Baidu, a Chinese Tech Giant released a conversational AI machine that works in both Chinese and English. The new machine ERNIE 2.0 claims to outperform Google’s BERT and XLNet in several language understanding benchmarks and in nine different Chinese natural language tasks. ERNIE 2.0 relies on a transformer encoder and the BookCorpus data set for training. The new technology has been relatively successful at analyzing sentiment in movie reviews and inferring meaning from sentences. ERNIE 2.0 is based on an open-sourced language understanding model. This new machine relies on something which Baidu refers to as continual pre-training. It’s a two step process. The first is unsupervised pre-training tasks using a lot of data and the second is multi-task learning.