Daily AI Papers — August 21, 2026

1. EnvHarness: Awakening Static Worlds for Agent Learning

Authors: Chengsong Huang, Zifeng Wang, Rujun Han, Jun Yan, Yanfei Chen, Zoey CuiZhu, Ke Jiang, Peng Xia, Han Yu, Yufan Zhuang, Yifei Ming, Jiaqi Pan, Bhavana Dalvi Mishra, Jiaxin Huang, Burak Gokturk, Tomas Pfister, Chen-Yu Lee arXiv: arxiv.org/abs/2608.19880 Summary: LLM agents learn by interacting with environments, yet these environments are hand-built and static: blind to an agent’s weaknesses, and quickly left behind as it improves. While recent environment generation methods attempt to address this, they require domain-specific pipelines, rely on expensive or unreliable verifiers, and still produce static environments. Trending because: 221 HuggingFace upvotes + surging interest in scalable environments for training capable AI agents.

2. FACET: Preserving Source Intent and Executable State in Terminal Task Synthesis

Authors: Kou Shi, Zun Wang, Qisheng Su, Shiting Huang, Ziao Zhang, Zhen Fang, Qingnan Ren, Jin Liu, Yu Zeng, Yiming Zhao, Lin Chen, Zehui Chen, Feng Zhao arXiv: arxiv.org/abs/2608.18580 Summary: Training terminal agents requires scalable executable supervision, yet synthesizing high-quality terminal tasks remains challenging. Each task couples an instruction, an initialized environment, a reference solution, and an executable verifier; if these artifacts are generated from inconsistent assumptions, the resulting task may be unsolvable or incorrectly evaluated. Trending because: 102 HuggingFace upvotes + surging interest in scalable environments for training capable AI agents.

3. SWE-bench Science: Can Coding Agents Resolve Engineering Tasks in Science?

Authors: Zhipeng Xu, Jiahao Lu, Yining Zheng, Yuxin Wang, Xipeng Qiu arXiv: arxiv.org/abs/2608.19799 Summary: Software increasingly functions as part of the scientific instrument itself, making failures in scientific code capable of compromising not only program behavior but also the evidence underlying scientific conclusions. Yet existing evaluations of coding agents largely emphasize aggregate task success, providing limited insight into why agents fail when repairing scientific software. Trending because: 50 HuggingFace upvotes + the community’s appetite for rigorous new benchmarks that expose model weaknesses.

4. 4DAnyone: Create Anyone in 4D from a Casual Monocular Video

Authors: Yudong Jin, Tao Xie, Qihang Zhang, Zehong Shen, Zhen Xu, Yujun Shen, Hujun Bao, Xiaowei Zhou, Yinghao Xu arXiv: arxiv.org/abs/2608.20335 Summary: We present 4DAnyone, a framework for reconstructing 4D humans from an uncalibrated monocular video by generating reconstruction-grade multiview-consistent videos and lifting them into 4D Gaussian Splatting (4DGS). Existing camera-controlled video diffusion models synthesize plausible novel-view videos but fail to maintain consistency when scaled to the tens of target views required for 4DGS reconstruction. Trending because: 47 HuggingFace upvotes + rapid progress in controllable generative video/4D content creation.

5. WithEveryone: Unified Planning and Identity Grounding for Group Image Generation

Authors: Hengyuan Xu, Qixun Wang, Yiji Cheng, Miles Yang, Zhao Zhong, Wei Cheng, Xingjun Ma, Yu-gang Jiang arXiv: arxiv.org/abs/2608.20336 Summary: Identity-preserving image generation becomes increasingly unreliable when a scene must contain many specified people. Beyond retaining each identity, the model must bind every reference to a distinct person and location, while training-time identity losses must establish correspondence among several noisy predicted faces. Trending because: 34 HuggingFace upvotes + the community’s appetite for rigorous new benchmarks that expose model weaknesses.

6. MemTrapBench: Benchmarking Cognitive Traps in LLM Memory Use

Authors: Mengru Wang, Haozhe Luo, Zhenqian Xu, Zhixiang Cui, Haoming Xu, Qu Yang, Jizhan Fang, Junfeng Fang, Ningyu Zhang arXiv: arxiv.org/abs/2608.20202 Summary: Memory has become a key component of large language models, enabling them to retain information and learn from long-term interactions. However, existing memory benchmarks mainly evaluate whether information is correctly extracted, stored, and retrieved, while largely overlooking how retrieved memories reshape model reasoning and affect performance on the current task. Trending because: 24 HuggingFace upvotes + the community’s appetite for rigorous new benchmarks that expose model weaknesses.

7. SkillEvo: Self-Renewing Evolution Gradients from Multi-Turn Interaction Feedback

Authors: Qianxi Yan, Chunrong Chen, Jiuzhou Zhao, Min Zhang, Yongzhou Xu, Xiaochuan Xu arXiv: arxiv.org/abs/2608.13120 Summary: Agent Skills are today either hand-authored or produced in a single LLM generation pass, and consequently possess no closed loop through which they might improve from the interaction failures they actually cause. Recent work does close this loop, but derives its feedback from single-turn question-answering evaluation. Trending because: 20 HuggingFace upvotes + intense focus on coding agents and automated software engineering.

8. ForgeWM: Progressive Causal Training for Few-Step Action-Conditioned Video World Models

Authors: Xinye Li, Lingshuai Lin, Lei Wang, Liuzhou Zhang, Jialin Cui, Qingshan Li, Guanchu Wang, Qingbin Liu, Xi Chen, Jiang Bian, Wai Lam arXiv: arxiv.org/abs/2608.14022 Summary: Action-conditioned video world models require low-latency causal generation and reliable responses to game-native controls. Although causal distillation enables one- or few-step video synthesis, extending it to interactive world models remains challenging, as discrete keyboard states and continuous mouse motion must remain aligned with temporally compressed latent chunks during causal training and autoregressive rollout. Trending because: 13 HuggingFace upvotes + rapid progress in controllable generative video/4D content creation.

9. Repo0: Design-Driven Zero-to-All Code Generation

Authors: Silin Chen, Haoyi Teng, Xiaodong Gu, Yuling Shi, Jiale Huang, Yongpan Wang, Hongyu Zhang, Haibing Guan arXiv: arxiv.org/abs/2608.19854 Summary: Large language model agents have made substantial progress in code generation, yet most existing systems assume a predefined repository architecture. This assumption does not hold in zero-to-all code generation, where an agent must construct an entire software project directly from natural-language requirements while maintaining a modular repository architecture throughout development. Trending because: 10 HuggingFace upvotes + surging interest in scalable environments for training capable AI agents.

10. FlashPrefill V2: Block-Sparse Prefill Attention for Long-Context LLM Serving

Authors: Qihang Fan, Huaibo Huang, Zhiying Wu, Bingning Wang, Ran He arXiv: arxiv.org/abs/2608.19758 Summary: Long-context modeling is a pivotal capability for Large Language Models, yet the quadratic complexity of attention remains a critical bottleneck, particularly during the compute-intensive prefilling phase. Our previous work, FlashPrefill, mitigates this cost through instantaneous pattern discovery and max-based dynamic thresholding; however, it remains an algorithmic prototype that is still distant from production deployment. Trending because: 9 HuggingFace upvotes + demand for efficient long-context inference at scale.

11. VA-Judger: Reward Modeling from Human Preference Feedback for Joint Video-Audio Generation

Authors: Yinming Huang, Shuyuan Tu, Xi Yan, Zihan Yang, Jianhua Han, Xu Hang, Yu-Gang Jiang, Zuxuan Wu arXiv: arxiv.org/abs/2608.18607 Summary: Using reinforcement learning to post-train joint video-audio generation models requires a reward signal. Existing methods construct this reward by combining metrics for individual quality dimensions, including audio quality, visual fidelity, and synchronization. Trending because: 9 HuggingFace upvotes + the community’s appetite for rigorous new benchmarks that expose model weaknesses.

12. Thinking in a Low-Resource Language: What SFT Builds, What RL Fixes, What Accuracy Cannot See

Authors: Ayoub Kirouane, Christos Petrocheilos arXiv: arxiv.org/abs/2608.17744 Summary: Take three frontier mixture-of-experts models (Alibaba, OpenAI, NVIDIA; 3.6-4.0B active parameters each) and fine-tune them to reason in a low-resource language. On accuracy benchmarks almost nothing happens, and the benchmark itself is noise at this scale: changing only the random seed moves the score by 7.7 points, more than every data and recipe effect we measured. Trending because: 7 HuggingFace upvotes + the community’s appetite for rigorous new benchmarks that expose model weaknesses.

13. EXIMO: VLM Guided Exploration of VLA Policies

Authors: Bhavya Sukhija, Oliver Groth, Mohit Shridhar, Tim Hertweck, Michael Bloesch, Markus Wulfmeier, Abbas Abdolmaleki, Martin Riedmiller arXiv: arxiv.org/abs/2608.19891 Summary: How to efficiently finetune robot policies to learn new tasks on the fly? State of the art robotic manipulation policies are based on behaviour cloning of large vision-language-action (VLA) models with billions of parameters on huge teleoperation datasets. Trending because: 4 HuggingFace upvotes + demand for efficient long-context inference at scale.

14. Inject, Align, Recover: Staged Post-Training for Retrieval-Free Document Knowledge Internalization

Authors: Qian Kou, Xiaofeng Shi, Xiaosong Qiu, Hua Zhou arXiv: arxiv.org/abs/2608.20281 Summary: Large language models often fail to answer questions about a bounded document collection when the source documents are not retrieved at inference time. We study this setting as document knowledge internalization: converting a fixed corpus into usable parametric knowledge for retrieval-free question answering. Trending because: 4 HuggingFace upvotes + the community’s appetite for rigorous new benchmarks that expose model weaknesses.

15. LLMs Get Smarter from Targeted Synthetic Multilingual Data

Authors: Ishika Agarwal, Arkajyoti Charaborty, Tanner Sorensen, Neha Gupta, Andreas Stolcke arXiv: arxiv.org/abs/2608.15964 Summary: Language-specific competency (LSC) is the phenomenon of a language model performing better or worse depending on the language of the prompt. In other words, a language model outputs different (and potentially incorrect) responses to the same semantic query when prompted in different languages. Trending because: 4 HuggingFace upvotes + intense focus on coding agents and automated software engineering.

16. Towards Quantifying Benchmark Optimization in ASR Models

Authors: Theo Lebryk, David Ayllon, Alice Baird, Jakub Piotr Cłapa, Jens Madsen, Panagiotis Tzirakis arXiv: arxiv.org/abs/2608.19936 Summary: Public benchmarks are important measures of Automatic Speech Recognition (ASR) model capabilities. However, by nature of being public, there is risk of models being optimized for these benchmarks in ways that do not generalize well to real-world data. Trending because: 4 HuggingFace upvotes + the community’s appetite for rigorous new benchmarks that expose model weaknesses.

17. SkillGate: Training In-Policy Skill Selection in Long-Horizon Agents

Authors: Qingyao Li, Wenxiang Jiao, Shuai Shao, Kangning Zhang, Yuan Lu, Yi Guo, Weiwen Liu, Weinan Zhang, Yong Yu arXiv: arxiv.org/abs/2608.18852 Summary: Agent frameworks increasingly package procedural knowledge as skills: instruction files an agent reads on demand, while public libraries now hold thousands of them. Which skill to read has thus become a decision the policy itself makes in the middle of an episode, yet no existing signal trains it. Trending because: 3 HuggingFace upvotes + the community’s appetite for rigorous new benchmarks that expose model weaknesses.

18. The Problem Is the Problem: Towards Scalable Mathematical Discovery

Authors: Zeyu Zheng, Shengtong Zhang, Jeremy Avigad, Prasad Tetali, Sean Welleck arXiv: arxiv.org/abs/2608.16977 Summary: AI systems are increasingly capable of contributing to mathematical research. In research practice, frontier-model reasoning is a limited resource, and expert mathematical review is even more sharply constrained. Trending because: 3 HuggingFace upvotes + demand for efficient long-context inference at scale.

19. Bounded Agents: Delegation Security for Multi-Agent AI Systems

Authors: Xabier Muruaga arXiv: arxiv.org/abs/2608.15888 Summary: LLM-based agents can act on behalf of a user to access cloud services, call tools, or invoke agents. At session start, the agent’s permissions are set but remain static, and each request is evaluated independently, without considering prior actions. Trending because: 2 HuggingFace upvotes + broad relevance to current LLM research directions.

20. Chain-of-Experience for Continual LLM Improvement

Authors: Haoqin Tu, Yunhao Fang, Yizhong Wang, Cihang Xie, Shen Yan arXiv: arxiv.org/abs/2608.18027 Summary: Humans continuously learn from experience, whereas conventional large language model (LLM) evaluations ignore the models’ ability to improve through inference-time interaction. In this paper, we study how LLMs learn from iterative experience at test time, a setting we refer to as Chain-of-Experience (CoE), where models accumulate experiential traces through iterative interactions with self or environmental feedback to form a continual improvement loop beyond zero-shot inference. Trending because: 2 HuggingFace upvotes + active work on reward modeling and post-training alignment.