The animated figure is excellent. Interesting to train on dialogue text to model: structure of statements and response, and a model for predicting good response

The animated figure is excellent. Interesting to train on dialogue text to model: structure of statements and response, and a model for predicting good responses. I am not surprised that this works really well. Getting data to train BERT is fairly simple: any text can be used to drop out words and predict them. The LaMDA training text must have been more difficult to curate.