Pith. sign in

REVIEW 4 cited by

Your Language Model May Think Too Rigidly: Achieving Reasoning Consistency with Symmetry-Enhanced Training

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2502.17800 v1 pith:CT6LME76 submitted 2025-02-25 cs.CL

Your Language Model May Think Too Rigidly: Achieving Reasoning Consistency with Symmetry-Enhanced Training

classification cs.CL
keywords reasoningquerymodelvariationsacrossapproachaugmentationimproves
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Large Language Models (LLMs) have demonstrated strong reasoning capabilities across various tasks. However, even minor variations in query phrasing, despite preserving the underlying semantic meaning, can significantly affect their performance. To address this, we focus on enhancing LLMs' awareness of symmetry in query variations and propose syMmetry-ENhanceD (MEND) Data Augmentation, a data-centric approach that improves the model's ability to extract useful information from context. Unlike existing methods that emphasize reasoning chain augmentation, our approach improves model robustness at the knowledge extraction stage through query augmentations, enabling more data-efficient training and stronger generalization to Out-of-Distribution (OOD) settings. Extensive experiments on both logical and arithmetic reasoning tasks show that MEND enhances reasoning performance across diverse query variations, providing new insight into improving LLM robustness through structured dataset curation.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Partial exploiters sustain cooperation: the hump-shaped strategy stably coexists with unconditional cooperators

    physics.soc-ph 2026-04 unverdicted novelty 7.0

    Hump-shaped partial exploiters stably coexist with unconditional cooperators to exclude defectors and sustain cooperation in agent-based models of social dilemmas.

  2. CAP-CoT: Cycle Adversarial Prompt for Improving Chain of Thoughts in LLM Reasoning

    cs.AI 2026-04 unverdicted novelty 6.0

    A cycle of solver, adversarial flawed-chain challenger, and feedback agent optimizes CoT prompts to raise accuracy and cut run-to-run answer variability within two to three cycles.

  3. Partial exploiters sustain cooperation: the hump-shaped strategy stably coexists with unconditional cooperators

    physics.soc-ph 2026-04 conditional novelty 5.0

    CAP-CoT improves LLM reasoning accuracy and stability by iteratively refining solver prompts via contrast with adversarially generated flawed reasoning chains.

  4. CAP-CoT: Cycle Adversarial Prompt for Improving Chain of Thoughts in LLM Reasoning

    cs.AI 2026-04 unverdicted novelty 5.0

    CAP-CoT uses iterative adversarial prompt cycles to improve CoT accuracy, stability, and robustness across six benchmarks and four LLM backbones.