The paper introduces 4DMemory, a benchmark for long‑horizon spatiotemporal memory that includes the LongVideo dataset—about 900 hours of video with over 5,000 manually reviewed questions—and the 4DMemoryBench. It also presents the 4DMemory Harness, a system that combines four complementary skills, more than 22 specialized tool models, and a retrieval‑augmented generation component. Evaluation of 13 leading models shows that standalone foundation models perform poorly on long‑horizon tasks, while the 4DMemory Harness boosts performance by over 50% and reveals a performance drop of more than 10% compared to previous benchmarks.
By Yanhao Jia, Yi Zhu, Yiran Zhong, Weigao Sun, Feida Zhu, Pengkun Jiao, Sihan Yang, Ming Ma, Steven Hoi, Erik Cambria
The paper introduces CONSTRAINTDIAL, a framework that automates the lifelong construction of smart contract security tasks through protocol mutation. It reconstructs contracts locally and uses attack execution to modify required calls or inputs, ensuring the original attack fails while maintaining exploitability and normal functionality. Applying this method to 203 public cases yielded 3,045 task inputs across detect, exploit, and patch categories, reducing reliance on copied answers by 10.9 percentage points and lowering the fraction of tasks solved by online copying by 9.3 points.
By Zifan Peng, Peng Huang, Jingyi Zheng, Yingjie Xue, Xuechao Wang, Xinlei He, Xia Hu, Dongrui Liu
The paper presents an ontology‑guided framework that uses specialized agents and large language models to extract, verify, and link factual triples from scientific publications, ensuring semantic consistency and traceability. Applied to 2,000 climate science papers, the system produced an ontology‑grounded knowledge graph with 133,990 nodes and 755,308 relations, achieving high agreement with human annotators and a 98.96% correctness rate among validated triples. The pipeline and resulting graph are publicly released for community use.
By Aayush Acharya, Qi Zhang, Eduard Dragut, Longin Jan Latecki
The paper introduces a pipeline that constructs a quote‑grounded knowledge graph of healthcare software adoption. It crawls vendor websites and uses a budgeted ReAct agent to search the web, extracting customer and integration relations that must be supported by verbatim quotes. The system achieves high precision (0.94) and recall (0.71) on a test set of 61 vendors, and its confidence scoring improves performance to 0.90 precision at 0.90 recall.
By Juan Felipe Duran, Arthur Tobler, Martin Valli\`eres
The paper introduces Uncertainty-Guided Latent Diagnostic Trajectory Learning (LDTL), a framework that uses a planning LLM agent and a diagnostic LLM agent to model sequential clinical diagnosis under uncertainty. By treating diagnostic action sequences as latent paths and applying a posterior distribution that favors trajectories with higher diagnostic value, the approach encourages coherent evidence acquisition that progressively reduces uncertainty. Experiments on the MIMIC-CDM benchmark show that LDTL improves diagnostic accuracy while requiring fewer tests, and ablation studies confirm the importance of trajectory-level posterior alignment.
By Xuyang Shen, Haoran Liu, Dongjin Song, Martin Renqiang Min
The paper introduces a model‑adaptive definition of tool necessity that accounts for each LLM’s empirical performance, contrasting it with observed tool‑call behavior across four models on arithmetic and factual QA datasets. It finds substantial mismatches—26.5‑54.0% for arithmetic and 30.8‑41.8% for factual QA—between when tools are needed and when they are actually used. By probing hidden states, the authors show that the main source of mismatch lies in the transition from internal cognition (recognizing tool necessity) to action (making a tool call), revealing a knowing‑doing gap in LLM tool use.
By Yize Cheng, Chenrui Fan, Mahdi JafariRaviz, Keivan Rezaei, Soheil Feizi
NeuroMAS treats a multi-agent language system as a trainable neural‑network‑like architecture, with large language model agents as nodes and textual signals as edges. The topology defines general information flow, while reinforcement learning determines how agents communicate, specialize, and coordinate, making the system role‑free but structure‑aware. Experiments show NeuroMAS outperforms both inference‑time and trained multi‑agent baselines, and that scaling larger systems is feasible when grown progressively from smaller trained systems.
By Haoran Lu, Luyang Fang, Wenxuan Zhong, Ping Ma
AlphaMemo is a self‑evolving alpha‑mining agent that uses Structured Search‑Process Memory to improve factor discovery. Instead of storing only final factors or full search trajectories, it records reusable evidence about which edit motifs succeed or fail in specific parent‑factor contexts, extracted from Abstract Syntax Tree differences. The system incorporates confidence‑gated residual memory, a search‑ledger prior, and asymmetric veto control to suppress high‑confidence failure patterns, leading to better out‑of‑sample performance and fixed‑budget discovery efficiency on CSI 500 and S&P 500 datasets.
By Hang Yu, Zifan Zheng, Chun Gan, Jeff Z. Pan, Tongliang Liu, Zhiyong Wang, Fengxiang He
The study examines how swapping members in large‑language‑model (LLM) teams affects performance. By replacing a team member with a newcomer who either carries the original team’s written notebook, a notebook from another team, or no notebook, the authors find that newcomers with the original notebook often outperform those without it in cooperative tasks such as cooking, wildfire response, and medical Q&A. However, when compared to the original team, these replacements consistently underperform and require more messages per unit of progress, with the deficit persisting across model families and decoding temperatures and narrowing only when planning‑only members are added.
By Jianxin Gao, Runze Li, Tianyi Yu, Liangwei Ren, Bohan Chen, Zining Wang
EvoSkill v2 is a three‑agent system that edits the instructions (skills) given to an agent while keeping its model and harness unchanged. The system comprises a benchmark adaptor that turns user‑supplied benchmarks and graders into plugins, a solver that attempts tasks, and a stateful evolver that rewrites skills each epoch based on the solver’s training rollouts. Compared to existing skill optimizers, EvoSkill v2’s evolver uses a single coding agent with a bash tool, accepts every epoch’s skills, and allows multiple attempts per training task, achieving higher held‑out pass rates and lower cost per performance gain on SpreadsheetBench v2 and $ au^3$‑bench banking.
whyItMatters":"EvoSkill v2 demonstrates that a single, stateful evolver can more efficiently improve agent skills, enabling even less capable models to achieve significant performance gains in benchmark tasks."
By Dastin Huang, Baran Nama, Salaheddin Alzubi
The paper argues that current AI systems are limited by their fixed representational frameworks and that true open‑ended intelligence requires the ability to extend its own semantic domain. By modeling cognitive states as partial evaluations over a formal language, the authors show that unresolved propositions create a frontier for extension that cannot be fully resolved by finite architectures. They propose that intelligence must transform unresolved structures into new spaces of meaning—generating new distinctions, questions, representations, and purposes—through persistent external memory and collective generative capacity.
By Simon Hegelich, Bardh Prenkaj, Habiba Sarhan, Morteza Shahrezaye, Kolja Hegelich, Erik Cambria, Christian Montag
The paper reports the first in‑space demonstration of using machine learning‑based warm starts to accelerate trajectory optimization for the Astrobee free‑flying robot on the International Space Station. A neural network is trained to learn the structure of the trajectory generation problem solved by sequential convex programming, and onboard predictions reduce solver iterations by 60% for rotational dynamics and 50% for obstacle‑laden scenarios. This approach enables real‑time trajectory planning while respecting safety constraints on limited flight hardware.
By Somrita Banerjee, Abhishek Cauligi, Marco Pavone
arXiv:2602.22402v2 Announce Type: replace-cross
Abstract: As large language models engage in extended reasoning tasks, they accumulate significant state -- architectural mappings, trade-off decisions...
By Cosmo Santoni
arXiv:2603.13023v3 Announce Type: replace-cross
Abstract: Training capable software engineering (SWE) agents demands large-scale, executable, and verifiable environments that provide dynamic feedback...
By Dayuan Fu, Shenyu Wu, Yunze Wu, Zerui Peng, Yaxing Huang, Jie Sun, Ji Zeng, Mohan Jiang, Lin Zhang, Yukun Li, Jiarui Hu, Liming Liu, Jinlong Hou, Pengfei Liu
arXiv:2608.25577v2 Announce Type: replace-cross
Abstract: With the widespread adoption of personal intelligent agents, users generate massive, heterogeneous data during long-term interactions. Levera...
By Yu Wang, Jiaheng Lu
arXiv:2610.20959v2 Announce Type: replace-cross
Abstract: Answering questions over heterogeneous data means combining evidence from files, databases, documents and the web. Data agents automate this,...
By Zhivar Sourati, Marianne Menglin Liu, Zheng Wang, Rongguang Wang, Sai Ashish Somayajula, Tao Sheng, Kyu J. Han, Sujith Ravi, Dan Roth
arXiv:2610.31820v1 Announce Type: new
Abstract: Reliable confidence estimation is essential for trustworthy deployment of large language models (LLMs), but dedicated confidence training stage require...
By Shujin Wu, Heng Ji
arXiv:2610.32333v1 Announce Type: new
Abstract: When LLM-powered agents interact, information is repeatedly shared, expanded, and forgotten, yet multi-agent evaluation typically focuses on final task...
By Haocheng Jiang, Herun Wan, Yongyuan Peng, Minnan Luo
arXiv:2610.32441v1 Announce Type: new
Abstract: Large language model (LLM) agents are increasingly capable of supporting scientific workflows, but their capabilities across diverse research activitie...
By Tianyun Zhong, Wei Wang, Jialin Li, Jianhong Tu, Liyao Li, Chuqiao Kuang, Tianshi Zheng, Songchi Zhou, Weiqi Zhai, Qile Jin, Jiajia Li, Jiasheng Zheng, Boxi Cao, Yaojie Lu, Hongyu Lin, Shiwei Ye, Bing Zhao, Hu Wei, Lin Qu, Xianpei Han
arXiv:2610.32802v1 Announce Type: new
Abstract: On-policy distillation aligns teacher feedback with a student's current behavior, yet acquiring useful supervision for multi-turn agents requires ident...
By Zheng Jiang, Yiming Chen, Houde Qian, Ling Li, Chaoyang Li, Lifeng Sun