A team uses a strong LLM as an automated judge to pick the better of two candidate responses, which are placed side by side within the same judging prompt. Across many comparisons, they notice the judge tends to favor whichever response sits in the first slot, even when the same two responses are swapped into the other slot and re-judged. What phenomenon does this describe, and what is a standard mitigation?
- This is a sign the underlying model being judged is broken, since a properly trained model should never receive different scores across two separately-run evaluations
- This is benchmark contamination, and the standard mitigation is to remove the affected test examples from the evaluation set entirely
- This is expected and requires no mitigation, because a judge's preference for the first-listed response reflects a genuine and stable quality signal about response ordering
- This is position bias in LLM-as-judge evaluation; a standard mitigation is to run each comparison twice with the two responses' positions swapped, and only count a response as preferred if it wins in both orderings (or average the results across both orderings)
Why D? And why not the others?
Correct answer: D. This is position bias in LLM-as-judge evaluation; a standard mitigation is to run each comparison twice with the two responses' positions swapped, and only count a response as preferred if it wins in both orderings (or average the results across both orderings)
This is position bias, a well-documented weakness of using an LLM as an automated judge: the judge's verdict shifts based on which slot a response is placed in, independent of the response's actual content, which is exactly what the swap-and-re-judge test in the scenario reveals. The standard mitigation is to run each pairwise comparison in both orderings and only treat a response as the winner if it wins regardless of which slot it occupies (or to average the two orderings' results), which cancels out the ordering effect. The option calling this a sign the judged model is broken is wrong because the inconsistency described is in the judge's verdicts, not in the model being judged, which produced the same two responses both times. The option calling this benchmark contamination is wrong because contamination refers to test data leaking into training data, not to a judge's sensitivity to answer placement. The option claiming no mitigation is needed is wrong because treating an artifact of response placement as a genuine quality signal would systematically bias the evaluation toward whichever response happens to be listed first.
Source: Zheng et al., 'Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena,' arXiv:2306.05685 (2023) -- documents position bias, verbosity bias, and self-enhancement bias in LLM-as-judge evaluation