Auto-Prompt Ensemble for LLM Judge

2025-10-07

AI Summary

Current LLM evaluation systems suffer from low reliability due to their inability to identify implicit dimensions embedded in human judgments. To address this, we propose the Auto-Prompt Ensemble framework coupled with Collective Confidence—a confidence-aware modeling method that adaptively discovers missing evaluation dimensions by analyzing failure cases, and integrates prompt engineering, ensemble learning, and test-time computation scaling for dynamic, multi-dimensional assessment. Our key contribution is the first integration of automated implicit-dimension discovery with confidence-driven evaluation, effectively bridging the gap between human and machine evaluation criteria. Under zero-shot settings, our method boosts GPT-4o’s human agreement rate on Reward Bench from 87.2% to 90.5%, significantly improving judgment consistency and robustness.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Safety and Robustness
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• User Modeling, Personalization and Recommendation: Metrics for user behavior and evaluating success

Abstract

We present a novel framework that improves the reliability of LLM judges by selectively augmenting LLM with auxiliary evaluation dimensions. Existing LLM judges often miss crucial evaluation dimensions because they fail to recognize the implicit standards underlying human assessments. To address this challenge, we propose the Auto-Prompt Ensemble (APE), an adaptive framework that automatically learns evaluation dimensions from its failure cases. APE incorporates a confidence-based ensemble mechanism to decide when to adopt the judgments from additional evaluation dimensions through a novel confidence estimation approach called Collective Confidence. Extensive experiments demonstrate that APE improves the reliability of LLM Judge across diverse standard benchmarks. For instance, APE enhances GPT-4o agreement rate on Reward Bench from 87.2% to 90.5% in the zero-shot setting. Overall, APE provides a principled approach for LLM Judge to leverage test-time computation, and bridge the evaluation gap between human and LLM judges.

Problem

• Improves reliability of LLM judges through selective augmentation
• Automatically learns evaluation dimensions from failure cases
• Bridges evaluation gap between human and LLM judges

Innovation

• Auto-Prompt Ensemble learns evaluation dimensions from failure cases
• Confidence-based ensemble decides when to adopt additional judgments
• Collective Confidence estimation improves LLM judge reliability