As you can see I have it defined it for when the output/prediction is [1 1]; I want to make my RNN smarter by having a Sequence output as can be seen below:

for me it looks like the shape of your last Keras Dense Layer is not correctly defined. Could this be?
What are you settings for the number of units and the activation function in the last Keras Dense Layer?

I would therefore like to take the opportunity to ask you another question regarding the RNNs in general.

(I have looked at and partially worked through those examples you are referring to, but my example is purely numeric and not NLP.)

My question is this:

I am doing Sequence to Sequence mapping (many to many as you call it). In my particular case there are always 18 entries in the sequence.

How is the accuracy of the output sequence determined? It is very likely that some numbers in the output vector is more random than others and therefore it would be nice to know both many to one and many to many probability.

To do that, would I need to build another many to one RNN and how would I compare the two RNNs sensible?

I am not sure my question is clear. Does the question makes sense?

Maybe you can help me by letting me know little bit more about your task.
What kind of problem are you trying to solve? Is it a sequence of class values or numerical values?
If it is a sequence on numerical values, are the values always between 0 and 1? If yes are these values probabilities for different classes?