Firehose

This paper introduces ICAE-Bench, a benchmark for evaluating coding agents that can build software from incomplete product intent, simulating real-world interactive project-building settings. Practitioners in AI and software development may care about this research because it aims to create more realistic and challenging tests for coding agents.

This paper introduces a new method for generating videos in multi-agent environments, where each agent has its own view of the world. It's useful for applications like video games or simulations where multiple agents need to interact with each other and the environment.

This paper introduces TableVerse, a large-scale dataset of tabletop environments that are realistic, physically consistent, and suitable for training generalizable robotic manipulation policies. Practitioners can use this dataset to improve the performance of robots in tasks like pick-and-place manipulation.

This paper introduces Tencent WorkBuddy Bench, a benchmark for coding agents that tests their performance across multiple domains, including code, web, office, and security. Practitioners might care about this benchmark because it provides a standardized way to evaluate and compare the performance of coding agents.

This paper introduces GraphVid, a model that allows for flexible and precise control over video generation through structured interaction graphs. Practitioners in the field of video generation may care about this work because it enables more intuitive and user-friendly control over generated videos.

This paper introduces AREX, a self-improving agent that recursively refines its answers by verifying intermediate results and using the partially verified state to guide subsequent refinement, aiming to improve the efficiency of deep research. Practitioners might care about AREX's approach to reducing the cost of discovery and verification in complex research tasks.

This paper introduces Visual Contrastive Self-Distillation, a method that removes the need for external teacher information and privileged answers in on-policy self-distillation, allowing for simpler and more efficient learning. Practitioners might care about this approach because it can lead to better performance in language models.

This paper develops a new architecture for implicit neural representations (INRs) that uses sinusoidal recurrence to improve the quality of the representations, and it demonstrates that this approach can lead to better image and 3D representations with fewer parameters and less computation.

This paper develops a framework to evaluate spatial cognition in image-generation models by asking them to draw or mark answers in a visual space, rather than relying on text-based inputs. Practitioners might care about this research to better understand how image-generation models think spatially and to improve their performance on tasks that require spatial reasoning.

This paper develops a method called Experience Distillation that allows agents to learn from their own interaction histories without needing additional environment interactions, making learning more sample-efficient. Practitioners might care about this because it can improve the performance of agents in complex environments with limited resources.

This paper introduces a new framework for training educational language models, specifically designed to evaluate their ability to understand curriculum knowledge and its visual presentation. Practitioners might care about this work because it aims to improve language models' performance in educational settings.

This paper learns how to separate the motion in videos into two parts: the camera's movement and the objects' movements, allowing for more accurate and meaningful analysis of the video content. Practitioners might care because this can improve the performance of computer vision models on tasks like object tracking and motion analysis.

This paper introduces a new video diffusion transformer called SANA-Video 2.0 that can generate high-quality videos efficiently, using a hybrid approach that combines linear attention with attention residuals. Practitioners might care about this paper because it presents a scalable and efficient method for generating high-resolution videos.

This paper introduces a new framework for building AI agents in Python, allowing developers to write agents that are both deterministic and model-agnostic. Practitioners might care because this framework could simplify the development of reliable AI agents for various applications.

This paper proposes a new approach for embodied visual tracking that first identifies a target described in natural language and then tracks it using a single camera. Practitioners in robotics and autonomous systems might care about this work because it shows promise for robust and reliable visual tracking in real-world applications.

This paper introduces Robostral Navigate, a vision-language model that enables robots to navigate using only a single monocular RGB camera, making it more scalable and cost-effective for deployment across various robotic platforms. Practitioners might care about this because it can simplify navigation tasks for robots in real-world environments.

This paper investigates how Large Language Models (LLMs) handle dynamic user intent in conversations, where users' goals change over time, and finds that current LLMs struggle with this capability, leading to significant performance drops when evaluated in a more realistic setting.

This paper investigates whether large language models, like Google's Gemma-4-E4B-it, represent scientific concepts and governing physics, and whether this representation affects their answers. Practitioners caring about the accuracy and reliability of language models in scientific domains might find this research valuable.

This paper introduces ATSplat, a new method for 3D Gaussian Splatting that can efficiently synthesize novel views of scenes from a few images. ATSplat is faster and more efficient than previous methods, allowing for high-quality novel views to be rendered quickly.

This paper develops a new framework for robotic grasping in complex scenes that uses language to specify task requirements, allowing robots to grasp objects with more precision and flexibility. Practitioners working on robotic grasping and manipulation might care about this approach because it could improve the performance and efficiency of robots in real-world applications.

This paper introduces a new framework for generating synchronized multi-view RGB-T data for aerial object detection, which can help practitioners improve their object detection models by training on more realistic data and by reducing the need for expensive real-world data collection.

This paper introduces a new method, Surrogate Latent Policy Optimization (SLPO), to improve the performance of autoregressive latent reasoners in reinforcement learning by combining outcome-reward RL with latent reasoning. Practitioners might care because SLPO can help scale the performance of these models at test time, which is currently a major challenge.

This paper optimizes the training of massive neural networks on a special-purpose hardware, the Ascend SuperPOD, to improve performance and stability. Practitioners in AI/ML model training might care about the techniques and results presented here for large-scale model training on non-GPU hardware.

This paper develops a new framework for evaluating and selecting document sets for AI agents, considering the interactions between documents, and proposes a training-free method that achieves the best downstream generation performance with fewer documents and search rounds.

This paper proposes a new training method for autoregressive video diffusion models called Self Gradient Forcing, which helps them better remember and use past information to generate future frames. Practitioners might care about this because it could lead to more realistic and stable video extrapolation.

This paper introduces a new environment called Trace, which allows vision-language models to reason across multiple domains and tasks, using a taxonomy-guided approach. Practitioners might care because this work could lead to more generalizable and transferable AI models.

This paper proposes a method to improve the explainability of natural-language autoencoders by making it harder for models to manipulate the explanations, and shows that this approach can increase the reliability of activation explanations and improve AI safety.

This paper introduces a benchmark framework called DocOps to evaluate autonomous agents' ability to manipulate digital documents. Practitioners may care about this work because it identifies limitations in current agents and provides insights for designing more robust agents.

This paper introduces a new dataset and tool to study contextual entrainment in vision-language models, which is the tendency for models to respond to irrelevant or false context in their inputs. Practitioners in AI and ML might care about this because it can affect the accuracy and reliability of vision-language models in real-world applications.

This paper develops a framework called NexForge that helps train more capable artificial agents by automatically generating a large number of tasks and training data, without requiring a lot of manual setup. Practitioners might care because it can improve the performance of their own agent models.

This paper develops a new optimization framework called Isospectral Optimization (ISO) that improves the performance of language models trained using reinforcement learning with verifiable rewards (RLVR) by reusing the base model's weight spectra while adapting the input and output singular frames. Practitioners might care about this because it could lead to faster and more accurate training of large language models.

This paper presents a video world model that can interact with a virtual environment in real-time, using a single desktop GPU. Practitioners might care about the potential applications of this technology in areas like virtual reality, gaming, and robotics.

This paper introduces a faster version of a generative world renderer that can run in real-time, allowing for interactive play and user control. A practitioner might care about this if they want to create immersive games or simulations.

This paper creates a benchmark for testing the ability of AI agents to understand and analyze complex financial documents, and uses it to evaluate the performance of different agents in this task. Practitioners in finance and AI research can care about this work because it aims to improve the accuracy and reliability of financial document analysis.

This paper introduces AutoIndex, a framework that learns to transform raw documents into representations for retrieval systems, allowing for more flexible and effective indexing. Practitioners may care about AutoIndex because it can improve the quality of search results in complex information retrieval tasks.

This paper investigates using hypernetworks for large-scale knowledge injection into language models, a technique that can improve their ability to answer factual questions. Practitioners may care because it could lead to more accurate and scalable language models for applications like customer service or question-answering systems.

This paper investigates how the quality of training data affects the performance of text-to-video models, and it provides a controlled testbed to study this relationship. Practitioners may care about the implications of data quality on model performance and training efficiency, especially when developing large-scale text-to-video models.

This paper introduces H^2SD, a hybrid hindsight self-distillation framework for reinforcement learning with verifiable rewards, which combines the strengths of different methods to improve large language models' reasoning capabilities. Practitioners may care about this work because it addresses limitations of existing methods and shows promising results on challenging reasoning benchmarks.

This paper introduces a way to control the output of diffusion transformers, a type of AI model used for image generation, by providing it with guidance on specific regions of the image. This can be useful for creative professionals who need to generate images with precise control over details like materials, objects, and spatial arrangements.

This paper develops a new way to control video models, allowing them to understand how objects move in the world and how robots interact with them, which can be useful for robots that need to perform tasks in real-world environments. Practitioners in robotics and AI may care about this work because it could improve the ability of robots to understand and interact with their surroundings.

This paper introduces Mage-Flow, a compact model for generating and editing high-resolution images, which can be trained efficiently and deployed on a single GPU. Practitioners might care about the potential applications of this model in interactive image editing and generation tasks.

This paper develops a new method to improve the stability of asynchronous reinforcement learning by adapting the trust region to account for staleness, which is a common problem in this field. Practitioners might care about this because stable reinforcement learning can lead to better performance and more efficient training.

This paper introduces a new framework to evaluate the factuality and completeness of long-form generation models, which is essential for ensuring that generated text is accurate and informative. Practitioners in natural language processing and artificial intelligence can benefit from this framework to assess the quality of their models.

This paper introduces HPD-Parsing, a new approach to document parsing that uses hierarchical parallel decoding to improve efficiency and throughput. Practitioners in natural language processing and computer vision might care because it could lead to faster and more accurate document parsing models.

This paper helps us understand how text-to-image diffusion transformers work by analyzing the role of "template tokens" in generating images from text prompts. Practitioners might care because it shows how to improve the efficiency of these models without sacrificing their performance.

This paper introduces AgentDebugX, an open-source toolkit that helps debug and recover from failures in large language model (LLM) agents, making it easier to identify and fix errors. Practitioners can benefit from AgentDebugX as it provides a framework for detecting, attributing, and recovering from errors, which can improve the accuracy and reliability of LLM agents.

This paper investigates where the optimizer state should be allocated in the context of mixture-of-experts training to reduce memory usage without sacrificing accuracy. Practitioners might care because optimizing memory usage can be crucial for large-scale language models.

This paper develops a new AI model, Delineate Anything v2, designed to accurately map large agricultural fields from satellite images. A practitioner might care about this because it can help ensure global food security and supply chain transparency by providing precise boundaries for fields.

This paper explores how to make AI systems understand and generate humor, particularly in visual formats like memes and comics, and why this is important for developing more human-like AI that can understand and create humor. Practitioners might care because humor is a key aspect of human communication and understanding.

This paper improves the accuracy of Automatic Speech Recognition (ASR) models by controlling the style of the transcription, allowing for more reliable word-level timing and disfluency detection. Practitioners can benefit from this work to improve the quality of ASR systems in real-world applications.

This paper develops a system called AlayaWorld that can generate interactive virtual worlds from text, images, or videos, allowing for customizable and evolving environments. Practitioners in areas like game development, virtual reality, or interactive storytelling might care about this research for its potential to streamline the creation of immersive experiences.

This paper evaluates how well six SLAM systems perform in low-light environments, finding that only one can track in all cases and that RGB-only SLAM can maintain stable tracking with both inertial fusion and global optimization, but still has limitations that require further improvement. Practitioners working on autonomous robots may care about this research as it helps them understand the challenges of SLAM in low-light conditions and how to address them.

This paper explores a new type of neural network architecture, Differentiable Logic Gate Networks (Diff-Logic), that can be used to classify EEG signals in real-time on edge devices with low latency. Practitioners may care about this approach because it could be used to develop more portable and efficient brain-computer interfaces.

This paper proposes a new framework for video spatial reasoning that focuses on learning geometric consistency to improve the accuracy and stability of models in tasks like navigation and question answering. Practitioners working on multimodal models and spatial reasoning tasks may benefit from this approach.

This paper develops a new framework, SciForma, to generate scientific diagrams that accurately represent research logic, which is crucial for scientific communication and methodology validation. Practitioners can benefit from SciForma's ability to ensure structural fidelity in diagram generation, which can improve the accuracy and reliability of scientific research.

This paper introduces EduPanel, a machine learning model that evaluates the quality of teaching videos in a more nuanced way than existing methods, and shows that it can provide reliable and interpretable assessments that complement human expertise.

This paper investigates how Diffusion Language Models (DLMs) internally represent time and how this representation can be used to modulate the model's behavior. Practitioners might care because understanding how DLMs process time could lead to more controllable and interpretable models.

This paper proposes a new way to improve the performance of diffusion models by aligning their inference process with the forward statistical structure of the model. This can lead to more realistic and diverse generated images, which is important for applications like image generation.

This paper proposes a new way for large language models to learn from feedback, allowing them to retain more detailed information about the quality of their responses and learn from it in a more nuanced way. Practitioners might care because this approach could lead to better performance on tasks where the model doesn't have a clear way to evaluate its own output.

This paper introduces RynnBrain 1.1, a family of large-scale embodied foundation models that can perform tasks like spatial reasoning, localization, and planning, and shows promising results in real-world robot experiments. Practitioners may care about the potential of these models for robot manipulation and control.