https://techgazette.com.br/research/#ai This week's papers in AI and agents, each with what its authors did and found, in sentences from its own abstract. en 180 Fri, 16 Oct 2026 12:00:00 GMT https://dl.acm.org/doi/10.1145/3859251 https://dl.acm.org/doi/10.1145/3859251 Fri, 16 Oct 2026 12:00:00 GMT ACM TIST (Intelligent Systems) AI and agents ACM TIST (Intelligent Systems) — Did: Large Language Models (LLMs) struggle with knowledge-intensive tasks due to hallucinations and fragmented reasoning over dispersed information. https://arxiv.org/abs/2610.22350 https://arxiv.org/abs/2610.22350 Thu, 15 Oct 2026 17:57:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: We introduce FDE-Gym, a benchmark of 95 end-to-end delivery tasks from 20 customer scenarios across 18 sectors, each adapted by industry experts from real-world delivery projects. Found: These results reveal a substantial gap between solving well-specified tasks and delivering effective AI systems for real customers. By Dongxu Yang, Hongda Yan, Renxin Wang, Baoli Zhang, Haojia Zuo, Ying Zhao. https://arxiv.org/abs/2610.22344 https://arxiv.org/abs/2610.22344 Thu, 15 Oct 2026 17:56:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: We propose a post-training method for autonomous long-horizon MLE optimization and apply it to train MLExpert, a 35B Mixture-of-Experts agentic model. Found: Further experiments show that MLExpert improves performance on NatureBench Lite, agent scaffold optimization, and algorithm optimization, suggesting that the learned optimization capability generalizes to other tasks. By Shangheng Du, Jinxin Shi, Shuaiyu Zhang, Yiming Wu, Junpeng Zhao, Shiyang Feng. https://arxiv.org/abs/2610.22333 https://arxiv.org/abs/2610.22333 Thu, 15 Oct 2026 17:53:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: We release all 47,880 judged responses to support further work. Found: These results suggest that safety evaluations in Indic languages may not reflect how models treat their everyday users. By Trizal Garg, Sibayan Mitra, Arya Karanjkar, Advait Singh, Jagat Sesh Challa, Yash Sinha. https://arxiv.org/abs/2610.22331 https://arxiv.org/abs/2610.22331 Thu, 15 Oct 2026 17:52:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: We propose KnowGPO, a reinforcement learning framework that aligns trajectories by the task-relevant knowledge they have acquired rather than by their surface observations. Found: Experiments on Xbench, BrowseComp-ZH, and GAIA show KnowGPO consistently outperforms outcome-only (GRPO) and process-supervision baselines (GiGPO, IGPO) by 3.8 to 4.2 points, with our 8B model rivaling or exceeding most 32B open-source systems. By Siqi Chen, Zhengxi Lu, Yuchen Yan, Xingyu Wu, Qipeng Chen, Xin Zhang. https://arxiv.org/abs/2610.22330 https://arxiv.org/abs/2610.22330 Thu, 15 Oct 2026 17:52:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: In this paper, we conduct a systematic study on Transformer architectures for CDLMs and introduce Language Diffusion Transformer (LangDiT), combining feature and query-key normalization, wider noise conditioning, likelihood-shortcut gradient routing and calibration, and confidence-gated training self-conditioning. Found: Our study identifies architectural design as an important source of improvement in continuous diffusion language models. By Weijian Luo, Yujian Chen, Hao Jiang, Guang Cheng, Tianyang Hu. https://arxiv.org/abs/2610.22322 https://arxiv.org/abs/2610.22322 Thu, 15 Oct 2026 17:51:00 GMT arXiv cs.MA AI and agents arXiv cs.MA — Did: We introduce market-based post-training (MPT), a framework for jointly learning action policies and self-orchestration in decentralized agent swarms. Found: Ablations show that both action and bidding benefit from learning, and that learned bids provide more effective action-training signals. By Guowei Xu, Zhenting Qi, Amani Kiruga, Himabindu Lakkaraju, Yilun Du. https://arxiv.org/abs/2610.22319 https://arxiv.org/abs/2610.22319 Thu, 15 Oct 2026 17:51:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: We describe the setup, how the run unfolded, the team's strategies (dividing the work, sharing what they learned, recovering each other's items), and the limits of the result. By Maxence Faldor, Satinder Singh, Roberta Raileanu. https://arxiv.org/abs/2610.22297 https://arxiv.org/abs/2610.22297 Thu, 15 Oct 2026 17:45:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: We propose a simple method to preserve datastore privacy in RAG systems via random projection. Found: Our results for the first time suggest that random projection can serve as a competitive and practical baseline for privacy-preserving RAG systems. By Dixi Yao, Tahseen Rabbani, Tian Li. https://arxiv.org/abs/2610.22290 https://arxiv.org/abs/2610.22290 Thu, 15 Oct 2026 17:43:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: We propose EliteReplay, a post-training recipe that pairs GRPO with an "elite" archive built by quality-diversity search. Found: It preserves most of the base model's correct strategies, outperforming existing methods in strategy retention. By Hanane Nour Moussa, Doeun Lee, Sachin Kumar. https://arxiv.org/abs/2610.22289 https://arxiv.org/abs/2610.22289 Thu, 15 Oct 2026 17:43:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: We investigate why this advantage persists and whether an environment-agnostic agent can be improved without environment-specific engineering. Found: A complementary AGI Maze experiment shows that additional tutorial experience can reduce performance. By Sergey Rodionov. https://arxiv.org/abs/2610.22280 https://arxiv.org/abs/2610.22280 Thu, 15 Oct 2026 17:40:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: To this end, we propose LEASH, a training-free framework that unleashes the strong agent only when the weak agent's latent uncertainty signals it is needed, and keeps it on a leash afterward. Found: With Qwen3.5-35B-A3B, LEASH achieves 73.66% task success at $0.4117 per task, matching always-on Claude Opus 5 (72.90%) while reducing inference cost by 19.2%. By Qingni Wang, Changdae Oh, Yuzhe Yang, Sharon Li, Jiachen Yang, Ang Li. https://arxiv.org/abs/2610.22277 https://arxiv.org/abs/2610.22277 Thu, 15 Oct 2026 17:39:00 GMT arXiv (vision, retrieval & HCI) cs.MA AI and agents arXiv (vision, retrieval & HCI) cs.MA — Did: We present a framework for multi-agent motion planning under legal requirements based on games with partially ordered preferences (posetal games). Found: These results indicate that guideline-based posetal games offer an auditable route from partially specified laws to multi-agent motion planning. By Florian Lercher, Karan Mahesh, Yujun Huang, Meshal Alharbi, Runyu Zhang, Hanna Krasowski. https://arxiv.org/abs/2610.22269 https://arxiv.org/abs/2610.22269 Thu, 15 Oct 2026 17:34:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: We propose eXplainable RIC (X-RIC), a novel framework integrating agentic AI into O-RAN architectures. Found: X-RIC achieves 89% operational cost reduction while maintaining SLA compliance, converging 75% faster with 50% fewer training steps and 46% lower variance. By Adhwaa Al-Chaab, Ayman Younis, Dario Pompili. https://arxiv.org/abs/2610.22245 https://arxiv.org/abs/2610.22245 Thu, 15 Oct 2026 17:19:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: In particular, we introduce a notion of model disagreement that can be measured directly from model-generated responses, without ground-truth supervision. Found: Moreover, even when ground-truth correctness signals are available for training, disagreement-based routing can outperform methods trained directly to predict correctness By Sima Noorani, Shayan Kiyani, George Pappas, Hamed Hassani. https://arxiv.org/abs/2610.22243 https://arxiv.org/abs/2610.22243 Thu, 15 Oct 2026 17:18:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: We introduce a three-level shortcut-auditing framework and audit completed LinuxArena trajectories and two TS-Bench-derived step datasets. Found: Both effects remain nonzero after subtracting the formatting-placebo effect, with exploratory 95% ranges excluding zero. By Hasan Demirkiran, Jens Ernstberger. https://arxiv.org/abs/2610.22214 https://arxiv.org/abs/2610.22214 Thu, 15 Oct 2026 17:00:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: To address these issues, we introduce a new conflict graph model capturing the previously overlooked constraints and develop a local refinement strategy powered by the graph neural network, enabling efficient and scalable layout decomposition. Found: Experiments demonstrate that our method achieves higher solution quality with faster convergence than state-of-the-art methods, while maintaining robustness across layout scales, design styles, technology nodes, and varying mask counts. By Yu Feng, Liyi Zhou, Haoyu Wang, Yiming Xu, Hao Geng, Shaozhi Deng. https://arxiv.org/abs/2610.22195 https://arxiv.org/abs/2610.22195 Thu, 15 Oct 2026 16:45:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: Therefore, we introduce HTN-Orchestrate, a two-layer framework that separates these responsibilities through a hierarchical task network (HTN). Found: On five benchmarks spanning web research, scientific reasoning, and research workflows, HTN-Orchestrate outperforms ReAct, Agent Team, and DeepSeek Harness using the same base model, achieving 89.73% accuracy on BrowseComp and 73.44% on SUPERChem, exceeding Agent Team by 2.45 and 4.69 percentage points, respectively. By Yanchen Deng, Jianguo Huang, Yao Long Teng, Yixuan Li, Handuo Zhang, Ning Wang. https://arxiv.org/abs/2610.22194 https://arxiv.org/abs/2610.22194 Thu, 15 Oct 2026 16:44:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: We design a data action space for these changes and ask whether an automated system can make the full range rather than one of them, and what structure that requires. Found: Our findings suggest that automating SFT data research depends more on covering diverse data actions than on a stronger single action. By Bingxiang He, Pengqi Sun, Shihao Peng, Shaoxuan Li, Huiyu Xiao, Hongjian Shi. https://arxiv.org/abs/2610.22192 https://arxiv.org/abs/2610.22192 Thu, 15 Oct 2026 16:43:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: We present Terminal-RSI, which alternates Recursive Self-Rewrite (RSR) of verified trajectories with critic-inheriting Recursive Policy-Refine (RPR). Found: On LHTB, process reward rises from 0.21 to 0.37 and strict completion from 0/46 to 5/46; SWR100 success doubles from 3.00% to 6.00%. By Junyao Yang, Zongxia Li, Yucheng Shi, Zhongzhi Li, Ruhan Wang, Ruiwen Zhou. https://arxiv.org/abs/2610.22163 https://arxiv.org/abs/2610.22163 Thu, 15 Oct 2026 16:26:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: To aid such students, we developed a general-purpose AI tutor system that can process student questions in foreign languages (currently focused on Mandarin) and provide bilingual responses. Found: The tutor also provided high-quality translations for 76.92% of the questions and generated technical term glossaries for 64.47% of them. By Yiyin Shen, Xinyi Ai, Weibing Wang, Rogers Jeffrey Leo John, Meenakshi Syamkumar. https://arxiv.org/abs/2610.22155 https://arxiv.org/abs/2610.22155 Thu, 15 Oct 2026 16:20:00 GMT arXiv cs.AI AI and agents arXiv cs.AI — Did: We introduce Realtime-Bench, a suite of carefully curated dynamic tasks where delayed actions directly affect success. Found: These findings suggest directions for improving real-time multimodal agents, including adaptive reasoning with temporal awareness and continuous perception and action. By Jingyu Tang, Xu Yang, Zhehao Zhang, Zheng Wang, Qi Cheng, Haozhe Zhao. https://arxiv.org/abs/2610.22061 https://arxiv.org/abs/2610.22061 Thu, 15 Oct 2026 15:24:00 GMT arXiv (systems & security) cs.AI AI and agents arXiv (systems & security) cs.AI — Did: We present Functional Agent Operating Procedures (FAOP), a formal theory that brings database-grade discipline to AI agent workflows. By Ali Mohammadi Shanghooshabad. https://dl.acm.org/doi/10.1145/3856806 https://dl.acm.org/doi/10.1145/3856806 Thu, 15 Oct 2026 12:00:00 GMT ACM TIST (Intelligent Systems) AI and agents ACM TIST (Intelligent Systems) — Did: Video Individual Counting (VIC), which aims to enumerate unique individuals in a video sequence, is a vital problem with key applications in urban analytics, including transportation management and public safety. https://dl.acm.org/doi/10.1145/3856808 https://dl.acm.org/doi/10.1145/3856808 Wed, 14 Oct 2026 12:00:00 GMT ACM TIST (Intelligent Systems) AI and agents ACM TIST (Intelligent Systems) — Did: The use of synthetic data is becoming increasingly widespread in a variety of applications.