Proposes human-AI system (HAS) framework for scientific teams, emphasizing dynamic feedback and collaboration to mitigate risks and enhance innovation.

Patrick Emami, Sameera Horawalavithana, Truc Nguyen et al.

PMMC shifts multimodal long-term memory from query-time retrieval to consolidation, improving answer accuracy and efficiency by 3.2-4.0 points on benchmarks.

Jingyu Sun, Yan Lin, Yuyang Xue et al.

Introducing 'isotropy cliffs' as a geometric signature of decision-making in LLMs; strong correlation (r=0.84) with downstream accuracy.

Okan S. Coskun, Florian Rottach, Carsten Eickhoff et al.

AgentSLABench is a resource-aware evaluation framework for autonomous AI agents, integrating correctness with latency, cost, memory, and network metrics.

Meher Bhaskar Madiraju, Meher Sai Preetam Madiraju

Proposed FinDeepIndicator benchmark evaluates deep research agents' four-stage end-to-end financial indicator construction, revealing significant gaps in data retrieval and numerical execution.

Chaoqun Yang, Fengbin Zhu, Xinyu Lin et al.

Proposes DGA₂D, a directed graph-guided framework for automated algorithm design, reducing normalized gaps by up to 10.96 percentage points across 12 COPs.

Jiale Zhao, Zimu Chen, Sirui Mao et al.

Proposed geometric self-supervised pretraining enhances neural TSP solvers, achieving 7.23% tour length improvement on large-scale instances.

David Aguado, Daniel Fuertes, Carlos R. del-Blanco et al.

Introduced CoCo method for expert model interpretation via contribution contrast, enhancing interpretability and accuracy.

Yifan Wang, Jinyi Mu, Mayank Jobanputra et al.

Decode-Branch Transformer decouples prompt processing from decoding, reducing inference costs and improving performance.

Liming Liu, Mingze Wang, Tuo Zhao

Q-CARE introduces query coverage and claim verifiability for reference-free, fine-grained RAG evaluation, outperforming existing metrics.

Jeonghwan Choi, Taewon Yun, Minjeong Ban et al.

WILC leverages complementarity in LLM collaboration, matching GPT-5.2 performance at 7× lower cost.

Yanbin Fang, Xuan Wei, Wei Chen

Study on inference-time scaling in local CUAs shows diminishing returns with increased computation.

Woongkyu Lee, Jungwook Choi

The study reveals benchmark scoring errors for computer-use agents, proposing a four-stage reliability framework with 15.3% incorrect failure verdicts.

Zihan Dong, Zhiyuan Ma, Zekun Wang et al.

SKIMIX integrates multi-agent skill retrieval, anti-dilution routing, and adaptive evolution, boosting open-ended reasoning by 33%.

Jia Luo

Proposed STEREODISCO framework applies semantic differential method to discover and quantify stereotypical axes in LLM internal representations, revealing new bias dimensions beyond prior social psychology research.

Farane Jalali Farahani, Corina Dima, Mojtaba Nayyeri et al.

SpatialCLI employs a three-stage framework to internalize spatial perception, boosting VLM reasoning from 29.3% to 84.6% accuracy using tools, surpassing GPT-5.6 Sol.

Yang Zhou, Zixuan Huang, Sunzhu Li et al.

GAMER constructs an Action-Centric Graph with dual-stream TD learning, boosting inference success by 20.81%.

Xu Zheng, Chaohao Lin, Zhuomin Chen et al.

Introduces MultivationBench, a benchmark based on Maslow and Reiss models, to evaluate multimodal sequential motivation reasoning; models perform poorly.

Kawai Chung, Chunkit Chan, Yauwai Yim et al.

HiSkill enhances LLM agents with hierarchical skill graphs, achieving a 17.33% success rate increase and 78.75% reduction in inference token consumption.

Yu Hao, Jinxuan Cai, Qi Zhang et al.

Using a video transformer encoder, the study explores concept bottleneck models in echocardiography, finding volume supervision reduces errors.

Hyunkyung Han, Min Jung Kim