Some of the below statements regarding RNNs are true, some of them are not. Mark those which you think true. Mark as many as necessary.

		1.With truncated backpropagation in time, which uses only K past frames, the vanilla RNN model at a given time step t cannot aggregate (remember) any information belonging to the time steps before t K.
		2.RNNs have a sequential structure therefore they are inconvenient for parallel implementation and hence slow to compute.
		3.The vanishing gradient problem in RNNs does not have anything to do with the length of the input sequence.
		4.One of the key ideas in LSTM model is to define a new state variable (cell state) through which the gradient flows with less interruption during backpropagation.
		5.The parameters of a typical RNN (vanilla or LSTM) are not necessarily shared between time steps, that is, the parameters of the recurrent cell may change through time.
		6.An RNN model can in theory be fed with sequential data of arbitrary length.
		7.LSTM model handles the vanishing gradient problem better when compared to vanilla RNNs; but does not provide any guarantee for it.