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.

Proposes a cross-lingual fairness evaluation framework for watermarking schemes, revealing structural disparities across 11 languages with specific metrics and experiments.

Alexander Nemecek, Osama Zafar, Debargha Ganguly et al.

Training only the projector achieves strong multimodal performance in 3D MLLMs, avoiding language model drift.

Nyx Iskandar, Saathvik Selvan, Slater Victoroff

Using Falcon-7B-Instruct with FAISS-based RAG, achieving ROUGE-1 scores over 0.33 and outperforming Lead-3 in financial news summarization.

Pranav Chandaliya

SPADE employs self-play with a single LLM generating and optimizing executable environments, achieving an average 5.3% improvement across benchmarks.

Bo Liu, Simon Yu, Yiding Jiang et al.

WhiteMatter dynamically mixes all layer states into KV channels; 16-layer full-cache PPL reaches 19.968, 8.2% below vanilla.

Wenbo Zhang, Xiang Ren

OmniAlign unifies multilingual word and sentence alignment, excelling in long-text scenarios.

Mengpeng Yang, Jingxu Yang, Chao Chen et al.

Fine-tuning poetry data improves Arabic idiom understanding by 2.33%; cultural fine-tuning shows no significant effect.

Mena Attia, Mona Diab, Thamar Solorio

D2-ScaleAgent employs dual-dimensional scaling with Verifier-driven routing, achieving logical closure over evidence chains, outperforming traditional methods.

Hao Zhang, Longrong Yang, Lunhao Duan et al.

KV-Cache mismatch during model rollback causes state residuals, enabling information leakage via a causal cache effect.

Guijia Zhang, Harry Yang

MicroVerse employs long-horizon multi-agent simulations with a 'soul file' and multi-layer memory to quantify identity drift, validated through threshold robustness tests.

Sky Ng, Brihi Joshi, Ishan Gupta et al.

Proposes associative and declarative tasks to probe model year encoding; finds mechanisms differ, with limited updateability.

Suze van Adrichem, Aditi Bhaskar, Diyi Yang et al.

This paper uncovers the 'curse of ambiguity' in large language models, analyzing its roots in model capacity, embedding size, and learning difficulty, highlighting practical limitations.

Nicolas Zucchet, Hyun Dong Lee, Scott Linderman

This study compares 192 artificial and natural languages, revealing that language models develop left-branching bias artificially but favor right-branching SVO in natural languages, driven by data resources.

Varvara Arzt, Allan Hanbury, Terra Blevins

Holistic evaluation of 11 memory substrates across models and tasks reveals task-dependent performance and scalability trade-offs, advocating multi-substrate routing.

Wei-Chieh Huang, Weizhi Zhang, Yuchen Wu et al.

This survey reviews decoding methods for LLMs and LVLMs, emphasizing contrastive, guided, and parallel paradigms that improve quality, control, and efficiency.

Haoran Wang, Xiongxiao Xu, Philip S. Yu et al.

SimpleOPD enables cross-tokenizer long-context reasoning transfer, boosting mathematical proof performance by 21.2 points on ProofBench.

Haonan He, Haodi Lei, Yun Luo et al.

RippleMem uses event-centric memory graphs and associative recall, improving accuracy by 11.87% and reducing graph construction costs by 30x.

Jingbo Ji, Lingyi Li, Xilong Cheng et al.

Decoupled Contrastive Decoding (DCD) accelerates generation by using an expert-aligned lightweight proposer and only applying amateurs in verification, achieving 1.65-1.95× speedup.

Zhixuan Liu, Zhichen Dong, Yuanfu Wang et al.