Totally agree, but RF's are more related to "CNN-ish problems" (image classification and...?), not RNNs, or generally, any graphical sequence model.

EDIT: to clarify: "j/k" with the thing in parenthesis ;-)