Training-free LLM Verification via Recycling Few-shot Examples

2025-06-08

arXiv.org

AI Summary

Large language models (LLMs) exhibit stochasticity and inconsistency in reasoning outputs, and existing verification methods—such as majority voting or external verifiers—are limited by poor generalization or reliance on additional training. Method: We propose Referi, a training-free, lightweight verification framework that uniquely reuses the same set of in-context few-shot examples for both response generation and internal verification. Leveraging Bayesian principles, Referi introduces a dual-scoring mechanism that jointly assesses response consistency and confidence, performing verification entirely within the LLM’s inference process. Contribution/Results: Referi requires zero fine-tuning and zero parameter updates. Evaluated across three mainstream LLMs and seven diverse reasoning tasks, it achieves an average accuracy improvement of 4.8%, significantly outperforming majority voting and Best-of-N baselines. Referi establishes a new, efficient, and broadly applicable paradigm for trustworthy LLM reasoning.

Technology Category

• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Although LLMs have achieved remarkable performance, the inherent stochasticity of their reasoning process and varying conclusions present significant challenges. Majority voting or Best-of-N with external verification models has been explored to find the most promising solution among multiple LLM outputs. However, these approaches have certain limitations, such as limited applicability or the cost of an additional training step. To address this problem, we propose a novel and effective framework that Recycles Few-shot examples to verify LLM outputs (Referi). Our key idea is to additionally utilize the given few-shot examples to evaluate the candidate outputs of the target query, not only using them to generate outputs as the conventional few-shot prompting setup. Specifically, Referi evaluates the generated outputs by combining two different scores, designed motivated from Bayes'rule, and subsequently selects the candidate that is both confidently determined and contextually coherent through a few additional LLM inferences. Experiments with three different LLMs and across seven diverse tasks demonstrate that our framework significantly improves the accuracy of LLMs-achieving an average gain of 4.8%-through effective response selection, without additional training.

Problem

• Verifying LLM outputs without training using few-shot examples
• Selecting most reliable responses via Bayesian-inspired scoring mechanism
• Improving LLM accuracy through recycling demonstration examples

Innovation

• Recycles few-shot examples for LLM output verification
• Combines two scores motivated by Bayes' rule
• Selects confident candidates through additional LLM inferences