REVIEW 11 cited by
Systematic Evaluation of LLM-as-a-Judge in LLM Alignment Tasks: Explainable Metrics and Diverse Prompt Templates
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
read the original abstract
LLM-as-a-Judge has been widely applied to evaluate and compare different LLM alignmnet approaches (e.g., RLHF and DPO). However, concerns regarding its reliability have emerged, due to LLM judges' biases and inconsistent decision-making. Previous research has developed evaluation frameworks to assess reliability of LLM judges and their alignment with human preferences. However, the employed evaluation metrics often lack adequate explainability and fail to address LLM internal inconsistency. Additionally, existing studies inadequately explore the impact of various prompt templates when applying LLM-as-a-Judge methods, leading to potentially inconsistent comparisons between different alignment algorithms. In this work, we systematically evaluate LLM-as-a-Judge on alignment tasks by defining more theoretically interpretable evaluation metrics and explicitly mitigating LLM internal inconsistency from reliability metrics. We develop an open-source framework to evaluate, compare, and visualize the reliability and alignment of LLM judges, which facilitates practitioners to choose LLM judges for alignment tasks. In the experiments, we examine effects of diverse prompt templates on LLM-judge reliability and also demonstrate our developed framework by comparing various LLM judges on two common alignment datasets (i.e., TL;DR Summarization and HH-RLHF-Helpfulness). Our results indicate a significant impact of prompt templates on LLM judge performance, as well as a mediocre alignment level between the tested LLM judges and human evaluators.
Forward citations
Cited by 11 Pith papers
-
SciVisAgentBench: A Benchmark for Evaluating Scientific Data Analysis and Visualization Agents
SciVisAgentBench provides 108 expert-crafted tasks and a mixed LLM-plus-deterministic evaluation pipeline for benchmarking AI agents that perform scientific visualization workflows.
-
MixAssist: An Audio-Language Dataset for Co-Creative AI Assistance in Music Mixing
MixAssist is the first audio-grounded, multi-turn conversational dataset for co-creative music mixing instruction, and fine-tuning Qwen-Audio on it yields human-comparable mixing advice.
-
Beyond Correctness: Enhancing Architectural Reasoning in Code LLMs via Scalable Labeling with Agentic Judgment
Filtering SFT trajectories with repository-grounded Architecture Complexity and Quality LLM judges yields up to 27.2% SWE-bench Verified resolve rate and better architectural patch conformance than full-data fine-tuning.
-
Multi-Modal Requirements Data-based Acceptance Criteria Generation using LLMs
Multi-modal RAG (text plus UI screenshots) with reward-based polishing generates acceptance criteria from user stories that three industry experts rated near 4/5 on relevance, correctness, and understandability.
-
LitBench: A Benchmark and Dataset for Reliable Evaluation of Creative Writing
Small reward models trained on LitBench reach 78% agreement with upvote-derived human preferences in creative writing, beating all zero-shot LLM judges tested.
-
Time To Impeach LLM-as-a-Judge: Programs are the Future of Evaluation
LLM-synthesized judging programs, aggregated with weak supervision, can replace direct LLM-as-a-judge scoring at far lower API cost, with better consistency and bias resistance in some settings.
-
IDEAlign: Comparing Large Language Models to Human Experts in Open-ended Interpretive Annotations
IDEAlign uses a pick-the-odd-one-out triplet task to measure idea-level similarity between LLMs and expert human annotations, and shows LLM judges using this protocol outperform lexical and vector-based baselines.
-
AI Propaganda factories with language models
Small language models sustain political personas and become more ideologically extreme when replying to counter-arguments, according to a language-model judge.
-
An Empirical Study of LLM-as-a-Judge: How Design Choices Impact Evaluation Reliability
The reliability of LLM-as-a-Judge depends strongly on scoring rubrics and reference answers; sampling with averaging outperforms greedy decoding, and chain-of-thought reasoning adds little when rubrics are clear.
-
Efficient Online RFT with Plug-and-Play LLM Judges: Unlocking State-of-the-Art Performance
A frozen 7B-8B LLM with a JSON rubric and a small LoRA adapter is claimed to outperform 27B-70B reward models and enable 92% GSM-8K exact match under online PPO.
-
Response-Level Rewards Are All You Need for Online Reinforcement Learning in LLMs: A Mathematical Perspective
Under the assumption that the response reward equals the discounted sum of token rewards, response-level rewards suffice for unbiased token-level policy gradients in LLMs.
Discussion (0). Sign in to comment.