arxiv arXiv cs.CL · 61d ago · research

ChLogic: Testing Logical Reasoning Robustness in Chinese Expressions

from English

ChLogic evaluates how well large language models maintain logical reasoning when English logical structures are expressed in Chinese. It reveals a persistent English-Chinese performance gap, with back-translation improving results on general items but harming performance on difficult problems. The benchmark highlights the impact of surface realization, translation artifacts, and model-specific behaviors on multilingual reasoning.

Importance 2/3 arXiv cs.CL Alibaba (Qwen) Mistral AI Zhipu AI Evaluation & benchmarks Multimodal Reasoning models

Read original

arxiv arXiv cs.LG · 59d ago

De-biased VLM-as-3D-Judge Protocol for Furniture Generation

A de-biased VLM-based judge protocol specializes TRELLIS on furniture generation using lightweight adaptation. The protocol addresses failure modes like image overload and geometry-hiding, with calibration showing 0.83–1.0 win rates and base-vs-base symmetry at 0.5. Among six adaptation methods, conditioner repair under severe degradation achieves parity with the base model, while no method exceeds a 65% win-rate target.

arxiv arXiv cs.CL · 61d ago · 1 view

Visuals Lie, Consistency Speaks: Disentangling Spatial Attention from Reliability in Vision-Language Models

A study challenges the assumption that visual attention signals reliability in vision-language models. It finds near-zero correlation between spatial attention and accuracy, showing instead that self-consistency across reasoning paths is a stronger predictor of truth. Reliability is better explained by generation dynamics and internal state distributions, not visual attention patterns.

arxiv arXiv cs.AI · 54d ago

LLMs Benchmarked for Web Vulnerability Detection

A study evaluates six LLMs on detecting real-world web vulnerabilities in WordPress plugins, finding detection rates vary by model and prompt design. Claude Opus 4.6 achieved the highest detection rate at 63%, while Qwen 3.5 only reached 35%, and no model consistently identified all baseline vulnerabilities across iterations.

arxiv arXiv cs.AI · 54d ago · 2 views

Benchmark Evaluation of Small Language Models for Arabic NLP

A benchmark of 240 Arabic test items across eight domains and ten skills assesses twelve small language models in zero-shot settings. Gemma 3 (12B) achieved the highest overall score (4.548/5), followed by Aya and C4AI Command Arabic, with performance linked more to Arabic alignment and instruction-following than model size. Common failure modes include prompt leakage, hallucination, and weak task adherence.

arxiv arXiv cs.CL · 55d ago · 6 views

Test-Time Steering Resolves Temporal Fact Conflicts in LLMs

Researchers identify parametric temporal conflicts in language models where outdated facts persist in parameters. They introduce Temporal Attractor Steering (TAS), a test-time method that resolves 29-57% of such conflicts without retraining, maintaining 85-99% accuracy on non-conflict queries and outperforming a baseline on three of four models.