output behavior of the system. And you can't deploy them without moving them off distribution. So in this legin spase, we might end up on something people call mesaomesrs. There's just a lot of ways that, once the thing is out there and it's like, doing its thing, everything changes. We're also training it to do a million other things. At some point possibly deployment, it realizes that the cognitive architecture that it has is not particularly well suited to playing chess,. If it wants to win at chess, which it does, it should write a programme that does like a minmax andalp alphabet