Challenge: Existing methods measure self-preference bias by comparing the scores a judge model assigns to its own responses with those assigned to other models.
Approach: They propose to use gold judgments as proxies for the actual quality of responses . they propose to measure self-preference bias as the difference between the judge model's own and other models' scores .
Outcome: The proposed method can assess self-preference bias across large language models . it uses gold judgments as proxies for the ground truth scores of the judge model .

Similar Papers

Humans or LLMs as the Judge? A Study on Judgement Bias (2024.emnlp-main)

Copied to clipboard

Challenge: Proprietary models such as GPT-4, Claude, Gemini-Pro and others are being democratized to improve evaluations of LLMs.
Approach: They propose a framework that is free from referencing groundtruth annotations for investigating **Misinformation Oversight Bias**, **Gender Bia**,**Authority Bia* and **Beauty Bia's** on LLM and human judges.
Outcome: The proposed framework investigates **Misinformation Oversight Bias**, **Gender Bia**,**Authority Bia* and **Beauty Bia' on LLM and human judges.
Think Again! The Effect of Test-Time Compute on Preferences, Opinions, and Beliefs of Large Language Models (2025.acl-industry)

Copied to clipboard

Challenge: Large Language Models exhibit subjective preferences, opinions, and beliefs, which may shape their behavior, influence advice and recommendations, and potentially reinforce certain viewpoints.
Approach: They developed a benchmark to assess LLMs’ subjective inclinations across societal, cultural, ethical, and personal domains.
Outcome: The proposed benchmark assesses LLMs’ subjective inclinations across societal, cultural, ethical, and personal domains.
Dissecting Human and LLM Preferences (2024.acl-long)

Copied to clipboard

Challenge: a recent study shows that human and Large Language Model preferences are important for model fine-tuning and evaluation.
Approach: They dissect the preferences of human and 32 different Large Language Models to understand their quantitative composition.
Outcome: The proposed model is compared with 32 different large language models using real-world user-model conversations.
JuStRank: Benchmarking LLM Judges for System Ranking (2025.acl-long)

Copied to clipboard

Challenge: Recent work has focused on instance-based evaluation of LLM judges, where a judge is evaluated over a set of responses, or response pairs, while being agnostic to their source systems.
Approach: They propose to validate the quality of the LLM judge itself by comparing system scores to a human-based ranking.
Outcome: The proposed model fails to validate the quality of the judge itself, ignoring critical factors affecting system-level ranking, such as a judge’s positive or negative bias towards certain systems.
Assistant-Guided Mitigation of Teacher Preference Bias in LLM-as-a-Judge (2025.findings-emnlp)

Copied to clipboard

Challenge: LLM-as-a-Judge uses large language models to evaluate the quality of LLM generated responses, but training proxy judge models using evaluation data generated by powerful teacher models introduces a critical yet previously overlooked issue: teacher preference bias.
Approach: They propose a new setting that incorporates an additional assistant model, which is not biased toward the teacher model’s responses, to complement the training data.
Outcome: The proposed model reduces teacher preference bias while maintaining strong performance across six evaluation benchmarks.
Beyond Performance: Quantifying and Mitigating Label Bias in LLMs (2024.naacl-long)

Copied to clipboard

Challenge: Large language models exhibit undesirable preference toward predicting certain answers over others, despite their adaptability to diverse tasks.
Approach: They propose a label bias calibration method that outperforms recent calibration approaches for improving performance and mitigating label bias.
Outcome: The proposed method outperforms calibration approaches for improving performance and mitigating label bias.
Rating Roulette: Self-Inconsistency in LLM-As-A-Judge Frameworks (2025.findings-emnlp)

Copied to clipboard

Challenge: Using large language models (LLMs) for evaluating natural language generation has gained traction . lm judges have low intra-rater reliability in their assigned scores, making it difficult to measure how good their judgments actually are.
Approach: They show that large language models align more closely with human preferences than n-grams . they quantify this variance and compare them to other NLG tasks and benchmarks based on the results .
Outcome: The proposed models align more closely with human preferences than n-gram or embedding-based metrics.
Don’t Judge Code by Its Cover: Exploring Biases in LLM Judges for Code Evaluation (2026.findings-eacl)

Copied to clipboard

Challenge: Large language models (LLMs) are increasingly used as evaluators for code evaluation tasks . however, whether they can handle superficial variations remains unclear .
Approach: They define six types of potential biases in code evaluation and reveal their impact on LLM judges.
Outcome: The proposed method can be used to evaluate semantically equivalent code with superficial variations without reference implementations.
What Factors Influence LLMs’ Judgments? A Case Study on Question Answering (2024.lrec-main)

Copied to clipboard

Challenge: Existing studies indicate that Large Language Models perform at a level comparable to humans with advantages of speed and cost-effectiveness in different fields.
Approach: They propose to introduce four unexplored factors and a new dimension of question difficulty to provide a more comprehensive understanding of LLMs’ judgments across varying question intricacies.
Outcome: The proposed dimensions of question difficulty and answer quantity provide valuable insights into optimizing LLMs’ performance as judges.
OffsetBias: Leveraging Debiased Data for Tuning Evaluators (2024.findings-emnlp)

Copied to clipboard

Challenge: Existing studies have identified that judge models are vulnerable to certain biases, such as favoring longer answers regardless of content.
Approach: They propose to use open-source LLMs to fine-tune judge models and propose a meta-evaluation collection of test cases for each bias type.
Outcome: The proposed meta-evaluation dataset shows that fine-tuning improves the robustness of judge models against biases and improves performance across most evaluation scenarios.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations