SkillForge continuously verifies and evolves skills, improving RL agent performance; achieves +6.3% success rate over SkillRL in ALFWorld.

Shidong Yang, Ziyu Ma, Tongwen Huang et al.

AgentSpec enhances LLM agent inference efficiency via structure-isolated drafting and redundancy-aware budget allocation.

Xin Wang, Ziming Miao, Yi Zhu et al.

This study critically evaluates automatic creativity metrics and LLM-based judgments, revealing weak correlation with human assessments and systemic biases, highlighting evaluation limitations.

Alessandro Tutone, Giorgio Franceschelli, Mirco Musolesi

Introduces SWE Refactor Bench with a three-stage evaluation for long-horizon repository migration; only 5.4% success rate across 520 runs, highlighting current model limitations.

Deyao Hong, Yizhe Chi, Wenyi Li et al.

ConvergeFlow uses flow matching in continuous embedding space to provably converge to valid token embeddings, enabling direct token prediction without a cross-entropy decoder.

Na Li, Yuchen Jiao, Changxiao Cai et al.

Proposes a claim-based cross-lingual biography enrichment framework leveraging non-English Wikipedia, achieving 75.17 F1 in claim extraction and 86.2% claim coverage.

Yifei Song, Ziyang Chen, Emil Sayilov et al.

FormuEvo employs LLM-guided evolutionary algorithms to optimize MIP formulations, achieving up to 5.5× speedup over baselines.

Haofeng Yuan, Jianing Peng, Jieyi Bi et al.

Introduces Credal Large Language Models (CLLM) using LoRA ensembles to quantify uncertainty via credal sets, improving QA accuracy and hallucination detection.

Shireen Kudukkil Manchingal, Sofiia Nikolenko, Fabio Cuzzolin

DRY method adjusts logits at sampling to reduce loops by 47% and improve lexical diversity.

Philipp Emanuel Weidmann, Allen Roush, Judah Goldfeder et al.

XTC improves generation diversity by excluding top choices, achieving 15% Distinct-2 increase, 47% trigram repetition reduction.

Philipp Emanuel Weidmann, Allen Roush, Judah Goldfeder et al.

Kernel Token Contradiction (KTC) offers a fast, efficient LLM claim uncertainty quantification, achieving 8.2x speedup.

Jérémie Dentan, Alexi Canesse, Mahammed El Sharkawy et al.

Bulbul dataset supports dialectal Arabic speech recognition across 11 Arab countries.

Ahmed Ashraf, Aisha Alansari, Fadel Al Abbas et al.

GUI-Primitives benchmark reveals limitations in spatial reasoning of vision-language models, with max accuracy of 32%.

Md Abrar Jahin, Md Rizwan Parvez

Using LaBSE and other encoders, the study examines framing differences in religious and scientific concepts across Wikipedia's language editions.

Hung-Hsuan Chen

SAC-Copula uses Gaussian copula to construct smooth correlated Gumbel fields for quality-preserving watermarking in diffusion language models.

Baixin Li, Haiyun He

Test-time scaling for scientific equation discovery, search width is key, improves efficiency and performance.

Haowei Lin, Hubert Lim, Xiangyu Wang et al.

LiLiCorr improves speculative decoding acceptance length by 9-19% using lightweight likelihood correlation.

Matan Rusanovsky, Yoav Miron, Roy Uziel et al.

ConceptGuard uses dual-use concepts to evaluate context-sensitive unlearning, revealing limitations of current methods in contextual separation.

Sahil Kale, Ian Harris

G-CARL employs retrieval-grounded claim verification and checklist-aligned reinforcement learning to enhance medical report interpretation accuracy and patient-centricity.

Shiao Xie, Siyu Chen, Jianwei Lv et al.

Proposes IAR three-stage framework for retrieval-free document knowledge internalization, boosting domain QA and general skills.

Qian Kou, Xiaofeng Shi, Xiaosong Qiu et al.