Daily AI Papers — September 2, 2026

1. StudentSim: Training LLM-based Student Simulators

Authors: Ke Yang, Chenglong Wang, Michel Galley, Chandan Singh, Jeevana Priya Inala, ChengXiang Zhai, Jianfeng Gao arXiv: arxiv.org/abs/2609.01591 Summary: AI tutors are most useful when they adapt to each student’s strengths, weaknesses, and preferred guidance, but evidence about which guidance works for which student is sparse, slow, and costly to collect from real learners. Student simulators can provide this signal as a proxy, yet existing approaches are limited: state-tracking models fit student behavior but struggle to process explanations or corrections, while LLM role-play follows guidance fluently but does not reliably match the competence of the student being imitated. Trending because: 484 HuggingFace upvotes + strong interest in realistic student simulation for adaptive AI tutoring

2. Qwen-Drive-1.0: An Initial Step towards a Vision-Language Foundation Model for Autonomous Driving

Authors: Xin Zhou, Zongchuang Zhao, Zhibo Yang, Mingsheng Li, Humen Zhong, Shuai Bai, Du Chu, Ruizhe Chen, Zhaohai Li, Jun Tang, Qiuyue Wang, Mingkun Yang, Jiazhao Zhang, Dayiheng Liu, Dingkang Liang, Xiang Bai arXiv: arxiv.org/abs/2609.00111 Summary: We present Qwen-Drive-1.0, an initial step towards a vision-language foundation model for autonomous driving. Qwen-Drive-1.0 retains the architecture of the pretrained vision-language model (VLM) and integrates 3D perception, visual question answering, and motion planning within a unified framework. Trending because: 377 HuggingFace upvotes + a major vision-language foundation model for autonomous driving

3. SMELT: Scaling Laws for Compute-Matched MoE Looped Transformers

Authors: Shaowen Wang, Ge Zhang, Kairong Luo, Yuhao Wu, Shaofan Liu, Jiaheng Liu, Wenhao Huang, Shen Yan, Jian Li arXiv: arxiv.org/abs/2609.01343 Summary: Looped Transformers increase effective depth by iterating a shared block of layers, but most evaluations compare at fixed model size, conflating architectural advantage with extra FLOPs. We study looping on Mixture-of-Experts Transformers while closely matching per-token FLOPs, total non-embedding parameters, and KV cache. Trending because: 99 HuggingFace upvotes + compute-matched scaling laws clarify the value of looped MoE architectures

4. UI-Venus-2 Technical Report

Authors: Venus Team, Zhuohan Cai, Haoxing Chen, Jiaxuan Chen, Weizhi Chen, Changlong Gao, Zhangxuan Gu, Yuan Guo, Yusong Hu, Jianrong Jiang, Jianguo Li, Runze Li, Jinzhen Lin, Zhenyu Ma, Changhua Meng, Han Peng, Xinyu Qiu, Shuheng Shen, Zhongyi Shui, Weiqiang Wang, Ming Wen, Zhuoer Xu, Hang Yan, Kaiwen Yang, Ruilin Yao, Nanjun Yu, Zhengwen Zeng, Lianrui Zhang, Yunzhu Zhang, Zhe Zhao, Beitong Zhou arXiv: arxiv.org/abs/2609.00028 Summary: Multimodal GUI agents have emerged as a promising paradigm for digital task automation, yet transitioning from benchmark-oriented models to dependable real-world applications remains challenging due to limited environment coverage, brittle task construction, and unreliable reward verification. In this work, we present UI-Venus-2, a general-purpose foundation GUI agent designed to operate across mobile, web, and desktop environments through a unified closed-loop reasoning-action framework. Trending because: 62 HuggingFace upvotes + cross-platform GUI agents are central to practical digital automation

5. ZimaBlue: Evolving Generalizable World Action Models through Scalable Video Pre-training

Authors: Xionghao Wu, Yijun Yang, Shiyang Zhou, Haoze Sun, Jianhui Liu, Songsong Yu, Jiyao Zhang, Wenbo Li, Bo Wang, Guoqing Ma, Lin Song, Renjie Liao, Shenghe Zheng, Wei Tang, Xiaojuan Qi, Yanwei Li, Yuan Zhang, Zhuotao Tian, Haoyang Huang, Nan Duan arXiv: arxiv.org/abs/2609.00188 Summary: Robotic manipulation faces a fundamental scaling challenge: robust generalization demands broad physical experience, yet action-labeled robot trajectories are expensive to collect and inherently limited in diversity. Egocentric videos offer a far more scalable source of embodied experience, capturing object interactions, contact dynamics, tool use, and long-horizon behaviors across diverse environments. Trending because: 51 HuggingFace upvotes + scalable video pre-training targets broader robotic manipulation generalization

6. H3-World: Turning Language Understanding into World Control

Authors: Danze Chen, Zeqing Wang, Ziyue Lin, Xingyi Yang, Yeying Jin arXiv: arxiv.org/abs/2609.01560 Summary: We present H3-World, an efficient framework that turns the 33B MiniMax-H3 video generator into an interactive world model. Our key finding is that, as large video generators become more capable, language is emerging as a natural interface for control. Trending because: 50 HuggingFace upvotes + language-controlled interactive world models connect generation with action

7. Hi-Q: Hierarchical Evidence-guided Query Refinement for Multi-Hop Question Answering

Authors: Jueun Kim, Sungho Park, Wook-Shin Han arXiv: arxiv.org/abs/2608.30468 Summary: A central bottleneck in multi-hop Question Answering (QA) is that the granularity at which a question is expressed often differs from the granularity at which corpus evidence is retrievable. Existing methods address this mismatch by imposing fixed graph structures over the corpus, by iteratively reformulating the query, or by executing a generated program over it, but these strategies do not explicitly decide when a query unit is already supported by evidence and when it should be refined. Trending because: 34 HuggingFace upvotes + evidence-guided query refinement addresses a core multi-hop QA bottleneck

8. From Production Traffic to Post-Training: Building a Self-Hosted LLM That Covers the Corporate Request Mix

Authors: Olga Tsymboi, Dmitrii Stoianov, Ramil Latypov, Danil Taranets, Daniil Dryabin, Mikhail Gashkov, Viktor Zelenkovskiy, Aleksandr Fida, Gleb Alektorov, Nikita Gulyakov, Arthur Babkin, Aleksandr Medvedev, Pavel Gein, Anatolii Potapov arXiv: arxiv.org/abs/2609.01572 Summary: Data-residency constraints force enterprises to self-host LLMs, but continuous adoption of newer models without decommissioning their predecessors expands the serving fleet, fragmenting a finite GPU pool. We consolidate traffic from over 200 internal applications onto a single model by closing quality gaps identified through production error analysis along three axes: instruction following, function-calling, and internal task distribution. Trending because: 33 HuggingFace upvotes + production-driven post-training tackles enterprise LLM fleet consolidation

9. Evaluating Multimodal LLMs as Generalist Vision-Language-Action Agents for Drone Control: Commanding, Approaching, Tracking and Searching

Authors: Jaewoo Park, Minyoung Lee, Sukmin Seo, Moonbin Yim, Hyunwook Yoon, Dohoon Ryu, Daehee Kim, Myungseo Song, Jihyuk Byun, Seunggyu Chang, Taeho Kil, Jiseob Kim, Bado Lee, Geewook Kim arXiv: arxiv.org/abs/2609.01404 Summary: Multimodal Large Language Models (MLLMs) are strong perceivers of images and video. We ask how far that reach extends into acting: dropping an MLLM directly into a drone’s control loop, with its entire action space declared solely in the prompt. Trending because: 27 HuggingFace upvotes + generalist multimodal control pushes VLMs into real drone action loops

10. Uncovering Understanding-Generation Synergy in Native Unified Multimodal Models: From Representation, Task to System

Authors: Penghao Wu, Haiwen Diao, Weichen Fan, Lewei Lu, Dahua Lin, Ziwei Liu arXiv: arxiv.org/abs/2609.01607 Summary: While unified multimodal models (UMMs) jointly perform visual understanding and generation within a single model, functional unification does not guarantee learning synergy: the two objectives may reinforce each other, compete for capacity, or merely coexist. We investigate their relationship at the representation, task, and system levels in a controlled, structurally native setting without pretrained vision priors. Trending because: 23 HuggingFace upvotes + it tests whether multimodal understanding and generation truly reinforce each other

11. DiagEvo: Diagnosis-Guided Self-Evolution via Hierarchical Error Memory

Authors: Xincheng Wei, Yifan Ding, Yoshua Li, Dongsheng Ma, Rongxiang Weng, Xunliang Cai, Wenjian Ding, Yao Zhang arXiv: arxiv.org/abs/2609.00768 Summary: Self-play is an effective paradigm for language-model self-evolution, but without guidance, solver performance can plateau or decline across rounds. Unguided methods steer question generation with signals such as difficulty, learnability, or diversity. Trending because: 20 HuggingFace upvotes + diagnosis-guided self-evolution aims to prevent language-model performance plateaus

12. Safin-1: Safety from Within through Memory-Native State Evolution

Authors: Ming Zhang, Kaisen Yang, Shu Yu, Ermo Hua, Zhekai Chen, Cheng Jin, Jingnan Zheng, Yi Zhang, Zhongtian Ma, Jiawei Zhou, Sirui Chen, Qiaosheng Zhang, Xiang Wang, Ning Ding, Xia Hu, Bowen Zhou, Youbang Sun, Chaochao Lu arXiv: arxiv.org/abs/2609.00092 Summary: Long-horizon complex tasks require foundation models to accumulate information, maintain internal states, and adapt over extended interactions. Safety should be an intrinsic property of the model itself, rather than a behavioral constraint relying solely on external safeguards or post-hoc alignment such as supervised fine-tuning. Trending because: 20 HuggingFace upvotes + memory-native safety targets intrinsic protection in long-horizon agents

13. AgentJudgeBench: A Multi-Difficulty Benchmark for Evaluating LLM Judges on Agentic Tool-Calling

Authors: Abhigya Verma, Amit Kumar Saha, Seganrasan Subramanian, Sai Harshitha Aluru arXiv: arxiv.org/abs/2608.26623 Summary: LLM judges are widely used to evaluate agentic tool-calling systems, yet their reliability on structured, dependency-driven workflows remains largely unexamined. We present AgentJudgeBench, the first benchmark to systematically study LLM-as-a-judge reliability for agentic tool-calling over workflow DAGs, as distinct from the broader LLM-as-a-judge task of open-ended text or preference evaluation. Trending because: 20 HuggingFace upvotes + reliable LLM judging is essential for evaluating tool-using agents

14. Harness-of-Harness: Multi-Day Autonomous Software Development with Continual Improvement

Authors: Haoyang Yan, Min-le Su, Hangfan Zhang, Zhanhao Li, Chen Zhang, Shao Zhang, Yang Chen, Lei Bai, Shuyue Hu arXiv: arxiv.org/abs/2609.01481 Summary: This paper studies autonomous software development, in which LLM-based coding agents transform high-level requirements into complete, functional, and usable software systems without human intervention. We introduce Harness-of-Harness (HoH), a framework that enables coding agents to continually improve software during autonomous development. Trending because: 17 HuggingFace upvotes + continual improvement over multi-day coding tasks advances autonomous software agents

15. E-Commerce Bench: Evaluating LLM Agents on Long-Horizon Autonomous Business Operation

Authors: Wei Fan, Xinjie Shen, Xudong Guo, Jianhong Tu, Yang Su, Yinger Zhang, Lianghao Deng, Fengyu Wang, Baohua Dong, Yangqiu Song, Dayiheng Liu arXiv: arxiv.org/abs/2608.30730 Summary: Long-horizon agentic tasks go beyond chaining short tasks over more interaction turns. Their evolving dynamic environments and long-range dependencies require Large Language Models (LLMs) to continually explore, learn from experience, and adapt their policies over thousands of steps. Trending because: 16 HuggingFace upvotes + long-horizon business operation is a demanding benchmark for adaptive agents

16. EM^2Mem: Event-Centric Multimodal Memory for Large Language Models

Authors: Yijun Chen, Yaqi Zheng, Yanya Li, Boyi Xiao, Buqiang Xu, Shuofei Qiao, Jizhan Fang, Xinle Deng, Yunzhi Yao, Xuehai Wang, Liuxin Zhang, Hui Li, Huajun Chen, Shumin Deng arXiv: arxiv.org/abs/2609.00551 Summary: Multimodal memory offers a scalable interface for long-video question answering, but existing methods often retrieve captions, frames, transcripts, summaries, or graph facts as isolated fragments. Although searchable, such fragments are not generation-ready: language models must reconstruct cross-modal and temporal alignments at inference time, when context is limited and attribution is difficult. Trending because: 14 HuggingFace upvotes + event-centric memory improves generation-ready retrieval from long multimodal streams

17. Agents in the Large: Perception-Centered Architecture for Persistent Agents

Authors: Shihan Dou, Haoxiang Jia, Shichun Liu, Feng Chen, Chenhao Huang, Yujiong Shen, Shaofan Liu, Jiayi Chen, Jiahang Lin, Honglin Guo, Qianyu He, Minghao Guo, Ziyi Ye, Pluto Zhou, Tao Gui, Qi Zhang, Xuanjing Huang arXiv: arxiv.org/abs/2608.30478 Summary: Cognitive language agents have achieved substantial progress by equipping language models with memory, tools, and decision-making procedures, enabling agents to reason and act in interactive environments. Existing frameworks largely cast these agents as systems for solving user-specified, bounded tasks. Trending because: 11 HuggingFace upvotes + persistent agents require architectures beyond bounded task execution

18. Learning Where Outcomes Change:Credit-Addressable Reasoning for Multimodal Geometry

Authors: Jiani Guo, Junjie Wang, Jie Wu, Pengxiang Zhao, Dongdong Zhang, Shaohan Huang, Yujiu Yang, Furu Wei arXiv: arxiv.org/abs/2608.30457 Summary: Multimodal geometry reasoning requires VLMs to extract precise visual relations and preserve them through multi-step deduction. Existing free-form traces obscure the decisions that determine the answer, and trajectory-level reinforcement learning distributes a single terminal signal across the entire response. Trending because: 9 HuggingFace upvotes + credit-addressable reasoning improves supervision for multimodal geometry

19. Knowledge Distillation During Mid-Training Favors Reasoning over Factual Recall

Authors: Jacqueline He, Howard Yen, Shuyue Stella Li, Margaret Li, Hanqing Zeng, Yinglong Xia, Benyu Zhang, Zhuokai Zhao, Qiang Zhang, Pang Wei Koh, Luke Zettlemoyer, Wen-tau Yih arXiv: arxiv.org/abs/2609.01532 Summary: Logit-based knowledge distillation (KD) is used to train smaller language models (LMs) via supervision from stronger teachers, but whether its benefits are consistent across training stages remains unclear. Through controlled experiments, we find that forward Kullback-Leibler (KL) distillation–the standard KD formulation–with post-trained teachers behaves fundamentally differently during mid-training, an intermediate phase of self-supervised learning on curated corpora. Trending because: 8 HuggingFace upvotes + the study distinguishes distillation effects on reasoning from factual recall

20. Control-Data Flow Separation: Stable Prompt Optimization in Multi-Agent LLMs

Authors: Wentao Zhang, Syed Shariyar Murtaza, Junaid Ahmad Bhatti, Utkarsh Soni, Yifan Nie, Eugene Wen, Yuntian Deng arXiv: arxiv.org/abs/2609.00621 Summary: Prompt optimization can improve multi-agent LLM systems, but the prompts being optimized often serve two entangled roles: generating task-relevant content and specifying execution-critical protocols, such as message routing, output formatting, and termination signals, on which the underlying code relies. As a result, a prompt edit intended to improve content generation can inadvertently corrupt the protocol and cause the entire agent pipeline to fail. Trending because: 8 HuggingFace upvotes + separating protocol control from content stabilizes multi-agent prompt optimization