diff --git a/feed.xml b/feed.xml index a6fdc4a..db40b5d 100644 --- a/feed.xml +++ b/feed.xml @@ -7,335 +7,265 @@ http://www.rssboard.org/rss-specification python-feedgen en - Wed, 15 Apr 2026 00:12:49 +0000 + Thu, 16 Apr 2026 00:12:50 +0000 - SPEED-Bench: A Unified and Diverse Benchmark for Speculative Decoding - https://arxiv.org/abs/2604.09557 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.09557.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Talor Abramovich, Maor Ashkenazi, Carl, Putterman, Benjamin Chislett, Tiyasa Mitra, Bita Darvish Rouhani, Ran Zilberstein, Yonatan Geifman</p><p><b>Upvotes:</b> 8</p><p><b>Summary:</b> Speculative Decoding (SD) has emerged as a critical technique for accelerating Large Language Model (LLM) inference. Unlike deterministic system optimizations, SD performance is inherently data-dependent, meaning that diverse and representative workloads are essential for accurately measuring its effectiveness. Existing benchmarks suffer from limited task diversity, inadequate support for throughput-oriented evaluation, and a reliance on high-level implementations that fail to reflect production environments. To address this, we introduce SPEED-Bench, a comprehensive suite designed to standardize SD evaluation across diverse semantic domains and realistic serving regimes. SPEED-Bench offers a carefully curated Qualitative data split, selected by prioritizing semantic diversity across the data samples. Additionally, it includes a Throughput data split, allowing speedup evaluation across a range of concurrencies, from latency-sensitive low-batch settings to throughput-oriented high-load scenarios. By integrating with production engines like vLLM and TensorRT-LLM, SPEED-Bench allows practitioners to analyze system behaviors often masked by other benchmarks. We highlight this by quantifying how synthetic inputs overestimate real-world throughput, identifying batch-size dependent optimal draft lengths and biases in low-diversity data, and analyzing the caveats of vocabulary pruning in state-of-the-art drafters. We release SPEED-Bench to establish a unified evaluation standard for practical comparisons of SD algorithms.</p> - https://arxiv.org/abs/2604.09557 - Tue, 10 Feb 2026 00:00:00 +0000 + Turing Test on Screen: A Benchmark for Mobile GUI Agent Humanization + https://arxiv.org/abs/2604.09574 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.09574.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Jiachen Zhu, Lingyu Yang, Rong Shan, Congmin Zheng, Zeyu Zheng, Weiwen Liu, Yong Yu, Weinan Zhang, Jianghao Lin</p><p><b>Upvotes:</b> 26</p><p><b>Summary:</b> The rise of autonomous GUI agents has triggered adversarial countermeasures from digital platforms, yet existing research prioritizes utility and robustness over the critical dimension of anti-detection. We argue that for agents to survive in human-centric ecosystems, they must evolve Humanization capabilities. We introduce the ``Turing Test on Screen,'' formally modeling the interaction as a MinMax optimization problem between a detector and an agent aiming to minimize behavioral divergence. We then collect a new high-fidelity dataset of mobile touch dynamics, and conduct our analysis that vanilla LMM-based agents are easily detectable due to unnatural kinematics. Consequently, we establish the Agent Humanization Benchmark (AHB) and detection metrics to quantify the trade-off between imitability and utility. Finally, we propose methods ranging from heuristic noise to data-driven behavioral matching, demonstrating that agents can achieve high imitability theoretically and empirically without sacrificing performance. This work shifts the paradigm from whether an agent can perform a task to how it performs it within a human-centric ecosystem, laying the groundwork for seamless coexistence in adversarial digital environments.</p> + https://arxiv.org/abs/2604.09574 + Tue, 24 Feb 2026 00:00:00 +0000 - QuanBench+: A Unified Multi-Framework Benchmark for LLM-Based Quantum Code Generation - https://arxiv.org/abs/2604.08570 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.08570.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Ali Slim, Haydar Hamieh, Jawad Kotaich, Yehya Ghosn, Mahdi Chehimi, Ammar Mohanna, Hasan Abed Al Kader Hammoud, Bernard Ghanem</p><p><b>Upvotes:</b> 109</p><p><b>Summary:</b> Large Language Models (LLMs) are increasingly used for code generation, yet quantum code generation is still evaluated mostly within single frameworks, making it difficult to separate quantum reasoning from framework familiarity. We introduce QuanBench+, a unified benchmark spanning Qiskit, PennyLane, and Cirq, with 42 aligned tasks covering quantum algorithms, gate decomposition, and state preparation. We evaluate models with executable functional tests, report Pass@1 and Pass@5, and use KL-divergence-based acceptance for probabilistic outputs. We additionally study Pass@1 after feedback-based repair, where a model may revise code after a runtime error or wrong answer. Across frameworks, the strongest one-shot scores reach 59.5% in Qiskit, 54.8% in Cirq, and 42.9% in PennyLane; with feedback-based repair, the best scores rise to 83.3%, 76.2%, and 66.7%, respectively. These results show clear progress, but also that reliable multi-framework quantum code generation remains unsolved and still depends strongly on framework-specific knowledge.</p> - https://arxiv.org/abs/2604.08570 - Wed, 25 Mar 2026 00:00:00 +0000 + Spatial Competence Benchmark + https://arxiv.org/abs/2604.09594 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.09594.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Jash Vira, Ashley Harris</p><p><b>Upvotes:</b> 1</p><p><b>Summary:</b> Spatial competence is the quality of maintaining a consistent internal representation of an environment and using it to infer discrete structure and plan actions under constraints. Prevailing spatial evaluations for large models are limited to probing isolated primitives through 3D transformations or visual question answering. We introduce the Spatial Competence Benchmark (SCBench), spanning three hierarchical capability buckets whose tasks require executable outputs verified by deterministic checkers or simulator-based evaluators. On SCBench, three frontier models exhibit monotonically decreasing accuracy up the capability ladder. Sweeping output-token caps shows that accuracy gains concentrate at low budgets and saturate quickly, and failures are dominated by locally plausible geometry that breaks global constraints. We release the task generators, verifiers, and visualisation tooling.</p> + https://arxiv.org/abs/2604.09594 + Thu, 05 Mar 2026 00:00:00 +0000 - TRACE: Capability-Targeted Agentic Training - https://arxiv.org/abs/2604.05336 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.05336.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Hangoo Kang, Tarun Suresh, Jon Saad-Falcon, Azalia Mirhoseini</p><p><b>Upvotes:</b> 11</p><p><b>Summary:</b> Large Language Models (LLMs) deployed in agentic environments must exercise multiple capabilities across different task instances, where a capability is performing one or more actions in a trajectory that are necessary for successfully solving a subset of tasks in the environment. Many existing approaches either rely on synthetic training data that is not targeted to the model's actual capability deficits in the target environment or train directly on the target environment, where the model needs to implicitly learn the capabilities across tasks. We introduce TRACE (Turning Recurrent Agent failures into Capability-targeted training Environments), an end-to-end system for environment-specific agent self-improvement. TRACE contrasts successful and failed trajectories to automatically identify lacking capabilities, synthesizes a targeted training environment for each that rewards whether the capability was exercised, and trains a LoRA adapter via RL on each synthetic environment, routing to the relevant adapter at inference. Empirically, TRACE generalizes across different environments, improving over the base agent by +14.1 points on τ^2-bench (customer service) and +7 perfect scores on ToolSandbox (tool use), outperforming the strongest baseline by +7.4 points and +4 perfect scores, respectively. Given the same number of rollouts, TRACE scales more efficiently than baselines, outperforming GRPO and GEPA by +9.2 and +7.4 points on τ^2-bench.</p> - https://arxiv.org/abs/2604.05336 - Tue, 07 Apr 2026 00:00:00 +0000 - - - ATANT: An Evaluation Framework for AI Continuity - https://arxiv.org/abs/2604.06710 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.06710.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Samuel Sameer Tanguturi</p><p><b>Upvotes:</b> 0</p><p><b>Summary:</b> We present ATANT (Automated Test for Acceptance of Narrative Truth), an open evaluation framework for measuring continuity in AI systems: the ability to persist, update, disambiguate, and reconstruct meaningful context across time. While the AI industry has produced memory components (RAG pipelines, vector databases, long context windows, profile layers), no published framework formally defines or measures whether these components produce genuine continuity. We define continuity as a system property with 7 required properties, introduce a 10-checkpoint evaluation methodology that operates without an LLM in the evaluation loop, and present a narrative test corpus of 250 stories comprising 1,835 verification questions across 6 life domains. We evaluate a reference implementation across 5 test suite iterations, progressing from 58% (legacy architecture) to 100% in isolated mode (250 stories) and 100% in 50-story cumulative mode, with 96% at 250-story cumulative scale. The cumulative result is the primary measure: when 250 distinct life narratives coexist in the same database, the system must retrieve the correct fact for the correct context without cross-contamination. ATANT is system-agnostic, model-independent, and designed as a sequenced methodology for building and validating continuity systems. The framework specification, example stories, and evaluation protocol are available at https://github.com/Kenotic-Labs/ATANT. The full 250-story corpus will be released incrementally.</p> - https://arxiv.org/abs/2604.06710 - Wed, 08 Apr 2026 00:00:00 +0000 - - - Efficient RL Training for LLMs with Experience Replay - https://arxiv.org/abs/2604.08706 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.08706.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Charles Arnal, Vivien Cabannes, Taco Cohen, Julia Kempe, Remi Munos</p><p><b>Upvotes:</b> 9</p><p><b>Summary:</b> While Experience Replay - the practice of storing rollouts and reusing them multiple times during training - is a foundational technique in general RL, it remains largely unexplored in LLM post-training due to the prevailing belief that fresh, on-policy data is essential for high performance. In this work, we challenge this assumption. We present a systematic study of replay buffers for LLM post-training, formalizing the optimal design as a trade-off between staleness-induced variance, sample diversity and the high computational cost of generation. We show that strict on-policy sampling is suboptimal when generation is expensive. Empirically, we show that a well-designed replay buffer can drastically reduce inference compute without degrading - and in some cases even improving - final model performance, while preserving policy entropy.</p> - https://arxiv.org/abs/2604.08706 - Thu, 09 Apr 2026 00:00:00 +0000 - - - Uni-ViGU: Towards Unified Video Generation and Understanding via A Diffusion-Based Video Generator - https://arxiv.org/abs/2604.08121 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.08121.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Luozheng Qin, Jia Gong, Qian Qiao, Tianjiao Li, Li Xu, Haoyu Pan, Chao Qu, Zhiyu Tan, Hao Li</p><p><b>Upvotes:</b> 39</p><p><b>Summary:</b> Unified multimodal models integrating visual understanding and generation face a fundamental challenge: visual generation incurs substantially higher computational costs than understanding, particularly for video. This imbalance motivates us to invert the conventional paradigm: rather than extending understanding-centric MLLMs to support generation, we propose Uni-ViGU, a framework that unifies video generation and understanding by extending a video generator as the foundation. We introduce a unified flow method that performs continuous flow matching for video and discrete flow matching for text within a single process, enabling coherent multimodal generation. We further propose a modality-driven MoE-based framework that augments Transformer blocks with lightweight layers for text generation while preserving generative priors. To repurpose generation knowledge for understanding, we design a bidirectional training mechanism with two stages: Knowledge Recall reconstructs input prompts to leverage learned text-video correspondences, while Capability Refinement fine-tunes on detailed captions to establish discriminative shared representations. Experiments demonstrate that Uni-ViGU achieves competitive performance on both video generation and understanding, validating generation-centric architectures as a scalable path toward unified multimodal intelligence. Project Page and Code: https://fr0zencrane.github.io/uni-vigu-page/.</p> - https://arxiv.org/abs/2604.08121 - Thu, 09 Apr 2026 00:00:00 +0000 - - - SPASM: Stable Persona-driven Agent Simulation for Multi-turn Dialogue Generation - https://arxiv.org/abs/2604.09212 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.09212.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Han Luo, Guy Laban</p><p><b>Upvotes:</b> 1</p><p><b>Summary:</b> Large language models are increasingly deployed in multi-turn settings such as tutoring, support, and counseling, where reliability depends on preserving consistent roles, personas, and goals across long horizons. This requirement becomes critical when LLMs are used to generate synthetic dialogues for training and evaluation, since LLM--LLM conversations can accumulate identity-related failures such as persona drift, role confusion, and "echoing", where one agent gradually mirrors its partner. We introduce SPASM (Stable Persona-driven Agent Simulation for Multi-turn dialogue generation), a modular, stability-first framework that decomposes simulation into (i) persona creation via schema sampling, plausibility validation, and natural-language persona crafting, (ii) Client--Responder dialogue generation, and (iii) termination detection for coherent stopping. To improve long-horizon stability without changing model weights, we propose Egocentric Context Projection (ECP): dialogue history is stored in a perspective-agnostic representation and deterministically projected into each agent's egocentric view before generation. Across three LLM backbones (GPT-4o-mini, DeepSeek-V3.2, Qwen-Plus) and nine Client--Responder pairings, we construct a dataset of 4,500 personas and 45,000 conversations (500 personas X 10 conversations per pairing). Ablations show ECP substantially reduces persona drift and, under human validation, eliminates echoing; embedding analyses recover persona structure and reveal strong responder-driven interaction geometry. Our code is available at https://github.com/lhannnn/SPASM.</p> - https://arxiv.org/abs/2604.09212 + CONSCIENTIA: Can LLM Agents Learn to Strategize? Emergent Deception and Trust in a Multi-Agent NYC Simulation + https://arxiv.org/abs/2604.09746 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.09746.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Aarush Sinha, Arion Das, Soumyadeep Nag, Charan Karnati, Shravani Nag, Chandra Vadhan Raj, Aman Chadha, Vinija Jain, Suranjana Trivedy, Amitava Das</p><p><b>Upvotes:</b> 0</p><p><b>Summary:</b> As large language models (LLMs) are increasingly deployed as autonomous agents, understanding how strategic behavior emerges in multi-agent environments has become an important alignment challenge. We take a neutral empirical stance and construct a controlled environment in which strategic behavior can be directly observed and measured. We introduce a large-scale multi-agent simulation in a simplified model of New York City, where LLM-driven agents interact under opposing incentives. Blue agents aim to reach their destinations efficiently, while Red agents attempt to divert them toward billboard-heavy routes using persuasive language to maximize advertising revenue. Hidden identities make navigation socially mediated, forcing agents to decide when to trust or deceive. We study policy learning through an iterative simulation pipeline that updates agent policies across repeated interaction rounds using Kahneman-Tversky Optimization (KTO). Blue agents are optimized to reduce billboard exposure while preserving navigation efficiency, whereas Red agents adapt to exploit remaining weaknesses. Across iterations, the best Blue policy improves task success from 46.0% to 57.3%, although susceptibility remains high at 70.7%. Later policies exhibit stronger selective cooperation while preserving trajectory efficiency. However, a persistent safety-helpfulness trade-off remains: policies that better resist adversarial steering do not simultaneously maximize task completion. Overall, our results show that LLM agents can exhibit limited strategic behavior, including selective trust and deception, while remaining highly vulnerable to adversarial persuasion.</p> + https://arxiv.org/abs/2604.09746 Fri, 10 Apr 2026 00:00:00 +0000 - TAIHRI: Task-Aware 3D Human Keypoints Localization for Close-Range Human-Robot Interaction - https://arxiv.org/abs/2604.08921 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.08921.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Ao Li, Yonggen Ling, Yiyang Lin, Yuji Wang, Yong Deng, Yansong Tang</p><p><b>Upvotes:</b> 2</p><p><b>Summary:</b> Accurate 3D human keypoints localization is a critical technology enabling robots to achieve natural and safe physical interaction with users. Conventional 3D human keypoints estimation methods primarily focus on the whole-body reconstruction quality relative to the root joint. However, in practical human-robot interaction (HRI) scenarios, robots are more concerned with the precise metric-scale spatial localization of task-relevant body parts under the egocentric camera 3D coordinate. We propose TAIHRI, the first Vision-Language Model (VLM) tailored for close-range HRI perception, capable of understanding users' motion commands and directing the robot's attention to the most task-relevant keypoints. By quantizing 3D keypoints into a finite interaction space, TAIHRI precisely localize the 3D spatial coordinates of critical body parts by 2D keypoint reasoning via next token prediction, and seamlessly adapt to downstream tasks such as natural language control or global space human mesh recovery. Experiments on egocentric interaction benchmarks demonstrate that TAIHRI achieves superior estimation accuracy for task-critical body parts. We believe TAIHRI opens new research avenues in the field of embodied human-robot interaction. Code is available at: https://github.com/Tencent/TAIHRI.</p> - https://arxiv.org/abs/2604.08921 + Hierarchical SVG Tokenization: Learning Compact Visual Programs for Scalable Vector Graphics Modeling + https://arxiv.org/abs/2604.05072 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.05072.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Ximing Xing, Ziteng Xue, Zhenxi Li, Weicong Liang, Linqing Wang, Zhantao Yang, Tiankai Hang, Zijin Yin, Qinglin Lu, Chunyu Wang, Qian Yu</p><p><b>Upvotes:</b> 2</p><p><b>Summary:</b> Recent large language models have shifted SVG generation from differentiable rendering optimization to autoregressive program synthesis. However, existing approaches still rely on generic byte-level tokenization inherited from natural language processing, which poorly reflects the geometric structure of vector graphics. Numerical coordinates are fragmented into discrete symbols, destroying spatial relationships and introducing severe token redundancy, often leading to coordinate hallucination and inefficient long-sequence generation. To address these challenges, we propose HiVG, a hierarchical SVG tokenization framework tailored for autoregressive vector graphics generation. HiVG decomposes raw SVG strings into structured atomic tokens and further compresses executable command--parameter groups into geometry-constrained segment tokens, substantially improving sequence efficiency while preserving syntactic validity. To further mitigate spatial mismatch, we introduce a Hierarchical Mean--Noise (HMN) initialization strategy that injects numerical ordering signals and semantic priors into new token embeddings. Combined with a curriculum training paradigm that progressively increases program complexity, HiVG enables more stable learning of executable SVG programs. Extensive experiments on both text-to-SVG and image-to-SVG tasks demonstrate improved generation fidelity, spatial consistency, and sequence efficiency compared with conventional tokenization schemes. Our code is publicly available at https://github.com/ximinng/HiVG</p> + https://arxiv.org/abs/2604.05072 Fri, 10 Apr 2026 00:00:00 +0000 - Strips as Tokens: Artist Mesh Generation with Native UV Segmentation - https://arxiv.org/abs/2604.09132 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.09132.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Rui Xu, Dafei Qin, Kaichun Qiao, Qiujie Dong, Huaijin Pi, Qixuan Zhang, Longwen Zhang, Lan Xu, Jingyi Yu, Wenping Wang, Taku Komura</p><p><b>Upvotes:</b> 44</p><p><b>Summary:</b> Recent advancements in autoregressive transformers have demonstrated remarkable potential for generating artist-quality meshes. However, the token ordering strategies employed by existing methods typically fail to meet professional artist standards, where coordinate-based sorting yields inefficiently long sequences, and patch-based heuristics disrupt the continuous edge flow and structural regularity essential for high-quality modeling. To address these limitations, we propose Strips as Tokens (SATO), a novel framework with a token ordering strategy inspired by triangle strips. By constructing the sequence as a connected chain of faces that explicitly encodes UV boundaries, our method naturally preserves the organized edge flow and semantic layout characteristic of artist-created meshes. A key advantage of this formulation is its unified representation, enabling the same token sequence to be decoded into either a triangle or quadrilateral mesh. This flexibility facilitates joint training on both data types: large-scale triangle data provides fundamental structural priors, while high-quality quad data enhances the geometric regularity of the outputs. Extensive experiments demonstrate that SATO consistently outperforms prior methods in terms of geometric quality, structural coherence, and UV segmentation.</p> - https://arxiv.org/abs/2604.09132 + Many-Tier Instruction Hierarchy in LLM Agents + https://arxiv.org/abs/2604.09443 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.09443.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Jingyu Zhang, Tianjian Li, William Jurayj, Hongyuan Zhan, Benjamin Van Durme, Daniel Khashabi</p><p><b>Upvotes:</b> 13</p><p><b>Summary:</b> Large language model agents receive instructions from many sources-system messages, user prompts, tool outputs, and more-each carrying different levels of trust and authority. When these instructions conflict, models must reliably follow the highest-privilege instruction to remain safe and effective. The dominant paradigm, instruction hierarchy (IH), assumes a fixed, small set of privilege levels (typically fewer than five) defined by rigid role labels (e.g., system > user). This is inadequate for real-world agentic settings, where conflicts can arise across far more sources and contexts. In this work, we propose Many-Tier Instruction Hierarchy (ManyIH), a paradigm for resolving instruction conflicts among instructions with arbitrarily many privilege levels. We introduce ManyIH-Bench, the first benchmark for ManyIH. ManyIH-Bench requires models to navigate up to 12 levels of conflicting instructions with varying privileges, comprising 853 agentic tasks (427 coding and 426 instruction-following). ManyIH-Bench composes constraints developed by LLMs and verified by humans to create realistic and difficult test cases spanning 46 real-world agents. Our experiments show that even the current frontier models perform poorly (~40% accuracy) when instruction conflict scales. This work underscores the urgent need for methods that explicitly target fine-grained, scalable instruction conflict resolution in agentic settings.</p> + https://arxiv.org/abs/2604.09443 Fri, 10 Apr 2026 00:00:00 +0000 - Counting to Four is still a Chore for VLMs - https://arxiv.org/abs/2604.10039 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10039.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Duy Le Dinh Anh, Patrick Amadeus Irawan, Tuan Van Vo</p><p><b>Upvotes:</b> 1</p><p><b>Summary:</b> Vision--language models (VLMs) have achieved impressive performance on complex multimodal reasoning tasks, yet they still fail on simple grounding skills such as object counting. Existing evaluations mostly assess only final outputs, offering limited insight into where these failures arise inside the model. In this work, we present an empirical study of VLM counting behavior through both behavioral and mechanistic analysis. We introduce COUNTINGTRICKS, a controlled evaluation suite of simple shape-based counting cases designed to expose vulnerabilities under different patchification layouts and adversarial prompting conditions. Using attention analysis and component-wise probing, we show that count-relevant visual evidence is strongest in the modality projection stage but degrades substantially in later language layers, where models become more susceptible to text priors. Motivated by this finding, we further evaluate Modality Attention Share (MAS), a lightweight intervention that encourages a minimum budget of visual attention during answer generation. Our results suggest that counting failures in VLMs stem not only from visual perception limits, but also from the underuse of visual evidence during language-stage reasoning. Code and dataset will be released at https://github.com/leduy99/-CVPRW26-Modality-Attention-Share.</p> - https://arxiv.org/abs/2604.10039 - Sat, 11 Apr 2026 00:00:00 +0000 + BERT-as-a-Judge: A Robust Alternative to Lexical Methods for Efficient Reference-Based LLM Evaluation + https://arxiv.org/abs/2604.09497 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.09497.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Hippolyte Gisserot-Boukhlef, Nicolas Boizard, Emmanuel Malherbe, Céline Hudelot, Pierre Colombo</p><p><b>Upvotes:</b> 21</p><p><b>Summary:</b> Accurate evaluation is central to the large language model (LLM) ecosystem, guiding model selection and downstream adoption across diverse use cases. In practice, however, evaluating generative outputs typically relies on rigid lexical methods to extract and assess answers, which can conflate a model's true problem-solving ability with its compliance with predefined formatting guidelines. While recent LLM-as-a-Judge approaches mitigate this issue by assessing semantic correctness rather than strict structural conformity, they also introduce substantial computational overhead, making evaluation costly. In this work, we first systematically investigate the limitations of lexical evaluation through a large-scale empirical study spanning 36 models and 15 downstream tasks, demonstrating that such methods correlate poorly with human judgments. To address this limitation, we introduce BERT-as-a-Judge, an encoder-driven approach for assessing answer correctness in reference-based generative settings, robust to variations in output phrasing, and requiring only lightweight training on synthetically annotated question-candidate-reference triplets. We show that it consistently outperforms the lexical baseline while matching the performance of much larger LLM judges, providing a compelling tradeoff between the two and enabling reliable, scalable evaluation. Finally, through extensive experimentation, we provide detailed insights into BERT-as-a-Judge's performance to offer practical guidance for practitioners, and release all project artifacts to foster downstream adoption.</p> + https://arxiv.org/abs/2604.09497 + Fri, 10 Apr 2026 00:00:00 +0000 - Zero-shot World Models Are Developmentally Efficient Learners - https://arxiv.org/abs/2604.10333 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10333.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Khai Loong Aw, Klemen Kotar, Wanhee Lee, Seungwoo Kim, Khaled Jedoui, Rahul Venkatesh, Lilian Naing Chen, Michael C. Frank, Daniel L. K. Yamins</p><p><b>Upvotes:</b> 6</p><p><b>Summary:</b> Young children demonstrate early abilities to understand their physical world, estimating depth, motion, object coherence, interactions, and many other aspects of physical scene understanding. Children are both data-efficient and flexible cognitive systems, creating competence despite extremely limited training data, while generalizing to myriad untrained tasks -- a major challenge even for today's best AI systems. Here we introduce a novel computational hypothesis for these abilities, the Zero-shot Visual World Model (ZWM). ZWM is based on three principles: a sparse temporally-factored predictor that decouples appearance from dynamics; zero-shot estimation through approximate causal inference; and composition of inferences to build more complex abilities. We show that ZWM can be learned from the first-person experience of a single child, rapidly generating competence across multiple physical understanding benchmarks. It also broadly recapitulates behavioral signatures of child development and builds brain-like internal representations. Our work presents a blueprint for efficient and flexible learning from human-scale data, advancing both a computational account for children's early physical understanding and a path toward data-efficient AI systems.</p> - https://arxiv.org/abs/2604.10333 - Sat, 11 Apr 2026 00:00:00 +0000 + SPPO: Sequence-Level PPO for Long-Horizon Reasoning Tasks + https://arxiv.org/abs/2604.08865 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.08865.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Tianyi Wang, Yixia Li, Long Li, Yibiao Chen, Shaohan Huang, Yun Chen, Peng Li, Yang Liu, Guanhua Chen</p><p><b>Upvotes:</b> 25</p><p><b>Summary:</b> Proximal Policy Optimization (PPO) is central to aligning Large Language Models (LLMs) in reasoning tasks with verifiable rewards. However, standard token-level PPO struggles in this setting due to the instability of temporal credit assignment over long Chain-of-Thought (CoT) horizons and the prohibitive memory cost of the value model. While critic-free alternatives like GRPO mitigate these issues, they incur significant computational overhead by requiring multiple samples for baseline estimation, severely limiting training throughput. In this paper, we introduce Sequence-Level PPO (SPPO), a scalable algorithm that harmonizes the sample efficiency of PPO with the stability of outcome-based updates. SPPO reformulates the reasoning process as a Sequence-Level Contextual Bandit problem, employing a decoupled scalar value function to derive low-variance advantage signals without multi-sampling. Extensive experiments on mathematical benchmarks demonstrate that SPPO significantly surpasses standard PPO and matches the performance of computation-heavy group-based methods, offering a resource-efficient framework for aligning reasoning LLMs.</p> + https://arxiv.org/abs/2604.08865 + Fri, 10 Apr 2026 00:00:00 +0000 - Not All Denoising Steps Are Equal: Model Scheduling for Faster Masked Diffusion Language Models - https://arxiv.org/abs/2604.02340 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.02340.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Ivan Sedykh, Nikita Sorokin, Valentin Malykh</p><p><b>Upvotes:</b> 6</p><p><b>Summary:</b> Recent advances in masked diffusion language models (MDLMs) narrow the quality gap to autoregressive LMs, but their sampling remains expensive because generation requires many full-sequence denoising passes with a large Transformer and, unlike autoregressive decoding, cannot benefit from KV caching. In this work, we exploit the flexibility of the diffusion framework and study model scheduling, where a smaller MDLM replaces the full model at a subset of denoising steps. Across models trained on OpenWebText and LM1B, we show that early and late denoising steps are substantially more robust to such replacement than middle steps, enabling up to a 17% reduction in FLOPs with only modest degradation in generative perplexity under both unconditional and prefix-conditional generation, while preserving sample diversity. We support these findings with a step-importance analysis based on loss and KL divergence between small and large models across timesteps, as well as an exhaustive search over coarse step segments, both of which identify the middle of the diffusion trajectory as most sensitive consistently across datasets. Our results suggest that simple, architecture-agnostic scheduling rules can significantly accelerate MDLM sampling while largely preserving generation quality.</p> - https://arxiv.org/abs/2604.02340 - Sat, 11 Apr 2026 00:00:00 +0000 - - - Prompt Relay: Inference-Time Temporal Control for Multi-Event Video Generation - https://arxiv.org/abs/2604.10030 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10030.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Gordon Chen, Ziqi Huang, Ziwei Liu</p><p><b>Upvotes:</b> 11</p><p><b>Summary:</b> Video diffusion models have achieved remarkable progress in generating high-quality videos. However, these models struggle to represent the temporal succession of multiple events in real-world videos and lack explicit mechanisms to control when semantic concepts appear, how long they persist, and the order in which multiple events occur. Such control is especially important for movie-grade video synthesis, where coherent storytelling depends on precise timing, duration, and transitions between events. When using a single paragraph-style prompt to describe a sequence of complex events, models often exhibit semantic entanglement, where concepts intended for different moments in the video bleed into one another, resulting in poor text-video alignment. To address these limitations, we propose Prompt Relay, an inference-time, plug-and-play method to enable fine-grained temporal control in multi-event video generation, requiring no architectural modifications and no additional computational overhead. Prompt Relay introduces a penalty into the cross-attention mechanism, so that each temporal segment attends only to its assigned prompt, allowing the model to represent one semantic concept at a time and thereby improving temporal prompt alignment, reducing semantic interference, and enhancing visual quality.</p> - https://arxiv.org/abs/2604.10030 - Sat, 11 Apr 2026 00:00:00 +0000 - - - Attention Sink in Transformers: A Survey on Utilization, Interpretation, and Mitigation - https://arxiv.org/abs/2604.10098 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10098.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Zunhai Su, Hengyuan Zhang, Wei Wu, Yifan Zhang, Yaxiu Liu, He Xiao, Qingyao Yang, Yuxuan Sun, Rui Yang, Chao Zhang, Keyu Fan, Weihao Ye, Jing Xiong, Hui Shen, Chaofan Tao, Taiqiang Wu, Zhongwei Wan, Yulei Qian, Yuchen Xie, Ngai Wong</p><p><b>Upvotes:</b> 55</p><p><b>Summary:</b> As the foundational architecture of modern machine learning, Transformers have driven remarkable progress across diverse AI domains. Despite their transformative impact, a persistent challenge across various Transformers is Attention Sink (AS), in which a disproportionate amount of attention is focused on a small subset of specific yet uninformative tokens. AS complicates interpretability, significantly affecting the training and inference dynamics, and exacerbates issues such as hallucinations. In recent years, substantial research has been dedicated to understanding and harnessing AS. However, a comprehensive survey that systematically consolidates AS-related research and offers guidance for future advancements remains lacking. To address this gap, we present the first survey on AS, structured around three key dimensions that define the current research landscape: Fundamental Utilization, Mechanistic Interpretation, and Strategic Mitigation. Our work provides a pivotal contribution by clarifying key concepts and guiding researchers through the evolution and trends of the field. We envision this survey as a definitive resource, empowering researchers and practitioners to effectively manage AS within the current Transformer paradigm, while simultaneously inspiring innovative advancements for the next generation of Transformers. The paper list of this work is available at https://github.com/ZunhaiSu/Awesome-Attention-Sink.</p> - https://arxiv.org/abs/2604.10098 - Sat, 11 Apr 2026 00:00:00 +0000 - - - IceCache: Memory-efficient KV-cache Management for Long-Sequence LLMs - https://arxiv.org/abs/2604.10539 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10539.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Yuzhen Mao, Qitong Wang, Martin Ester, Ke Li</p><p><b>Upvotes:</b> 0</p><p><b>Summary:</b> Key-Value (KV) cache plays a crucial role in accelerating inference in large language models (LLMs) by storing intermediate attention states and avoiding redundant computation during autoregressive generation. However, its memory footprint scales linearly with sequence length, often leading to severe memory bottlenecks on resource-constrained hardware. Prior work has explored offloading KV cache to the CPU while retaining only a subset on the GPU, but these approaches often rely on imprecise token selection and suffer performance degradation in long-generation tasks such as chain-of-thought reasoning. In this paper, we propose a novel KV cache management strategy, IceCache, which integrates semantic token clustering with PagedAttention. By organizing semantically related tokens into contiguous memory regions managed by a hierarchical, dynamically updatable data structure, our method enables more efficient token selection and better utilization of memory bandwidth during CPU-GPU transfers. Experimental results on LongBench show that, with a 256-token budget, IceCache maintains 99% of the original accuracy achieved by the full KV cache model. Moreover, compared to other offloading-based methods, IceCache attains competitive or even superior latency and accuracy while using only 25% of the KV cache token budget, demonstrating its effectiveness in long-sequence scenarios. The code is available on our project website at https://yuzhenmao.github.io/IceCache/.</p> - https://arxiv.org/abs/2604.10539 + When Reasoning Models Hurt Behavioral Simulation: A Solver-Sampler Mismatch in Multi-Agent LLM Negotiation + https://arxiv.org/abs/2604.11840 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11840.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Sandro Andric</p><p><b>Upvotes:</b> 1</p><p><b>Summary:</b> Large language models are increasingly used as agents in social, economic, and policy simulations. A common assumption is that stronger reasoning should improve simulation fidelity. We argue that this assumption can fail when the objective is not to solve a strategic problem, but to sample plausible boundedly rational behavior. In such settings, reasoning-enhanced models can become better solvers and worse simulators: they can over-optimize for strategically dominant actions, collapse compromise-oriented terminal behavior, and sometimes exhibit a diversity-without-fidelity pattern in which local variation survives without outcome-level fidelity. We study this solver-sampler mismatch in three multi-agent negotiation environments adapted from earlier simulation work: an ambiguous fragmented-authority trading-limits scenario, an ambiguous unified-opposition trading-limits scenario, and a new-domain grid-curtailment case in emergency electricity management. We compare three reflection conditions, no reflection, bounded reflection, and native reasoning, across two primary model families and then extend the same protocol to direct OpenAI runs with GPT-4.1 and GPT-5.2. Across all three experiments, bounded reflection produces substantially more diverse and compromise-oriented trajectories than either no reflection or native reasoning. In the direct OpenAI extension, GPT-5.2 native ends in authority decisions in 45 of 45 runs across the three experiments, while GPT-5.2 bounded recovers compromise outcomes in every environment. The contribution is not a claim that reasoning is generally harmful. It is a methodological warning: model capability and simulation fidelity are different objectives, and behavioral simulation should qualify models as samplers, not only as solvers.</p> + https://arxiv.org/abs/2604.11840 Sun, 12 Apr 2026 00:00:00 +0000 - Audio-Omni: Extending Multi-modal Understanding to Versatile Audio Generation and Editing - https://arxiv.org/abs/2604.10708 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10708.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Zeyue Tian, Binxin Yang, Zhaoyang Liu, Jiexuan Zhang, Ruibin Yuan, Hubery Yin, Qifeng Chen, Chen Li, Jing Lv, Wei Xue, Yike Guo</p><p><b>Upvotes:</b> 1</p><p><b>Summary:</b> Recent progress in multimodal models has spurred rapid advances in audio understanding, generation, and editing. However, these capabilities are typically addressed by specialized models, leaving the development of a truly unified framework that can seamlessly integrate all three tasks underexplored. While some pioneering works have explored unifying audio understanding and generation, they often remain confined to specific domains. To address this, we introduce Audio-Omni, the first end-to-end framework to unify generation and editing across general sound, music, and speech domains, with integrated multi-modal understanding capabilities. Our architecture synergizes a frozen Multimodal Large Language Model for high-level reasoning with a trainable Diffusion Transformer for high-fidelity synthesis. To overcome the critical data scarcity in audio editing, we construct AudioEdit, a new large-scale dataset comprising over one million meticulously curated editing pairs. Extensive experiments demonstrate that Audio-Omni achieves state-of-the-art performance across a suite of benchmarks, outperforming prior unified approaches while achieving performance on par with or superior to specialized expert models. Beyond its core capabilities, Audio-Omni exhibits remarkable inherited capabilities, including knowledge-augmented reasoning generation, in-context generation, and zero-shot cross-lingual control for audio generation, highlighting a promising direction toward universal generative audio intelligence. The code, model, and dataset will be publicly released on https://zeyuet.github.io/Audio-Omni.</p> - https://arxiv.org/abs/2604.10708 + PokeRL: Reinforcement Learning for Pokemon Red + https://arxiv.org/abs/2604.10812 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10812.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Dheeraj Mudireddy, Sai Patibandla</p><p><b>Upvotes:</b> 1</p><p><b>Summary:</b> Pokemon Red is a long-horizon JRPG with sparse rewards, partial observability, and quirky control mechanics that make it a challenging benchmark for reinforcement learning. While recent work has shown that PPO agents can clear the first two gyms using heavy reward shaping and engineered observations, training remains brittle in practice, with agents often degenerating into action loops, menu spam, or unproductive wandering. In this paper, we present PokeRL, a modular system that trains deep reinforcement learning agents to complete early game tasks in Pokemon Red, including exiting the player's house, exploring Pallet Town to reach tall grass, and winning the first rival battle. Our main contributions are a loop-aware environment wrapper around the PyBoy emulator with map masking, a multi-layer anti-loop and anti-spam mechanism, and a dense hierarchical reward design. We argue that practical systems like PokeRL, which explicitly model failure modes such as loops and spam, are a necessary intermediate step between toy benchmarks and full Pokemon League champion agents. Code is available at https://github.com/reddheeraj/PokemonRL</p> + https://arxiv.org/abs/2604.10812 Sun, 12 Apr 2026 00:00:00 +0000 - SciPredict: Can LLMs Predict the Outcomes of Scientific Experiments in Natural Sciences? - https://arxiv.org/abs/2604.10718 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10718.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Udari Madhushani Sehwag, Elaine Lau, Haniyeh Ehsani Oskouie, Shayan Shabihi, Erich Liang, Andrea Toledo, Guillermo Mangialardi, Sergio Fonrouge, Ed-Yeremai Hernandez Cardona, Paula Vergara, Utkarsh Tyagi, Chen Bo Calvin Zhang, Pavi Bhatter, Nicholas Johnson, Furong Huang, Ernesto Gabriel Hernandez Montoya, Bing Liu</p><p><b>Upvotes:</b> 2</p><p><b>Summary:</b> Accelerating scientific discovery requires the identification of which experiments would yield the best outcomes before committing resources to costly physical validation. While existing benchmarks evaluate LLMs on scientific knowledge and reasoning, their ability to predict experimental outcomes - a task where AI could significantly exceed human capabilities - remains largely underexplored. We introduce SciPredict, a benchmark comprising 405 tasks derived from recent empirical studies in 33 specialized sub-fields of physics, biology, and chemistry. SciPredict addresses two critical questions: (a) can LLMs predict the outcome of scientific experiments with sufficient accuracy? and (b) can such predictions be reliably used in the scientific research process? Evaluations reveal fundamental limitations on both fronts. Model accuracies are 14-26% and human expert performance is approx20%. Although some frontier models exceed human performance model accuracy is still far below what would enable reliable experimental guidance. Even within the limited performance, models fail to distinguish reliable predictions from unreliable ones, achieving only approx20% accuracy regardless of their confidence or whether they judge outcomes as predictable without physical experimentation. Human experts, in contrast, demonstrate strong calibration: their accuracy increases from approx5% to approx80% as they deem outcomes more predictable without conducting the experiment. SciPredict establishes a rigorous framework demonstrating that superhuman performance in experimental science requires not just better predictions, but better awareness of prediction reliability. For reproducibility all our data and code are provided at https://github.com/scaleapi/scipredict</p> - https://arxiv.org/abs/2604.10718 + Rethinking the Diffusion Model from a Langevin Perspective + https://arxiv.org/abs/2604.10465 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10465.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Candi Zheng, Yuan Lan</p><p><b>Upvotes:</b> 9</p><p><b>Summary:</b> Diffusion models are often introduced from multiple perspectives, such as VAEs, score matching, or flow matching, accompanied by dense and technically demanding mathematics that can be difficult for beginners to grasp. One classic question is: how does the reverse process invert the forward process to generate data from pure noise? This article systematically organizes the diffusion model from a fresh Langevin perspective, offering a simpler, clearer, and more intuitive answer. We also address the following questions: how can ODE-based and SDE-based diffusion models be unified under a single framework? Why are diffusion models theoretically superior to ordinary VAEs? Why is flow matching not fundamentally simpler than denoising or score matching, but equivalent under maximum-likelihood? We demonstrate that the Langevin perspective offers clear and straightforward answers to these questions, bridging existing interpretations of diffusion models, showing how different formulations can be converted into one another within a common framework, and offering pedagogical value for both learners and experienced researchers seeking deeper intuition.</p> + https://arxiv.org/abs/2604.10465 Sun, 12 Apr 2026 00:00:00 +0000 - DiningBench: A Hierarchical Multi-view Benchmark for Perception and Reasoning in the Dietary Domain - https://arxiv.org/abs/2604.10425 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10425.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Song Jin, Juntian Zhang, Xun Zhang, Zeying Tian, Fei Jiang, Guojun Yin, Wei Lin, Yong Liu, Rui Yan</p><p><b>Upvotes:</b> 2</p><p><b>Summary:</b> Recent advancements in Vision-Language Models (VLMs) have revolutionized general visual understanding. However, their application in the food domain remains constrained by benchmarks that rely on coarse-grained categories, single-view imagery, and inaccurate metadata. To bridge this gap, we introduce DiningBench, a hierarchical, multi-view benchmark designed to evaluate VLMs across three levels of cognitive complexity: Fine-Grained Classification, Nutrition Estimation, and Visual Question Answering. Unlike previous datasets, DiningBench comprises 3,021 distinct dishes with an average of 5.27 images per entry, incorporating fine-grained "hard" negatives from identical menus and rigorous, verification-based nutritional data. We conduct an extensive evaluation of 29 state-of-the-art open-source and proprietary models. Our experiments reveal that while current VLMs excel at general reasoning, they struggle significantly with fine-grained visual discrimination and precise nutritional reasoning. Furthermore, we systematically investigate the impact of multi-view inputs and Chain-of-Thought reasoning, identifying five primary failure modes. DiningBench serves as a challenging testbed to drive the next generation of food-centric VLM research. All codes are released in https://github.com/meituan/DiningBench.</p> - https://arxiv.org/abs/2604.10425 + The Blind Spot of Agent Safety: How Benign User Instructions Expose Critical Vulnerabilities in Computer-Use Agents + https://arxiv.org/abs/2604.10577 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10577.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Xuwei Ding, Skylar Zhai, Linxin Song, Jiate Li, Taiwei Shi, Nicholas Meade, Siva Reddy, Jian Kang, Jieyu Zhao</p><p><b>Upvotes:</b> 11</p><p><b>Summary:</b> Computer-use agents (CUAs) can now autonomously complete complex tasks in real digital environments, but when misled, they can also be used to automate harmful actions programmatically. Existing safety evaluations largely target explicit threats such as misuse and prompt injection, but overlook a subtle yet critical setting where user instructions are entirely benign and harm arises from the task context or execution outcome. We introduce OS-BLIND, a benchmark that evaluates CUAs under unintended attack conditions, comprising 300 human-crafted tasks across 12 categories, 8 applications, and 2 threat clusters: environment-embedded threats and agent-initiated harms. Our evaluation on frontier models and agentic frameworks reveals that most CUAs exceed 90% attack success rate (ASR), and even the safety-aligned Claude 4.5 Sonnet reaches 73.0% ASR. More interestingly, this vulnerability becomes even more severe, with ASR rising from 73.0% to 92.7% when Claude 4.5 Sonnet is deployed in multi-agent systems. Our analysis further shows that existing safety defenses provide limited protection when user instructions are benign. Safety alignment primarily activates within the first few steps and rarely re-engages during subsequent execution. In multi-agent systems, decomposed subtasks obscure the harmful intent from the model, causing safety-aligned models to fail. We will release our OS-BLIND to encourage the broader research community to further investigate and address these safety challenges.</p> + https://arxiv.org/abs/2604.10577 Sun, 12 Apr 2026 00:00:00 +0000 - BMdataset: A Musicologically Curated LilyPond Dataset - https://arxiv.org/abs/2604.10628 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10628.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Matteo Spanio, Ilay Guler, Antonio Rodà</p><p><b>Upvotes:</b> 2</p><p><b>Summary:</b> Symbolic music research has relied almost exclusively on MIDI-based datasets; text-based engraving formats such as LilyPond remain unexplored for music understanding. We present BMdataset, a musicologically curated dataset of 393 LilyPond scores (2,646 movements) transcribed by experts directly from original Baroque manuscripts, with metadata covering composer, musical form, instrumentation, and sectional attributes. Building on this resource, we introduce LilyBERT (weights can be found at https://huggingface.co/csc-unipd/lilybert), a CodeBERT-based encoder adapted to symbolic music through vocabulary extension with 115 LilyPond-specific tokens and masked language model pre-training. Linear probing on the out-of-domain Mutopia corpus shows that, despite its modest size (~90M tokens), fine-tuning on BMdataset alone outperforms continuous pre-training on the full PDMX corpus (~15B tokens) for both composer and style classification, demonstrating that small, expertly curated datasets can be more effective than large, noisy corpora for music understanding. Combining broad pre-training with domain-specific fine-tuning yields the best results overall (84.3% composer accuracy), confirming that the two data regimes are complementary. We release the dataset, tokenizer, and model to establish a baseline for representation learning on LilyPond.</p> - https://arxiv.org/abs/2604.10628 - Sun, 12 Apr 2026 00:00:00 +0000 - - - Advancing Polish Language Modeling through Tokenizer Optimization in the Bielik v3 7B and 11B Series - https://arxiv.org/abs/2604.10799 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10799.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Krzysztof Ociepa, Łukasz Flis, Remigiusz Kinas, Krzysztof Wróbel, Adrian Gwoździej</p><p><b>Upvotes:</b> 4</p><p><b>Summary:</b> The development of the Bielik v3 PL series, encompassing both the 7B and 11B parameter variants, represents a significant milestone in the field of language-specific large language model (LLM) optimization. While general-purpose models often demonstrate impressive multilingual capabilities, they frequently suffer from a fundamental architectural inefficiency: the use of universal tokenizers. These tokenizers, typically designed to cover a broad spectrum of languages, often fail to capture the morphological nuances of specific languages like Polish, leading to higher fertility ratios, increased inference costs, and restricted effective context windows. This report details the transition from the universal Mistral-based tokenization to a dedicated Polish-optimized vocabulary for the Bielik v3 models, exploring the FOCUS-based embedding initialization, the multi-stage pretraining curriculum, and the subsequent post-training alignment involving Supervised Fine-Tuning, Direct Preference Optimization, and Reinforcement Learning through Group Relative Policy Optimization with verifiable rewards.</p> - https://arxiv.org/abs/2604.10799 - Sun, 12 Apr 2026 00:00:00 +0000 - - - TorchUMM: A Unified Multimodal Model Codebase for Evaluation, Analysis, and Post-training - https://arxiv.org/abs/2604.10784 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10784.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Yinyi Luo, Wenwen Wang, Hayes Bai, Hongyu Zhu, Hao Chen, Pan He, Marios Savvides, Sharon Li, Jindong Wang</p><p><b>Upvotes:</b> 5</p><p><b>Summary:</b> Recent advances in unified multimodal models (UMMs) have led to a proliferation of architectures capable of understanding, generating, and editing across visual and textual modalities. However, developing a unified framework for UMMs remains challenging due to the diversity of model architectures and the heterogeneity of training paradigms and implementation details. In this paper, we present TorchUMM, the first unified codebase for comprehensive evaluation, analysis, and post-training across diverse UMM backbones, tasks, and datasets. TorchUMM supports a broad spectrum of models covering a wide range of scales and design paradigms. Our benchmark encompasses three core task dimensions: multimodal understanding, generation, and editing, and integrates both established and novel datasets to evaluate perception, reasoning, compositionality, and instruction-following abilities. By providing a unified interface and standardized evaluation protocols, TorchUMM enables fair and reproducible comparisons across heterogeneous models and fosters deeper insights into their strengths and limitations, facilitating the development of more capable unified multimodal systems. Code is available at: https://github.com/AIFrontierLab/TorchUMM.</p> - https://arxiv.org/abs/2604.10784 - Sun, 12 Apr 2026 00:00:00 +0000 - - - SCOPE: Signal-Calibrated On-Policy Distillation Enhancement with Dual-Path Adaptive Weighting - https://arxiv.org/abs/2604.10688 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10688.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Binbin Zheng, Xing Ma, Yiheng Liang, Jingqing Ruan, Xiaoliang Fu, Kepeng Lin, Benchang Zhu, Ke Zeng, Xunliang Cai</p><p><b>Upvotes:</b> 5</p><p><b>Summary:</b> On-policy reinforcement learning has become the dominant paradigm for reasoning alignment in large language models, yet its sparse, outcome-level rewards make token-level credit assignment notoriously difficult. On-Policy Distillation (OPD) alleviates this by introducing dense, token-level KL supervision from a teacher model, but typically applies this supervision uniformly across all rollouts, ignoring fundamental differences in signal quality. We propose Signal-Calibrated On-Policy Distillation Enhancement (SCOPE), a dual-path adaptive training framework that routes on-policy rollouts by correctness into two complementary supervision paths. For incorrect trajectories, SCOPE performs teacher-perplexity-weighted KL distillation to prioritize instances where the teacher demonstrates genuine corrective capability, while down-weighting unreliable guidance. For correct trajectories, it applies student-perplexity-weighted MLE to concentrate reinforcement on low-confidence samples at the capability boundary rather than over-reinforcing already mastered ones. Both paths employ a group-level normalization to adaptively calibrate weight distributions, accounting for the intrinsic difficulty variance across prompts. Extensive experiments on six reasoning benchmarks show that SCOPE achieves an average relative improvement of 11.42% in Avg@32 and 7.30% in Pass@32 over competitive baselines, demonstrating its consistent effectiveness.</p> - https://arxiv.org/abs/2604.10688 - Sun, 12 Apr 2026 00:00:00 +0000 - - - Tracing the Roots: A Multi-Agent Framework for Uncovering Data Lineage in Post-Training LLMs - https://arxiv.org/abs/2604.10480 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10480.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Yu Li, Xiaoran Shang, Qizhi Pei, Yun Zhu, Xin Gao, Honglin Lin, Zhanping Zhong, Zhuoshi Pan, Zheng Liu, Xiaoyang Wang, Conghui He, Dahua Lin, Feng Zhao, Lijun Wu</p><p><b>Upvotes:</b> 14</p><p><b>Summary:</b> Post-training data plays a pivotal role in shaping the capabilities of Large Language Models (LLMs), yet datasets are often treated as isolated artifacts, overlooking the systemic connections that underlie their evolution. To disentangle these complex relationships, we introduce the concept of data lineage to the LLM ecosystem and propose an automated multi-agent framework to reconstruct the evolutionary graph of dataset development. Through large-scale lineage analysis, we characterize domain-specific structural patterns, such as vertical refinement in math-oriented datasets and horizontal aggregation in general-domain corpora. Moreover, we uncover pervasive systemic issues, including structural redundancy induced by implicit dataset intersections and the propagation of benchmark contamination along lineage paths. To demonstrate the practical value of lineage analysis for data construction, we leverage the reconstructed lineage graph to create a lineage-aware diversity-oriented dataset. By anchoring instruction sampling at upstream root sources, this approach mitigates downstream homogenization and hidden redundancy, yielding a more diverse post-training corpus. We further highlight lineage-centric analysis as an efficient and robust topological alternative to sample-level dataset comparison for large-scale data ecosystems. By grounding data construction in explicit lineage structures, our work advances post-training data curation toward a more systematic and controllable paradigm.</p> - https://arxiv.org/abs/2604.10480 - Sun, 12 Apr 2026 00:00:00 +0000 - - - SHARE: Social-Humanities AI for Research and Education - https://arxiv.org/abs/2604.11152 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11152.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> João Gonçalves, Sonia de Jager, Petr Knoth, David Pride, Nick Jelicic</p><p><b>Upvotes:</b> 0</p><p><b>Summary:</b> This intermediate technical report introduces the SHARE family of base models and the MIRROR user interface. The SHARE models are the first causal language models fully pretrained by and for the social sciences and humanities (SSH). Their performance in modelling SSH texts is close to that of general purpose models (Phi-4) which use 100 times more tokens, as shown by our custom SSH Cloze benchmark. The MIRROR user interface is designed for reviewing text inputs from the SSH disciplines while preserving critical engagement. By prototyping a generative AI interface that does not generate any text, we propose a way to harness the capabilities of the SHARE models without compromising the integrity of SSH principles and norms.</p> - https://arxiv.org/abs/2604.11152 + 3DTV: A Feedforward Interpolation Network for Real-Time View Synthesis + https://arxiv.org/abs/2604.11211 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11211.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Stefan Schulz, Fernando Edelstein, Hannah Dröge, Matthias B. Hullin, Markus Plack</p><p><b>Upvotes:</b> 1</p><p><b>Summary:</b> Real-time free-viewpoint rendering requires balancing multi-camera redundancy with the latency constraints of interactive applications. We address this challenge by combining lightweight geometry with learning and propose 3DTV, a feedforward network for real-time sparse-view interpolation. A Delaunay-based triplet selection ensures angular coverage for each target view. Building on this, we introduce a pose-aware depth module that estimates a coarse-to-fine depth pyramid, enabling efficient feature reprojection and occlusion-aware blending. Unlike methods that require scene-specific optimization, 3DTV runs feedforward without retraining, making it practical for AR/VR, telepresence, and interactive applications. Our experiments on challenging multi-view video datasets demonstrate that 3DTV consistently achieves a strong balance of quality and efficiency, outperforming recent real-time novel-view baselines. Crucially, 3DTV avoids explicit proxies, enabling robust rendering across diverse scenes. This makes it a practical solution for low-latency multi-view streaming and interactive rendering. Project Page: https://stefanmschulz.github.io/3DTV_webpage/</p> + https://arxiv.org/abs/2604.11211 Mon, 13 Apr 2026 00:00:00 +0000 - How Alignment Routes: Localizing, Scaling, and Controlling Policy Circuits in Language Models - https://arxiv.org/abs/2604.04385 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.04385.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Gregory N. Frank</p><p><b>Upvotes:</b> 0</p><p><b>Summary:</b> This paper localizes the policy routing mechanism in alignment-trained language models. An intermediate-layer attention gate reads detected content and triggers deeper amplifier heads that boost the signal toward refusal. In smaller models the gate and amplifier are single heads; at larger scale they become bands of heads across adjacent layers. The gate contributes under 1% of output DLA, but interchange testing (p<0.001) and knockout cascade confirm it is causally necessary. Interchange screening at n>=120 detects the same motif in twelve models from six labs (2B to 72B), though specific heads differ by lab. Per-head ablation weakens up to 58x at 72B and misses gates that interchange identifies; interchange is the only reliable audit at scale. Modulating the detection-layer signal continuously controls policy from hard refusal through evasion to factual answering. On safety prompts the same intervention turns refusal into harmful guidance, showing the safety-trained capability is gated by routing rather than removed. Thresholds vary by topic and by input language, and the circuit relocates across generations within a family while behavioral benchmarks register no change. Routing is early-commitment: the gate commits at its own layer before deeper layers finish processing the input. Under an in-context substitution cipher, gate interchange necessity collapses 70 to 99% across three models and the model switches to puzzle-solving. Injecting the plaintext gate activation into the cipher forward pass restores 48% of refusals in Phi-4-mini, localizing the bypass to the routing interface. A second method, cipher contrast analysis, uses plain/cipher DLA differences to map the full cipher-sensitive routing circuit in O(3n) forward passes. Any encoding that defeats detection-layer pattern matching bypasses the policy regardless of whether deeper layers reconstruct the content.</p> - https://arxiv.org/abs/2604.04385 + Seeing Through Touch: Tactile-Driven Visual Localization of Material Regions + https://arxiv.org/abs/2604.11579 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11579.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Seongyu Kim, Seungwoo Lee, Hyeonggon Ryu, Joon Son Chung, Arda Senocak</p><p><b>Upvotes:</b> 1</p><p><b>Summary:</b> We address the problem of tactile localization, where the goal is to identify image regions that share the same material properties as a tactile input. Existing visuo-tactile methods rely on global alignment and thus fail to capture the fine-grained local correspondences required for this task. The challenge is amplified by existing datasets, which predominantly contain close-up, low-diversity images. We propose a model that learns local visuo-tactile alignment via dense cross-modal feature interactions, producing tactile saliency maps for touch-conditioned material segmentation. To overcome dataset constraints, we introduce: (i) in-the-wild multi-material scene images that expand visual diversity, and (ii) a material-diversity pairing strategy that aligns each tactile sample with visually varied yet tactilely consistent images, improving contextual localization and robustness to weak signals. We also construct two new tactile-grounded material segmentation datasets for quantitative evaluation. Experiments on both new and existing benchmarks show that our approach substantially outperforms prior visuo-tactile methods in tactile localization.</p> + https://arxiv.org/abs/2604.11579 Mon, 13 Apr 2026 00:00:00 +0000 - ADD for Multi-Bit Image Watermarking - https://arxiv.org/abs/2604.11491 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11491.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> An Luo, Jie Ding</p><p><b>Upvotes:</b> 1</p><p><b>Summary:</b> As generative models enable rapid creation of high-fidelity images, societal concerns about misinformation and authenticity have intensified. A promising remedy is multi-bit image watermarking, which embeds a multi-bit message into an image so that a verifier can later detect whether the image is generated by someone and further identify the source by decoding the embedded message. Existing approaches often fall short in capacity, resilience to common image distortions, and theoretical justification. To address these limitations, we propose ADD (Add, Dot, Decode), a multi-bit image watermarking method with two stages: learning a watermark to be linearly combined with the multi-bit message and added to the image, and decoding through inner products between the watermarked image and the learned watermark. On the standard MS-COCO benchmark, we demonstrate that for the challenging task of 48-bit watermarking, ADD achieves 100\% decoding accuracy, with performance dropping by at most 2\% under a wide range of image distortions, substantially smaller than the 14\% average drop of state-of-the-art methods. In addition, ADD achieves substantial computational gains, with 2-fold faster embedding and 7.4-fold faster decoding than the fastest existing method. We further provide a theoretical analysis explaining why the learned watermark and the corresponding decoding rule are effective.</p> - https://arxiv.org/abs/2604.11491 + Beyond Perception Errors: Semantic Fixation in Large Vision-Language Models + https://arxiv.org/abs/2604.12119 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.12119.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Md Tanvirul Alam</p><p><b>Upvotes:</b> 1</p><p><b>Summary:</b> Large vision-language models (VLMs) often rely on familiar semantic priors, but existing evaluations do not cleanly separate perception failures from rule-mapping failures. We study this behavior as semantic fixation: preserving a default interpretation even when the prompt specifies an alternative, equally valid mapping. To isolate this effect, we introduce VLM-Fix, a controlled benchmark over four abstract strategy games that evaluates identical terminal board states under paired standard and inverse rule formulations. Across 14 open and closed VLMs, accuracy consistently favors standard rules, revealing a robust semantic-fixation gap. Prompt interventions support this mechanism: neutral alias prompts substantially narrow the inverse-rule gap, while semantically loaded aliases reopen it. Post-training is strongly rule-aligned: training on one rule improves same-rule transfer but hurts opposite-rule transfer, while joint-rule training improves broader transfer. To test external validity beyond synthetic games, we evaluate analogous defamiliarization interventions on VLMBias and observe the same qualitative pattern. Finally, late-layer activation steering partially recovers degraded performance, indicating that semantic-fixation errors are at least partly editable in late representations. Project page, code, and dataset available at https://maveryn.github.io/vlm-fix/.</p> + https://arxiv.org/abs/2604.12119 Mon, 13 Apr 2026 00:00:00 +0000 - Polyglot Teachers: Evaluating Language Models for Multilingual Synthetic Data Generation - https://arxiv.org/abs/2604.11290 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11290.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Lester James V. Miranda, Ivan Vulić, Anna Korhonen</p><p><b>Upvotes:</b> 1</p><p><b>Summary:</b> Synthesizing supervised finetuning (SFT) data from language models (LMs) to teach smaller models multilingual tasks has become increasingly common. However, teacher model selection is often ad hoc, typically defaulting to the largest available option, even though such models may have significant capability gaps in non-English languages. This practice can result in poor-quality synthetic data and suboptimal student downstream performance. In this work, we systematically characterize what makes an effective multilingual teacher. We measure intrinsic measures of data quality with extrinsic student model performance in a metric we call Polyglot Score; evaluating 10 LMs across 6 typologically diverse languages, generating over 1.4M SFT examples and training 240 student models. Among the models tested, Gemma 3 27B and Aya Expanse 32B emerge as consistently effective teachers across different student base model families. Further analyses reveal that model scale alone does not significantly predict teacher effectiveness; instead, data qualities such as prompt diversity, length, and response fluency capture over 93.3% of variance in intrinsic data quality and predict student performance. Finally, we provide practical recommendations, including matching the model families of teacher-student pairs and translating from or responding to existing prompts, which can yield improvements for less-resourced languages. We hope that our work advances data-centric research in multilingual synthetic data and LM development.</p> - https://arxiv.org/abs/2604.11290 + LASA: Language-Agnostic Semantic Alignment at the Semantic Bottleneck for LLM Safety + https://arxiv.org/abs/2604.12710 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.12710.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Junxiao Yang, Haoran Liu, Jinzhe Tu, Jiale Cheng, Zhexin Zhang, Shiyao Cui, Jiaqi Weng, Jialing Tao, Hui Xue, Hongning Wang, Han Qiu, Minlie Huang</p><p><b>Upvotes:</b> 2</p><p><b>Summary:</b> Large language models (LLMs) often demonstrate strong safety performance in high-resource languages, yet exhibit severe vulnerabilities when queried in low-resource languages. We attribute this gap to a mismatch between language-agnostic semantic understanding ability and language-dominant safety alignment biased toward high-resource languages. Consistent with this hypothesis, we empirically identify the semantic bottleneck in LLMs, an intermediate layer in which the geometry of model representations is governed primarily by shared semantic content rather than language identity. Building on this observation, we propose Language-Agnostic Semantic Alignment (LASA), which anchors safety alignment directly in semantic bottlenecks. Experiments show that LASA substantially improves safety across all languages: average attack success rate (ASR) drops from 24.7% to 2.8% on LLaMA-3.1-8B-Instruct and remains around 3-4% across Qwen2.5 and Qwen3 Instruct models (7B-32B). Together, our analysis and method offer a representation-level perspective on LLM safety, suggesting that safety alignment requires anchoring safety understanding not in surface text, but in the model's language-agnostic semantic space.</p> + https://arxiv.org/abs/2604.12710 Mon, 13 Apr 2026 00:00:00 +0000 - Panoptic Pairwise Distortion Graph - https://arxiv.org/abs/2604.11004 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11004.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Muhammad Kamran Janjua, Abdul Wahab, Bahador Rashidi</p><p><b>Upvotes:</b> 1</p><p><b>Summary:</b> In this work, we introduce a new perspective on comparative image assessment by representing an image pair as a structured composition of its regions. In contrast, existing methods focus on whole image analysis, while implicitly relying on region-level understanding. We extend the intra-image notion of a scene graph to inter-image, and propose a novel task of Distortion Graph (DG). DG treats paired images as a structured topology grounded in regions, and represents dense degradation information such as distortion type, severity, comparison and quality score in a compact interpretable graph structure. To realize the task of learning a distortion graph, we contribute (i) a region-level dataset, PandaSet, (ii) a benchmark suite, PandaBench, with varying region-level difficulty, and (iii) an efficient architecture, Panda, to generate distortion graphs. We demonstrate that PandaBench poses a significant challenge for state-of-the-art multimodal large language models (MLLMs) as they fail to understand region-level degradations even when fed with explicit region cues. We show that training on PandaSet or prompting with DG elicits region-wise distortion understanding, opening a new direction for fine-grained, structured pairwise image assessment.</p> - https://arxiv.org/abs/2604.11004 + Do Thought Streams Matter? Evaluating Reasoning in Gemini Vision-Language Models for Video Scene Understanding + https://arxiv.org/abs/2604.11177 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11177.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Shivam Sharma, Sankalp Nagaonkar, Ashish Choithani, Ashutosh Trivedi</p><p><b>Upvotes:</b> 4</p><p><b>Summary:</b> We benchmark how internal reasoning traces, which we call thought streams, affect video scene understanding in vision-language models. Using four configurations of Google's Gemini 2.5 Flash and Flash Lite across scenes extracted from 100 hours of video, we ask three questions: does more thinking lead to better outputs, where do the gains stop, and what do these models actually think about? We introduce three evaluation metrics. Contentfulness measures how much of the thought stream is useful scene content versus meta-commentary. Thought-Final Coverage measures how faithfully the thought stream translates into the final output. Dominant Entity Analysis identifies which subjects, actions, and settings the model focuses on. GPT-5 serves as an independent judge. We find that quality gains from additional thinking plateau quickly, with most improvement occurring in the first few hundred tokens. Flash Lite offers the best balance between quality and token usage. Tight reasoning budgets cause the model to add content in the final output that it never reasoned about, a form of compression-step hallucination. Despite being different model tiers, Flash and Flash Lite produce similar thought streams, though they differ in style: Flash discusses its reasoning process, while Lite focuses on describing the scene.</p> + https://arxiv.org/abs/2604.11177 Mon, 13 Apr 2026 00:00:00 +0000 - Time is Not a Label: Continuous Phase Rotation for Temporal Knowledge Graphs and Agentic Memory - https://arxiv.org/abs/2604.11544 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11544.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Weixian Waylon Li, Jiaxin Zhang, Xianan Jim Yang, Tiejun Ma, Yiwen Guo</p><p><b>Upvotes:</b> 1</p><p><b>Summary:</b> Structured memory representations such as knowledge graphs are central to autonomous agents and other long-lived systems. However, most existing approaches model time as discrete metadata, either sorting by recency (burying old-yet-permanent knowledge), simply overwriting outdated facts, or requiring an expensive LLM call at every ingestion step, leaving them unable to distinguish persistent facts from evolving ones. To address this, we introduce RoMem, a drop-in temporal knowledge graph module for structured memory systems, applicable to agentic memory and beyond. A pretrained Semantic Speed Gate maps each relation's text embedding to a volatility score, learning from data that evolving relations (e.g., "president of") should rotate fast while persistent ones (e.g., "born in") should remain stable. Combined with continuous phase rotation, this enables geometric shadowing: obsolete facts are rotated out of phase in complex vector space, so temporally correct facts naturally outrank contradictions without deletion. On temporal knowledge graph completion, RoMem achieves state-of-the-art results on ICEWS05-15 (72.6 MRR). Applied to agentic memory, it delivers 2-3x MRR and answer accuracy on temporal reasoning (MultiTQ), dominates hybrid benchmark (LoCoMo), preserves static memory with zero degradation (DMR-MSC), and generalises zero-shot to unseen financial domains (FinTMMBench).</p> - https://arxiv.org/abs/2604.11544 + You Only Judge Once: Multi-response Reward Modeling in a Single Forward Pass + https://arxiv.org/abs/2604.10966 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10966.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Yinuo Yang, Zixian Ma, Manasi Ganti, Jieyu Zhang, Ranjay Krishna</p><p><b>Upvotes:</b> 6</p><p><b>Summary:</b> We present a discriminative multimodal reward model that scores all candidate responses in a single forward pass. Conventional discriminative reward models evaluate each response independently, requiring multiple forward passes, one for each potential response. Our approach concatenates multiple responses with separator tokens and applies cross-entropy over their scalar scores, enabling direct comparative reasoning and efficient N-way preference learning. The multi-response design also yields up to Ntimes wall-clock speedup and FLOPs reduction over conventional single-response scoring. To enable N-way reward evaluation beyond existing pairwise benchmarks, we construct two new benchmarks: (1) MR^2Bench-Image contains human-annotated rankings over responses from 8 diverse models; (2) MR^2Bench-Video is a large-scale video-based reward benchmark derived from 94K crowdsourced pairwise human judgments over video question-answering spanning 19 models, denoised via preference graph ensemble. Both benchmarks provide 4-response evaluation variants sampled from the full rankings. Built on a 4B vision-language backbone with LoRA fine-tuning and a lightweight MLP value head, our model achieves state-of-the-art results on six multimodal reward benchmarks, including MR^2Bench-Image, MR^2Bench-Video, and four other existing benchmarks. Our model outperforms existing larger generative and discriminative reward models. We further demonstrate that our reward model, when used in reinforcement learning with GRPO, produces improved policy models that maintain performance across standard multimodal benchmarks while substantially improving open-ended generation quality, outperforming a single-response discriminative reward model (RM) baseline by a large margin in both training stability and open-ended generation quality.</p> + https://arxiv.org/abs/2604.10966 Mon, 13 Apr 2026 00:00:00 +0000 - Learning Long-term Motion Embeddings for Efficient Kinematics Generation - https://arxiv.org/abs/2604.11737 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11737.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Nick Stracke, Kolja Bauer, Stefan Andreas Baumann, Miguel Angel Bautista, Josh Susskind, Björn Ommer</p><p><b>Upvotes:</b> 2</p><p><b>Summary:</b> Understanding and predicting motion is a fundamental component of visual intelligence. Although modern video models exhibit strong comprehension of scene dynamics, exploring multiple possible futures through full video synthesis remains prohibitively inefficient. We model scene dynamics orders of magnitude more efficiently by directly operating on a long-term motion embedding that is learned from large-scale trajectories obtained from tracker models. This enables efficient generation of long, realistic motions that fulfill goals specified via text prompts or spatial pokes. To achieve this, we first learn a highly compressed motion embedding with a temporal compression factor of 64x. In this space, we train a conditional flow-matching model to generate motion latents conditioned on task descriptions. The resulting motion distributions outperform those of both state-of-the-art video models and specialized task-specific approaches.</p> - https://arxiv.org/abs/2604.11737 + LARY: A Latent Action Representation Yielding Benchmark for Generalizable Vision-to-Action Alignment + https://arxiv.org/abs/2604.11689 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11689.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Dujun Nie, Fengjiao Chen, Qi Lv, Jun Kuang, Xiaoyu Li, Xuezhi Cao, Xunliang Cai</p><p><b>Upvotes:</b> 7</p><p><b>Summary:</b> While the shortage of explicit action data limits Vision-Language-Action (VLA) models, human action videos offer a scalable yet unlabeled data source. A critical challenge in utilizing large-scale human video datasets lies in transforming visual signals into ontology-independent representations, known as latent actions. However, the capacity of latent action representation to derive robust control from visual observations has yet to be rigorously evaluated. We introduce the Latent Action Representation Yielding (LARY) Benchmark, a unified framework for evaluating latent action representations on both high-level semantic actions (what to do) and low-level robotic control (how to do). The comprehensively curated dataset encompasses over one million videos (1,000 hours) spanning 151 action categories, alongside 620K image pairs and 595K motion trajectories across diverse embodiments and environments. Our experiments reveal two crucial insights: (i) General visual foundation models, trained without any action supervision, consistently outperform specialized embodied latent action models. (ii) Latent-based visual space is fundamentally better aligned to physical action space than pixel-based space. These results suggest that general visual representations inherently encode action-relevant knowledge for physical control, and that semantic-level abstraction serves as a fundamentally more effective pathway from vision to action than pixel-level reconstruction.</p> + https://arxiv.org/abs/2604.11689 Mon, 13 Apr 2026 00:00:00 +0000 - Low-rank Optimization Trajectories Modeling for LLM RLVR Acceleration - https://arxiv.org/abs/2604.11446 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11446.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Zhipeng Chen, Tao Qian, Wayne Xin Zhao, Ji-Rong Wen</p><p><b>Upvotes:</b> 3</p><p><b>Summary:</b> Recently, scaling reinforcement learning with verifiable rewards (RLVR) for large language models (LLMs) has emerged as an effective training paradigm for significantly improving model capabilities, which requires guiding the model to perform extensive exploration and learning, leading to substantial computational overhead and becoming a key challenge. To reduce the number of training steps, Prior work performs linear extrapolation of model parameters. However, the dynamics of model parameter updates during RLVR training remain insufficiently understood. To further investigate the evolution of LLMs during RLVR training, we conduct empirical experiments and find that the rank-1 subspace of the model does not evolve linearly, and its dominance over the original parameters is further amplified during LoRA training. Based on the above insights, we propose the Nonlinear Extrapolation of low-rank trajectories (NExt), a novel framework that models and extrapolates low-rank parameter trajectories in a nonlinear manner. Concretely, we first train the model using LoRA and extract the rank-1 subspace of parameter differences at multiple training steps, which is then used for the subsequent nonlinear extrapolation. Afterward, we utilized the extracted rank-1 subspace to train a predictor, which can model the trajectory of parameter updates during RLVR, and then perform the predict-extend process to extrapolate model parameters, achieving the acceleration of RLVR. To further study and understand NExt, we conduct comprehensive experiments that demonstrate the effectiveness and robustness of the method. Our method reduces computational overhead by approximately 37.5\% while remaining compatible with a wide range of RLVR algorithms and tasks. We release our code in https://github.com/RUCAIBox/NExt.</p> - https://arxiv.org/abs/2604.11446 + ClawGUI: A Unified Framework for Training, Evaluating, and Deploying GUI Agents + https://arxiv.org/abs/2604.11784 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11784.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Fei Tang, Zhiqiong Lu, Boxuan Zhang, Weiming Lu, Jun Xiao, Yueting Zhuang, Yongliang Shen</p><p><b>Upvotes:</b> 120</p><p><b>Summary:</b> GUI agents drive applications through their visual interfaces instead of programmatic APIs, interacting with arbitrary software via taps, swipes, and keystrokes, reaching a long tail of applications that CLI-based agents cannot. Yet progress in this area is bottlenecked less by modeling capacity than by the absence of a coherent full-stack infrastructure: online RL training suffers from environment instability and closed pipelines, evaluation protocols drift silently across works, and trained agents rarely reach real users on real devices. We present ClawGUI, an open-source framework addressing these three gaps within a single harness. ClawGUI-RL provides the first open-source GUI agent RL infrastructure with validated support for both parallel virtual environments and real physical devices, integrating GiGPO with a Process Reward Model for dense step-level supervision. ClawGUI-Eval enforces a fully standardized evaluation pipeline across 6 benchmarks and 11+ models, achieving 95.8\% reproduction against official baselines. ClawGUI-Agent brings trained agents to Android, HarmonyOS, and iOS through 12+ chat platforms with hybrid CLI-GUI control and persistent personalized memory. Trained end to end within this pipeline, ClawGUI-2B achieves 17.1\% Success Rate on MobileWorld GUI-Only, outperforming the same-scale MAI-UI-2B baseline by 6.0\%.</p> + https://arxiv.org/abs/2604.11784 Mon, 13 Apr 2026 00:00:00 +0000 - SWE-AGILE: A Software Agent Framework for Efficiently Managing Dynamic Reasoning Context - https://arxiv.org/abs/2604.11716 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11716.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Shuquan Lian, Juncheng Liu, Yazhe Chen, Yuhong Chen, Hui Li</p><p><b>Upvotes:</b> 3</p><p><b>Summary:</b> Prior representative ReAct-style approaches in autonomous Software Engineering (SWE) typically lack the explicit System-2 reasoning required for deep analysis and handling complex edge cases. While recent reasoning models demonstrate the potential of extended Chain-of-Thought (CoT), applying them to the multi-turn SWE task creates a fundamental dilemma: retaining full reasoning history leads to context explosion and ``Lost-in-the-Middle'' degradation, while discarding it would force the agent to redundantly re-reason at every step. To address these challenges, we propose SWE-AGILE, a novel software agent framework designed to bridge the gap between reasoning depth, efficiency, and context constraints. SWE-AGILE introduces a Dynamic Reasoning Context strategy, maintaining a ``sliding window'' of detailed reasoning for immediate continuity to prevent redundant re-analyzing, while compressing historical reasoning content into concise Reasoning Digests. Empirically, SWE-AGILE sets a new standard for 7B-8B models on SWE-Bench-Verified using only 2.2k trajectories and 896 tasks. Code is available at https://github.com/KDEGroup/SWE-AGILE.</p> - https://arxiv.org/abs/2604.11716 - Mon, 13 Apr 2026 00:00:00 +0000 + SpotSound: Enhancing Large Audio-Language Models with Fine-Grained Temporal Grounding + https://arxiv.org/abs/2604.13023 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.13023.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Luoyi Sun, Xiao Zhou, Zeqian Li, Ya Zhang, Yanfeng Wang, Weidi Xie</p><p><b>Upvotes:</b> 0</p><p><b>Summary:</b> Large Audio-Language Models (ALMs) have recently demonstrated remarkable capabilities in holistic audio understanding, yet they remain unreliable for temporal grounding, i.e., the task of pinpointing exactly when an event occurs within long-form audio. This limitation stems from two factors: training data dominated by clip-level supervision lacking precise timestamps, and benchmarks that fail to simulate real-world scenarios where short events are obscured by dense background sounds. In this paper, we introduce SpotSound, an audio language model designed for grounding audio events. SpotSound incorporates a novel training objective, specifically designed to suppress hallucinated timestamps for events absent from the input. Additionally, we present SpotSound-Bench, a challenging temporal grounding benchmark where target events occupy less than ~10\% of each clip, creating a rigorous `needle-in-a-haystack' evaluation. Experiments demonstrate that SpotSound achieves state-of-the-art results on temporal grounding benchmarks while maintaining robust performance across general downstream audio-language tasks. Code, models and benchmark are released on https://loiesun.github.io/spotsound/</p> + https://arxiv.org/abs/2604.13023 + Tue, 14 Apr 2026 00:00:00 +0000 - Playing Along: Learning a Double-Agent Defender for Belief Steering via Theory of Mind - https://arxiv.org/abs/2604.11666 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11666.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Hanqi Xiao, Vaidehi Patil, Zaid Khan, Hyunji Lee, Elias Stengel-Eskin, Mohit Bansal</p><p><b>Upvotes:</b> 3</p><p><b>Summary:</b> As large language models (LLMs) become the engine behind conversational systems, their ability to reason about the intentions and states of their dialogue partners (i.e., form and use a theory-of-mind, or ToM) becomes increasingly critical for safe interaction with potentially adversarial partners. We propose a novel privacy-themed ToM challenge, ToM for Steering Beliefs (ToM-SB), in which a defender must act as a Double Agent to steer the beliefs of an attacker with partial prior knowledge within a shared universe. To succeed on ToM-SB, the defender must engage with and form a ToM of the attacker, with a goal of fooling the attacker into believing they have succeeded in extracting sensitive information. We find that strong frontier models like Gemini3-Pro and GPT-5.4 struggle on ToM-SB, often failing to fool attackers in hard scenarios with partial attacker prior knowledge, even when prompted to reason about the attacker's beliefs (ToM prompting). To close this gap, we train models on ToM-SB to act as AI Double Agents using reinforcement learning, testing both fooling and ToM rewards. Notably, we find a bidirectionally emergent relationship between ToM and attacker-fooling: rewarding fooling success alone improves ToM, and rewarding ToM alone improves fooling. Across four attackers with different strengths, six defender methods, and both in-distribution and out-of-distribution (OOD) evaluation, we find that gains in ToM and attacker-fooling are well-correlated, highlighting belief modeling as a key driver of success on ToM-SB. AI Double Agents that combine both ToM and fooling rewards yield the strongest fooling and ToM performance, outperforming Gemini3-Pro and GPT-5.4 with ToM prompting on hard scenarios. We also show that ToM-SB and AI Double Agents can be extended to stronger attackers, demonstrating generalization to OOD settings and the upgradability of our task.</p> - https://arxiv.org/abs/2604.11666 - Mon, 13 Apr 2026 00:00:00 +0000 + Domain-Specific Latent Representations Improve the Fidelity of Diffusion-Based Medical Image Super-Resolution + https://arxiv.org/abs/2604.12152 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.12152.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Sebastian Cajas, Ashaba Judith, Rahul Gorijavolu, Sahil Kapadia, Hillary Clinton Kasimbazi, Leo Kinyera, Emmanuel Paul Kwesiga, Sri Sri Jaithra Varma Manthena, Luis Filipe Nakayama, Ninsiima Doreen, Leo Anthony Celi</p><p><b>Upvotes:</b> 1</p><p><b>Summary:</b> Latent diffusion models for medical image super-resolution universally inherit variational autoencoders designed for natural photographs. We show that this default choice, not the diffusion architecture, is the dominant constraint on reconstruction quality. In a controlled experiment holding all other pipeline components fixed, replacing the generic Stable Diffusion VAE with MedVAE, a domain-specific autoencoder pretrained on more than 1.6 million medical images, yields +2.91 to +3.29 dB PSNR improvement across knee MRI, brain MRI, and chest X-ray (n = 1,820; Cohen's d = 1.37 to 1.86, all p < 10^{-20}, Wilcoxon signed-rank). Wavelet decomposition localises the advantage to the finest spatial frequency bands encoding anatomically relevant fine structure. Ablations across inference schedules, prediction targets, and generative architectures confirm the gap is stable within plus or minus 0.15 dB, while hallucination rates remain comparable between methods (Cohen's h < 0.02 across all datasets), establishing that reconstruction fidelity and generative hallucination are governed by independent pipeline components. These results provide a practical screening criterion: autoencoder reconstruction quality, measurable without diffusion training, predicts downstream SR performance (R^2 = 0.67), suggesting that domain-specific VAE selection should precede diffusion architecture search. Code and trained model weights are publicly available at https://github.com/sebasmos/latent-sr.</p> + https://arxiv.org/abs/2604.12152 + Tue, 14 Apr 2026 00:00:00 +0000 - Eliciting Medical Reasoning with Knowledge-enhanced Data Synthesis: A Semi-Supervised Reinforcement Learning Approach - https://arxiv.org/abs/2604.11547 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11547.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Haolin Li, Shuyang Jiang, Ruipeng Zhang, Jiangchao Yao, Ya Zhang, Yanfeng Wang</p><p><b>Upvotes:</b> 4</p><p><b>Summary:</b> While large language models hold promise for complex medical applications, their development is hindered by the scarcity of high-quality reasoning data. To address this issue, existing approaches typically distill chain-of-thought reasoning traces from large proprietary models via supervised fine-tuning, then conduct reinforcement learning (RL). These methods exhibit limited improvement on underrepresented domains like rare diseases while incurring substantial costs from generating complex reasoning chains. To efficiently enhance medical reasoning, we propose MedSSR, a Medical Knowledge-enhanced data Synthesis and Semi-supervised Reinforcement learning framework. Our framework first employs rare disease knowledge to synthesize distribution-controllable reasoning questions. We then utilize the policy model itself to generate high-quality pseudo-labels. This enables a two-stage, intrinsic-to-extrinsic training paradigm: self-supervised RL on the pseudo-labeled synthetic data, followed by supervised RL on the human-annotated real data. MedSSR scales model training efficiently without relying on costly trace distillation. Extensive experiments on Qwen and Llama demonstrate that our method outperforms existing methods across ten medical benchmarks, achieving up to +5.93% gain on rare-disease tasks. Our code is available at https://github.com/tdlhl/MedSSR.</p> - https://arxiv.org/abs/2604.11547 - Mon, 13 Apr 2026 00:00:00 +0000 + VideoFlexTok: Flexible-Length Coarse-to-Fine Video Tokenization + https://arxiv.org/abs/2604.12887 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.12887.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Andrei Atanov, Jesse Allardice, Roman Bachmann, Oğuzhan Fatih Kar, R Devon Hjelm, David Griffiths, Peter Fu, Afshin Dehghan, Amir Zamir</p><p><b>Upvotes:</b> 2</p><p><b>Summary:</b> Visual tokenizers map high-dimensional raw pixels into a compressed representation for downstream modeling. Beyond compression, tokenizers dictate what information is preserved and how it is organized. A de facto standard approach to video tokenization is to represent a video as a spatiotemporal 3D grid of tokens, each capturing the corresponding local information in the original signal. This requires the downstream model that consumes the tokens, e.g., a text-to-video model, to learn to predict all low-level details "pixel-by-pixel" irrespective of the video's inherent complexity, leading to high learning complexity. We present VideoFlexTok, which represents videos with a variable-length sequence of tokens structured in a coarse-to-fine manner -- where the first tokens (emergently) capture abstract information, such as semantics and motion, and later tokens add fine-grained details. The generative flow decoder enables realistic video reconstructions from any token count. This representation structure allows adapting the token count according to downstream needs and encoding videos longer than the baselines with the same budget. We evaluate VideoFlexTok on class- and text-to-video generative tasks and show that it leads to more efficient training compared to 3D grid tokens, e.g., achieving comparable generation quality (gFVD and ViCLIP Score) with a 5x smaller model (1.1B vs 5.2B). Finally, we demonstrate how VideoFlexTok can enable long video generation without prohibitive computational cost by training a text-to-video model on 10-second 81-frame videos with only 672 tokens, 8x fewer than a comparable 3D grid tokenizer.</p> + https://arxiv.org/abs/2604.12887 + Tue, 14 Apr 2026 00:00:00 +0000 - Continuous Adversarial Flow Models - https://arxiv.org/abs/2604.11521 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11521.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Shanchuan Lin, Ceyuan Yang, Zhijie Lin, Hao Chen, Haoqi Fan</p><p><b>Upvotes:</b> 5</p><p><b>Summary:</b> We propose continuous adversarial flow models, a type of continuous-time flow model trained with an adversarial objective. Unlike flow matching, which uses a fixed mean-squared-error criterion, our approach introduces a learned discriminator to guide training. This change in objective induces a different generalized distribution, which empirically produces samples that are better aligned with the target data distribution. Our method is primarily proposed for post-training existing flow-matching models, although it can also train models from scratch. On the ImageNet 256px generation task, our post-training substantially improves the guidance-free FID of latent-space SiT from 8.26 to 3.63 and of pixel-space JiT from 7.17 to 3.57. It also improves guided generation, reducing FID from 2.06 to 1.53 for SiT and from 1.86 to 1.80 for JiT. We further evaluate our approach on text-to-image generation, where it achieves improved results on both the GenEval and DPG benchmarks.</p> - https://arxiv.org/abs/2604.11521 - Mon, 13 Apr 2026 00:00:00 +0000 + Learning Versatile Humanoid Manipulation with Touch Dreaming + https://arxiv.org/abs/2604.13015 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.13015.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Yaru Niu, Zhenlong Fang, Binghong Chen, Shuai Zhou, Revanth Senthilkumaran, Hao Zhang, Bingqing Chen, Chen Qiu, H. Eric Tseng, Jonathan Francis, Ding Zhao</p><p><b>Upvotes:</b> 2</p><p><b>Summary:</b> Humanoid robots promise general-purpose assistance, yet real-world humanoid loco-manipulation remains challenging because it requires whole-body stability, dexterous hands, and contact-aware perception under frequent contact changes. In this work, we study dexterous, contact-rich humanoid loco-manipulation. We first develop an RL-based whole-body controller that provides stable lower-body and torso execution during complex manipulation. Built on this controller, we develop a whole-body humanoid data collection system that combines VR-based teleoperation with human-to-humanoid motion mapping, enabling efficient collection of real-world demonstrations. We then propose Humanoid Transformer with Touch Dreaming (HTD), a multimodal encoder--decoder Transformer that models touch as a core modality alongside multi-view vision and proprioception. HTD is trained in a single stage with behavioral cloning augmented by touch dreaming: in addition to predicting action chunks, the policy predicts future hand-joint forces and future tactile latents, encouraging the shared Transformer trunk to learn contact-aware representations for dexterous interaction. Across five contact-rich tasks, Insert-T, Book Organization, Towel Folding, Cat Litter Scooping, and Tea Serving, HTD achieves a 90.9% relative improvement in average success rate over the stronger baseline. Ablation results further show that latent-space tactile prediction is more effective than raw tactile prediction, yielding a 30% relative gain in success rate. These results demonstrate that combining robust whole-body execution, scalable humanoid data collection, and predictive touch-centered learning enables versatile, high-dexterity humanoid manipulation in the real world. Project webpage: humanoid-touch-dream.github.io.</p> + https://arxiv.org/abs/2604.13015 + Tue, 14 Apr 2026 00:00:00 +0000 - General365: Benchmarking General Reasoning in Large Language Models Across Diverse and Challenging Tasks - https://arxiv.org/abs/2604.11778 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11778.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Junlin Liu, Shengnan An, Shuang Zhou, Dan Ma, Shixiong Luo, Ying Xie, Yuan Zhang, Wenling Yuan, Yifan Zhou, Xiaoyu Li, Ziwen Wang, Xuezhi Cao, Xunliang Cai</p><p><b>Upvotes:</b> 6</p><p><b>Summary:</b> Contemporary large language models (LLMs) have demonstrated remarkable reasoning capabilities, particularly in specialized domains like mathematics and physics. However, their ability to generalize these reasoning skills to more general and broader contexts--often termed general reasoning--remains under-explored. Unlike domain-specific reasoning, general reasoning relies less on expert knowledge but still presents formidable reasoning challenges, such as complex constraints, nested logical branches, and semantic interference. To address this gap, we introduce General365, a benchmark specifically designed to assess general reasoning in LLMs. By restricting background knowledge to a K-12 level, General365 explicitly decouples reasoning from specialized expertise. The benchmark comprises 365 seed problems and 1,095 variant problems across eight categories, ensuring both high difficulty and diversity. Evaluations across 26 leading LLMs reveal that even the top-performing model achieves only 62.8% accuracy, in stark contrast to the near-perfect performances of LLMs in math and physics benchmarks. These results suggest that the reasoning abilities of current LLMs are heavily domain-dependent, leaving significant room for improvement in broader applications. We envision General365 as a catalyst for advancing LLM reasoning beyond domain-specific tasks toward robust, general-purpose real-world scenarios. Code, Dataset, and Leaderboard: https://general365.github.io</p> - https://arxiv.org/abs/2604.11778 - Mon, 13 Apr 2026 00:00:00 +0000 + Parcae: Scaling Laws For Stable Looped Language Models + https://arxiv.org/abs/2604.12946 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.12946.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Hayden Prairie, Zachary Novack, Taylor Berg-Kirkpatrick, Daniel Y. Fu</p><p><b>Upvotes:</b> 2</p><p><b>Summary:</b> Traditional fixed-depth architectures scale quality by increasing training FLOPs, typically through increased parameterization, at the expense of a higher memory footprint, or data. A potential alternative is looped architectures, which instead increase FLOPs by sending activations through a block of layers in a loop. While promising, existing recipes for training looped architectures can be unstable, suffering from residual explosion and loss spikes. We address these challenges by recasting looping as a nonlinear time-variant dynamical system over the residual stream. Via a linear approximation to this system, we find that instability occurs in existing looped architectures as a result of large spectral norms in their injection parameters. To address these instability issues, we propose Parcae, a novel stable, looped architecture that constrains the spectral norm of the injection parameters via discretization of a negative diagonal parameterization. As a result, Parcae achieves up to 6.3% lower validation perplexity over prior large-scale looped models. Using our stable looped architecture, we investigate the scaling properties of looping as a medium to improve quality by increasing FLOPs in training and test-time. For training, we derive predictable power laws to scale FLOPs while keeping parameter count fixed. Our initial scaling laws suggest that looping and data should be increased in tandem, given a fixed FLOP budget. At test-time, we find that Parcae can use looping to scale compute, following a predictable, saturating exponential decay. When scaled up to 1.3B parameters, we find that Parcae improves CORE and Core-Extended quality by 2.99 and 1.18 points when compared to strong Transformer baselines under a fixed parameter and data budget, achieving a relative quality of up to 87.5% a Transformer twice the size.</p> + https://arxiv.org/abs/2604.12946 + Tue, 14 Apr 2026 00:00:00 +0000 - From Reasoning to Agentic: Credit Assignment in Reinforcement Learning for Large Language Models - https://arxiv.org/abs/2604.09459 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.09459.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Chenchen Zhang</p><p><b>Upvotes:</b> 9</p><p><b>Summary:</b> Reinforcement learning (RL) for large language models (LLMs) increasingly relies on sparse, outcome-level rewards -- yet determining which actions within a long trajectory caused the outcome remains difficult. This credit assignment (CA) problem manifests in two regimes: reasoning RL, where credit must be distributed across tokens and steps within a single chain-of-thought generation (500--30K+ tokens); and agentic RL, where multi-turn environment interaction introduces stochastic transitions, partial observability, and horizons of 100+ turns (100K--1M tokens), making episode-level credit increasingly uninformative. We survey 47 CA methods (41 core, 6 adjacent enablers) published between 2024 and early 2026, organizing them in a two-dimensional taxonomy by assignment granularity (token, segment, step, turn, multi-agent) and methodology (Monte Carlo, temporal difference, model-based, game-theoretic, information-theoretic). Beyond the survey itself, we contribute three reusable resources: (1) a structured, machine-readable paper inventory with taxonomy labels, baseline families, and evidence levels; (2) a reporting checklist for future CA papers, validated against the reviewed literature to identify systematic methodological gaps; and (3) a benchmark protocol specification with task families, metadata requirements, and controlled bifurcation tasks, accompanied by a method selection decision tree. Our synthesis suggests that the shift from reasoning to agentic RL complicates and reshapes the credit assignment landscape: reasoning CA is maturing around process reward models and critic-free group comparison, while agentic CA is driving genuinely new approaches -- hindsight counterfactual analysis, privileged asymmetric critics, and turn-level MDP reformulations -- that have no direct precedent in reasoning RL.</p> - https://arxiv.org/abs/2604.09459 - Mon, 13 Apr 2026 00:00:00 +0000 + GlotOCR Bench: OCR Models Still Struggle Beyond a Handful of Unicode Scripts + https://arxiv.org/abs/2604.12978 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.12978.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Amir Hossein Kargaran, Nafiseh Nikeghbal, Jana Diesner, François Yvon, Hinrich Schütze</p><p><b>Upvotes:</b> 4</p><p><b>Summary:</b> Optical character recognition (OCR) has advanced rapidly with the rise of vision-language models, yet evaluation has remained concentrated on a small cluster of high- and mid-resource scripts. We introduce GlotOCR Bench, a comprehensive benchmark evaluating OCR generalization across 100+ Unicode scripts. Our benchmark comprises clean and degraded image variants rendered from real multilingual texts. Images are rendered using fonts from the Google Fonts repository, shaped with HarfBuzz and rasterized with FreeType, supporting both LTR and RTL scripts. Samples of rendered images were manually reviewed to verify correct rendering across all scripts. We evaluate a broad suite of open-weight and proprietary vision-language models and find that most perform well on fewer than ten scripts, and even the strongest frontier models fail to generalize beyond thirty scripts. Performance broadly tracks script-level pretraining coverage, suggesting that current OCR systems rely on language model pretraining as much as on visual recognition. Models confronted with unfamiliar scripts either produce random noise or hallucinate characters from similar scripts they already know. We release the benchmark and pipeline for reproducibility. Pipeline Code: https://github.com/cisnlp/glotocr-bench, Benchmark: https://hf.co/datasets/cis-lmu/glotocr-bench.</p> + https://arxiv.org/abs/2604.12978 + Tue, 14 Apr 2026 00:00:00 +0000 - Mobile GUI Agent Privacy Personalization with Trajectory Induced Preference Optimization - https://arxiv.org/abs/2604.11259 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11259.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Zhixin Lin, Jungang Li, Dongliang Xu, Shidong Pan, Yibo Shi, Yuchi Liu, Yuecong Min, Yue Yao</p><p><b>Upvotes:</b> 9</p><p><b>Summary:</b> Mobile GUI agents powered by Multimodal Large Language Models (MLLMs) can execute complex tasks on mobile devices. Despite this progress, most existing systems still optimize task success or efficiency, neglecting users' privacy personalization. In this paper, we study the often-overlooked problem of agent personalization. We observe that personalization can induce systematic structural heterogeneity in execution trajectories. For example, privacy-first users often prefer protective actions, e.g., refusing permissions, logging out, and minimizing exposure, leading to logically different execution trajectories from utility-first users. Such variable-length and structurally different trajectories make standard preference optimization unstable and less informative. To address this issue, we propose Trajectory Induced Preference Optimization (TIPO), which uses preference-intensity weighting to emphasize key privacy-related steps and padding gating to suppress alignment noise. Results on our Privacy Preference Dataset show that TIPO improves persona alignment and distinction while preserving strong task executability, achieving 65.60% SR, 46.22 Compliance, and 66.67% PD, outperforming existing optimization methods across various GUI tasks. The code and dataset will be publicly released at https://github.com/Zhixin-L/TIPO.</p> - https://arxiv.org/abs/2604.11259 - Mon, 13 Apr 2026 00:00:00 +0000 + Lightning OPD: Efficient Post-Training for Large Reasoning Models with Offline On-Policy Distillation + https://arxiv.org/abs/2604.13010 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.13010.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Yecheng Wu, Song Han, Hai Cai</p><p><b>Upvotes:</b> 4</p><p><b>Summary:</b> On-policy distillation (OPD) has emerged as an efficient post-training paradigm for large language models. However, standard OPD requires a live teacher inference server throughout training, resulting in substantial infrastructure overhead. In this work, we investigate whether on-policy distillation can be performed offline. A natural approach is to precompute teacher log-probabilities once over SFT rollouts and reuse them during training. In practice, however, this offline variant fails to reliably match the performance of standard OPD. To understand this discrepancy, we identify a previously overlooked condition that is critical for any OPD pipeline, which we term teacher consistency. This condition requires that the same teacher model be used for both supervised fine-tuning and OPD. We show that violating teacher consistency introduces an irreducible gradient bias, causing both offline and online OPD to converge to a suboptimal fixed point regardless of training duration. Building on this insight, we propose Lightning OPD, an offline on-policy distillation framework that enforces teacher consistency by precomputing teacher log-probabilities over SFT rollouts. This design eliminates the need for a live teacher server entirely. We further show that, under teacher consistency, Lightning OPD shares the same optimum as standard OPD, with bounded gradient discrepancy and an implicit regularization effect that helps prevent policy drift. Extensive experiments on mathematical reasoning and code generation demonstrate that Lightning OPD achieves state-of-the-art performance with significantly improved efficiency. Starting from an SFT-initialized Qwen3-8B-Base model, Lightning OPD reaches 69.9% on AIME 2024 in just 30 GPU hours, achieving a 4.0x speedup over standard OPD and substantially lowering the barrier to entry for academic research on LLM post-training.</p> + https://arxiv.org/abs/2604.13010 + Tue, 14 Apr 2026 00:00:00 +0000 - Agentic Aggregation for Parallel Scaling of Long-Horizon Agentic Tasks - https://arxiv.org/abs/2604.11753 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11753.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Yoonsang Lee, Howard Yen, Xi Ye, Danqi Chen</p><p><b>Upvotes:</b> 10</p><p><b>Summary:</b> We study parallel test-time scaling for long-horizon agentic tasks such as agentic search and deep research, where multiple rollouts are generated in parallel and aggregated into a final response. While such scaling has proven effective for chain-of-thought reasoning, agentic tasks pose unique challenges: trajectories are long, multi-turn, and tool-augmented, and outputs are often open-ended. Aggregating only final answers discards rich information from trajectories, while concatenating all trajectories exceeds the model's context window. To address this, we propose AggAgent, an aggregation agent that treats parallel trajectories as an environment. We equip it with lightweight tools to inspect candidate solutions and search across trajectories, enabling it to navigate and synthesize information on demand. Across six benchmarks and three model families (GLM-4.7, Qwen3.5, MiniMax-M2.5), AggAgent outperforms all existing aggregation methods-by up to 5.3% absolute on average and 10.3% on two deep research tasks-while adding minimal overhead, as the aggregation cost remains bounded by a single agentic rollout. Our findings establish agentic aggregation as an effective and cost-efficient approach to parallel test-time scaling.</p> - https://arxiv.org/abs/2604.11753 - Mon, 13 Apr 2026 00:00:00 +0000 + Masked by Consensus: Disentangling Privileged Knowledge in LLM Correctness + https://arxiv.org/abs/2604.12373 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.12373.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Tomer Ashuach, Liat Ein-Dor, Shai Gretz, Yoav Katz, Yonatan Belinkov</p><p><b>Upvotes:</b> 4</p><p><b>Summary:</b> Humans use introspection to evaluate their understanding through private internal states inaccessible to external observers. We investigate whether large language models possess similar privileged knowledge about answer correctness, information unavailable through external observation. We train correctness classifiers on question representations from both a model's own hidden states and external models, testing whether self-representations provide a performance advantage. On standard evaluation, we find no advantage: self-probes perform comparably to peer-model probes. We hypothesize this is due to high inter-model agreement of answer correctness. To isolate genuine privileged knowledge, we evaluate on disagreement subsets, where models produce conflicting predictions. Here, we discover domain-specific privileged knowledge: self-representations consistently outperform peer representations in factual knowledge tasks, but show no advantage in math reasoning. We further localize this domain asymmetry across model layers, finding that the factual advantage emerges progressively from early-to-mid layers onward, consistent with model-specific memory retrieval, while math reasoning shows no consistent advantage at any depth.</p> + https://arxiv.org/abs/2604.12373 + Tue, 14 Apr 2026 00:00:00 +0000 - Solving Physics Olympiad via Reinforcement Learning on Physics Simulators - https://arxiv.org/abs/2604.11805 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11805.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Mihir Prabhudesai, Aryan Satpathy, Yangmin Li, Zheyang Qin, Nikash Bhardwaj, Amir Zadeh, Chuan Li, Katerina Fragkiadaki, Deepak Pathak</p><p><b>Upvotes:</b> 13</p><p><b>Summary:</b> We have witnessed remarkable advances in LLM reasoning capabilities with the advent of DeepSeek-R1. However, much of this progress has been fueled by the abundance of internet question-answer (QA) pairs, a major bottleneck going forward, since such data is limited in scale and concentrated mainly in domains like mathematics. In contrast, other sciences such as physics lack large-scale QA datasets to effectively train reasoning-capable models. In this work, we show that physics simulators can serve as a powerful alternative source of supervision for training LLMs for physical reasoning. We generate random scenes in physics engines, create synthetic question-answer pairs from simulated interactions, and train LLMs using reinforcement learning on this synthetic data. Our models exhibit zero-shot sim-to-real transfer to real-world physics benchmarks: for example, training solely on synthetic simulated data improves performance on IPhO (International Physics Olympiad) problems by 5-10 percentage points across model sizes. These results demonstrate that physics simulators can act as scalable data generators, enabling LLMs to acquire deep physical reasoning skills beyond the limitations of internet-scale QA data. Code available at: https://sim2reason.github.io/.</p> - https://arxiv.org/abs/2604.11805 - Mon, 13 Apr 2026 00:00:00 +0000 + Accelerating Speculative Decoding with Block Diffusion Draft Trees + https://arxiv.org/abs/2604.12989 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.12989.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Liran Ringel, Yaniv Romano</p><p><b>Upvotes:</b> 4</p><p><b>Summary:</b> Speculative decoding accelerates autoregressive language models by using a lightweight drafter to propose multiple future tokens, which the target model then verifies in parallel. DFlash shows that a block diffusion drafter can generate an entire draft block in a single forward pass and achieve state-of-the-art speculative decoding performance, outperforming strong autoregressive drafters such as EAGLE-3. Vanilla DFlash, however, still verifies only a single drafted trajectory per round, potentially limiting its acceptance length. We introduce DDTree (Diffusion Draft Tree), a method that constructs a draft tree directly from the per-position distributions of a block diffusion drafter. Under a fixed node budget, DDTree uses a simple best-first heap algorithm to select the continuations that are most likely to match the target model according to a surrogate defined by the draft model's output. The resulting tree is verified efficiently in a single target model forward pass using an ancestor-only attention mask. Because DDTree builds on DFlash, a leading draft model for speculative decoding, these gains place DDTree among the leading approaches to speculative decoding.</p> + https://arxiv.org/abs/2604.12989 + Tue, 14 Apr 2026 00:00:00 +0000 - Introspective Diffusion Language Models - https://arxiv.org/abs/2604.11035 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11035.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Yifan Yu, Yuqing Jian, Junxiong Wang, Zhongzhu Zhou, Donglin Zhuang, Xinyu Fang, Sri Yanamandra, Xiaoxia Wu, Qingyang Wu, Shuaiwen Leon Song, Tri Dao, Ben Athiwaratkun, James Zou, Fan Lai, Chenfeng Xu</p><p><b>Upvotes:</b> 14</p><p><b>Summary:</b> Diffusion language models promise parallel generation, yet still lag behind autoregressive (AR) models in quality. We stem this gap to a failure of introspective consistency: AR models agree with their own generations, while DLMs often do not. We define the introspective acceptance rate, which measures whether a model accepts its previously generated tokens. This reveals why AR training has a structural advantage: causal masking and logit shifting implicitly enforce introspective consistency. Motivated by this observation, we introduce Introspective Diffusion Language Model (I-DLM), a paradigm that retains diffusion-style parallel decoding while inheriting the introspective consistency of AR training. I-DLM uses a novel introspective strided decoding (ISD) algorithm, which enables the model to verify previously generated tokens while advancing new ones in the same forward pass. From a systems standpoint, we build I-DLM inference engine on AR-inherited optimizations and further customize it with a stationary-batch scheduler. To the best of our knowledge, I-DLM is the first DLM to match the quality of its same-scale AR counterpart while outperforming prior DLMs in both model quality and practical serving efficiency across 15 benchmarks. It reaches 69.6 on AIME-24 and 45.7 on LiveCodeBench-v6, exceeding LLaDA-2.1-mini (16B) by more than 26 and 15 points, respectively. Beyond quality, I-DLM is designed for the growing demand of large-concurrency serving, delivering about 3x higher throughput than prior state-of-the-art DLMs.</p> - https://arxiv.org/abs/2604.11035 - Mon, 13 Apr 2026 00:00:00 +0000 + Generative Refinement Networks for Visual Synthesis + https://arxiv.org/abs/2604.13030 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.13030.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Jian Han, Jinlai Liu, Jiahuan Wang, Bingyue Peng, Zehuan Yuan</p><p><b>Upvotes:</b> 5</p><p><b>Summary:</b> While diffusion models dominate the field of visual generation, they are computationally inefficient, applying a uniform computational effort regardless of different complexity. In contrast, autoregressive (AR) models are inherently complexity-aware, as evidenced by their variable likelihoods, but are often hindered by lossy discrete tokenization and error accumulation. In this work, we introduce Generative Refinement Networks (GRN), a next-generation visual synthesis paradigm to address these issues. At its core, GRN addresses the discrete tokenization bottleneck through a theoretically near-lossless Hierarchical Binary Quantization (HBQ), achieving a reconstruction quality comparable to continuous counterparts. Built upon HBQ's latent space, GRN fundamentally upgrades AR generation with a global refinement mechanism that progressively perfects and corrects artworks -- like a human artist painting. Besides, GRN integrates an entropy-guided sampling strategy, enabling complexity-aware, adaptive-step generation without compromising visual quality. On the ImageNet benchmark, GRN establishes new records in image reconstruction (0.56 rFID) and class-conditional image generation (1.81 gFID). We also scale GRN to more challenging text-to-image and text-to-video generation, delivering superior performance on an equivalent scale. We release all models and code to foster further research on GRN.</p> + https://arxiv.org/abs/2604.13030 + Tue, 14 Apr 2026 00:00:00 +0000 - Audio Flamingo Next: Next-Generation Open Audio-Language Models for Speech, Sound, and Music - https://arxiv.org/abs/2604.10905 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10905.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Sreyan Ghosh, Arushi Goel, Kaousheik Jayakumar, Lasha Koroshinadze, Nishit Anand, Zhifeng Kong, Siddharth Gururani, Sang-gil Lee, Jaehyeon Kim, Aya Aljafari, Chao-Han Huck Yang, Sungwon Kim, Ramani Duraiswami, Dinesh Manocha, Mohammad Shoeybi, Bryan Catanzaro, Ming-Yu Liu, Wei Ping</p><p><b>Upvotes:</b> 17</p><p><b>Summary:</b> We present Audio Flamingo Next (AF-Next), the next-generation and most capable large audio-language model in the Audio Flamingo series, designed to advance understanding and reasoning over speech, environmental sounds and music. Compared to Audio Flamingo 3, AF-Next introduces: (i) a stronger foundational audio-language model that significantly improves accuracy across diverse audio understanding tasks; (ii) scalable strategies for constructing large-scale audio understanding and reasoning data beyond existing academic benchmarks; (iii) support for long and complex audio inputs up to 30 minutes; and (iv) Temporal Audio Chain-of-Thought, a new reasoning paradigm that explicitly grounds intermediate reasoning steps to timestamps in long audio, enabling fine-grained temporal alignment and improved interpretability. To enable these capabilities, we first conduct a systematic analysis of Audio Flamingo 3 to identify key gaps in audio understanding and reasoning. We then curate and scale new large-scale datasets totaling over 1 million hours to address these limitations and expand the existing AudioSkills-XL, LongAudio-XL, AF-Think and AF-Chat datasets. AF-Next is trained using a curriculum-based strategy spanning pre-training, mid-training and post-training stages. Extensive experiments across 20 audio understanding and reasoning benchmarks, including challenging long-audio tasks, show that AF-Next outperforms similarly sized open models by large margins and remains highly competitive with and sometimes surpasses, much larger open-weight and closed models. Beyond benchmark performance, AF-Next exhibits strong real-world utility and transfers well to unseen tasks, highlighting its robustness and generalization ability. In addition to all data, code and methods, we open-source 3 variants of AF-Next, including AF-Next-Instruct, AF-Next-Think and AF-Next-Captioner.</p> - https://arxiv.org/abs/2604.10905 - Mon, 13 Apr 2026 00:00:00 +0000 + Self-Adversarial One Step Generation via Condition Shifting + https://arxiv.org/abs/2604.12322 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.12322.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Deyuan Liu, Peng Sun, Yansen Han, Zhenglin Cheng, Chuyan Chen, Tao Lin</p><p><b>Upvotes:</b> 11</p><p><b>Summary:</b> The push for efficient text to image synthesis has moved the field toward one step sampling, yet existing methods still face a three way tradeoff among fidelity, inference speed, and training efficiency. Approaches that rely on external discriminators can sharpen one step performance, but they often introduce training instability, high GPU memory overhead, and slow convergence, which complicates scaling and parameter efficient tuning. In contrast, regression based distillation and consistency objectives are easier to optimize, but they typically lose fine details when constrained to a single step. We present APEX, built on a key theoretical insight: adversarial correction signals can be extracted endogenously from a flow model through condition shifting. Using a transformation creates a shifted condition branch whose velocity field serves as an independent estimator of the model's current generation distribution, yielding a gradient that is provably GAN aligned, replacing the sample dependent discriminator terms that cause gradient vanishing. This discriminator free design is architecture preserving, making APEX a plug and play framework compatible with both full parameter and LoRA based tuning. Empirically, our 0.6B model surpasses FLUX-Schnell 12B (20times more parameters) in one step quality. With LoRA tuning on Qwen-Image 20B, APEX reaches a GenEval score of 0.89 at NFE=1 in 6 hours, surpassing the original 50-step teacher (0.87) and providing a 15.33times inference speedup. Code is available https://github.com/LINs-lab/APEX.</p> + https://arxiv.org/abs/2604.12322 + Tue, 14 Apr 2026 00:00:00 +0000 - CocoaBench: Evaluating Unified Digital Agents in the Wild - https://arxiv.org/abs/2604.11201 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11201.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> CocoaBench Team, Shibo Hao, Zhining Zhang, Zhiqi Liang, Tianyang Liu, Yuheng Zha, Qiyue Gao, Jixuan Chen, Zilong Wang, Zhoujun Cheng, Haoxiang Zhang, Junli Wang, Hexi Jin, Boyuan Zheng, Kun Zhou, Yu Wang, Feng Yao, Licheng Liu, Yijiang Li, Zhifei Li, Zhengtao Han, Pracha Promthaw, Tommaso Cerruti, Xiaohan Fu, Ziqiao Ma, Jingbo Shang, Lianhui Qin, Julian McAuley, Eric P. Xing, Zhengzhong Liu, Rupesh Kumar Srivastava, Zhiting Hu</p><p><b>Upvotes:</b> 28</p><p><b>Summary:</b> LLM agents now perform strongly in software engineering, deep research, GUI automation, and various other applications, while recent agent scaffolds and models are increasingly integrating these capabilities into unified systems. Yet, most evaluations still test these capabilities in isolation, which leaves a gap for more diverse use cases that require agents to combine different capabilities. We introduce CocoaBench, a benchmark for unified digital agents built from human-designed, long-horizon tasks that require flexible composition of vision, search, and coding. Tasks are specified only by an instruction and an automatic evaluation function over the final output, enabling reliable and scalable evaluation across diverse agent infrastructures. We also present CocoaAgent, a lightweight shared scaffold for controlled comparison across model backbones. Experiments show that current agents remain far from reliable on CocoaBench, with the best evaluated system achieving only 45.1% success rate. Our analysis further points to substantial room for improvement in reasoning and planning, tool use and execution, and visual grounding.</p> - https://arxiv.org/abs/2604.11201 - Mon, 13 Apr 2026 00:00:00 +0000 + Habitat-GS: A High-Fidelity Navigation Simulator with Dynamic Gaussian Splatting + https://arxiv.org/abs/2604.12626 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.12626.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Ziyuan Xia, Jingyi Xu, Chong Cui, Yuanhong Yu, Jiazhao Zhang, Qingsong Yan, Tao Ni, Junbo Chen, Xiaowei Zhou, Hujun Bao, Ruizhen Hu, Sida Peng</p><p><b>Upvotes:</b> 12</p><p><b>Summary:</b> Training embodied AI agents depends critically on the visual fidelity of simulation environments and the ability to model dynamic humans. Current simulators rely on mesh-based rasterization with limited visual realism, and their support for dynamic human avatars, where available, is constrained to mesh representations, hindering agent generalization to human-populated real-world scenarios. We present Habitat-GS, a navigation-centric embodied AI simulator extended from Habitat-Sim that integrates 3D Gaussian Splatting scene rendering and drivable gaussian avatars while maintaining full compatibility with the Habitat ecosystem. Our system implements a 3DGS renderer for real-time photorealistic rendering and supports scalable 3DGS asset import from diverse sources. For dynamic human modeling, we introduce a gaussian avatar module that enables each avatar to simultaneously serve as a photorealistic visual entity and an effective navigation obstacle, allowing agents to learn human-aware behaviors in realistic settings. Experiments on point-goal navigation demonstrate that agents trained on 3DGS scenes achieve stronger cross-domain generalization, with mixed-domain training being the most effective strategy. Evaluations on avatar-aware navigation further confirm that gaussian avatars enable effective human-aware navigation. Finally, performance benchmarks validate the system's scalability across varying scene complexity and avatar counts.</p> + https://arxiv.org/abs/2604.12626 + Tue, 14 Apr 2026 00:00:00 +0000 - CodeTracer: Towards Traceable Agent States - https://arxiv.org/abs/2604.11641 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11641.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Han Li, Yifan Yao, Letian Zhu, Rili Feng, Hongyi Ye, Jiaming Wang, Yancheng He, Pengyu Zou, Lehan Zhang, Xinping Lei, Haoyang Huang, Ken Deng, Ming Sun, Zhaoxiang Zhang, He Ye, Jiaheng Liu</p><p><b>Upvotes:</b> 29</p><p><b>Summary:</b> Code agents are advancing rapidly, but debugging them is becoming increasingly difficult. As frameworks orchestrate parallel tool calls and multi-stage workflows over complex tasks, making the agent's state transitions and error propagation hard to observe. In these runs, an early misstep can trap the agent in unproductive loops or even cascade into fundamental errors, forming hidden error chains that make it hard to tell when the agent goes off track and why. Existing agent tracing analyses either focus on simple interaction or rely on small-scale manual inspection, which limits their scalability and usefulness for real coding workflows. We present CodeTracer, a tracing architecture that parses heterogeneous run artifacts through evolving extractors, reconstructs the full state transition history as a hierarchical trace tree with persistent memory, and performs failure onset localization to pinpoint the failure origin and its downstream chain. To enable systematic evaluation, we construct CodeTraceBench from a large collection of executed trajectories generated by four widely used code agent frameworks on diverse code tasks (e.g., bug fixing, refactoring, and terminal interaction), with supervision at both the stage and step levels for failure localization. Experiments show that CodeTracer substantially outperforms direct prompting and lightweight baselines, and that replaying its diagnostic signals consistently recovers originally failed runs under matched budgets. Our code and data are publicly available.</p> - https://arxiv.org/abs/2604.11641 - Mon, 13 Apr 2026 00:00:00 +0000 + Towards Long-horizon Agentic Multimodal Search + https://arxiv.org/abs/2604.12890 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.12890.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Yifan Du, Zikang Liu, Jinbiao Peng, Jie Wu, Junyi Li, Jinyang Li, Wayne Xin Zhao, Ji-Rong Wen</p><p><b>Upvotes:</b> 15</p><p><b>Summary:</b> Multimodal deep search agents have shown great potential in solving complex tasks by iteratively collecting textual and visual evidence. However, managing the heterogeneous information and high token costs associated with multimodal inputs over long horizons remains a critical challenge, as existing methods often suffer from context explosion or the loss of crucial visual signals. To address this, we propose a novel Long-horizon MultiModal deep search framework, named LMM-Searcher, centered on a file-based visual representation mechanism. By offloading visual assets to an external file system and mapping them to lightweight textual identifiers (UIDs), our approach mitigates context overhead while preserving multimodal information for future access. We equip the agent with a tailored fetch-image tool, enabling a progressive, on-demand visual loading strategy for active perception. Furthermore, we introduce a data synthesis pipeline designed to generate queries requiring complex cross-modal multi-hop reasoning. Using this pipeline, we distill 12K high-quality trajectories to fine-tune Qwen3-VL-Thinking-30A3B into a specialized multimodal deep search agent. Extensive experiments across four benchmarks demonstrate that our method successfully scales to 100-turn search horizons, achieving state-of-the-art performance among open-source models on challenging long-horizon benchmarks like MM-BrowseComp and MMSearch-Plus, while also exhibiting strong generalizability across different base models. Our code will be released in https://github.com/RUCAIBox/LMM-Searcher.</p> + https://arxiv.org/abs/2604.12890 + Tue, 14 Apr 2026 00:00:00 +0000 - Pseudo-Unification: Entropy Probing Reveals Divergent Information Patterns in Unified Multimodal Models - https://arxiv.org/abs/2604.10949 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.10949.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Songlin Yang, Xianghao Kong, Anyi Rao</p><p><b>Upvotes:</b> 36</p><p><b>Summary:</b> Unified multimodal models (UMMs) were designed to combine the reasoning ability of large language models (LLMs) with the generation capability of vision models. In practice, however, this synergy remains elusive: UMMs fail to transfer LLM-like reasoning to image synthesis and exhibit divergent response behaviors. We term this phenomenon pseudo-unification. Diagnosing its internal causes is important, but existing probing methods either lack model-internal insight or ignore prompt-response dependencies. To address these limitations, we propose an information-theoretic probing framework that jointly analyzes how UMMs encode inputs and generate outputs. Applied to ten representative UMMs, our framework reveals that pseudo-unification stems from a dual divergence: (i) Modality-Asymmetric Encoding, where vision and language follow different entropy trajectories, and (ii) Pattern-Split Response, where text generation exhibits high-entropy creativity while image synthesis enforces low-entropy fidelity. Only models that unify both sides (e.g., via contextual prediction) achieve more genuine unification, enabling stronger reasoning-based text-to-image generation even with fewer parameters. Our work provides the first model-internal probing of unification, demonstrating that real multimodal synergy requires consistency in information flow, not just shared parameters.</p> - https://arxiv.org/abs/2604.10949 - Mon, 13 Apr 2026 00:00:00 +0000 + Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning + https://arxiv.org/abs/2604.12374 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.12374.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> NVIDIA, Aakshita Chandiramani, Aaron Blakeman, Abdullahi Olaoye, Abhibha Gupta, Abhilash Somasamudramath, Abhinav Khattar, Adeola Adesoba, Adi Renduchintala, Adil Asif, Aditya Agrawal, Aditya Vavre, Ahmad Kiswani, Aishwarya Padmakumar, Ajay Hotchandani, Akanksha Shukla, Akhiad Bercovich, Aleksander Ficek, Aleksandr Shaposhnikov, Alex Gronskiy, Alex Kondratenko, Alex Neefus, Alex Steiner, Alex Yang, Alexander Bukharin, Alexander Young, Ali Hatamizadeh, Ali Taghibakhshi, Alina Galiautdinova, Alisa Liu, Alok Kumar, Ameya Sunil Mahabaleshwarkar, Amir Klein, Amit Zuker, Amnon Geifman, Anahita Bhiwandiwalla, Ananth Subramaniam, Andrew Tao, Anjaney Shrivastava, Anjulie Agrusa, Ankur Srivastava, Ankur Verma, Ann Guan, Anna Shors, Annamalai Chockalingam, Anubhav Mandarwal, Aparnaa Ramani, Arham Mehta, Arti Jain, Arun Venkatesan, Asha Anoosheh, Ashwath Aithal, Ashwin Poojary, Asif Ahamed, Asit Mishra, Asli Sabanci Demiroz, Asma Kuriparambil Thekkumpate, Atefeh Sohrabizadeh, Avinash Kaur, Ayush Dattagupta, Barath Subramaniam Anandan, Bardiya Sadeghi, Barnaby Simkin, Ben Lanir, Benedikt Schifferer, Benjamin Chislett, Besmira Nushi, Bilal Kartal, Bill Thiede, Bita Darvish Rouhani, Bobby Chen, Boris Ginsburg, Brandon Norick, Branislav Kisacanin, Brian Yu, Bryan Catanzaro, Buvaneswari Mani, Carlo del Mundo, Chankyu Lee, Chanran Kim, Chantal Hwang, Chao Ni, Charles Wang, Charlie Truong, Cheng-Ping Hsieh, Chenhan Yu, Chenjie Luo, Cherie Wang, Chetan Mungekar, Chintan Patel, Chris Alexiuk, Chris Holguin, Chris Wing, Christian Munley, Christopher Parisien, Chuck Desai, Chunyang Sheng, Collin Neale, Cyril Meurillon, Dakshi Kumar, Dan Gil, Dan Su, Dane Corneil, Daniel Afrimi, Daniel Burkhardt Eliuth Triana, Daniel Egert, Daniel Fatade, Daniel Lo, Daniel Rohrer, Daniel Serebrenik, Daniil Sorokin, Daria Gitman, Daria Levy, Darko Stosic, David Edelsohn, David Messina, David Mosallanezhad, David Tamok, Deena Donia, Deepak Narayanan, Devin O'Kelly, Dheeraj Peri, Dhruv Nathawani, Di Wu, Dima Rekesh, Dina Yared, Divyanshu Kakwani, Dmitry Konyagin Brandon Tuttle, Dong Ahn, Dongfu Jiang, Dorrin Poorkay, Douglas O'Flaherty, Duncan Riach, Dusan Stosic, Dustin Van Stee, Edgar Minasyan, Edward Lin, Eileen Peters Long, Elad Segal, Elena Lantz, Elena Lewis, Ellie Evans, Elliott Ning, Eric Chung, Eric Harper, Eric Pham-Hung, Eric W. Tramel, Erick Galinkin, Erik Pounds, Esti Etrog, Evan Briones, Evan Wu, Evelina Bakhturina, Evgeny Tsykunov, Ewa Dobrowolska, Farshad Saberi Movahed, Farzan Memarian, Fay Wang, Fei Jia, Felipe Soares, Felipe Vieira Frujeri, Feng Chen, Fengguang Lin, Ferenc Galko, Fortuna Zhang, Frankie Siino, Frida Hou, Gantavya Bhatt, Gargi Prasad, Geethapriya Venkataramani, Geetika Gupta, George Armstrong, Gerald Shen, Giulio Borghesi, Gordana Neskovic, Gorkem Batmaz, Grace Lam, Grace Wu, Greg Pauloski, Greyson Davis, Grigor Nalbandyan, Guoming Zhang, Guy Farber, Guyue Huang, Haifeng Qian, Haran Kumar Shiv Kumar, Harry Kim, Harsh Sharma, Hayate Iso, Hayley Ross, Herbert Hum, Herman Sahota, Hexin Wang, Himanshu Soni, Hiren Upadhyay, Huy Nguyen, Iain Cunningham, Ido Galil, Ido Shahaf, Igino Padovani, Igor Gitman, Igor Shovkun, Ikroop Dhillon, Ilya Loshchilov, Ingrid Kelly, Itamar Schen, Itay Levy, Ivan Moshkov, Izik Golan, Izzy Putterman, Jain Tu, Jan Baczek, Jan Kautz, Jane Polak Scowcroft, Janica Rosenberg, Jared Casper, Jarrod Pflum, Jason Grant, Jason Sewall, Jatin Mitra, Jeffrey Glick, Jenny Chen, Jesse Oliver, Jiacheng Xu, Jiafan Zhu, Jialin Song, Jian Zhang, Jiaqi Zeng, Jie Lou, Jill Milton, Jim Chow, Jimmy Zhang, Jinhang Choi, Jining Huang, Jocelyn Huang, Joel Caruso, Joey Conway, Joey Guman, Johan Jatko, John Kamalu, Johnny Greco, Jonathan Cohen, Jonathan Raiman, Joseph Jennings, Joyjit Daw, Juan Yu, Julio Tapia, Junkeun Yi, Jupinder Parmar, Jyothi Achar, Kari Briski, Kartik Mattoo, Katherine Cheung, Katherine Luna, Keith Wyss, Kevin Shih, Kezhi Kong, Khanh Nguyen, Khushi Bhardwaj, Kirill Buryak, Kirthi Shankar Sivamani, Konstantinos Krommydas, Kris Murphy, Krishna C. Puvvada, Krzysztof Pawelec, Kumar Anik, Laikh Tewari, Laya Sleiman, Leo Du, Leon Derczynski, Li Ding, Lilach Ilan, Lingjie Wu, Lizzie Wei, Luis Vega, Lun Su, Maarten Van Segbroeck, Maer Rodrigues de Melo, Magaret Zhang, Mahan Fathi, Makesh Narsimhan Sreedhar, Makesh Sreedhar, Makesh Tarun Chandran, Manuel Reyes Gomez, Maor Ashkenazi, Marc Cuevas, Marc Romeijn, Margaret Zhang, Mark Cai, Mark Gabel, Markus Kliegl, Martyna Patelka, Maryam Moosaei, Matthew Varacalli, Matvei Novikov, Mauricio Ferrato, Mehrzad Samadi, Melissa Corpuz, Meng Xin, Mengdi Wang, Mengru Wang, Meredith Price, Micah Schaffer, Michael Andersch, Michael Boone, Michael Evans, Michael Z Wang, Miguel Martinez, Mikail Khona, Mike Chrzanowski, Mike Hollinger, Mingyuan Ma, Minseok Lee, Mohammad Dabbah, Mohammad Shoeybi, Mostofa Patwary, Nabin Mulepati, Nader Khalil, Najeeb Nabwani, Nancy Agarwal, Nanthini Balasubramaniam, Narimane Hennouni, Narsi Kodukula, Natalie Hereth, Nathaniel Pinckney, Nave Assaf, Negar Habibi, Nestor Qin, Neta Zmora, Netanel Haber, Nick Reamaroon, Nickson Quak, Nidhi Bhatia, Nikhil Jukar, Nikki Pope, Nikolai Ludwig, Nima Tajbakhsh, Nir Ailon, Nirmal Juluru, Nirmalya De, Nowel Pitt, Oleg Rybakov, Oleksii Hrinchuk, Oleksii Kuchaiev, Olivier Delalleau, Oluwatobi Olabiyi, Omer Ullman Argov, Omri Almog, Omri Puny, Oren Tropp, Otavio Padovani, Ouye Xie, Parth Chadha, Pasha Shamis, Paul Gibbons, Pavlo Molchanov, Peter Belcak, Peter Jin, Pinky Xu, Piotr Januszewski, Pooya Jannaty, Prachi Shevate, Pradeep Thalasta, Pranav Prashant Thombre, Prasoon Varshney, Prerana Gambhir, Pritam Gundecha, Przemek Tredak, Qing Miao, Qiyu Wan, Quan Tran Minh, Rabeeh Karimi Mahabadi, Rachel Oberman, Rachit Garg, Rahul Kandu, Raina Zhong, Ran El-Yaniv, Ran Zilberstein, Rasoul Shafipour, Renee Yao, Renjie Pi, Richard Mazzarese, Richard Wang, Rick Izzo, Ridhima Singla, Rima Shahbazyan, Rishabh Garg, Ritika Borkar, Ritu Gala, Riyad Islam, Robert Clark, Robert Hesse, Roger Waleffe, Rohit Varma Kalidindi, Rohit Watve, Roi Koren, Ron Fan, Ruchika Kharwar, Ruisi Cai, Ruoxi Zhang, Russell J. Hewett, Ryan Prenger, Ryan Timbrook, Ryota Egashira, Sadegh Mahdavi, Sagar Singh Ashutosh Joshi, Sahil Modi, Samuel Kriman, Sandeep Pombra, Sanjay Kariyappa, Sanjeev Satheesh, Santiago Pombo, Saori Kaji, Satish Pasumarthi, Saurav Mishra, Saurav Muralidharan, Scott Hara, Sean Narenthiran, Sebastian Rogawski, Seonjin Na, Seonmyeong Bak, Sepehr Sameni, Seth Poulos, Shahar Mor, Shantanu Acharya, Shaona Ghosh Adam Lord, Sharath Turuvekere Sreenivas, Shaun Kotek, Shaya Gharghabi, Shelby Thomas, Sheng-Chieh Lin, Shibani Likhite, Shiqing Fan, Shiyang Chen, Shreya Gopal, Shrimai Prabhumoye, Shubham Pachori, Shubham Toshniwal, Shuo Zhang, Shuoyang Ding, Shyam Renjith, Shyamala Prayaga, Siddhartha Jain, Simeng Sun, Sirisha Rella, Sirshak Das, Smita Ithape, Sneha Harishchandra S, Somshubra Majumdar, Soumye Singhal, Sri Harsha Singudasu, Sriharsha Niverty, Stas Sergienko, Stefana Gloginic, Stefania Alborghetti, Stephen Ge, Stephen McCullough, Sugam Dipak Devare, Suguna Varshini Velury, Sukrit Rao, Sumeet Kumar Barua, Sunny Gai, Suseella Panguluri, Sushil Koundinyan, Swathi Patnam, Sweta Priyadarshi, Swetha Bhendigeri, Syeda Nahida Akter, Sylendran Arunagiri, Tailling Yuan, Talor Abramovich, Tan Bui, Tan Yu, Terry Kong, Thanh Do, Thomas Gburek, Thorgane Marques, Tiffany Moore, Tijmen Blankevoort, Tim Moon, Timothy Ma, Tiyasa Mitra, Tomasz Grzegorzek, Tomer Asida, Tomer Bar Natan, Tomer Keren, Tomer Ronen, Traian Rebedea, Trenton Starkey, Tugrul Konuk, Twinkle Vashishth, Tyler Condensa, Udi Karpas, Ushnish De, Vahid Noorozi, Vahid Noroozi, Vanshil Atul Shah, Veena Vaidyanathan, Venkat Srinivasan, Venmugil Elango, Victor Cui, Vijay Korthikanti, Vikas Mehta, Virginia Adams, Virginia Wu, Vitaly Kurin, Vitaly Lavrukhin, Vladimir Anisimov, Wan Seo, Wanli Jiang, Wasi Uddin Ahmad, Wei Du, Wei Ping, Wei-Ming Chen, Wendy Quan, Wenliang Dai, Wenwen Gao, Will Jennings, William Zhang, Xiaowei Ren, Xiaowen Xin, Xin Li, Yang Yu, Yangyi Chen, Yaniv Galron, Yashaswi Karnati, Yejin Choi, Yev Meyer, Yi-Fu Wu, Yian Zhang, Ying Lin, Yonatan Geifman, Yonggan Fu, Yoshi Suhara, Youngeun Kwon, Yuan Zhang, Yuki Huang, Zach Moshe, Zhilin Wang, Zhiyu Cheng, Zhongbo Zhu, Zhuolin Yang, Zihan Liu, Zijia Chen, Zijie Yan, Zuhair Ahmed</p><p><b>Upvotes:</b> 16</p><p><b>Summary:</b> We describe the pre-training, post-training, and quantization of Nemotron 3 Super, a 120 billion (active 12 billion) parameter hybrid Mamba-Attention Mixture-of-Experts model. Nemotron 3 Super is the first model in the Nemotron 3 family to 1) be pre-trained in NVFP4, 2) leverage LatentMoE, a new Mixture-of-Experts architecture that optimizes for both accuracy per FLOP and accuracy per parameter, and 3) include MTP layers for inference acceleration through native speculative decoding. We pre-trained Nemotron 3 Super on 25 trillion tokens followed by post-training using supervised fine tuning (SFT) and reinforcement learning (RL). The final model supports up to 1M context length and achieves comparable accuracy on common benchmarks, while also achieving up to 2.2x and 7.5x higher inference throughput compared to GPT-OSS-120B and Qwen3.5-122B, respectively. Nemotron 3 Super datasets, along with the base, post-trained, and quantized checkpoints, are open-sourced on HuggingFace.</p> + https://arxiv.org/abs/2604.12374 + Tue, 14 Apr 2026 00:00:00 +0000 - OmniShow: Unifying Multimodal Conditions for Human-Object Interaction Video Generation - https://arxiv.org/abs/2604.11804 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11804.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Donghao Zhou, Guisheng Liu, Hao Yang, Jiatong Li, Jingyu Lin, Xiaohu Huang, Yichen Liu, Xin Gao, Cunjian Chen, Shilei Wen, Chi-Wing Fu, Pheng-Ann Heng</p><p><b>Upvotes:</b> 55</p><p><b>Summary:</b> In this work, we study Human-Object Interaction Video Generation (HOIVG), which aims to synthesize high-quality human-object interaction videos conditioned on text, reference images, audio, and pose. This task holds significant practical value for automating content creation in real-world applications, such as e-commerce demonstrations, short video production, and interactive entertainment. However, existing approaches fail to accommodate all these requisite conditions. We present OmniShow, an end-to-end framework tailored for this practical yet challenging task, capable of harmonizing multimodal conditions and delivering industry-grade performance. To overcome the trade-off between controllability and quality, we introduce Unified Channel-wise Conditioning for efficient image and pose injection, and Gated Local-Context Attention to ensure precise audio-visual synchronization. To effectively address data scarcity, we develop a Decoupled-Then-Joint Training strategy that leverages a multi-stage training process with model merging to efficiently harness heterogeneous sub-task datasets. Furthermore, to fill the evaluation gap in this field, we establish HOIVG-Bench, a dedicated and comprehensive benchmark for HOIVG. Extensive experiments demonstrate that OmniShow achieves overall state-of-the-art performance across various multimodal conditioning settings, setting a solid standard for the emerging HOIVG task.</p> - https://arxiv.org/abs/2604.11804 - Mon, 13 Apr 2026 00:00:00 +0000 + Lyra 2.0: Explorable Generative 3D Worlds + https://arxiv.org/abs/2604.13036 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.13036.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Tianchang Shen, Sherwin Bahmani, Kai He, Sangeetha Grama Srinivasan, Tianshi Cao, Jiawei Ren, Ruilong Li, Zian Wang, Nicholas Sharp, Zan Gojcic, Sanja Fidler, Jiahui Huang, Huan Ling, Jun Gao, Xuanchi Ren</p><p><b>Upvotes:</b> 17</p><p><b>Summary:</b> Recent advances in video generation enable a new paradigm for 3D scene creation: generating camera-controlled videos that simulate scene walkthroughs, then lifting them to 3D via feed-forward reconstruction techniques. This generative reconstruction approach combines the visual fidelity and creative capacity of video models with 3D outputs ready for real-time rendering and simulation. Scaling to large, complex environments requires 3D-consistent video generation over long camera trajectories with large viewpoint changes and location revisits, a setting where current video models degrade quickly. Existing methods for long-horizon generation are fundamentally limited by two forms of degradation: spatial forgetting and temporal drifting. As exploration proceeds, previously observed regions fall outside the model's temporal context, forcing the model to hallucinate structures when revisited. Meanwhile, autoregressive generation accumulates small synthesis errors over time, gradually distorting scene appearance and geometry. We present Lyra 2.0, a framework for generating persistent, explorable 3D worlds at scale. To address spatial forgetting, we maintain per-frame 3D geometry and use it solely for information routing -- retrieving relevant past frames and establishing dense correspondences with the target viewpoints -- while relying on the generative prior for appearance synthesis. To address temporal drifting, we train with self-augmented histories that expose the model to its own degraded outputs, teaching it to correct drift rather than propagate it. Together, these enable substantially longer and 3D-consistent video trajectories, which we leverage to fine-tune feed-forward reconstruction models that reliably recover high-quality 3D scenes.</p> + https://arxiv.org/abs/2604.13036 + Tue, 14 Apr 2026 00:00:00 +0000 - The Past Is Not Past: Memory-Enhanced Dynamic Reward Shaping - https://arxiv.org/abs/2604.11297 - <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.11297.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Yang Liu, Enxi Wang, Yufei Gao, Weixin Zhang, Bo Wang, Zhiyuan Zeng, Yikai Zhang, Yining Zheng, Xipeng Qiu</p><p><b>Upvotes:</b> 80</p><p><b>Summary:</b> Despite the success of reinforcement learning for large language models, a common failure mode is reduced sampling diversity, where the policy repeatedly generates similar erroneous behaviors. Classical entropy regularization encourages randomness under the current policy, but does not explicitly discourage recurrent failure patterns across rollouts. We propose MEDS, a Memory-Enhanced Dynamic reward Shaping framework that incorporates historical behavioral signals into reward design. By storing and leveraging intermediate model representations, we capture features of past rollouts and use density-based clustering to identify frequently recurring error patterns. Rollouts assigned to more prevalent error clusters are penalized more heavily, encouraging broader exploration while reducing repeated mistakes. Across five datasets and three base models, MEDS consistently improves average performance over existing baselines, achieving gains of up to 4.13 pass@1 points and 4.37 pass@128 points. Additional analyses using both LLM-based annotations and quantitative diversity metrics show that MEDS increases behavioral diversity during sampling.</p> - https://arxiv.org/abs/2604.11297 - Mon, 13 Apr 2026 00:00:00 +0000 + Toward Autonomous Long-Horizon Engineering for ML Research + https://arxiv.org/abs/2604.13018 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.13018.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Guoxin Chen, Jie Chen, Lei Chen, Jiale Zhao, Fanzhe Meng, Wayne Xin Zhao, Ruihua Song, Cheng Chen, Ji-Rong Wen, Kai Jia</p><p><b>Upvotes:</b> 22</p><p><b>Summary:</b> Autonomous AI research has advanced rapidly, but long-horizon ML research engineering remains difficult: agents must sustain coherent progress across task comprehension, environment setup, implementation, experimentation, and debugging over hours or days. We introduce AiScientist, a system for autonomous long-horizon engineering for ML research built on a simple principle: strong long-horizon performance requires both structured orchestration and durable state continuity. To this end, AiScientist combines hierarchical orchestration with a permission-scoped File-as-Bus workspace: a top-level Orchestrator maintains stage-level control through concise summaries and a workspace map, while specialized agents repeatedly re-ground on durable artifacts such as analyses, plans, code, and experimental evidence rather than relying primarily on conversational handoffs, yielding thin control over thick state. Across two complementary benchmarks, AiScientist improves PaperBench score by 10.54 points on average over the best matched baseline and achieves 81.82 Any Medal% on MLE-Bench Lite. Ablation studies further show that File-as-Bus protocol is a key driver of performance, reducing PaperBench by 6.41 points and MLE-Bench Lite by 31.82 points when removed. These results suggest that long-horizon ML research engineering is a systems problem of coordinating specialized work over durable project state, rather than a purely local reasoning problem.</p> + https://arxiv.org/abs/2604.13018 + Tue, 14 Apr 2026 00:00:00 +0000 + + + Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipe + https://arxiv.org/abs/2604.13016 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.13016.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Yaxuan Li, Yuxin Zuo, Bingxiang He, Jinqian Zhang, Chaojun Xiao, Cheng Qian, Tianyu Yu, Huan-ang Gao, Wenkai Yang, Zhiyuan Liu, Ning Ding</p><p><b>Upvotes:</b> 58</p><p><b>Summary:</b> On-policy distillation (OPD) has become a core technique in the post-training of large language models, yet its training dynamics remain poorly understood. This paper provides a systematic investigation of OPD dynamics and mechanisms. We first identify that two conditions govern whether OPD succeeds or fails: (i) the student and teacher should share compatible thinking patterns; and (ii) even with consistent thinking patterns and higher scores, the teacher must offer genuinely new capabilities beyond what the student has seen during training. We validate these findings through weak-to-strong reverse distillation, showing that same-family 1.5B and 7B teachers are distributionally indistinguishable from the student's perspective. Probing into the token-level mechanism, we show that successful OPD is characterized by progressive alignment on high-probability tokens at student-visited states, a small shared token set that concentrates most of the probability mass (97%-99%). We further propose two practical strategies to recover failing OPD: off-policy cold start and teacher-aligned prompt selection. Finally, we show that OPD's apparent free lunch of dense token-level reward comes at a cost, raising the question of whether OPD can scale to long-horizon distillation.</p> + https://arxiv.org/abs/2604.13016 + Tue, 14 Apr 2026 00:00:00 +0000 + + + KnowRL: Boosting LLM Reasoning via Reinforcement Learning with Minimal-Sufficient Knowledge Guidance + https://arxiv.org/abs/2604.12627 + <p><img src="https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2604.12627.png" alt="Paper thumbnail" style="max-width: 300px; height: auto;" /></p><p><b>Authors:</b> Linhao Yu, Tianmeng Yang, Siyu Ding, Renren Jin, Naibin Gu, Xiangzhao Hao, Shuaiyi Nie, Deyi Xiong, Weichong Yin, Yu Sun, Hua Wu</p><p><b>Upvotes:</b> 82</p><p><b>Summary:</b> RLVR improves reasoning in large language models, but its effectiveness is often limited by severe reward sparsity on hard problems. Recent hint-based RL methods mitigate sparsity by injecting partial solutions or abstract templates, yet they typically scale guidance by adding more tokens, which introduce redundancy, inconsistency, and extra training overhead. We propose KnowRL (Knowledge-Guided Reinforcement Learning), an RL training framework that treats hint design as a minimal-sufficient guidance problem. During RL training, KnowRL decomposes guidance into atomic knowledge points (KPs) and uses Constrained Subset Search (CSS) to construct compact, interaction-aware subsets for training. We further identify a pruning interaction paradox -- removing one KP may help while removing multiple such KPs can hurt -- and explicitly optimize for robust subset curation under this dependency structure. We train KnowRL-Nemotron-1.5B from OpenMath-Nemotron-1.5B. Across eight reasoning benchmarks at the 1.5B scale, KnowRL-Nemotron-1.5B consistently outperforms strong RL and hinting baselines. Without KP hints at inference, KnowRL-Nemotron-1.5B reaches 70.08 average accuracy, already surpassing Nemotron-1.5B by +9.63 points; with selected KPs, performance improves to 74.16, establishing a new state of the art at this scale. The model, curated training data, and code are publicly available at https://github.com/Hasuer/KnowRL.</p> + https://arxiv.org/abs/2604.12627 + Tue, 14 Apr 2026 00:00:00 +0000