> the next question is if they can be made fun without too much additional work

They can't. If you think about how these things are trained it's blatantly obvious fun is an impossible metric to optimize them for

Training impossibilities aside, how would you even make an optimization loop for "fun"?