In 2018 we saw the rise of pretraining and finetuning in natural language processing. Large neural networks have been trained on general tasks like language modeling and then fine-tuned for classification tasks. One of the latest milestones in this development is the release of BERT.