Bingran You, Wenjun Ke, Ria Rosenauer, Bo Liu, Wenbo Chen, Min Chen, Shuqi Xu, Yue Ma 等 72 位作者
AI agents are playing an increasingly important role in physics research. However, existing physics benchmarks only evaluate fixed questions, isolated sub-tasks, or paper reproduction with given solutions, and rarely check research quality alongside correctness. We introduce FrontierPhysics, an end-to-end benchmark with 56 tasks made by 42 physicists based on their own research, covering nine domains across theoretical, computational, and experimental physics. Each task requires agents to review the literature, develop a research plan, conduct research and experiments, and produce structured outputs and scientific manuscripts. Combining deterministic verifiers with expert-authored rubrics, FrontierPhysics evaluates whether agents' research is correct and publishable. Across 22 models in 26 configurations, no configuration passes more than 46 of its 168 rollouts. Manual review of agent trajectories shows that frontier agents' main weaknesses are insufficiently rigorous reasoning and limited expert intuition, rather than gaps in physics knowledge or execution errors.
Streaming video understanding requires models to repeatedly decide whether to respond or continue observing. Recently, Group Relative Policy Optimization (GRPO) has been applied to this setting, with existing approaches emphasizing reward design while retaining a single trajectory-level advantage across turns. Such shared credit can reinforce redundant replies after earlier successes and discourage useful follow-ups after earlier failures. We propose TurnWise, a turn-level GRPO framework that aligns learning with individual streaming decisions. TurnWise addresses the mismatch between trajectory-level feedback and turn-level interaction, promoting informative responses and reducing redundancy. Specifically, turn-level credit assignment uses silence-completed prefix scoring to construct group-relative advantages from each decision's remaining return, separating its learning signal from earlier rewards. Turn-mean policy optimization averages token losses within each turn and then across turns, giving response and waiting decisions equal weight regardless of output length. These complementary designs improve streaming policy learning without a learned critic or additional rollouts. Extensive experiments show our method outperforms prior RL approaches for streaming video understanding.
Qi Xu, Zhigang Hua, Zijie Lei, Yan Xie, Jizhe Zhang, Lynn Lu, Delun Du, Sha Hua 等 13 位作者
Whole-page ranking jointly optimizes advertisements and organic content to balance business outcomes and user engagement under complex layout and business constraints. Existing non-autoregressive (NAR) approaches enable efficient page generation but typically score candidate items independently of previously selected items, limiting their ability to capture order-dependent interactions. Autoregressive (AR) methods address this limitation through sequential conditioning, but incur substantial inference overhead that challenges large-scale deployment. Bridging this quality-efficiency gap remains a challenge in industrial whole-page ranking. We propose Multi-Item Scoring (MIS), a prefix-aware NAR-style framework that incorporates sequential context into efficient multi-item scoring. MIS integrates global candidate interactions, localized prefix conditioning, and position-aware scoring within a shared representation architecture. By computing reusable representations once per request and performing lightweight prefix-conditioned scoring during constrained beam search, MIS captures important sequential dependencies without repeated full-model inference. Experiments on KuaiRand-1K and RL4RS demonstrate that MIS approaches the generation quality of an autoregressive CGA baseline while achieving 4.07x and 3.54x faster CPU decoding, respectively. In a randomized evaluation on a large-scale social-video platform, MIS improves the realization metric by 0.104% and watch time by 0.058%, while reducing ranking-service CPU consumption by 9.73%. These results demonstrate that prefix-aware multi-item scoring can narrow the quality gap between NAR and AR generation while retaining the efficiency required for industrial scalability.
Video multimodal large language models achieve strong results on existing benchmarks, but answer accuracy alone does not establish whether they can locate the evidence needed to answer a question. We introduce VideoZeroBench, a challenging long-video benchmark with manually annotated question-answer pairs spanning 13 video domains. Questions target fine-grained cues, fleeting events, and evidence distributed across multiple segments. Temporal intervals and key-frame boxes are annotated where applicable. All questions undergo two rounds of cross-verification for answer validity and evidence quality. Our five-level diagnostic protocol compares answering with and without evidence hints, then combines answer correctness with independently evaluated temporal and spatial grounding. Across 19 evaluated models, the best standard QA accuracy is 24.8% (Level-3), achieved by Gemini-3.7-Flash. No model exceeds 1.8% when correct answers and accurate spatio-temporal localization are jointly required (Level-5). Analyses of atomic abilities, evidence spans, input modalities, and thinking-with-videos inference further characterize where the evaluated systems struggle. These findings motivate more precise evidence search and localization for long-video question answering. Our code and data are publicly released.
Wenjie Huang, Jun Zhang, Hongyaoxing Gu, Haoyang Huang, Tianqi Xu, Xinyue Hao, Xinyi Hu, Xinle Du 等 13 位作者
Multi-turn conversations with multimodal large language models (MLLMs) reuse input context, but long multimodal inputs increase time to first token (TTFT). Query-aware hard pruning reduces prefill tokens but permanently excludes information that later questions may require. We propose Raincheck, a framework for deferred multimodal prefill that decouples initial token selection from later information availability. An existing pruning method selects tokens for initial prefill; Raincheck preserves and progressively prefills unselected tokens in source order using unused compute capacity during decode. Unfinished prefill continues after answer return during available scheduling gaps. Attention visibility control and affine state mapping enable context updates after generation begins in full-attention and hybrid recurrent models. Prefill-decode fusion reduces repeated weight reads. We evaluate Qwen3-VL-8B and Qwen3.5-9B with 30% initial multimodal-token retention. Across both models and four pruning methods, Raincheck recovers on average 77.7% and 60.4% of the quality lost to matched hard pruning on MultiVerse and ConvBench, respectively. On MultiVerse, the SGLang implementation reduces mean first-turn TTFT by up to 24.1% relative to full prefill, with less than 0.5% increase in mean time per output token (TPOT).
A model is only half of an AI agent system. The other half is the inference strategy: the reasoning procedure that turns model calls into a response. Today that second half is written by hand, and the experience of building one strategy never carries to the next. We introduce Strategist, a meta-agent that evolves inference strategies across tasks. Each strategy is written as a blueprint,executable code that other strategies can reuse, and together the blueprints form a library. Given a task and a small development set, Strategist retrieves related blueprints, writes candidate strategies, tests them on real samples, and revises them from the results. Every candidate then enters the library, so later tasks build on earlier discoveries. On 30 benchmarks, Strategist beats the strongest baseline on 24 and matches its average accuracy (74.7% vs.74.0%) at 39% lower inference cost. Its strategies also transfer to unseen tasks and other models without further search. The part of an AI system that is not improved on its own now does.
Xinlei Yu, Gen Li, Ji Zhao, Zhangquan Chen, Congcong Wang, Shuai Dong, Zelin Peng, Yuqi Xu 等 18 位作者
Despite recent advances in CUAs, their development remains constrained by the limited diversity, controllability, and reliability of interaction data. Moreover, conventional optimization pipelines lack iterative feedback, while existing recursive improvement approaches often rely on self-generated experiences, limiting ability expansion and yielding diminishing returns. To address these challenges, we introduce CUA-Master, a nested double-loop recursive improvement framework. At the system level, an outer world-data-model flywheel continuously incorporates open-world corpus and leverages evolving model weaknesses to guide subsequent data construction and optimization iterations. At the data level, an inner flywheel couples open-world seeding and autocatalytic propagation. The former mines authentic workflows as seed tasks from instructional videos, and employs visual-semantic dual-path trajectory reconstruction; The latter transforms these seeds into reusable primitives for controllable synthesis of diverse, complex, and verifiable tasks. To enable efficient and reliable trajectory collection, we further introduce a feedback-guided snapshot-rollback retry mechanism in various real environments. Extensive experiments demonstrate that CUA-Master-27B achieves competitive performance across GUI-only and GUI-CLI hybrid benchmarks, scoring 80.2 on OSWorld Verified, 36.9 on OSWorld 2.0, 43.9 on WeaveBench, and 25.0 on Agents' Last Exam. Our framework also substantially improves the quality and efficiency of data synthesis, and exhibits continuous gains across recursive iterations.
Tianyu Liu, Jun Zhang, Wenjie Huang, Yuhao Shen, Tianqi Xu, Xinyue Hao, Hongyaoxing Gu, Xuyang Liu 等 15 位作者
Speculative decoding (SD) accelerates large language model inference by reducing the sequential cost of token-by-token generation. Advanced SD methods reduce proposal cost with parallel drafters that predict entire blocks in one forward pass, but overall speedup still depends on acceptance length. On-policy distillation (OPD) can push well-trained drafters beyond acceptance-length plateaus, yet conventional pipelines collect supervision through sequential draft-and-verify rollouts. Our key insight is that on-policy supervision need not inherit these dependencies: a draft block can be supervised using only its reference prefix and draft-sampled history. We propose SpecOPD, a branch-parallel OPD framework that parallelizes draft sampling and target supervision across independent branches drawn from the evolving drafter policy at reference anchors. We pack these branches into proposal trees to share target-prefix computation, using tree masks for full attention, branch-consistent indexer groups for sparse attention, and indexed state replay for linear attention to preserve each branch's history. We train the drafter with an E2E-TV surrogate that directly targets acceptance length. Our study spans 8B full-attention (GQA), 27B full-linear (GQA--GDN), and 125B sparse-linear (QSA--GDN) targets, three parallel draft architectures (DFlash, DSpark, and DFlash V2), and four task domains (dialogue, coding, mathematics, and agentic). SpecOPD achieves up to 41.45% higher mean acceptance length, $16.09\times$ supervised-token throughput over sequential-rollout OPD under native configurations, and up to $3.56\times$ generation speedup over autoregressive decoding.
Boao Kong, Xu Huang, Yuqi Xu, Yixuan Liang, Bin Wang, Kun Yuan
Pipeline-parallel distributed optimization is essential for large-scale machine learning but is challenged by significant communication overhead from transmitting high-dimensional activations and gradients between workers. Existing approaches often depend on impractical unbiased gradient assumptions or incur sample-size memory overhead. This paper introduces Clapping, a Communication compression algorithm with LAzy samPling for Pipeline-parallel learnING. Clapping adopts a lazy sampling strategy that reuses data samples across steps, breaking sample-wise memory barrier and supporting convergence in few-epoch or online regimes. Clapping comprises two variants including Clapping-FC and Clapping-FU, both of which achieve convergence without unbiased assumption for compressed gradient, effectively addressing compression error propagation in multi-worker settings. Numerical experiments validate the performance of Clapping across different tasks.
Modern large language models (LLMs) often exhibit overconfident errors, assigning high confidence to incorrect predictions and thereby undermining reliability in real-world use. Existing calibration methods typically rely on post-hoc adjustment or retraining, operating at the output level without addressing the underlying causes of miscalibration. We propose CoCHI, (Class-oriented Causal Head Intervention), a framework that improves calibration by identifying and intervening on internal components responsible for distinct confidence behaviours. By analysing predictions through correctness-confidence patterns, CoCHI enables targeted intervention on model internals, reducing overconfident errors while preserving correct predictions. Our approach provides a mechanistic perspective on calibration, linking reliability to identifiable structures within the model rather than treating it as an output-level issue. Experiments across multiple LLMs show consistent improvements in calibration metrics, including Expected Calibration Error, Negative Log-Likelihood, and Brier Score, while maintaining accuracy on five multiple-choice question answering benchmarks. These results suggest that effective calibration requires not only adjusting outputs, but also addressing the underlying causal mechanisms within the model.
Tang Da Huang, Weidong Tang, Wen Qi Xu, Xianpeng Guo
Multimodal Large Language Models typically assume linguistic context invariably enhances visual understanding. We study this assumption in semantic adversarial scenarios, specifically magic tricks, where narration deliberately diverges from physical reality. We introduce MagicBench, a diagnostic benchmark of 402 videos for evaluating MLLMs under hierarchical linguistic interference, together with a Physical Constraint Set (PCS) protocol for assessing adherence to physical laws. Evaluation uncovers a Semantic Dependency Paradox: (1) Semantic anchoring: Entity nouns act as anchors aiding localization, paradoxically boosting performance despite false predicates. (2) Visual Agency Loss: In semantic vacuums, multimodal performance collapses 12.4% (p < 0.01) below the vision-only “capability probe”. This gap persists under symmetric prompting, suggesting a form of functional perception suppression in which autonomous visual search may be under-utilized in multimodal settings without linguistic triggers. Causal interventions via spatial prompting and signal magnification provide evidence that internal reasoning remains functional, supporting the interpretation of a perceptual access bottleneck. Our findings suggest MLLMs function as “language-guided passive observers”, advocating for perceptually-independent architectures that decouple sensory agency from linguistic dominance. Code and dataset are available at https://github.com/Ink-Dawn/MagicBench
尚无 AI 解读，展示原文摘要
cs.CL · cs.CV
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Sizhe Wang Washington University in St. Louis, Ziqi Xu, Claire Najjuuko Washington University, Saint Louis, Charles Alba, Chenyang Lu Washington University, Saint Louis
Clinical language models (LMs) are increasingly applied to support clinical risk prediction from free-text notes, yet their uncertainty estimates often remain poorly calibrated and clinically unreliable. In this work, we propose Clinical Uncertainty Risk Alignment (CURA), a framework that aligns clinical LM-based risk estimates and uncertainty with both individual error likelihoods and cohort-level ambiguities. CURA first fine-tunes domain-specific clinical LMs to obtain task-adapted patient embeddings, and then performs uncertainty fine-tuning of a multi-head classifier using a bi-level uncertainty objective. Specifically, an individual-level calibration term aligns predictive uncertainty with each patient’s likelihood of error, while a cohort-aware regularizer pulls risk estimates toward event rates in their local neighborhoods in the embedding space and places extra weight on ambiguous cohorts near the decision boundary. We further show that this cohort-aware term can be interpreted as a cross-entropy loss with neighborhood-informed soft labels, providing a label-smoothing view of our method. Extensive experiments on MIMIC-IV clinical risk prediction tasks across various clinical LMs show that CURA consistently improves calibration metrics without substantially compromising discrimination. Further analysis illustrates that CURA reduces overconfident false reassurance and yields more trustworthy uncertainty estimates for downstream clinical decision support.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表