Prépublication
· 2026 arXiv
EN
Reinforcement learning (RL) is an effective way to strengthen the long-context reasoning of large language models (LLMs), but it needs training data that are long, diverse, challenging, and verifiable. Human annotation of such data is expensive, naturally occu...
Zhennan Wan, Jianfei Chen
#cs.CL
Prépublication
· 2026 arXiv
EN
Through a tangle of real-world interactions, from synthetic training pipelines to data shared on the open web, frontier LLMs absorb what their predecessors and competitors create. How well does each model play its role in this ecosystem, as a creator and as a...
Shibo Jin, Zhiqiu Xu et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Quantitative research is an iterative loop: a researcher proposes a market hypothesis, implements it as an executable signal, tests it on historical data, and decides what to keep. Existing benchmarks score formulas, code, or individual trading decisions, leav...
Houde Qian, Zheng Jiang et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Many NP-complete (NPC) problems have a polynomial-time (P) twin that differs in a single structural detail, such as 3- versus 2-colouring a graph. A model that reasons rather than recalls should notice which side of this boundary a problem falls on, yet existi...
Nurul Labib Sayeedi, Md. Faiyaz Abdullah Sayeedi et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Picture-description dementia classification uses a shared visual scene, but a transcript-level prediction alone does not identify the scene evidence underlying its inputs. NarraSpace links words and speech times to named picture elements and retrieves utteranc...
Qingming Li, Ruiyan Sun et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Mathematical reasoning remains underexplored for low-resource languages, including Dialectal Arabic. To bridge this gap, we introduce the DAR-MGSM (Dialectal Arabic Multilingual Grade School Math) Benchmark, the first manually curated benchmark for evaluating...
Noor Abo Mokh, Teresa Lynn et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Agentic on-policy distillation (OPD) trains a student agent on its own trajectories while a teacher scores every token. A natural explanation of where it goes wrong is compounding error: the student's mistakes carry its trajectories beyond the region where the...
Shu Zhao, Tan Yu
#cs.CL
Prépublication
· 2026 arXiv
EN
Multimodal large language models (MLLMs) have made rapid progress on visual reasoning, yet many tasks require evidence that is difficult to obtain through internal reasoning alone: our controlled study on 22 benchmarks shows that simply increasing the inferenc...
Yu Zhou, Cheng-Fu Yang et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Systems that combine several large language models (LLMs) often use one context for many questions. When the context is passed between models as text, the receiving model processes all of its tokens again for every question. Avoiding this cost requires a conte...
Huawei Lin, Jinting Liu et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Many modern language models have adopted hybrid architectures that incorporate both linear-time recurrence and softmax attention. Modern recurrence layers provide a highly expressive mixing mechanism while ensuring high throughput by compressing contextual inf...
Xingyu Shen, Yingfa Chen et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Context compression improves long-context efficiency but may introduce faithfulness issues through an information bottleneck. Existing evaluations rely on QA metrics, but a reader can still answer correctly from an unfaithful compression, so these metrics unde...
Ziqing Man, Di Wu et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Fine-tuning large language models (LLMs) can induce behaviors beyond those explicitly targeted. For example, training a model to respond more warmly can make it more sycophantic, and training it to write insecure code can cause broad misalignment on prompts un...
Dongyoon Hahm, Woogyeol Jin et al.
#cs.CL