Study finds LLM-generated skills do not significantly improve performance in data science workflows.

Wei-Jung Huang

EvoSOP enables self-evolving LLM agents by iteratively synthesizing atomic actions into reusable SOPs, boosting success rates by 3-13% and reducing interaction rounds.

Haipeng Ding, Yuexiang Xie, Zhewei Wei et al.

Proposes a concept-based interpretability framework using unsupervised dictionary learning to improve transparency and performance of end-to-end autonomous driving models.

Franz Motzkus, Sebastian Bernhard

TurnOPD employs adaptive turn-depth control and progressive loss normalization to enhance long-horizon agent distillation efficiency, improving validation accuracy.

Yuhang Zhou, Kai Zheng, Haoling Li et al.

Proposes Heading-Specific Activation Steering to causally control tool invocation in five open-source models, validated through geometric and causal analysis.

Yuqi Chen, Vincent Siu, Yang Liu et al.

Proposes in-loop memory for language agents, reducing latency to 100μs and improving task efficiency and accuracy.

Yusuf Khan, Carlo Lipizzi

This study introduces M3Bench, a benchmark for evaluating model editing in medical VLMs, measuring reliability, locality, and generalization across 16,276 clinical questions.

Guli Zhu, Chenwei Wu, Liyue Shen

MetaSkill-Evolve introduces recursive self-improvement via two-timescale meta-skill evolution, boosting task accuracy by +23.54 points on OfficeQA.

Zefeng Wang, Minxi Yan, Jinhe Bi et al.

EvoAgentBench evaluates agent self-evolution via ability transfer across four domains, highlighting limitations of automatic methods.

Xingze Gao, Chuanrui Hu, Hongda Chen et al.

Proposes LLM-as-a-Tutor framework using pairwise comparison and atomic constraints to enhance non-verifiable RL.

Yujin Kim, Namgyu Ho, Sangmin Hwang et al.

The study explores GUI agents' reliance on pixels vs. structure, introducing the Perception-Fusion Gap metric.

Guijia Zhang, Yuxun Chen, Yuheng Qi et al.

Agentic IoT transforms IoT into cognitive systems via autonomous AI agents, enhancing real-time reasoning and adaptive planning.

Rümeysa Hilal Sevinç, Bahaeddin Türkoğlu, İbrahim Kök

BRAID models multi-modal reasoning as a unified MDP, jointly optimizing text and image generation, boosting multi-turn reasoning performance.

Zican Hu, Xuyang Hu, Yiming Liu et al.

Proposed cross-survey transfer framework; zero-shot LLMs achieve 52% accuracy on unseen items, nearing supervised models.

Chan-Tung Ku, Chan Hsu, Pei-Cing Huang et al.

Proposes Raven-Agent, a modular trading layer for prediction markets, achieving +15.9% risk-adjusted return in replay tests.

Yishu Wang, Yuxuan Wang, Jiaqi Deng et al.

Oyster-II uses Zero-RL and SERL for constructive safety, raising Chinese long-query safety by 14.65% over joint training.

Jiyang Guan, Yong Xie, Jun Chen et al.

Proposes EXT++ and PPR to enhance hypergraph RAG, significantly improving fact extraction and chunk retrieval.

Houda Khrouf, Pedro Fillastre, Sebastiao Correia

EvoPolicyGym benchmarks autonomous policy evolution; GPT-5.5 achieves top performance across 16 RL environments with limited interactions.

Zhilin Wang, Han Song, Runzhe Zhan et al.

ScopeEdit employs a dual-branch scope-aware mechanism with orthogonal low-rank geometry to control knowledge propagation in online multimodal model editing.

Siyuan Li, Youyuan Zhang, Ruitong Liu et al.

SkillCoach improves agentic skill-use evaluation with self-evolving rubrics, significantly enhancing assessment quality.

Jiayin Zhu, Kelong Mao, Yudong Guo et al.