Hi Dwinterbello, thank you for raising this. This is a valid argument. This approach is explained in Section 5.4 in the LSTM chapter in, “Understanding Deep Learning.” You can download here:

https://www.understandingdeeplearning.com