# Towards Understanding the Robustness of LLM-based Evaluations under Perturbations

arXiv: https://arxiv.org/abs/2412.09269 · HTML: https://mlsanity.com/paper/2412.09269/

## Key facts

- Cited by 10 papers in the map (data through 14 Oct 2026).
- Builds on 16 of its 26 references that are in the map.
- First posted on arXiv 12 Dec 2024.

## Builds on (most cited first, 16 of 16)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [BERTScore: Evaluating Text Generation with BERT](https://mlsanity.com/paper/1904.09675/) (2019)
- [Finetuned Language Models Are Zero-Shot Learners](https://mlsanity.com/paper/2109.01652/) (2021)
- [Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone](https://mlsanity.com/paper/2404.14219/) (2024)
- [Gemma: Open Models Based on Gemini Research and Technology](https://mlsanity.com/paper/2403.08295/) (2024)
- [Gemini: A Family of Highly Capable Multimodal Models](https://mlsanity.com/paper/2312.11805/) (2023)
- [SummEval: Re-evaluating Summarization Evaluation](https://mlsanity.com/paper/2007.12626/) (2020)
- [Generated Knowledge Prompting for Commonsense Reasoning](https://mlsanity.com/paper/2110.08387/) (2021)
- [Re-evaluating Evaluation in Text Summarization](https://mlsanity.com/paper/2010.07100/) (2020)
- [Is LLM-as-a-Judge Robust? Investigating Universal Adversarial Attacks on Zero-shot LLM Assessment](https://mlsanity.com/paper/2402.14016/) (2024)
- [A Comprehensive Assessment of Dialog Evaluation Metrics](https://mlsanity.com/paper/2106.03706/) (2021)
- [Towards Understanding Jailbreak Attacks in LLMs: A Representation Space Analysis](https://mlsanity.com/paper/2406.10794/) (2024)
- [Scaling Trends in Language Model Robustness](https://mlsanity.com/paper/2407.18213/) (2024)

## Built on by (most cited first, 10 of 10)

- [The Fragility Of Moral Judgment In Large Language Models](https://mlsanity.com/paper/2603.05651/) (2026)
- [Evaluating Robustness of Large Language Models in Enterprise Applications: Benchmarks for Perturbation Consistency Across Formats and Languages](https://mlsanity.com/paper/2601.06341/) (2026)
- [Framing the Game: How Context Shapes LLM Decision-Making](https://mlsanity.com/paper/2503.04840/) (2025)
- [DialogGuard: Multi-Agent Psychosocial Safety Evaluation of Sensitive LLM Responses](https://mlsanity.com/paper/2512.02282/) (2025)
- [I Spy With My Model's Eye: Visual Search as a Behavioural Test for MLLMs](https://mlsanity.com/paper/2510.19678/) (2025)
- [Evaluating LLM-Based Grant Proposal Review via Structured Perturbations](https://mlsanity.com/paper/2603.08281/) (2026)
- [From Performance to Purpose: A Sociotechnical Taxonomy for Evaluating Large Language Model Utility](https://mlsanity.com/paper/2602.20513/) (2026)
- [Learning When to Trust in Contextual Social Bandits](https://mlsanity.com/paper/2603.13356/) (2026)
- [Prompt Perturbation for Reliable LLM Evaluation over Comparison Graphs](https://mlsanity.com/paper/2606.17634/) (2026)
- [Noisy Test-Time Reinforcement Learning for Code LLMs](https://mlsanity.com/paper/2609.32172/) (2026)

## Abstract (arXiv)

Traditional evaluation metrics like BLEU and ROUGE fall short when capturing the nuanced qualities of generated text, particularly when there is no single ground truth. In this paper, we explore the potential of Large Language Models (LLMs), specifically Google Gemini 1, to serve as automatic evaluators for non-standardized metrics in summarization and dialog-based tasks. We conduct experiments across multiple prompting strategies to examine how LLMs fare as quality evaluators when compared with human judgments on the SummEval and USR datasets, asking the model to generate both a score as well as a justification for the score. Furthermore, we explore the robustness of the LLM evaluator by using perturbed inputs. Our findings suggest that while LLMs show promise, their alignment with human evaluators is limited, they are not robust against perturbations and significant improvements are required for their standalone use as reliable evaluators for subjective metrics.