Zachary T. Rewolinski, Jan Simson, Frauke Kreuter, Christoph Kern, Bin Yu
Human judgment calls made throughout the data science life cycle can change the conclusions drawn from data. Independent analysts often make different choices when given the same data, leading to often unacceptable variations in results and contributing to the ongoing reproducibility crisis across scientific disciplines. The rapid adoption of agentic tools threatens to amplify this crisis. Several approaches have been developed to address this variation by following the same strategy: repeating an analysis under different combinations of analytical decisions and inspecting the distribution of outcomes. Examples include multiverse analysis, specification curve analysis, and vibration of effects. More broadly, the Predictability-Computability-Stability (PCS) framework assesses how documented judgment calls across the data science life cycle affect conclusions, with predictability serving as a reality-check that a model captures signal. Each of these approaches results in a "multiverse" object. While the distribution of outcomes reflects the overall variability of conclusions, estimating the influence of each individual decision requires additional tools. A common heuristic is to train a model to predict the outcome from these decisions and treat its coefficients or feature importances as proxies for each decision's influence. Although this strategy may work in certain cases, it assumes the model is correctly specified and accurately predicts the outcome, which we show is not always the case. In this work, we introduce DECIDE, a model-free framework that quantifies each decision's importance from the actual outcomes across the plausible decision space, avoiding issues of predictability by design. We demonstrate the effectiveness of DECIDE by applying it to real-world multiverse analyses. We further provide what is, to our knowledge, the first formal definition of the multiverse.
Language models increasingly create 3D content through code, with practical workflows often relying on editing existing programs rather than regenerating them, for speed and precise control. Such edits require understanding what a program builds and foreseeing how a change reshapes it, yet prior benchmarks mostly evaluate generation or simple edits against a reference. In this paper, we introduce 3DCodeEdit, a benchmark of 2,592 instruction-guided editing tasks over 108 Blender programs that construct objects entirely in code. Its edits range from parameter to structural changes, style redesign, and scene composition, and are judged from renders against human-reviewed checklists. Across 17 models, the average success rate is only 37%, and even the strongest models complete fewer than 60% of structural and style edits. As models become more capable, their failures shift to geometric foresight, where the code runs but builds the wrong shape. We therefore study whether multi-turn feedback can compensate for this missing foresight. Feedback from a separate reviewer model that describes how the renders differ from the request helps most, more than maximum reasoning effort or an agent harness, and can let a 27B model surpass the single-shot rate of a 2.8T-parameter model.
Visual world models increasingly support spatial reasoning by generating RGB views of unobserved states. However, RGB has become the default interface largely because it is well suited for visual generation, rather than necessarily being the most effective representation for downstream reasoning. Different spatial questions may benefit from different representations of the same imagined state: top-down geometry can facilitate global distance comparisons, while panoramic views provide broader context for reasoning about open space. To isolate the role of representation in spatial imagination, we introduce ImagiRep, a simulator-grounded benchmark of 922 questions spanning seven question types across two settings, Viewpoint Imagination and Cross-view Integration, providing oracle RGB and six alternative representations (e.g., depth, panoramic views, and top-down geometry) of the same unobserved state. This controlled comparison reveals that even accurate imagined RGB does not guarantee accurate spatial reasoning, and different questions benefit from different representations. Similar trends hold for representations derived from world-model-estimated states. Notably, estimated RGB alone can mislead spatial reasoning, while alternative representations retain complementary benefits. Building on these findings, we introduce GEARS, a lightweight test-time strategy for adaptive representation selection, outperforming existing spatial imagination methods and showing that reasoning about unseen states requires more than generating RGB views.
Muqing Cao, Yunwoo Lee, Junbin Yuan, Lorenzo Schenk, Sebastian Scherer
Multi-robot informative path planning (IPP) for persistent target monitoring requires robots to reason about spatial uncertainty, temporal evolution, and practical sensing and communication constraints. Recent learning-based multi-robot IPP methods use Gaussian Processes (GPs) for target uncertainty, but often rely on simplified sensing models and centralized belief updates. We propose a grid-based spatio-temporal GP-Kalman filtering framework for learning-based multi-robot IPP. Instead of maintaining one GP per target, we represent anonymous target presence as a single latent field over a discrete workspace grid. The proposed recursive update assimilates footprint-level camera measurements and supports arbitrary fields of view and range-dependent noise. A GP-derived temporal process update inflates uncertainty over time to account for moving targets. For decentralized deployment, each robot maintains its own mapper and exchanges fixed-size belief summaries rather than raw measurements. Received beliefs are fused using diagonal covariance intersection to remain conservative under unknown inter-robot correlations. We integrate the mapper with a reinforcement-learning policy for graph-based neighbor selection. In simulation, the centralized deployment improves target visitation and attains about 20% lower posterior target uncertainty than learning-based and classical auction/coverage baselines, a large share of which stems from the belief representation itself. A real-world two-UAV experiment demonstrates transfer to outdoor multi-robot search over a large field (>7000 m^2). Code is available at https://github.com/caomuqing/multirobot-ipp.
Qiaoling Chen, Jiaxiang Cheng, Xueqing Deng, Tiankai Hang, Ji Hou, Jiaqi Li, Yifan Liu, Tao Lu 等 38 位作者
Interactive world generation requires models to follow successive commands while preserving relevant history and action consequences. We present Hy-Nomas, a video world model with an 8B diffusion backbone that supports free-form text instructions and continuous camera control across multiple interaction turns. It generates variable-length segments conditioned on current inputs and cached history. Our central design aligns structured video captions with action-aware causal attention: descriptive context remains accessible across segments, while action instructions are restricted to their intended temporal windows. A vision-language Prompt Enhancer (PE) translates user requests and multimodal interaction histories into structured conditions for the next segment. A multi-stage curriculum progressively introduces causal generation, text interaction, camera control, and training on model-generated histories, followed by diffusion distillation. In human evaluation, Hy-Nomas shows superior performance over existing world models. We release model weights, code and benchmarks, together with a lightweight adaptation workflow, to support reproducible research on interactive world generation.
Coding agents are trained to complete tasks, and they keep trying even when the environment makes completion impossible. The evidence that only outside help can restore progress usually appears within a few steps, yet agents rarely act on it: they over-persist, attempt undesired workarounds, and claim success the record does not support -- a phenomenon we call evidence-action gap. To measure it, we introduce ImpasseBench, which injects twelve realistic environment blockers into otherwise solvable repair tasks and marks, in each trajectory, the step at which the collected evidence becomes sufficient. Scoring actions relative to that step separates appropriate effort from over-persistence and premature handoff. Across eight frontier models and 19,371 trajectories, we find that nearly every run that needs outside help collects sufficient evidence, yet a third still exhaust their budget, and every unsafe request we detect comes after that point. Fortunately, an escalation tool, or an external monitor for models that ignore the tool, already guards against much of this behavior: the tool cuts budget exhaustion from 34.0% to 6.7%, and the monitor nearly triples task success after environment restoration. Even then, a fired escalation is not yet a useful handoff: tickets reliably name the blocker but often omit the agent's own progress and verification status. Taken together, we argue that knowing when to stop therefore needs explicit support, both in the tool that lets an agent hand off and in the report it leaves behind.
Mingyu Wang, Bin Yu, Wenjian Lu, Zhi Wang, Gao Kefeng, Cheng Wen Xi'an University of Electronic Science and Technology, Xu Lu Xidian University, Cong Tian
Linux kernel device drivers are tightly coupled with hardware, making them difficult to execute and test without physical devices. This heavily limits automated code analysis and vulnerability discovery. While manual modeling is unscalable, Large Language Models (LLMs) offer a new approach to scale virtual device construction across the Linux driver ecosystem. In this paper, we present DevGen, an LLM-powered tool that generates QEMU-based virtual devices directly from Linux driver source code. DevGen combines static analysis to gather necessary context, guides the LLM through step-by-step prompting, and uses an automated self-correction loop driven by compilation and execution feedback. To further reduce errors, similar fixes are retrieved from a library of common modeling failures and incorporated into the repair prompt, which supports more targeted corrections in later iterations. The generated devices finally integrate with QEMU and Syzkaller, enabling driver fuzzing without physical hardware. DevGen is evaluated on 50 PCI/PCIe drivers from Linux 6.18 using three mainstream LLMs, and successfully generates usable models for 44 drivers. In these drivers, 24% of them achieve significant improvements in fuzzing coverage, and 7 previously unknown crashes are triggered with 1 CVE assigned. These results demonstrate the practical capability of LLMs to automate complex, system-level code generation tasks.
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 3: System Demonstrations) · 2026 · 正式发表
Cheng Wen Xi'an University of Electronic Science and Technology, Hu Junjie, YiKun Hu, Jie Su Xi'an University of Electronic Science and Technology, Bin Yu, Dugang Liu Shenzhen University, Zhiwu Xu Shenzhen University, Weidi Sun Peking University 等 10 位作者
Formal verification can provide strong mathematical guarantees about software correctness, but it typically requires developers to write detailed formal specifications (e.g., contracts and loop invariants), which is costly and error-prone. We introduce AutoSpec+, an LLM-driven neuro-symbolic demonstration system that reframes specification writing as constrained structured synthesis: large language models generate candidate specifications at the granularity of proof-relevant program components, while a symbolic verifier acts as a deterministic critic that checks legality, satisfiability, and proof adequacy, rejecting or repairing candidates in an iterative loop. This design turns unconstrained text generation into constrained structured synthesis, substantially reducing hallucinations and producing proof-ready annotations. We evaluate AutoSpec+ on seven benchmark suites, showing strong effectiveness. We release an open-source, Dockerized system with ensemble LLM backends and inter-modular verification support for reproducible demonstration and deployment
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 3: System Demonstrations) · 2026 · 正式发表
Lingrui Mei, Shenghua Liu Institute of Computing Technology, Chinese Academy of Sciences, Yiwei Wang University of California, Merced, Baolong Bi, Ruibin Yuan, Xueqi Cheng Institute of Computing Technology, Chinese Academy
As Large Language Models (LLMs) grow increasingly powerful, ensuring their safety and alignment with human values remains a critical challenge. Current alignment approaches predominantly rely on refusal alignment, such as training models to refuse harmful prompts or implementing filters at various stages to block certain responses. These methods are designed toward a binary outcome: either denying to answer the question entirely or answering with full access to the model’s parametric knowledge. The binary nature of current alignment approaches presents significant limitations. These methods often fail to balance safety and utility, resulting in either overly cautious responses or overlooking subtle harmful content. They also prevent users from accessing benign information when it’s mixed with harmful content. For instance, a model might refuse to provide basic, public information about a medication’s composition due to misuse concerns. Furthermore, these approaches struggle with context-dependent sensitivity, potentially over-censoring harmless content or missing nuanced harmful outputs. Ideally, LLMs should offer informative responses while avoiding the disclosure of harmful and sensitive information. To address these challenges, we introduce HiddenGuard, a novel framework for fine-grained safe generation in LLMs. Our method incorporates PRISM (rePresentation Router for In-Stream Moderation), a specialized moudule that operates alongside the LLM architecture. By leveraging intermediate hidden states, HiddenGuard enables real-time, token-level harmfulness detection and redaction, without loss in capability. This approach captures deeper semantic information, allowing for more nuanced and context-aware content control compared to traditional filtering techniques. Consequently, the model can generate informative responses while selectively redacting or replacing sensitive information, rather than refusing to answer outright. We also contribute a comprehensive dataset with token-level fine-grained annotations of potentially harmful information across diverse contexts. Our experiments demonstrate that HiddenGuard achieves over 90% in F1 score for detecting and redacting harmful content while preserving the overall utility and informativeness of the model’s responses.
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
Zheng Jia, Shengbin Yue, Wei Chen Huazhong University of Science and Technology, Siyuan Wang The Chinese University of Hong Kong, Yidong Liu Harbin Institute of Technology and Midu Technology CO.,Ltd., Zejun Li, Yun Song, Zhongyu Wei Fudan University
The gap between existing benchmarks and the dynamic nature of real-world legal practice poses a key barrier to advancing legal intelligence. To this end, we introduce J1-ENVS, the first interactive and dynamic legal environment tailored for LLM-based agents. Guided by legal experts, it comprises six representative scenarios from Chinese legal practices at three levels of environmental complexity. We further introduce J1-EVAL, a dual-metric evaluation framework, designed to assess both task performance and procedural compliance across varying levels of legal proficiency. Extensive experiments on 17 LLM agents reveal that while many models demonstrate solid legal knowledge, they struggle with procedural execution in dynamic settings. Even the SOTA model is below 60% overall performance . These findings highlight persistent challenges in achieving dynamic legal intelligence and offer valuable insights to guide future research.
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
Junyou Li, Qin Zhang, Yangbin Yu, Qiang Fu, Deheng Ye
We find that, simply via a sampling-and-voting method, the performance of large language models (LLMs) scales with the number of agents instantiated. Also, this method, termed as Agent Forest, is orthogonal to existing complicated methods to further enhance LLMs, while the degree of enhancement is correlated to the task difficulty. We conduct comprehensive experiments on a wide range of LLM benchmarks to verify the presence of our finding, and to study the properties that can facilitate its occurrence. Our code is publicly available at: https://github.com/MoreAgentsIsAllYouNeed/AgentForest