aiqbench analysis
The paper argues that alignment evaluations usually check only whether a model's outputs comply with stated values, and misses whether the same model judges its own outputs consistently. It introduces a benchmark and data-construction method that pairs generation with evaluation under identical principles, and reports that many models with compliant outputs still give inconsistent judgments. The mechanistic probing and representation-level intervention are early-stage results, so readers should check how well the gains transfer beyond the authors' benchmarks.
VALCON: Benchmarking and Enhancing Value Consistency in Large Language Models via Dual-Role Variation
arXiv:2610.21304v1 Announce Type: new Abstract: Current value alignment evaluations for large language models (LLMs) heavily rely on output compliance, overlooking whether LLMs maintain value consistency when alternating between generator and evaluator roles. Such consistency is increasingly important because modern LLM agents and alignment pipelines often rely on the same LLM to both generate actions and evaluate their acceptability. To mitigate this limitation, we propose VALCON, a role-driven data construction framework that operationalizes value consistency into three benchmark suites (VAB-Consistency, VAB-Boundary, VAB-Challenge) via dual-role variation. By systematically pairing an LLM's generative behavior with its evaluative judgments under identical principles, VALCON synthesizes target evaluation data that exposes hidden value discrepancies. Experiments on 28 LLMs reveal pervasive cross-role inconsistency: models with highly compliant outputs frequently fail to align their judgments with their own generations. To understand and address this gap, we perform mechanistic probing that uncovers a role-dependent representational decoupling associated with this behavioral asymmetry. Guided by these insights, we propose a representation-directed intervention that effectively mitigates consistency failures. Our work establishes value consistency as a distinct alignment dimension and provides a complete evaluation-analysis-mitigation pipeline. Code and data are available at https://github.com/tjunlp-lab/VALCON.
Abstract:Current value alignment evaluations for large language models (LLMs) heavily rely on output compliance, overlooking whether LLMs maintain value consistency when alternating between generator and evaluator roles. Such consistency is increasingly important because modern LLM agents and alignment pipelines often rely on the same LLM to both generate actions and evaluate their acceptability. To mitigate this limitation, we propose VALCON, a role-driven data construction framework that operationalizes value consistency into three benchmark suites (VAB-Consistency, VAB-Boundary, VAB-Challenge) via dual-role variation. By systematically pairing an LLM's generative behavior with its evaluative judgments under identical principles, VALCON synthesizes target evaluation data that exposes hidden value discrepancies. Experiments on 28 LLMs reveal pervasive cross-role inconsistency: models with highly compliant outputs frequently fail to align their judgments with their own generations. To understand and address this gap, we perform mechanistic probing that uncovers a role-dependent representational decoupling associated with this behavioral asymmetry. Guided by these insights, we propose a representation-directed intervention that effectively mitigates consistency failures. Our work establishes value consistency as a distinct alignment dimension and provides a complete evaluation-analysis-mitigation pipeline. Code and data are available at this https URL.
24 pages, 16 figures. Accepted to Findings of the Association for Computational Linguistics: EMNLP 2026. Code and data: this https URL
Computation and Language (cs.CL)
arXiv:2610.21304 [cs.CL]
(or arXiv:2610.21304v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2610.21304 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 302 words
Related coverage
Links #3: 2026/06 Part 1
LessWrong · Jun 15, 2026
Stepwise Reasoning Enhancement for LLMs via External Subgraph Generation
arXiv · Jun 4, 2026
Digital States 2026: AI Moves From Experiment to Infrastructure
govtech.com · Sep 24, 2026
TAHB: A Comprehensive Benchmark for Text-Attributed Hypergraph Learning
arXiv · Aug 18, 2026