Are Generative Models Underconfident? An Embarrassingly Simple Quality Estimation Approach

2025-02-16

AI Summary

Generative models often exhibit “under-confidence” in multi-solution tasks due to probability dispersion across semantically equivalent outputs, causing reference-free quality estimation (QE) based on raw sequence probability to severely underestimate true quality. To address this, we propose Dominant Quality Probability (DMP), a zero-overhead, training-free, and fine-tuning-free QE metric that statistically reconstructs and recalibrates model confidence over equivalent correct outputs. DMP operates solely on native token-level probabilities—introducing no additional parameters, modules, or architectural modifications—and is compatible with diverse autoregressive models (e.g., Whisper, Llama). Evaluated across translation, summarization, and other generation tasks, DMP improves average Pearson correlation with human ratings by 0.208 over baseline sequence probability, significantly enhancing alignment with human judgment. DMP establishes a simple, general, and robust new paradigm for reference-free QE.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Natural Language Processing: Generation
• Reasoning under Uncertainty: Relational Probabilistic Models

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: Fairness-aware retrieval and ranking

Abstract

Quality Estimation (QE) is estimating the quality of model output when the ground truth reference is not available. Looking at model uncertainty from its own output probabilities is the most trivial and low-effort way to estimate the output quality. However, for generative model, output probabilities might not be the best quality estimator. At an output step, there can be multiple correct options, making the probability distribution spread out more. Thus, lower token probability does not necessarily mean lower output quality. In other words, the model can be considered underconfident. In this paper, we propose a QE approach called Dominant Mass Probability (DMP}, that boosts the model confidence in cases where there are multiple viable output options. We show that, with no increase in complexity, DMP is notably better than sequence probability when estimating the quality of different models (Whisper, Llama, etc.) on different tasks (translation, summarization, etc.). Compared to sequence probability, DMP achieves on average +0.208 improvement in Pearson correlation to ground-truth quality.

Problem

• Evaluates generative models' confidence levels.
• Proposes Dominant Mass Probability for quality estimation.
• Improves correlation with ground-truth quality metrics.

Innovation

• Dominant Mass Probability introduced
• Boosts model confidence effectively
• Improves Pearson correlation significantly