mlsanityData to 7 Oct 2026

Pseudo-Deliberation in Language Models: When Reasoning Fails to Align Values and Actions

arXiv 2605.09893 ↗First posted 11 May 2026cs.CL cs.AIcited by 0

On the map: LLM reasoning → Reasoning benchmarks and evaluation

Abstract (arXiv)

Large language models (LLMs) are often evaluated based on their stated values, yet these do not reliably translate into their actions, a discrepancy termed "value-action gap." In this work, we argue that this gap persists even under explicit reasoning, revealing a deeper failure mode we call "Pseudo-Deliberation": the appearance of principled reasoning without corresponding behavioral alignment. To study this systematically, we introduce VALDI, a framework for measuring alignment between stated values and generated dialogue. VALDI includes 4,941 human-centered scenarios across five domains, three tasks that elicit value articulation, reasoning, and action, and five metrics for quantifying value adherence. Across both proprietary and open-source LLMs, we observe consistent misalignment between expressed values and downstream dialogues. To investigate intervention strategies, we propose VIVALDI, a multi-agent value auditor that intervenes at different stages of generation.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning. Builds on 9 of its 37 references that are in the map. First posted on arXiv 11 May 2026.

Builds on

9 of 37 references.

Closest ancestors · 3

Same direction, newest first

  1. 2026Social-R1: Towards Human-like Social Reasoning in LLMs10 Mar
  2. 2025Are they lovers or friends? Evaluating LLMs' Social Reasoning in English and Korean Dialogues21 Oct
  3. 2025Qwen3 Technical Report14 May

Other references · 6

  1. 2024GPT-4o System Cardcited by 4,467
  2. 2022Constitutional AI: Harmlessness from AI Feedbackcited by 2,950
  3. 2023DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelinescited by 803
  4. 2023Towards Measuring the Representation of Subjective Global Opinions in Language Modelscited by 424
  5. 2025Will AI Tell Lies to Save Sick Children? Litmus-Testing AI Values Prioritization with AIRiskDilemmascited by 17
  6. 2026ValueFlow: Measuring the Propagation of Value Perturbations in Multi-Agent LLM Systemscited by 4

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Context-Value-Action Architecture for Value-Driven Large Language Model Agents
  2. 2025VAL-Bench: Belief Consistency as a measure for Value Alignment in Language Models
  3. 2025Beyond Single-Sentence Prompts: Upgrading Value Alignment Benchmarks with Dialogues and Stories
  4. 2024Are the Values of LLMs Structurally Aligned with Humans? A Causal Perspective
  5. 2024Towards Dialogues for Joint Human-AI Reasoning and Value Alignment
  6. 2026Beyond Value Benchmarks: Measuring Value-Structure Alignment in Large Language Models via Symmetric Q-Sorts
  7. 2026Do LLMs Have Values? A Quantitative Analysis and Alignment Framework for Values in Large Language Models
  8. 2026Toward a Theory of Value in AI Alignment