MOSS-Image is an open-weight image generation system that combines a 5B single-stream diffusion transformer with an intention-aware prompt enhancement module to produce high-fidelity, text-rich images with precise layout control. The system supports bilingual long-form text rendering and introduces evaluation benchmarks LongLongTextBench and RealLayoutArena. It achieves state‑of‑the‑art performance on LLTB and ranks highly on RLA, with plans to release the model, training recipe, and evaluation framework for the community.
By MOSI. AI
Agentic Physics-Grounded Video Generation via Memory-Augmented In-Context Reasoning introduces PhysRepair, a framework that repairs physical inconsistencies in generated videos after synthesis. It uses spatiotemporal causal constraints to pinpoint invalid states and guides localized, physically plausible regeneration, while experience-augmented in-context learning adapts repair strategies based on prior failures. Experiments across solid mechanics, fluid dynamics, optics, thermodynamics, and magnetism show that PhysRepair consistently improves physical plausibility over existing methods.
By Xikang Guan, Shaocong Yang, Yanqi Bao, Can Wang, Zheng Gu, Zhengqing Chen, Kai Chen, Jing Huo, Yang Gao
The paper introduces an ENF-based forensic framework for authenticating videos against generative AI. It extracts electric network frequency signals from both audio and rolling‑shutter video, uses a synthetic‑injection training method grounded in real recordings, and verifies claimed recording times via reference matching. The authors find that ENF flicker alone cannot prove authenticity; instead, consistency with an independent reference is required, achieving high accuracy while reducing computational cost.
By Ryan Wyton (Southampton Solent University), Bacha Rehman (Southampton Solent University)
GatewayBench is a new evaluation framework that systematically studies how the design of Model Context Protocol (MCP) gateways influences AI agent capability, efficiency, and security. It categorizes gateways into flattened and hierarchical architectures and tests them on 524 tasks across six domains using a pool of 11,118 real-world tools from 251 MCP servers, including task-irrelevant tools to simulate large-scale deployments. Experiments on seven widely deployed gateways reveal significant differences in task completion, tool-use efficiency, and attack resilience, showing strong interactions with the underlying model and agent harness.
By Yitian Zheng, Zhaorun Chen, Haoyu Li, Bo Li
LDDR (Linear DPP Dynamic Resolution) is a training‑free, budget‑aware video frame sampling framework designed for multimodal large language models (MLLMs). It uses query‑aware Determinantal Point Process (DPP) selection in a task‑conditioned feature space, achieving a 3× runtime speedup over standard DPP baselines. By assigning a non‑uniform token budget—allocating more tokens to informative, non‑redundant frames and downscaling or pruning less useful ones—LDDR improves performance by 2.5 points in low‑budget settings and 1.6 points in high‑budget scenarios across four video benchmarks and multiple MLLM families.
By Jingfeng Chen, Jiawen Qian, Wendi Deng, Yinuo Guo, Jiaqi Yu, Sicong Leng, Raghuveer Thirukovalluru, Bhuwan Dhingra
The paper discusses the limitations of LASSO in high‑dimensional regression models when cointegrated regressors are present, noting that the convex penalty can overly shrink their coefficients and lead to inconsistent estimation. It proposes using folded concave penalties (FCP), such as SCAD and MCP, to overcome this issue, establishing both local and global oracle properties for the penalized regression. Simulation results and an application to euro area unemployment forecasting demonstrate that FCP improves variable selection and out‑of‑sample prediction compared to LASSO.
By Hongqi Chen, Chengwang Liao, Ziwei Mei, Zhentao Shi
The paper introduces TIDER, a time‑aware retrieval framework that uses a reinforcement‑learning trained temporal interval locator to identify relevant time windows for evidence retrieval in long‑horizon agent memory. By combining temporally grounded candidates with semantic retrieval and reranking them by proximity to the inferred windows, TIDER mitigates the displacement of target memories caused by temporally mismatched yet semantically similar entries. The lightweight 1.7B locator achieves state‑of‑the‑art accuracy on LifeBench and LoCoMo benchmarks, demonstrating the importance of temporal cues for effective evidence retrieval.
By Dabin Sheng, Zhe Wu, Jinming Zhao, Junliang Xing, Yuanchun Shi
PRIME (Precommitted Rules for Inference from Minimal Evidence) is a framework designed to improve the reliability of large language model agents when verifying records against organizational requirements. It predefines evidence routes for each possible conclusion, records check results in a ledger, and refuses any conclusion that is not fully justified by the recorded evidence. In experiments on 480 public‑record tasks, PRIME eliminated unjustified conclusions entirely and increased the average score for justified conclusions from 30.1 to 59.2 out of 100.
By Hamed Khosravi, Xiaoming Huo
The paper evaluates the planning abilities of large language model (LLM) agents by comparing their performance against formal semantics defined in hidden PDDL models. Across five domains, four LLMs, six prompting strategies, and three agent architectures, the best setup solved 78% of solvable tasks offline but only 40% online, with limited recognition of unsolvable tasks and no clear benefit from more elaborate prompting. The study concludes that claims of LLM planning competence require validated action traces rather than end‑to‑end success rates.
By Oriol Miro-Lopez-Feliu, Oscar Mir\'o L\'opez-Feliu, Victor Gimenez-Abalos, Javier Vazquez-Salceda, Victor Carballo Araruna, Sergio Alvarez-Napagao
The paper reports on an AI system that autonomously discovers and formally verifies new mathematical rules—called accelerations—for Turing machines in the Busy Beaver (BB) domain, specifically for eleven machines in BB(6). Using a Gemini-based pipeline that combines pattern recognition, conjecturing, informal reasoning, and Lean theorem proving, the system identified Collatz-like accelerations that provide compact descriptions of machine executions while leaving their halting behavior unresolved. These findings introduce a new class of machines, termed Cryptids, and push forward the analysis of the uncomputable BB frontier.
By Marco Dos Santos, Vivek Veeriah, Alex Havrilla, Andrew Bolt, Goran Zuzic, Satinder Singh, Tom Zahavy
Structured Authority Injection (SAI) is a protocol that enables persistent human control over large language model (LLM) agents by routing operator corrections through typed, schema‑validated Model Context Protocol (MCP) tool calls and maintaining a ranked authority context in each prompt. It defines three override classes—Constraint, Priority, and Strategy—each with distinct persistence semantics and a strict precedence order enforced via a directed authority channel. Evaluations across nine LLMs and eight scenarios show reduced ambiguous directive interpretation, mitigated authority decay, and lower prompt overhead in the strongest models, though benefits vary by model.
By Ryan Tran, YueMing Zhang, Lingfeng Tao, Nan Yang, Shengjie Xu, Yu Yin, Shuai Xu, Wenlu Zhang, Hailu Xu
The paper introduces RECAP, a framework for deciding when to retain, compress, or discard parts of a long-horizon agent’s history to improve future performance. RECAP uses matched context interventions, execution-relational prediction, and independent policy certification to evaluate the value of a failure block relative to the current execution. Experiments across five backbones show that RECAP can edit a significant portion of retained checkpoints while maintaining high continuation success and reducing prompt costs.
By Chenle Chen, Yangbo Wei, Zhijie Zheng, Zhenyu Su, Qifan Wang, Junhong Qian, Chao Yao, Chen Wu, Lei He
The paper introduces PrefMind, an online dual‑agent reinforcement learning framework that jointly trains a response Generator and a preference Critic to explicitly learn and apply human preference judgments. By having the agents discuss and correct each other’s evaluations of candidate responses, PrefMind embeds preference understanding directly into the generation process rather than relying solely on a scalar reward model. Experiments on models of varying sizes demonstrate gains in both preference judgment accuracy and downstream response quality.
By Ziyun Zou, Jiwoo Hong, Yifan Lu, Shao Tang, Weiyi Wu, Aman Gupta, Jin Tian, Imran Razzak
The paper investigates failures in large language model (LLM) safety systems where runtime monitors detect policy‑relevant evidence but the evidence never reaches the enforcement decision point. Using canary contracts, the authors identified 18 evidence‑path failures across 11 safety projects and four ecosystems, categorizing them into established‑boundary repairs, new‑surface policy gaps, and boundary‑only evidence cases. The study demonstrates that ordinary application behavior can continue while safety‑specific evidence is lost, highlighting the need for regression tests on evolving monitor schemas, streams, tools, and adapters.
By Junchi Liao
The paper introduces SEER‑Repro, a benchmark of 20 SEER‑based studies with data and resources for reproduction. An LLM agent reproduces 77.4% of 2,190 atomic claims and provides executable baselines. An agentic analysis pipeline then tests robustness by identifying analytical choices, proposing alternatives, and rerunning analyses, flagging potentially consequential choices in 9.5% of fully reproduced findings for expert review.
By Shicheng Liu, Krista Opsahl-Ong, Andrew Drozdov, Julian Killingback, Arnav Singhvi, Ronak Pradeep, Michael Bendersky, James C. Dickerson, Monica S. Lam
The paper introduces BRIA (BRanch-Inspect-Adopt), an agent that isolates tool‑candidate investigations into separate branches, executing each candidate independently and then inspecting the execution evidence to adopt only useful results. BRIA dynamically retrieves additional tools as new needs arise during task execution and is evaluated on PlanBench‑XL and MCP‑Atlas with a library of 1,897 tools, outperforming or matching the strongest retrieval baselines. On MCP‑Atlas, BRIA achieves an average claim score of 89.77 versus 75.40 for MCP‑Zero, and it lies on the Pareto frontier for task performance and input‑token cost.
By Anna Mazhar, Seiji Maekawa, Nikita Bhutani
The paper introduces Self-HPO, a recursive framework that alternates between optimizing the harness and the policy of large language model agents for coding tasks. By letting the current policy propose new harness candidates based on its failures, the method aligns harness evolution with policy improvement, leading to consistent performance gains across three benchmarks. Experiments show that Self-HPO raises the average resolve rate of Qwen3.5-35B-A3B from 57.8% to 64.8% and improves reliability and capability metrics, with benefits that generalize to out‑of‑distribution settings.
By Zhepei Wei, Wenlin Yao, Baolin Peng, Danqing Wang, Qianhui Wu, Hao Cheng, Yu Meng, Jianfeng Gao
QD-Rec is a quality–diversity approach that expands recommendation systems by generating diverse, high‑quality suggestions paired with natural‑language hypotheses. It uses a contextual bandit to select items and a foundation model to craft reasons, storing new recommendations in an archive when they are novel or superior to existing ones. Experiments on mergers & acquisitions, customer targeting, and content recommendation show that QD‑Rec achieves higher mean quality and larger archives than baseline methods.
By Takashi Ishida, Taro Makino, Issa Sugiura
The paper evaluates seven open‑source small language models (20 M–1.5 B parameters) on two post‑exploitation code‑generation tasks—Directory Enumeration and Backdoor Generation—in both Python and Lua. It finds that offensive code‑generation performance does not scale linearly with model size; instead, task formulation and prompting strategy play a larger role. Using algorithmic formulation and few‑shot prompting, Qwen2.5‑Coder 1.5B achieves near‑perfect success rates on both tasks in Python and high success in Lua.
By Gabriel Hortea, Simon Pietro Romano, Juan Tapiador, Mario Varlese
SIT‑Eval is a framework that evaluates conversational agents by sampling diverse situations at each turn and judging responses against situational rules. It uses an LLM‑interlocutor to generate uniformly sampled contexts and a context‑aware LLM‑judge to assess rule compliance. Benchmarking eight large language models on 1,292 turns uncovered hidden failure modes, showing that models with similar overall scores can differ greatly in rule adherence, especially in ambiguity clarification and Socratic questioning.
By Ravil Mussabayev, Saveliy Zavertyaev, Dmitriy Parpulov, Andrey Savchenko