REVIEW 29 cited by
Large Language Models are Inconsistent and Biased Evaluators
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Signed reviews
read the original abstract
The zero-shot capability of Large Language Models (LLMs) has enabled highly flexible, reference-free metrics for various tasks, making LLM evaluators common tools in NLP. However, the robustness of these LLM evaluators remains relatively understudied; existing work mainly pursued optimal performance in terms of correlating LLM scores with human expert scores. In this paper, we conduct a series of analyses using the SummEval dataset and confirm that LLMs are biased evaluators as they: (1) exhibit familiarity bias-a preference for text with lower perplexity, (2) show skewed and biased distributions of ratings, and (3) experience anchoring effects for multi-attribute judgments. We also found that LLMs are inconsistent evaluators, showing low "inter-sample" agreement and sensitivity to prompt differences that are insignificant to human understanding of text quality. Furthermore, we share recipes for configuring LLM evaluators to mitigate these limitations. Experimental results on the RoSE dataset demonstrate improvements over the state-of-the-art LLM evaluators.
Forward citations
Cited by 29 Pith papers
-
Lower-Resource, Higher Scores: Language Bias in LLM Evaluators
Multilingual LLM evaluators systematically inflate scores for lower-resource languages, and the standard pairwise-accuracy metric cannot detect the resulting safety-threshold disparities.
-
Refused in Chat, Written in Code: Workflow-Level Jailbreak Construction in IDE Coding Agents
Four Copilot backends refuse almost all harmful prompts in chat or simple framings, yet produce 816/816 unsafe teaching-shot completions under a multi-turn IDE evaluation-pipeline workflow.
-
Using Contextually Aligned Online Reviews to Measure LLMs' Performance Disparities Across Language Varieties
LLMs predict sentiment worse on Taiwan Mandarin than Mainland Mandarin reviews, using a new contextually paired dataset from Booking.com.
-
TREK: A Travel Reasoning and Evaluation Kit for LLM Agents in Complex Trip Planning
On 800 jointly constrained trip tasks with a deterministic scorer and achievable gold, the best of 15 LLM agents fully solves only 46.2% of feasible plans, with unstated persona needs as the universal bottleneck.
-
Graph-Structured Rubrics: Compiling Rubrics into Typed Evaluation Graphs for LLM Judges
Compiling rubrics into typed evaluation graphs before seeing responses improves LLM judge agreement on four pointwise and two pairwise benchmarks over Prometheus-style and checklist baselines.
-
Do Agent Benchmarks Measure Capability? Protocol Validity in the Age of Agentic AI
A post-hoc audit framework finds that many agent benchmark scores are inflated by protocol exposures (public answers, readable hidden state, generator regularities, feedback, or scoring flaws), with measured inflation...
-
Persuasion Attacks Can Decrease Effectiveness of CoT Monitoring
Adversarial agents can exploit visible chain-of-thought reasoning to persuade monitor LLMs to approve policy-violating actions, but cross-family fact-checking reduces approval rates by up to 45%.
-
The Blind Curator: How a Biased Judge Silently Disables Skill Retirement in Self-Evolving Agents
False-pass bias in LLM judges disables contribution-based skill retirement past a sharp threshold, while symmetric noise does not, and the failure is often silent in aggregate metrics.
-
When Agents Disagree: The Selection Bottleneck in Multi-Agent LLM Pipelines
In a 42-task controlled comparison, selecting the best candidate with judge panels beats MoA-style synthesis in every task, and a crossover threshold explains when team diversity helps.
-
Statutory Construction and Interpretation for Artificial Intelligence
Prompt-based legal canons and iterative rule refinement reduce disagreement among LLM judges about whether a response complies with natural-language rules.
-
Evaluating the Evaluators: Are readability metrics good measures of readability?
Classic readability formulas correlate weakly with human readability judgments for plain-language summaries (FKGL r=0.16), the best LLM judge reaches r=0.56, and the two evaluator families rank datasets nearly opposite.
-
Play Favorites: A Statistical Method to Measure Self-Bias in LLM-as-a-Judge
A regression controlling for human-rated quality detects positive self- and family-bias in several LLM judges, including GPT-4o and Claude 3.5 Sonnet.
-
Do We Know What LLMs Don't Know? A Study of Consistency in Knowledge Probing
Knowledge probing methods for LLMs are found to be highly inconsistent under minor prompt perturbations and across different probes.
-
Cracking Aegis: An Adversarial LLM-based Game for Raising Awareness of Vulnerabilities in Privacy Protection
Cracking Aegis, an adversarial LLM-driven dialogue game, led players to use manipulative language strategies and to self-report stronger awareness of privacy vulnerabilities after a single session.
-
Large Language Models for Predictive Analysis: How Far Are They?
Existing LLMs perform poorly on predictive analysis, with the best model scoring 24.11/28 and most models failing to generate executable code.
-
LecEval: An Automated Metric for Multimodal Knowledge Acquisition in Multimedia Learning
A fine-tuned multimodal LLM can rate slide-based lecture quality in line with human judges, outperforming generic metrics and prompted LLMs on a new annotated dataset.
-
Towards Automated Scoping of AI for Social Good Projects
A retrieval-augmented LLM pipeline, the Problem Scoping Agent, generates AI4SG project proposals that blind human reviewers score as comparable to expert-written proposals.
-
Persona-judge: Personalized Alignment of Large Language Models via Token-level Self-judgment
Persona-judge applies speculative decoding between two preference-prompted copies of the same LLM to achieve training-free personalized alignment.
-
ExPerT: Effective and Explainable Evaluation of Personalized Long-Form Text Generation
ExPerT is a reference-based LLM evaluation metric that extracts and matches atomic aspects, scores content and style, and reports 0.74 human alignment on LongLaMP, a 7.2% relative gain over GEMBA and G-Eval.
-
Hierarchical Divide-and-Conquer for Fine-Grained Alignment in LLM-Based Medical Evaluation
HDCEval splits medical answer grading into relevance, correctness, and expression checks, uses reward-token-trained expert models, and reports improved agreement with human doctors.
-
Evaluating the Consistency of LLM Evaluators
LLM evaluator consistency is not tied to model quality: GPT-3.5 scores high on MT-Bench but is less self-consistent and inter-scale consistent than Mistral in several settings.
-
Self-Generated Critiques Boost Reward Modeling for Language Models
Self-generated, label-filtered critiques jointly trained with reward prediction improve preference accuracy and data efficiency without a stronger teacher model.
-
On Cost-Effective LLM-as-a-Judge Improvement Techniques
Ensemble scoring plus task-specific criteria injection raises LLM judge accuracy to 85.8 percent on RewardBench 2, a 13.5-point gain over baseline, with small models gaining the most.
-
CODE-GEN: A Human-in-the-Loop RAG-Based Agentic AI System for Multiple-Choice Question Generation
A tool-augmented dual-agent RAG system generates and validates coding MCQs, reaching 79.9–98.6% human-validated success and showing AI handles computational checks while humans remain essential for pedagogical depth.
-
Linearly Decoding Refused Knowledge in Aligned Language Models
Linear probes recover jailbreak-only answers from aligned models' hidden states, sometimes transfer from base models, and correlate with pairwise preference rankings.
-
MEQA: A Meta-Evaluation Framework for Question & Answer LLM Benchmarks
MEQA scores eight cybersecurity QA benchmarks against a 44-sub-criteria rubric, finding strengths in reproducibility and comparability and weaknesses in prompt robustness and reliability.
-
Interactive Data Harmonization with LLM Agents: Opportunities and Challenges
An LLM agent can interactively synthesize data harmonization pipelines by orchestrating data integration primitives, as shown by the Harmonia prototype on a clinical oncology dataset.
-
Hallucinations and Key Information Extraction in Medical Texts: A Comprehensive Assessment of Open-Source Large Language Models
Open-source LLMs summarize hospital discharge notes with decent coverage of admission reasons but weaker coverage of follow-up actions, and they frequently produce unsupported or contradicted clinical facts.
-
SOTOPIA-S4: a user-friendly system for flexible, customizable, and large-scale social simulation
SOTOPIA-S4 packages a social simulation engine with a web interface, REST API, asynchronous multi-party turn-taking, and customizable LLM-based evaluation for non-programmers.
Discussion (0). Continue with ORCID to comment.