What is a primary reason for the "unreasonable effectiveness" of recurrent neural networks (RNNs) in sequence prediction tasks?
a. RNNs have a built-in mechanism for parallel processing of inputs.
b. RNNs utilize convolutional layers to handle spatial data.
c. RNNs can remember and use information from previous time steps through their internal state.
d. RNNs are effective due to their use of fixed-size input and output vectors.