> We measure a model’s quality in two ways. First, we evaluate the model using a well-defined accuracy metric we compute during model training. This measure tells us how well the model performs its task – specifically, giving engaging Tweets a high score.

Very curious to know what this metric would be. Maybe some combination of likes, retweets, and viewing time? Or maybe including a general liking term (i.e. liking a person's tweet gives a small boost to all their tweets).