Question 100
A small fine-tuning experiment shows higher quality on one specialist benchmark but major regression on ordinary help requests. What promotion criterion is appropriate?
Explanation
Correct Answer
Customization can cause catastrophic forgetting or narrow overfitting; overall task suitability must be tested.
Incorrect Answers
One metric cannot justify broad production use.
Removing baselines hides regressions.
Randomness does not restore lost capabilities.