SkillTV-Bench benchmarks trajectory verification, improving accuracy by 14.8 points via automated JudgeSkill evolution.

Zhi Han, Chenxi Zeng, Liuhaichen Yang et al.

SCP-NL2TL combines semantic verification and conformal risk control to abstain from unsafe NL-to-temporal-logic translations.

Yixuan Wang, Licheng Luo, Yu Fu et al.

This paper applies multidimensional Item Response Theory (IRT) to analyze 192 language models across 8 safety benchmarks, revealing three core latent abilities and enabling cost-effective evaluation and auditing.

Joshua Fonseca Rivera, Neil Shah, David Demitri Africa et al.

Proposes TriQua framework, combining base triples with qualifiers for fine-grained factuality evaluation, achieving high correlation (r=0.89) with human scores.

Jin Liu, Steffen Thoma, Achim Rettinger

ABD method achieves 59.43% accuracy on LiveCodeBench-v6, outperforming Single9 and HAC.

Ruitong Li, Binjie Guo, Aisheng Mo et al.

Proposed FinProBench with RGRC extracts role-based standards from real deliverables, boosting financial AI evaluation efficiency.

Ben Wang, Kang Zhou, Lifan Guo et al.

Proposes black-box dispersion–revision diagnostic using CI to verify if output diversity correlates with genuine epistemic revision.

Molood Arman

LLaDA MoE v2 scales mixture-of-experts diffusion models, trained on 23.5T tokens, nearing Qwen3 performance.

Fengqi Zhu, Shaoxuan Xu, Jingyang Ou et al.

Proposes CSP, a complex-valued state-space model relying solely on state propagation, achieving 100% accuracy on deterministic tasks.

Xiaohe Li, Yang Lu

Proposes DocTrace, a hierarchical evidence graph reasoning model for LongDocVQA, achieving 14.4% accuracy improvement and explicit evidence provenance.

Le Xiang, Zhicheng Guan, Hong Chen et al.

ParEvalLayer combines coverage and bootstrap tests; three benchmarks match full-evaluation decisions after only 15–25% of tasks.

Wei-Jung Huang, Bonan Shen

Learn feasible inventory policies via differentiable projection, achieving an average optimality gap below 1%.

Patrick Helm, Jan-Niklas Doerr, Joren Gijsbrechts et al.

Introduces IBA-Bench for implicit behavior alignment, combining deep retrieval and trajectory-level synthesis, significantly improving personalized LLM agent performance.

Jiajia Song, Bobo Li, Haiwen Yi et al.

HPFA uses hypergraphs and paired trajectories to efficiently localize LLM reasoning failures, achieving 64.6% accuracy.

Runchuan Zhu, Hongbin Lai, Bowen Jiang et al.

Proposes PCSD, leveraging local persistent teacher signals to improve self-distillation in RL, achieving +15.6% success rate on ALFWorld.

Chunji Lv, Yangguang Wei, Junlin Liu et al.

SearchMaster improves search agent accuracy from 38.19% to 51.52% through self-play.

Wentao Tan, Qiong Cao, Jiaqi Wang et al.

Proposed CoNav-UAV models dual-altitude UAV cooperation as a Stackelberg game, achieving up to 30.8 success rate improvement.

Junru Song, Wenhao Zhang, Yang Yang et al.

DAPD significantly improves privilege illusion by dual-path and dual-source anchoring, with an average gain of 2.00 points.

Jianyu Wu, Yizhou Wang, Encheng Su et al.

PNPO replaces cumulative importance ratios with prefix geometric mean, improving off-policy correction in large language model RL.

Wenhao Zhang, Yibo Xie, Rui Wang et al.

KoVRE model enhances Korean document retrieval efficiency using bilingual supervision and hard-negative mining.

Yongbin Choi, Gyuho Shim, Youngjoon Jang