This AI Research Diagnoses Problems in Recurrent Neural Networks RNN-based Language Models and Corrects them to Outperform Transformer-based Models on Long Sequence Tasks