Using a capability-based framework, the study assesses how close generative world models are to being true simulators, highlighting gaps in physical laws and state feedback.

Tong Wang, Huan Deng, Mucheng Yang et al.

GSAR combines self-evolving data synthesis and goal-state anchoring to improve GUI agent training with over 90% accuracy in trajectory verification.

Long Zhang, Yuhan Chen, Chaoran Zhang et al.

VIALS benchmark evaluates scientific artifact interpretation, with top models achieving only 26.5% accuracy, highlighting major gaps.

Elaine Lau, Thanuka Udumulla, Lee Izhaki-Tavor et al.

AUSO unifies skill internalization and utilization via action-aware progressive reinforcement learning, improving long-horizon task performance.

Huizu Lin, Chengkai Huang, Tianqi Gao et al.

CLEAR uses input-conditioned latent gating to reduce HarmBench ASR from 32.3% to 0.5%, maintaining high utility on GSM8K and other tasks.

Chengxiao Wang, Enyi Jiang, Xiaojing Liao et al.

GPU fine-grain parallelization of GPX partitioning accelerates large-scale TSP solving by 48-625×, reducing memory use significantly.

Swetha Varadarajan, Darrell Whitley

Proposed PSL framework combines semi-structured political records' stance and structural signals, significantly improving predictive political QA.

Yinan Liu, Zihan Zhou, Zichun Jin et al.

COTA enhances runtime intervention in LLM agents by comparing alternatives, improving performance in environments like WebShop.

Yanze Jiang, Mingxuan Li, Yuhao Wang et al.

UpgradeBench offers a decision-centric benchmark for upgrading LLM specialists, with a mean quality regret of 0.37pp.

Ye Chen, Weining Zhang

Proposes ingest-time semantic compilation (ISC) with maintained semantic index, outperforming query-time interpretation in accuracy and cost efficiency.

Kyle Wild, Yusuke Takahashi, Asako Uraki

This paper systematically evaluates diffusion-based block-parallel decoding in multimodal models, highlighting potential speedups up to 3.6× with current architectures.

Yantao Li, Huanlin Gao, Fang Zhao et al.

Proposed a multi-agent framework integrating conversational data collection, structured processing, and large language model prediction for weather-sensitive travel behavior, achieving up to 71.5% accuracy.

Narges Ahmadi, Yubo Jiao, Jônatas Augusto Manzolli et al.

AI4AI-Bench evaluates LLM agents' recursive self-improvement in training algorithms; mean score 0.166, top 0.250, across 10 algorithm families.

Yizhe Chi, Wenyi Li, Deyao Hong et al.

Introduces null-model based exact tests and multi-criteria trajectory analysis to reliably measure genuine self-improvement in language models, avoiding measurement artifacts.

Cheng Xu, Nan Yan, Liming Chen et al.

This study compares task-level and subtask-level skill induction, finding subtask and text-format skills transfer more reliably, and introduces a skill utility score.

Yiyang Feng, Biddut Sarker Bijoy, Niranjan Balasubramanian et al.

Using XGBoost with first 5-minute trading data, predict Solana memecoin rug pulls within 1 hour.

Jianghai Li, Pavel Kuznetsov, Yury Yanovich et al.

Proposes ADAPT, a physics-informed diffusion-based indoor model, achieving 7.3% energy savings and 30.2% occupant comfort improvement.

Xu Yang, Kailai Sun, Dianyu Zhong et al.

Proposes telemetry-guided adaptive compression for edge RAG, reducing GPU energy by up to 53.2% and SoC energy by 48.2%, with negligible quality loss.

Zlatan Feric, Amir Taherin, Yanzhi Wang et al.

Empirical analysis of 1,338 LLM post-training trajectories reveals strategies are locked in early, lacking in-execution self-reassessment mechanisms, limiting AI self-improvement.

Joy Jia Yin Lim, Xin Huang, Hao Peng et al.

Verifiable abstention makes AI leak diagnosis accountable, with 214 correct actions out of 550 events.

Tianwei Mu, Yue Wang, Mingzhe Yuan et al.