Semantic Voting: A Self-Evaluation-Free Approach for Efficient LLM Self-Improvement on Unverifiable Open-ended Tasks

2025-09-26

AI Summary

To address the high computational cost and self-assessment bias inherent in large language models (LLMs) performing self-improvement on unverifiable open-ended tasks (e.g., machine translation), this paper proposes a lightweight semantic voting method that eliminates explicit self-evaluation. The core innovation extends conventional hard-matching majority voting to soft voting based on semantic similarity, leveraging efficient sentence embedding models to generate high-quality pseudo-labels—thereby avoiding the computational overhead and overconfidence issues associated with self-scoring and entropy minimization. Experiments across diverse model architectures and translation benchmarks demonstrate significant improvements: average inference speedup of 2.3×, and gains of +1.8 BLEU and +2.4 COMET scores. Moreover, the method exhibits strong generalizability across tasks and models. This work establishes a more reliable and scalable unsupervised paradigm for LLM self-optimization.

Technology Category

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

Application Category

• 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
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

The rising cost of acquiring supervised data has driven significant interest in self-improvement for large language models (LLMs). Straightforward unsupervised signals like majority voting have proven effective in generating pseudo-labels for verifiable tasks, while their applicability to unverifiable tasks (e.g., translation) is limited by the open-ended character of responses. As a result, self-evaluation mechanisms (e.g., self-judging and entropy minimization) are predominantly used to derive pseudo-labels. However, self-evaluation relying on LLMs typically incurs high computational overhead and introduces overconfidence issues due to intrinsic biases. To address these challenges, we propose a novel self-evaluation-free approach for unverifiable tasks, designed for lightweight yet effective self-improvement. Inspired by majority voting commonly employed in verifiable tasks, we propose semantic voting as a novel mechanism that relaxes the principle of hard matching (i.e., exact matching) toward soft matching (i.e., semantic similarity). Soft matching is achieved by leveraging a lightweight sentence embedding model to quantify semantic similarity, thereby mitigating excessive computational burden and intrinsic bias-associated limitations of self-evaluation. Comprehensive experiments demonstrate that our method achieves substantial gains in computational efficiency and overall better performance than self-evaluation methods across diverse model architectures and tasks.

Problem

• Reducing computational costs in LLM self-improvement for unverifiable tasks
• Eliminating overconfidence bias from self-evaluation mechanisms in LLMs
• Enabling efficient pseudo-labeling for open-ended tasks using semantic similarity

Innovation

• Semantic voting replaces hard matching with soft matching
• Uses lightweight sentence embedding for semantic similarity
• Eliminates self-evaluation to reduce computational overhead