Analyzed 14.8 million prompts to assess how different demographic cues affect LLM responses, revealing inconsistent effects across cues.

Manuel Tonneau, Neil K. R. Seghal, Niyati Malhotra et al.

RPEval benchmark reveals personalized memory's impact on LLM behavior; RP-Reasoner significantly reduces irrational personalization.

Xueyang Feng, Weinan Gan, Xu Chen et al.

Deep binding of virtual personas with rich backgrounds and time grounding enhances LLMs' accuracy in social behavior simulation, matching human data closely.

Suhong Moon, Minwoo Kang, Joseph Suh et al.

CorpusQA benchmark challenges LLMs' global reasoning over 10 million tokens.

Zhiyuan Lu, Chenliang Li, Yingcheng Shi et al.

Proposes execution-guided automated AI research using evolutionary search and RL, achieving 69.4% success rate and reducing pretraining time to 19.7 min.

Chenglei Si, Zitong Yang, Yejin Choi et al.

Habibi introduces an open-source multi-dialect Arabic TTS framework using multi-step data curation and curriculum learning, covering 12+ dialects with performance surpassing specialized models.

Yushen Chen, Junzhe Liu, Yujie Tu et al.

CommunityBench evaluates community-level model alignment, revealing LLMs' limitations in modeling community preferences.

Jiayu Lin, Zhongyu Wei

This paper redefines tokenization as a core model design decision, proposing a context-aware co-design framework with standardized evaluation metrics.

Sawsan Alqahtani, Mir Tafseer Nayeem, Md Tahmid Rahman Laskar et al.

WorldMind aligns agentic world models via experiential learning, achieving 48.0% success rate on EB-ALFRED.

Baochang Ren, Yunzhi Yao, Rui Sun et al.

LongPAS method enhances long-context reasoning, significantly outperforming RLVR baselines.

Miao Peng, Weizhou Shen, Nuo Chen et al.

DPWriter combines semi-structured long Chain-of-Thought with diversity-guided RL, achieving over 10% improvement in output diversity metrics on creative writing benchmarks.

Qian Cao, Yahui Liu, Wei Bi et al.

Utilizing perspectivist modeling to analyze annotator disagreement in NLP, enhancing model fairness.

Yinuo Xu, David Jurgens

Introduces adversarial attack-based evaluation of role consistency in Vicuna LLMs for virtual counseling.

Eric Rudolph, Natalie Engert, Jens Albrecht

Proposes hyperspherical contrastive learning to embed emotions in a circular structure, improving interpretability and psychological alignment.

Yusuke Yamauchi, Akiko Aizawa

Proposes a unified attribution pipeline with four modules to mitigate hallucinations in RAG systems, significantly improving response fidelity.

Yuqing Zhao, Ziyao Liu, Yongsen Zheng et al.

AlignXplore+ uses text as a universal interface for cross-task personalization, outperforming larger models on nine benchmarks with an 8B model.

Yuting Liu, Jian Guan, Jia-Nan Li et al.

Proposes StealthGraph, using knowledge graphs to generate implicit harmful prompts; employs dual rewriting strategies to enhance stealth, achieving 84.92% attack success.

Huawei Zheng, Xinqi Jiang, Sen Yang et al.

Proposes TabTrim, transforming table pruning from sequential revisions to gold trajectory-supervised parallel search, boosting TableQA accuracy by 3.2%.

Yu Guo, Shenghao Ye, Shuangwu Chen et al.

This study evaluates whether LLM self-explanations improve users' ability to predict model behavior via counterfactual perturbations, showing consistent gains especially with pragmatic strategies.

Pingjun Hong, Benjamin Roth

O-Researcher uses multi-agent distillation and RL to generate high-quality research data, boosting open-source LLMs' performance.

Yi Yao, He Zhu, Piaohong Wang et al.