Investigating Pre-trained Language Models on Cross-Domain Datasets, a Step Closer to General AI