C4W4_Logging for ML Monitoring - 5min43sec

“Log data is of course the basis for your next training data set”.

I don’t understand this statement. In what way ?

The video lecture makes sense to me right up to this point. But it seems tangential and maybe would have benefited from one more sentence here to help explain it and the apparent shift in topic.

Any thoughts/insights please ?

Logs can be mined to extract features of interest of your model. If you consider a recommender system, the rating provided by user is a feature that can be extracted from logs and stored in your feature store.
Once a model is deployed, should a user rate a new movie, you can compare that rating with the predicted rating from your model to improve model performance. See movielens