Proposed a cost-aware single-generation architecture achieving 91.7% accuracy on the DevRev benchmark.

Yoga Sri Varshan Varadharajan, Ajay Yadav, Ritesh Goru et al.

This study reveals instability in black-box LLM observers on shared endpoints, validating that fixed thresholds often fail in real-world settings.

Haoyaun Zhu, Jie Zhang

Applying Ostrom’s governance principles, analyzing 100-agent research group’s cheating and whistleblowing dynamics.

Davide Paglieri, Logan Cross, Tim Genewein et al.

Proposes SIGNBALANCE to eliminate spurious advantage in GRPO, improving math and search task performance.

Jiamian Wang, Samyadeep Basu, Koustava Goswami et al.

Discriminative world models trained via predicted-state matching improve web agent decision-making and task success.

Kelvin Li, Dhruv Pendharkar, Anish Pahilajani et al.

AICOME framework uses AI-generated respondent-level measures to recover individual and group effects, validated on CFPS data.

Wenxin Jiang, Xuyang Wang, Yuxiao Wu

SafeEvolve co-evolves harness and policy via trajectory experience, significantly enhancing safety and utility.

Qinghua Mao, Wanying Qu, Dadi Guo et al.

Measurement-driven sub-network selection using hardware-aware multi-objective optimization enables efficient deployment of compressed LLMs on industrial edge devices, recovering 95% of unpruned quality.

Vasileios Rizeakos, Georgios Paisios, Alexandros Machairas et al.

SALA framework uses dynamic time warping for semantic alignment, enhancing complex reasoning in in-context learning by 5% on average.

Zhao Ji, Wenqing Chen, Zhixuan Chu et al.

Introduces SCILAWS-BENCH, a real-data benchmark evaluating LLMs' ability to discover scientific laws, covering 118 problems, 291 candidate laws, and 8M data points.

Yiming Huang, Ziche Liu, Zhuohang Wu et al.

EdiTikZ leverages DaEdiTikZ dataset and RL to achieve scientific figure editing; 9B model outperforms GPT-5.6-Sol.

Christian Greisinger, Zhixue Zhao, Steffen Eger

FigTree uses recursive SVG program generation to improve scientific figure editability and quality.

Yepeng Liu, Dasen Dai, Chengzhi Liu et al.

Introduces invalidation contracts to enhance cache validity and compliance with version stamps and cacheability hints.

Michael Wu, Arquimedes Canedo

ReDeck enhances slide generation with step-level render feedback, improving spatial accuracy and design quality.

Muzhao Tian, Zezi Zeng, Yifan Yang et al.

Proposes adaptive data cracking for LLMs, reducing reasoning costs by over 50% through dynamic structure extraction.

Milad Rezaei Hajidehi, Qitong Wang, Stratos Idreos

Introduced Answer Probing-Guided Tree Search (APTS) to enhance solution diversity across three reasoning tasks.

Yi Fang, Que Shen, Chengpeng Li et al.

RACER combines reinforcement learning and four-agent collaboration, improving KG-enhanced reasoning by about 5% on average.

Yuwei Lou, Hao Hu, Yuzhou Jiang et al.

GCJR recovers minimal repair families with 1.000 FEM and cuts replay by 55.1% on 90 controlled cases.

Bingjie Li, Yumeng Song, Zhongming Yao et al.

Logos extends spatiotemporal composability to multi-process systems using append-only transcripts, ensuring fault tolerance and reversibility.

Hanzhang Jia, Liheng Zeng, Hao Cheng et al.

InstructMesh employs latent-space region operations combined with multimodal interaction to enable non-experts to selectively repair generative 3D models for fabrication.

Faraz Faruqi, Ahmed Katary, Demircan Tas et al.