References Improve LLM Alignment in Non-Verifiable Domains

• URL: http://arxiv.org/abs/2602.16802v1
• Date: Wed, 18 Feb 2026 19:03:34 GMT
• Title: References Improve LLM Alignment in Non-Verifiable Domains
• Authors: Kejian Shi, Yixin Liu, Peifeng Wang, Alexander R. Fabbri, Shafiq Joty, Arman Cohan,
• Abstract summary: We investigate whether reference-guided LLM-evaluators can bridge the gap by serving as soft "verifiers"<n>We show that a reference-guided approach substantially improves the accuracy of less capable LLM-judges using references from frontier models.<n>We show that reference-guided self-improvement yields clear gains over both direct SFT on reference outputs and self-improvement with reference-free judges.
• Score: 118.26447686644808
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: While Reinforcement Learning with Verifiable Rewards (RLVR) has shown strong effectiveness in reasoning tasks, it cannot be directly applied to non-verifiable domains lacking ground-truth verifiers, such as LLM alignment. In this work, we investigate whether reference-guided LLM-evaluators can bridge this gap by serving as soft "verifiers". First, we design evaluation protocols that enhance LLM-based evaluators for LLM alignment using reference outputs. Through comprehensive experiments, we show that a reference-guided approach substantially improves the accuracy of less capable LLM-judges using references from frontier models; stronger LLM-judges can also be enhanced by high-quality (i.e., human-written) references. Building on these improved judges, we demonstrate the utility of high-quality references in alignment tuning, where LLMs guided with references are used as judges to self-improve. We show that reference-guided self-improvement yields clear gains over both direct SFT on reference outputs and self-improvement with reference-free judges, achieving performance comparable to training with ArmoRM, a strong finetuned reward model. Specifically, our method achieves 73.1% and 58.7% on AlpacaEval and Arena-Hard with Llama-3-8B-Instruct, and 70.0% and 74.1% with Qwen2.5-7B, corresponding to average absolute gains of +20.2 / +17.1 points over SFT distillation and +5.3 / +3.6 points over reference-free self-improvement on AlpacaEval / Arena-Hard. These results highlight the potential of using reference-guided LLM-evaluators to enable effective LLM post-training in non-verifiable domains.

Related papers

• On Evaluating LLM Alignment by Evaluating LLMs as Judges [68.15541137648721]
  evaluating large language models' (LLMs) alignment requires them to be helpful, honest, safe, and to precisely follow human instructions.<n>We examine the relationship between LLMs' generation and evaluation capabilities in aligning with human preferences.<n>We propose a benchmark that assesses alignment without directly evaluating model outputs.
  arXiv  Detail & Related papers  (2025-11-25T18:33:24Z)
• LLM2: Let Large Language Models Harness System 2 Reasoning [65.89293674479907]
  Large language models (LLMs) have exhibited impressive capabilities across a myriad of tasks, yet they occasionally yield undesirable outputs.<n>We introduce LLM2, a novel framework that combines an LLM with a process-based verifier.<n>LLMs2 is responsible for generating plausible candidates, while the verifier provides timely process-based feedback to distinguish desirable and undesirable outputs.
  arXiv  Detail & Related papers  (2024-12-29T06:32:36Z)
• Bayesian Calibration of Win Rate Estimation with LLM Evaluators [20.588104799661014]
  We propose two methods to improve the accuracy of win rate estimation using large language models (LLMs) as evaluators. We empirically validate our methods on six datasets covering story generation, summarization, and instruction following tasks.
  arXiv  Detail & Related papers  (2024-11-07T04:32:40Z)
• LLM Self-Correction with DeCRIM: Decompose, Critique, and Refine for Enhanced Following of Instructions with Multiple Constraints [86.59857711385833]
  We introduce RealInstruct, the first benchmark designed to evaluate LLMs' ability to follow real-world multi-constrained instructions. To address the performance gap between open-source and proprietary models, we propose the Decompose, Critique and Refine (DeCRIM) self-correction pipeline. Our results show that DeCRIM improves Mistral's performance by 7.3% on RealInstruct and 8.0% on IFEval even with weak feedback.
  arXiv  Detail & Related papers  (2024-10-09T01:25:10Z)
• Measuring and Enhancing Trustworthiness of LLMs in RAG through Grounded Attributions and Learning to Refuse [27.26121507279163]
  We introduce Trust-Score, a holistic metric that evaluates the trustworthiness of LLMs within the RAG framework.<n>We propose Trust-Align, a method to align LLMs for improved Trust-Score performance.<n>26 out of 27 models aligned using Trust-Align substantially outperform competitive baselines on ASQA, QAMPARI, and ELI5.
  arXiv  Detail & Related papers  (2024-09-17T14:47:33Z)
• Systematic Evaluation of LLM-as-a-Judge in LLM Alignment Tasks: Explainable Metrics and Diverse Prompt Templates [11.948519516797745]
  We develop an open-source framework to evaluate, compare, and visualize the reliability and alignment of LLM judges.<n>Our results indicate a significant impact of prompt templates on LLM judge performance, as well as a mediocre alignment level between the tested LLM judges and human evaluators.
  arXiv  Detail & Related papers  (2024-08-23T11:49:01Z)
• Large Language Models are Inconsistent and Biased Evaluators [2.136983452580014]
  We show that Large Language Models (LLMs) are biased evaluators as they exhibit familiarity bias and show skewed distributions of ratings. We also found that LLMs are inconsistent evaluators, showing low "inter-sample" agreement and sensitivity to prompt differences that are insignificant to human understanding of text quality.
  arXiv  Detail & Related papers  (2024-05-02T20:42:28Z)
• Enabling Weak LLMs to Judge Response Reliability via Meta Ranking [38.63721941742435]
  We propose a novel cross-query-comparison-based method called $textitMeta Ranking$ (MR) MR assesses reliability by pairwisely ranking the target query-response pair with multiple reference query-response pairs. We show that MR can enhance strong LLMs' performance in two practical applications: model cascading and instruction tuning.
  arXiv  Detail & Related papers  (2024-02-19T13:57:55Z)
• LLMRec: Benchmarking Large Language Models on Recommendation Task [54.48899723591296]
  The application of Large Language Models (LLMs) in the recommendation domain has not been thoroughly investigated. We benchmark several popular off-the-shelf LLMs on five recommendation tasks, including rating prediction, sequential recommendation, direct recommendation, explanation generation, and review summarization. The benchmark results indicate that LLMs displayed only moderate proficiency in accuracy-based tasks such as sequential and direct recommendation.
  arXiv  Detail & Related papers  (2023-08-23T16:32:54Z)
• On Learning to Summarize with Large Language Models as References [101.79795027550959]
  Large language models (LLMs) are favored by human annotators over the original reference summaries in commonly used summarization datasets. We study an LLM-as-reference learning setting for smaller text summarization models to investigate whether their performance can be substantially improved.
  arXiv  Detail & Related papers  (2023-05-23T16:56:04Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.