AI and LLMs · Senior · Question 72 of 75
Why is it risky to use the same model as both generator and only judge?
- 1It may favour its own style and share the same blind spots
- 2Models can't evaluate text written by any model
- 3Using one model for both is always more expensive
Show the answer
Correct answer: It may favour its own style and share the same blind spots
Explanation
Self-preference bias inflates scores. Use a different judge, rubrics and human calibration.