Do they infer anything? Correct me if I'm wrong but having the types right there in the source for training data just means more context.