Human-Centered Design Recommendations for LLM-as-a-Judge

2024/07/03 by Qian Pan, Zahra Ashktorab, Pan, Qian +13 · 11 citations
Social Sciences · Computer Science · #Legal Education and Practice Innovations #Artificial Intelligence in Law #Law, AI, and Intellectual Property

paper · pdf · doi:10.48550/arxiv.2407.03479

Abstract

Traditional reference-based metrics, such as BLEU and ROUGE, are less effective for assessing outputs from Large Language Models (LLMs) that produce highly creative or superior-quality text, or in situations where reference outputs are unavailable. While human evaluation remains an option, it is costly and difficult to scale. Recent work using LLMs as evaluators (LLM-as-a-judge) is promising, but trust and reliability remain a significant concern. Integrating human input is crucial to ensure criteria used to evaluate are aligned with the human's intent, and evaluations are robust and consistent. This paper presents a user study of a design exploration called EvaluLLM, that enables users to leverage LLMs as customizable judges, promoting human involvement to balance trust and cost-saving potential with caution. Through interviews with eight domain experts, we identified the need for assistance in developing effective evaluation criteria aligning the LLM-as-a-judge with practitioners' preferences and expectations. We offer findings and design recommendations to optimize human-assisted LLM-as-judge systems.

Cited by

• LeechHijack: Covert Computational Resource Exploitation in Intelligent Agent Systems
• Generate, Evaluate, Iterate: Synthetic Data for Human-in-the-Loop Refinement of LLM Judges
• User Perceptions of Privacy and Helpfulness in LLM Responses to Privacy-Sensitive Scenarios
• A Principle-based Framework for the Development and Evaluation of Large Language Models for Health and Wellness
• Multi-Agent Debate for LLM Judges with Adaptive Stability Detection
• Interpreting LLM-as-a-Judge Policies via Verifiable Global Explanations
• Vipera: Blending Visual and LLM-Driven Guidance for Systematic Auditing of Text-to-Image Generative AI
• Interactive Recommendation Agent with Active User Commands
• Guidelines for Empirical Studies in Software Engineering involving Large Language Models
• LLM-as-a-Judge for Privacy Evaluation? Exploring the Alignment of Human and LLM Perceptions of Privacy in Textual Data
• LLM-Crowdsourced: A Benchmark-Free Paradigm for Mutual Evaluation of Large Language Models

Related

• A Survey on LLM-as-a-Judge
• EvalAssist: A Human-Centered Tool for LLM-as-a-Judge
• LLMs-as-Judges: A Comprehensive Survey on LLM-based Evaluation Methods
• Calibrating LLM-Based Evaluator
• Aligning Human and LLM Judgments: Insights from EvalAssist on Task-Specific Evaluations and AI-assisted Assessment Strategy Preferences