{"ok":true,"snapshot":{"date":"2026-11-01","capturedAt":"2026-11-01T14:15:25.473Z","total_papers":30,"raw_count":50,"papers":[{"paperId":"2610.36671","title":"Look Only When Needed: Certificate-Guided Visual Routing for Efficient Diffusion VLMs","summary":"Diffusion vision-language models (DVLMs) process their entire input at every denoising step. Since visual tokens make up 89-97% of this input on the benchmarks we study, most of the decoding computation is spent repeatedly processing the same image. Visual token pruning methods partially relieve this burden by removing part of the visual tokens, but they lose the removed information and still process the remaining image at every step. To measure how often the image is actually needed, we compare LLaDA-V's predictions with and without the image at every decoding step and find that removing the image changes a token being unmasked at only 22.8% of steps on average and at as few as 7.9%. We therefore propose LOWN, a decoding method that decides at every step whether to process the image and …","authors":["Donghyeon Cho","Joowon Kim","Seungho Shin","Eunho Yang"],"publishedAt":"2026-10-27T00:00:00.000Z","submittedAt":null,"upvotes":101,"num_comments":3,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.36671.png","hf_url":"https://huggingface.co/papers/2610.36671","arxiv_url":"https://arxiv.org/abs/2610.36671","github_repo":null,"github_stars":null,"ai_keywords":["diffusion vision-language models","visual token pruning","LLaDA-V","LOWN","masked tokens","key-value caching","decoding FLOPs"]},{"paperId":"2610.38568","title":"A-RSI: Adaptive Task Refresh for Zero-Data Multimodal Self-Improvement","summary":"Recursive self-improvement requires models not only to acquire useful training experience, but also to decide when their current experience has ceased to be useful. Existing methods often focus on what experience to generate next, while paying less attention to when fresh experience should be introduced. Many also construct training targets through self-consistency or majority voting, making supervision depend on the model's own uncertain predictions. We address these two questions separately: training dynamics determine when to acquire new experience, while the answers to new tasks remain independently checkable. A-RSI monitors reward variation and accuracy in training logs to refresh a task pool when its learning signal declines, without consulting validation or test scores. A static li…","authors":["Yu Liu","Youxing Li","Yuedong Tan","Zhiwei Yang","Zihan Su","Hanqi Feng","Shaojie Zhang","Libin Liao"],"publishedAt":"2026-10-28T00:00:00.000Z","submittedAt":null,"upvotes":68,"num_comments":4,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.38568.png","hf_url":"https://huggingface.co/papers/2610.38568","arxiv_url":"https://arxiv.org/abs/2610.38568","github_repo":null,"github_stars":null,"ai_keywords":["recursive self-improvement","training dynamics","reward variation","task pool","multimodal questions","model-authored programs","self-consistency","verification","A-RSI"]},{"paperId":"2610.39302","title":"Proving and Improving the Platonic Representation Hypothesis with JEPAs","summary":"Whether models trained on different modalities converge to the same representation of the world is central to AI. The Platonic Representation Hypothesis (PRH) conjectures that they do as they scale, but without a proof it offers neither a guarantee nor a recipe. We turn the hypothesis into a theorem. First, we prove a Platonic Representation Theorem: Joint-Embedding Predictive Architectures (JEPAs) trained separately on any number of modalities learn the same representation up to one rotation whenever their positive pairs are two observations of one scene--good views--and their misalignment is bounded otherwise. Second, we show that next-token prediction is a JEPA with bad views, parts of a single document, which explains why language models align with vision only through scale. Third, we…","authors":["Randall Balestriero","Yann LeCun"],"publishedAt":"2026-10-29T00:00:00.000Z","submittedAt":null,"upvotes":60,"num_comments":4,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.39302.png","hf_url":"https://huggingface.co/papers/2610.39302","arxiv_url":"https://arxiv.org/abs/2610.39302","github_repo":null,"github_stars":null,"ai_keywords":["Platonic Representation Hypothesis","Joint-Embedding Predictive Architectures","JEPAs","positive pairs","good views","next-token prediction","scaling law of alignment","DINOv2","CLIP","text JEPA"]},{"paperId":"2610.40614","title":"L3G-World: Real-Time Infinite World Generation with Latent 3D Gaussian Memory","summary":"Interactive world models generate visual content conditioned on user actions, yet maintaining spatial consistency over long-horizon exploration in real time remains challenging. Existing approaches rely on either temporal context or explicit 3D memory, both of which scale poorly because memory grows with the length of the generated stream. Moreover, explicit 3D memories condition latent video generators by re-encoding RGB renderings, which precludes real-time generation. In this paper, we introduce a Latent 3D Gaussian Memory (L3G) that combines the efficiency of feed-forward reconstruction with a compact, rapidly renderable scene representation. Building on this memory, we propose L3G-World, in which a memory updater and a video diffusion transformer (DiT) alternate autoregressively to g…","authors":["Jinhyeok Choi","JoungBin Lee","Jongmin Lee","Minkyeong Jeon","Jiyoung Kim","Kaeun Han","Yoonjin Yoon","Seungryong Kim"],"publishedAt":"2026-10-29T00:00:00.000Z","submittedAt":null,"upvotes":39,"num_comments":5,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.40614.png","hf_url":"https://huggingface.co/papers/2610.40614","arxiv_url":"https://arxiv.org/abs/2610.40614","github_repo":null,"github_stars":null,"ai_keywords":["Latent 3D Gaussian Memory","L3G-World","3D Gaussian","VAE latent feature","video diffusion transformer","DiT","autoregressive","spatial consistency"]},{"paperId":"2610.37161","title":"WorkFactory: Turning Professional Intents into Executable Work for Scalable Agent Training","summary":"Scaling supervised training for working agents requires diverse professional tasks that yield reliable expert demonstrations. However, missing or inconsistent task information can obstruct teacher execution and weaken the resulting supervision. We propose WorkFactory, a scalable framework for constructing coherent and executable professional tasks. Starting from professional intents, information-level planning builds an information support graph to guide task instantiation, explicitly representing the dependencies and locations of information required for the intended outputs. Probe executions then identify support failures and guide repairs while preserving the intended work. We construct 112,686 tasks spanning 612 occupations and use the collected teacher trajectories to fine-tune Qwen …","authors":["Ziyang Zeng","Tingyang Chen","Jia Deng","Qiyao Wang","Shuo Tang","Yike Yang","Tao Ren","Wayne Xin Zhao"],"publishedAt":"2026-10-27T00:00:00.000Z","submittedAt":null,"upvotes":36,"num_comments":6,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.37161.png","hf_url":"https://huggingface.co/papers/2610.37161","arxiv_url":"https://arxiv.org/abs/2610.37161","github_repo":null,"github_stars":null,"ai_keywords":["information support graph","probe executions","WorkFactory","WorkPilot","supervised training","fine-tune","Qwen","Elo","scaling analyses"]},{"paperId":"2610.32411","title":"A Broader Look into MLLM's Retrieval Heads","summary":"Locating relevant evidence in long inputs is a core capability of Multimodal Large Language Models. Prior studies attribute this capability to retrieval heads, a sparse set of attention heads that transfer information from the context to the output. However, these analyses are confined to a limited set of modalities and the corresponding models. Moreover, how retrieval heads can be leveraged to enhance the model calls for further exploration. In this paper, we ask whether this mechanism expands together with the input modalities of the model. Taking audio-visual input as a newly added modality, we construct paired inputs in which the same answer is either spoken or shown on screen. With the evidence content fixed, retrieval heads remain sparse in the new modality and are shared to a large…","authors":["Jia Wang","Yuqing Li","Jiangnan Li","Zheng Lin","Mo Yu"],"publishedAt":"2026-10-23T00:00:00.000Z","submittedAt":null,"upvotes":28,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.32411.png","hf_url":"https://huggingface.co/papers/2610.32411","arxiv_url":"https://arxiv.org/abs/2610.32411","github_repo":null,"github_stars":null,"ai_keywords":["retrieval heads","attention heads","multimodal large language models","audio-visual input","omni-modal models","cross-modal sharing","attention ablation","ranking objective","document retrieval"]},{"paperId":"2610.40647","title":"Read, Allocate, Route: Role-Coupled Tokenization for Pixel-Space Generative Transformers","summary":"Pixel-space generative Transformers typically use one token per image patch, so capturing finer detail requires more tokens and makes global attention increasingly expensive. We observe that finer detail does not necessarily require more global communication: low-frequency structure benefits from global exchange, while high-frequency detail can be modeled locally. Based on this insight, we introduce SpecSlot, a pixel-space rectified-flow Transformer that decouples output detail from the length of the global attention sequence. SpecSlot expands each spatial anchor into multiple slots, each assigned a disjoint band of DCT coefficients; this separates anchor density from spectral subdivision. Only the lowest-frequency slot from each anchor participates in global attention, while all slots in…","authors":["Jiongfan Yang","Weichen Fan","Haiwen Diao","Zhongang Cai","Ziwei Liu"],"publishedAt":"2026-10-29T00:00:00.000Z","submittedAt":null,"upvotes":24,"num_comments":6,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.40647.png","hf_url":"https://huggingface.co/papers/2610.40647","arxiv_url":"https://arxiv.org/abs/2610.40647","github_repo":null,"github_stars":null,"ai_keywords":["rectified-flow Transformer","DCT coefficients","global attention","spectral slots","spatial head","FID","text-to-image generation"]},{"paperId":"2610.40599","title":"HairDeck: Animation-Ready Stylized Hair Cards from a Single Image","summary":"The hair of stylized characters in games and avatars is built from hair cards: artist-placed ribbons whose geometry is the visible hair, animated by spring bones. Existing image-based methods produce realistic hair or a single fused mesh, while CHARM, the only generator of stylized hair cards, decodes a hairstyle autoregressively as a long sequence of control points. Observing that a card hairstyle is an unordered set of artist-defined primitives, we propose HairDeck, which generates it as one: each card is a single token, and a set diffusion transformer denoises all cards together. From a single image, HairDeck generates the cards and the spring-bone chains that drive them, yielding an animatable asset that exports to ready-to-use downstream formats. On CHARM's own test set, it lowers Ch…","authors":["Yuze He","Yixiang Wei","Wang Zhao","Chuhao Chen","Yanning Zhou","Yong-Jin Liu"],"publishedAt":"2026-10-29T00:00:00.000Z","submittedAt":null,"upvotes":23,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.40599.png","hf_url":"https://huggingface.co/papers/2610.40599","arxiv_url":"https://arxiv.org/abs/2610.40599","github_repo":null,"github_stars":null,"ai_keywords":["hair cards","spring bones","set diffusion transformer","control points","Chamfer distance","card-orientation error","image-to-3D"]},{"paperId":"2610.33106","title":"VibeGame: Prompt-to-Game Development with AI-Native Engine and Self-Evolving Adversarial Agent Team","summary":"Large language model agents have enabled software development through natural-language interaction, but extending this paradigm to end-to-end game development remains challenging. Building a playable game requires coordinating code, scenes, assets, animations, physics, and configurations, while allowing developers to validate their integrated behavior through interactive testing. We introduce VibeGame, a framework for creating complete 2D game projects from natural-language prompts and optional reference images. Its AI-native game engine represents project state as schema-validated structured text and provides frame-synchronous runtime control, enabling agents to inspect game states, execute actions, and verify behavior at their own pace. Built upon this interface, we design an Adversaria…","authors":["Wenbo Hu","Ken Li","Jiazhe Wei","Yukang Cao","Weiyi Hong","Jiayi Dai","Chenjun Bai","Jiajun Liang"],"publishedAt":"2026-10-22T00:00:00.000Z","submittedAt":null,"upvotes":22,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.33106.png","hf_url":"https://huggingface.co/papers/2610.33106","arxiv_url":"https://arxiv.org/abs/2610.33106","github_repo":null,"github_stars":null,"ai_keywords":["large language model agents","adversarial agent team","AI-native game engine","schema-validated structured text","frame-synchronous runtime control","self-evolution mechanism","skeletons","modules","collaboration contracts","PA2G"]},{"paperId":"2610.29927","title":"Your Benchmark Is an Empirical Measure Over Difficulty","summary":"A benchmark is not merely a collection of test questions; it naturally induces an empirical measure over difficulty. In this work, we develop this measure-based view of benchmark evaluation. This perspective allows us to identify and formalize two central failure modes: measure mismatch, where the benchmark difficulty distribution differs from the intended difficulty distribution, and coverage gaps, where uncovered regions along the difficulty axis prevent benchmarks from detecting meaningful localized training progress. We propose two diagnostic metrics that capture these two failure modes and provide rigorous theoretical guarantees. Empirically, we analyze 25 popular benchmarks spanning 7 domains using carefully estimated item difficulties. Our results show that many benchmarks implicit…","authors":["Yifan Sun","Naicheng Yu","Jiawen Gong","Jingyan Shen","Huan Zhang"],"publishedAt":"2026-10-21T00:00:00.000Z","submittedAt":null,"upvotes":21,"num_comments":3,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.29927.png","hf_url":"https://huggingface.co/papers/2610.29927","arxiv_url":"https://arxiv.org/abs/2610.29927","github_repo":null,"github_stars":null,"ai_keywords":["measure mismatch","coverage gaps","diagnostic metrics","item difficulty","BenchFill","difficulty-targeted item rewriting"]},{"paperId":"2610.35138","title":"Co-Design Bench: Multimodal Collaborative Coding Evaluation for Agentic Models","summary":"Evaluating human-AI co-design requires testing whether coding agents can translate real users' multimodal requests into working application revisions. We introduce Co-Design Bench, a reference-free benchmark of 273 tasks curated from 37,463 deduplicated queries across 26,314 real user coding sessions. Its two tracks, Multimodal Editing (ME) and Repository Editing (RE), cover localized edits with optional image, audio, video, or document inputs and coordinated repository revisions. Tasks span three capability dimensions: Grounded Understanding, UI Transformation, and Functional Implementation. The tasks pair normalized requests and runnable pre-edit applications with 2,331 hidden expert-authored acceptance criteria. Our judge-then-refute protocol grades criteria using source and browser ev…","authors":["Xilin Wei","Shengyuan Ding","Xinyu Fang","Tiancheng Gu","Yuchi Wang","Zhixun Li","Ziyu Liu","Yuhang Zang"],"publishedAt":"2026-10-26T00:00:00.000Z","submittedAt":null,"upvotes":20,"num_comments":3,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.35138.png","hf_url":"https://huggingface.co/papers/2610.35138","arxiv_url":"https://arxiv.org/abs/2610.35138","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2610.39633","title":"Intrinsic Distribution Matching","summary":"One-step and few-step text-to-image generation must accurately render the details specified in the prompt despite limited opportunities for refinement. Distribution matching distillation (DMD) uses score differences to bring the generated distribution closer to a reference distribution. Following prior work, we estimate this difference with internal velocity branches, avoiding a separate fake-score network. We ask whether an explicit moment constraint improves this learned score supervision. A decomposition of the reverse-KL field into a Gaussian moment component and a non-Gaussian residual identifies statistics that can be constrained explicitly. We introduce Intrinsic Distribution Matching (IDM), which retains the full DMD correction and matches feature means and per-coordinate standard…","authors":["Deyuan Liu","Yansen Han","Kaiwen Zheng","Jun Xie","Jiacheng Liu","Zhenglin Cheng","Peng Sun","Jintao Zhang"],"publishedAt":"2026-10-29T00:00:00.000Z","submittedAt":null,"upvotes":18,"num_comments":4,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.39633.png","hf_url":"https://huggingface.co/papers/2610.39633","arxiv_url":"https://arxiv.org/abs/2610.39633","github_repo":null,"github_stars":null,"ai_keywords":["Distribution Matching Distillation","score differences","internal velocity branches","reverse-KL","Gaussian moment","non-Gaussian residual","Intrinsic Distribution Matching","exponential-moving-average","LoRA","LongText-Bench"]},{"paperId":"2610.22507","title":"RealCLI: Toward a Unified Evaluation of Human-Agent Collaboration for CLI Agents","summary":"Command-line interface (CLI) agents increasingly execute longer-horizon tasks, yet existing benchmarks either assume that user instructions remain fixed throughout execution or evaluate task outcomes without capturing runtime collaboration quality. In this paper, we present REALCLI, a paired benchmark that unifies the evaluation of task execution and human-agent collaboration quality on realistic CLI tasks. To comprehensively measure collaboration, REALCLI builds a five-dimensional framework, operationalized into 22 sub-dimensions and 51 interaction scenario templates. REALCLI provides two modes for paired comparison: REALCLI-EXEC supplies complete specifications upfront, whereas REALCLI-COLLAB introduces a scripted user that delivers progressive information, corrections, constraint revis…","authors":["Yunqi Liu","Peiyu Wang","Bingli Wang","Leo Luo","Xuanwu Song","Xudong Ma","Hongxin Li","Yaxin Du"],"publishedAt":"2026-10-15T00:00:00.000Z","submittedAt":null,"upvotes":18,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.22507.png","hf_url":"https://huggingface.co/papers/2610.22507","arxiv_url":"https://arxiv.org/abs/2610.22507","github_repo":null,"github_stars":null,"ai_keywords":["CLI agents","REALCLI","collaboration framework","progressive information","steerability","trust calibration","safety boundary management","joint collaboration score"]},{"paperId":"2610.38352","title":"WeVisCut: The Whole Editing Room in One Agent From Narration to Long-Form Video","summary":"Producing long-form videos from narration requires interdependent decisions that remain coherent across shots. Existing agentic editors often use multi-stage workflows driven by general models, leaving the quality-cost trade-off of compact, task-adapted policies unclear. We address this gap with WeVisCut, a long-form script-to-video system that connects tool execution with policy learning. Its agent harness maintains shared production state so that tool feedback informs subsequent decisions and repairs. Because finished videos do not expose these decisions, our four-stage data pipeline combines editing inversion with verified replay to turn recovered editing intent into executable supervision. Applied to 3,750 finished videos, this pipeline yields over 130K training examples. We use this …","authors":["Chenyu Mu","Wenming Weng","Qinrui Wang","Xingsong Ye","Hao Yu","Jun He","Chong Sun","Chen Li"],"publishedAt":"2026-10-28T00:00:00.000Z","submittedAt":null,"upvotes":18,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.38352.png","hf_url":"https://huggingface.co/papers/2610.38352","arxiv_url":"https://arxiv.org/abs/2610.38352","github_repo":null,"github_stars":null,"ai_keywords":["agentic editors","policy learning","shared production state","editing inversion","verified replay","executable supervision","9B policy","100-task benchmark","token efficiency"]},{"paperId":"2610.22990","title":"PersonaCUABench: Benchmarking Computer-Use Agents Serving Diverse Personas","summary":"Computer-use agents (CUAs) are commonly evaluated against fixed task specifications, but users may omit requirements, add constraints, or require approval before consequential actions. Completing the benchmark instruction does not guarantee serving the person who asked. We introduce PersonaCUABench, which holds 74 tasks across 11 domains fixed and varies only who asks. A persona agent, playing one of 44 personas in four groups, rewrites each instruction in that user's own words and answers the CUA's questions, while the CUA sees only the dialogue, never the profile or the original instruction. Across nine backbones (3,330 runs), mean task completion falls by 36% relative to the original requests. The losses have two sources: in 59.9% of cells that lose points, the agent engages with what …","authors":["Haoming Li","Defu Cao","Xu Hu","Yingtao Luo","Yiyang Wang","Xianghe Xu","Jiashu He","Wanting Jiang"],"publishedAt":"2026-10-16T00:00:00.000Z","submittedAt":null,"upvotes":17,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.22990.png","hf_url":"https://huggingface.co/papers/2610.22990","arxiv_url":"https://arxiv.org/abs/2610.22990","github_repo":null,"github_stars":null,"ai_keywords":["computer-use agents","PersonaCUABench","persona agent","dialogue evaluation","task completion","user satisfaction"]},{"paperId":"2610.38331","title":"Multi-Token Prediction Done Right","summary":"Multi-Token Prediction (MTP) is now a common ingredient in language-model pre-training, as it improves downstream performance and enables faster inference through speculative decoding. We identify a flaw in standard MTP training, in which a residual component of the MTP gradient interferes with both next-token and multi-token learning, increasingly so with scale. Specifically, the MTP cross-entropy gradient can be decomposed exactly into a KL self-distillation term that pulls each MTP head toward the model's own next-token prediction (NTP) distribution and improves speculative decoding, and a residual that has a low signal-to-noise ratio and buries the useful KL signal. We introduce the KL gradient-recovery trick: when two distributions share a linear output head, the gradient of their KL…","authors":["Jayden Teoh"],"publishedAt":"2026-10-28T00:00:00.000Z","submittedAt":null,"upvotes":15,"num_comments":3,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.38331.png","hf_url":"https://huggingface.co/papers/2610.38331","arxiv_url":"https://arxiv.org/abs/2610.38331","github_repo":null,"github_stars":null,"ai_keywords":["Multi-Token Prediction","speculative decoding","cross-entropy gradient","KL self-distillation","straight-through gradient routing","Soft-MTP","next-token prediction"]},{"paperId":"2610.38425","title":"IntReal: A Real-World Dataset and Benchmark for Multimodal Interleaved Generation","summary":"Humans routinely consume multimodal content, where textual and visual information is naturally interleaved. Although Unified Multimodal Models (UMMs) can jointly generate text and images, multimodal interleaved generation remains largely underexplored. Progress in this area is hindered by two key limitations: (1) the lack of well-defined interleaved generation tasks and corresponding training data, and (2) the absence of fine-grained evaluation protocols that can reliably diagnose model performance across different capability dimensions. To this end, we present IntReal, comprising two complementary components: IntReal-100K, a real-world dataset of multimodal interleaved samples spanning four major categories, and the IntReal Evaluation Suite, which evaluates model performance across seven…","authors":["Shaohan Yu","Jie Qin","Wende Tan","Haiwen Diao","Limeng Qiao","Yueming Lyu","Chenyang Si"],"publishedAt":"2026-10-28T00:00:00.000Z","submittedAt":null,"upvotes":14,"num_comments":4,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.38425.png","hf_url":"https://huggingface.co/papers/2610.38425","arxiv_url":"https://arxiv.org/abs/2610.38425","github_repo":null,"github_stars":null,"ai_keywords":["Unified Multimodal Models","interleaved generation","IntReal-100K","IntReal Evaluation Suite","expert models","Spearman's correlation"]},{"paperId":"2610.39301","title":"Youtu-Parsing-Omni: One Encoder, One Schema, All Modalities","summary":"We present Youtu-Parsing-Omni, an omni-modal parser that converts documents, images, audio, and video into structured records grounded in space and time, providing a perception layer for agents that read, locate, and cite evidence across modalities. Its design rests on one encoder for every modality and one schema for every task. Youtu-Omni-Encoder, initialized from a pre-trained language model to strengthen text reading, encodes all modalities with shared weights in a unified space-time coordinate system, and a few fusion layers let video frames attend to concurrent audio, bringing cross-modal fusion and temporal alignment into perception itself. OmniSchema unifies the outputs of all ten parsing tasks, allowing a single self-evolving post-training procedure, OmniSchema-Aware On-Policy Di…","authors":["Yunhang Shen","Zhangquan Chen","Xiaotian Li","Zhongpeng Cai","Qing Zhou","Zekai Zhang","Kun Zhang","Zezhong Wu"],"publishedAt":"2026-10-29T00:00:00.000Z","submittedAt":null,"upvotes":13,"num_comments":3,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.39301.png","hf_url":"https://huggingface.co/papers/2610.39301","arxiv_url":"https://arxiv.org/abs/2610.39301","github_repo":null,"github_stars":null,"ai_keywords":["omni-modal parser","shared weights","cross-modal fusion","temporal alignment","self-evolving post-training","on-policy distillation","token-level targets"]},{"paperId":"2610.40453","title":"PhysInsert: Physically Plausible Object Insertion with Geometric Priors and Bidirectional Object-Scene Interactions","summary":"We present PhysInsert, a framework for physically plausible reference-based object insertion that jointly adapts objects and their surrounding scenes. Existing methods often rely on masked inputs, generation-based pairs, or training pairs with limited viewpoint variation, leaving pose and illumination adaptation and bidirectional object-scene interactions insufficiently constrained. We construct the Insert World dataset with strictly aligned Blender-rendered counterfactual pairs and object references across viewpoints and lighting conditions. Building on these data, Geometry-Guided Adaptation couples a VGGT geometry stream with image and text streams under target-scene feature supervision to guide spatially consistent insertion. Complementary 2D Object-Scene Spatial Supervision jointly su…","authors":["Yue Gong","Le Zhang","Rui Chen","Song Chang","Xiaochao Qu","Luoqi Liu","Ting Liu","Yebin Liu"],"publishedAt":"2026-10-29T00:00:00.000Z","submittedAt":null,"upvotes":13,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.40453.png","hf_url":"https://huggingface.co/papers/2610.40453","arxiv_url":"https://arxiv.org/abs/2610.40453","github_repo":null,"github_stars":null,"ai_keywords":["PhysInsert","Geometry-Guided Adaptation","VGGT","cross-attention","object-scene spatial supervision","counterfactual pairs","illumination adaptation","pose adaptation"]},{"paperId":"2610.21143","title":"CoSTA: Compressed Speech Token Adaptation for Efficient Spoken Dialogue Models","summary":"Spoken Dialogue Models (SDMs) typically share a speech-token space for generation and synthesis, leading to long autoregressive sequences and costly inference. Lower-rate speech tokenizers shorten sequences but may introduce compression-quality trade-offs and costly replacement without breaking this coupling. To address this limitation, we propose CoSTA, a compatibility-preserving post-hoc framework that decouples generation and synthesis token spaces while retaining the original speech tokenizer and decoder. CoSTA extends the generation token space with compressed representations via mean-initialized vocabulary extension, pronunciation adaptation, and dialogue adaptation, followed by deterministic expansion back to the original speech-token space before synthesis. On bilingual speech-dia…","authors":["Yingzhuo Deng","Chong Li","Bingxin Liu","Yi Shu","Tianyu Peng","Can Xie","Siyao Song","Shuo Ren"],"publishedAt":"2026-10-15T00:00:00.000Z","submittedAt":null,"upvotes":12,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.21143.png","hf_url":"https://huggingface.co/papers/2610.21143","arxiv_url":"https://arxiv.org/abs/2610.21143","github_repo":null,"github_stars":null,"ai_keywords":["Spoken Dialogue Models","speech-token space","autoregressive sequences","speech tokenizers","CoSTA","vocabulary extension","pronunciation adaptation","dialogue adaptation","cross-architecture transfer","low-rate tokenization"]},{"paperId":"2610.37786","title":"WebAnything: From the Wild Web to Scalable Agent Environments","summary":"Web agents require diverse, executable, and resettable website environments for reproducible training and evaluation. Human-assisted mirrors demand per-site engineering, while generative construction can diverge from specified real websites. Capture-and-replay preserves observations, with interaction coverage constrained by captured content and replay capabilities.\n  We present WebAnything, a fully automated framework that reconstructs source-grounded, reusable Web environments from real URLs without per-site manual implementation or repair after initial setup and authorization. An Explorer-Builder collects browser-observable structures, resources, and interaction outcomes; an independent verifier checks frozen candidates and guides local repair or targeted source recapture.\n  We construc…","authors":["Tianci Xie","Zikun Cai","Chaoyun Zhang","Pinxue Guo","Bingxuan Li","Rui Zheng","Hang Yan"],"publishedAt":"2026-10-27T00:00:00.000Z","submittedAt":null,"upvotes":12,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.37786.png","hf_url":"https://huggingface.co/papers/2610.37786","arxiv_url":"https://arxiv.org/abs/2610.37786","github_repo":null,"github_stars":null,"ai_keywords":["WebAnything","Explorer-Builder","frozen candidates","source recapture","interaction trajectories","Qwen3.5-9B","REAL success","WebArena-Verified"]},{"paperId":"2610.38342","title":"Investor's Last Exam: Can AI Agents Do Equity Investment Without Seeing the Future?","summary":"Financial agents must form forecasts and investment judgments from decision-time evidence, with predictions evaluated against subsequent outcomes. Existing benchmarks assess financial knowledge, research workflows, or forecasting, but evaluating the full research process requires a strictly frozen information set and professional evaluation criteria. We introduce Investor's Last Exam (ILE), an equity-research benchmark led by experienced buy-side investors and fund managers. Its three components are a versioned financial database spanning 3.43 million document versions and nine source categories, supplemented by structured market data, with validated timestamps governing retrieval and computation; a task taxonomy organized around professional analyst deliverables; and an expert-led framew…","authors":["Huangfei Jiang","Anning Hu","Kai Sheng","Yuchen Cai","Rui Ge","Xiangyu Kong","Jianghao Lin","Shuzi Chen"],"publishedAt":"2026-10-28T00:00:00.000Z","submittedAt":null,"upvotes":11,"num_comments":4,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.38342.png","hf_url":"https://huggingface.co/papers/2610.38342","arxiv_url":"https://arxiv.org/abs/2610.38342","github_repo":null,"github_stars":null,"ai_keywords":["financial dependency graphs","versioned financial database","forecast accuracy","simulated investment performance","evidence-assumption-judgment chain"]},{"paperId":"2610.30161","title":"TauIndianBankBench: Procedure-Aware, Action-Oriented Evaluation of Tool-Using Conversational Agents","summary":"Many Indian banks serve customers through FAQ pages and scripted chatbots. Assistants based on language models could respond faster and carry out requests such as blocking a card, but only if they give correct information and follow the bank's rules. These rules prescribe a procedure, a set of required steps, such as obtaining a closure quote before closing a deposit. Adapting models to this domain requires measuring how well they follow the procedure, before and after training. Benchmarks usually judge only the records left at the end of a conversation. Many required steps, however, leave the records unchanged, so an assistant that skips them can still be judged successful. We present TauIndianBankBench, an action-oriented benchmark of 1,000 Indian retail-banking tasks where an assistant…","authors":["Krishanu Adhikary","Prashant Devadiga"],"publishedAt":"2026-09-30T00:00:00.000Z","submittedAt":null,"upvotes":11,"num_comments":3,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.30161.png","hf_url":"https://huggingface.co/papers/2610.30161","arxiv_url":"https://arxiv.org/abs/2610.30161","github_repo":null,"github_stars":null,"ai_keywords":["TauIndianBankBench","action-oriented benchmark","procedure adherence","required steps","unrequested change","Indian retail banking","language models"]},{"paperId":"2610.40454","title":"Glance: The Generator Is Its Own Reranker","summary":"Retrieval-augmented generation commonly serves separate models for reranking and answering, duplicating model residency and passage processing. We introduce Glance, which equips the generator with a lightweight decision adapter and pointer head to score candidate passages from a shared prefill. Short relevance probes branch from the cached context, producing one score per passage without separately prefilling each query-passage pair. A single resident 9B backbone supports generation, general decisions and reranking, with approximately 0.1 GiB of additional memory per decision skill. After 39 minutes of MS MARCO fine-tuning, Glance achieves nDCG@10 of 0.741 on SciFact and 0.345 on NFCorpus, compared with 0.721 and 0.326 for bge-reranker-v2-m3. On 300 HotpotQA questions, its selected contex…","authors":["Adilet Metinov","Chingiz Arziev"],"publishedAt":"2026-10-29T00:00:00.000Z","submittedAt":null,"upvotes":11,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.40454.png","hf_url":"https://huggingface.co/papers/2610.40454","arxiv_url":"https://arxiv.org/abs/2610.40454","github_repo":null,"github_stars":null,"ai_keywords":["retrieval-augmented generation","reranking","pointer head","decision adapter","shared prefill","cross-encoder","nDCG@10","answer F1"]},{"paperId":"2610.38678","title":"AskStream: When and What to Ask Proactively under Uncertainty in Streaming Video Understanding","summary":"Recent advances in multimodal large language models have increased interest in interaction paradigms for video understanding. The evolution from offline to streaming and proactive interaction has gradually shifted models from passive responders toward more proactive agents that can decide when to answer. However, existing proactive paradigms typically assume that the information required to answer a query is complete, leaving models to only wait or answer. In the real world, however, information is often incomplete due to uncertainty from either the video or the user. In proactive paradigms, such uncertainty can cause the model to wait forever or generate highly hallucinatory responses. We therefore argue that proactive interaction should be redefined with a third action: ask, allowing mo…","authors":["Lu Wang","Zhuoran Jin","Longxiang Wang","Junhong Chen","Yupu Hao","Yao Xu","Daojian Zeng","Yubo Chen"],"publishedAt":"2026-10-28T00:00:00.000Z","submittedAt":null,"upvotes":10,"num_comments":4,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.38678.png","hf_url":"https://huggingface.co/papers/2610.38678","arxiv_url":"https://arxiv.org/abs/2610.38678","github_repo":null,"github_stars":null,"ai_keywords":["multimodal large language models","streaming video understanding","proactive interaction","AskStream","Ask-F1 metric","video uncertainty","user uncertainty"]},{"paperId":"2610.39664","title":"EditAgent: Rethinking Agentic Image Editing","summary":"Agentic image editors decompose complex requests into steps, but when a step fails, they mostly retry with the same narrow set of tools and a judge that scores each output by eye directly. We argue that such agents are bounded by two spaces: the action space of what the agent can do, and the observation space of what its judge can see. On our expert-annotated benchmark, a general-purpose editor yields no usable result in over 40% of the cases, and no judge is reliable across edit types. We present EditAgent, which expands both spaces. It routes each step to general editors and specialized methods, and attaches to each candidate a verification record. With these records, off-the-shelf VLM judges outperform finetuned reward models. Candidates are generated in parallel and judged in one list…","authors":["Haiyang Xu","Tianyu Wang","Jialu Li","Jiteng Mu","Yuanjun Xiong","Jianming Zhang","Zhuowen Tu","Mingze Xu"],"publishedAt":"2026-10-29T00:00:00.000Z","submittedAt":null,"upvotes":10,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.39664.png","hf_url":"https://huggingface.co/papers/2610.39664","arxiv_url":"https://arxiv.org/abs/2610.39664","github_repo":null,"github_stars":null,"ai_keywords":["agentic image editors","action space","observation space","VLM judges","verification records","listwise judging","reward models","open-weight editors"]},{"paperId":"2610.20832","title":"Amortizing Sequential Experimental Design in Reasoning LLMs","summary":"Despite their extensive pretraining knowledge and strong reasoning capabilities, large language models (LLMs) struggle to adaptively gather information in experimental design tasks. We study how to amortize sequential Bayesian experimental design (BED) into an LLM by explicitly training the reasoning that guides action selection using information-gain rewards. Our approach specifies the experimental generative model independently of the LLM's own beliefs, allowing us to optimize expected information gain under a model of our choosing. We investigate the design choices for amortization, including alternative generative-model constructions, reward estimation, hypothesis-space specification, multi-step objectives, and computational costs relative to training- and inference-time BED baselines…","authors":["Anita Kriz","Antonin Berthon","Nicolás Astorga","Mihaela van der Schaar"],"publishedAt":"2026-10-14T00:00:00.000Z","submittedAt":null,"upvotes":9,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.20832.png","hf_url":"https://huggingface.co/papers/2610.20832","arxiv_url":"https://arxiv.org/abs/2610.20832","github_repo":null,"github_stars":null,"ai_keywords":["sequential Bayesian experimental design","information-gain rewards","amortization","generative model","expected information gain","multi-step objectives","Twenty Questions","Medi-Q"]},{"paperId":"2610.39264","title":"Online Dense Reinforcement Learning with Planning-Grounding Agent for UI Navigation on Live IOS Apps","summary":"While frontier vision-language models demonstrate strong capabilities in user interface (UI) navigation, their high inference costs prohibit everyday use. Conversely, smaller models trained via supervised fine-tuning (SFT) or offline reinforcement learning often memorize static trajectories, leading to brittle execution and an inability to recover from errors. To address these limitations, we propose a novel online reinforcement learning framework that trains UI navigation agents directly on live iOS applications. Our framework decouples the agent into two specialized components: a Planning agent and a UI Grounding model. We rely on an LLM-as-a-Judge to provide comprehensive reward signals, combining dense, step-level feedback to evaluate intermediate progress with a terminal reward for o…","authors":["Peitong Duan","Khashayar Gatmiry","Jeffrey Nichols","Samy Bengio","Emmanuel Abbe"],"publishedAt":"2026-10-29T00:00:00.000Z","submittedAt":null,"upvotes":9,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.39264.png","hf_url":"https://huggingface.co/papers/2610.39264","arxiv_url":"https://arxiv.org/abs/2610.39264","github_repo":null,"github_stars":null,"ai_keywords":["online reinforcement learning","UI navigation","Planning agent","UI Grounding model","LLM-as-a-Judge","Proximal Policy Optimization","step-level advantage estimation","GiGPPO","neighbor-conditioned value estimation","concurrent virtual iOS devices"]},{"paperId":"2610.36420","title":"SO-101 Bench: Accessible Spatial Manipulation","summary":"Low-cost arms such as the SO-101 have become the most common entry point to robot learning, yet no released policy performs spatially constrained, language-conditioned tasks on them out of the box, and existing SO-101 benchmarks run only on physical hardware with simple tasks and few held-out objects. We introduce SO-101 Bench, a simulated benchmark for language-conditioned spatial manipulation built on an Isaac Lab digital twin of a real SO-101 setup. It has four tasks scored as five families, 138 objects in seen, unseen-instance and unseen-class splits, 916 evaluation episodes with subsets for held-out appearances, rotate-to-fit planning and occlusion memory, and 3,629 simulated demonstrations on which every learned policy is fine-tuned under an equal compute budget. The twin ranks poli…","authors":["Truman Hickok"],"publishedAt":"2026-10-27T00:00:00.000Z","submittedAt":null,"upvotes":8,"num_comments":3,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.36420.png","hf_url":"https://huggingface.co/papers/2610.36420","arxiv_url":"https://arxiv.org/abs/2610.36420","github_repo":null,"github_stars":null,"ai_keywords":["Isaac Lab","digital twin","VLAs","vision-language-action models","world-action models","LLM agent","spatial manipulation","rotate-to-fit planning","occlusion memory","SO-101 Bench"]},{"paperId":"2610.40564","title":"Training Foundation Models with a Single Algorithm","summary":"Foundation models are typically trained in stages: pretraining, midtraining, supervised fine-tuning (SFT), and reinforcement learning (RL), with changes in both data and learning algorithms. We show that this algorithmic separation is not fundamental. By deriving suitable RL objectives as maximum-likelihood fitting to model-generated target distributions, we place all of these stages under a single likelihood-based training algorithm, unified training. This perspective turns the foundation-model training lifecycle into a problem of choosing which data to train on, in what proportions, and when, with conventional staged training as a special case. We instantiate this framework in a unified training system that mixes static data and RL generations within a single trainer. Across controlled …","authors":["Yuda Song","Alvin Zhang","Fahim Tajwar","Daman Arora","Jeff Schneider","Daniel Khashabi","Ruslan Salakhutdinov","J. Andrew Bagnell"],"publishedAt":"2026-10-29T00:00:00.000Z","submittedAt":null,"upvotes":8,"num_comments":3,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2610.40564.png","hf_url":"https://huggingface.co/papers/2610.40564","arxiv_url":"https://arxiv.org/abs/2610.40564","github_repo":null,"github_stars":null,"ai_keywords":["maximum-likelihood fitting","RL objectives","unified training","foundation models","reinforcement learning","continual learning"]}],"summary":{"by_keyword":[{"keyword":"next-token prediction","count":2},{"keyword":"multimodal large language models","count":2},{"keyword":"diffusion vision-language models","count":1},{"keyword":"visual token pruning","count":1},{"keyword":"LLaDA-V","count":1},{"keyword":"LOWN","count":1},{"keyword":"masked tokens","count":1},{"keyword":"key-value caching","count":1},{"keyword":"decoding FLOPs","count":1},{"keyword":"recursive self-improvement","count":1},{"keyword":"training dynamics","count":1},{"keyword":"reward variation","count":1},{"keyword":"task pool","count":1},{"keyword":"multimodal questions","count":1},{"keyword":"model-authored programs","count":1}],"most_upvoted":{"paperId":"2610.36671","title":"Look Only When Needed: Certificate-Guided Visual Routing for Efficient Diffusion VLMs","upvotes":101},"most_discussed":{"paperId":"2610.37161","title":"WorkFactory: Turning Professional Intents into Executable Work for Scalable Agent Training","comments":6}}}}