Question: In a concise analytical response, explain why adding more training examples cannot by itself deliver the target performance when the model already performs too poorly on the training set.

Sample answer: If the model's training error is already above the target, more data alone will not push it below that target. With additional training examples, training error typically stays flat or rises slightly, because the model is not being changed to fit the current data better. Also, development error is usually worse than training error. So if the model cannot reach the goal on training data, it is even less likely to reach it on development data by adding data alone; the underlying model or training procedure must also improve.

Key points:

• The training-set error starts above the desired level.
• Adding more data does not normally reduce training error.
• Development error is usually higher than training error.
• Therefore, more data by itself is not enough to reach the target.

Rubric: A strong response identifies the starting gap on training data, notes that training error does not fall just because more data is added, states that development error is usually higher, and concludes that another change besides data is needed.