Filtered to continual learning · clear filter
This paper proposes a method to scale automatic research agents by replacing environment execution with a world model, which can reduce training costs and improve performance. Practitioners might care about this approach because it can accelerate training times and lead to better results for complex AI tasks.
This paper introduces Atria Dawn Preview, a new type of AI model designed to work alongside humans in scientific research and engineering. Practitioners might care about this because it shows how AI can collaborate with humans more effectively, potentially leading to better research outcomes and more autonomous AI development.
This paper develops a method to create personalized AI tutors that can adapt to individual students' strengths and weaknesses, using a combination of pooled training and per-student fine-tuning. Practitioners in education and AI development may care about this work as it could lead to more effective and personalized learning experiences.
This paper introduces a method to compile neural functions from natural-language specifications, allowing for faster and more reliable execution without relying on remote models. Practitioners may care about this approach for building efficient and flexible AI systems that can perform complex tasks without the need for expensive model calls.
This paper develops a vision-language model for autonomous driving that combines 3D perception, question answering, and motion planning. A practitioner might care because it demonstrates a promising approach to integrating multiple tasks in autonomous driving.
This paper introduces MJev-Spatial, a new AI model that can quickly understand and act in the physical world by using a unified classification framework to answer spatial intelligence tasks. Practitioners may care about this model because it can solve real-world problems faster and more efficiently than existing models.
This paper addresses a problem in training spatial vision-language models (VLMs) where the model becomes specialized in answering certain types of questions, but loses its ability to perform well on others. A new method called ChronoReplay is proposed to improve the model's ability to generalize across different question types.
This paper proposes a method for recursive self-improvement in AI systems, where a model can learn from its own performance and use that knowledge to improve itself. Practitioners might care about this approach because it could lead to more efficient and effective AI systems.
This paper creates a new way to teach AI agents to use tools by generating realistic interactions that show how to use tools in a step-by-step way. Practitioners might care because it helps agents learn to use tools more effectively, especially in situations where the agent needs to remember previous actions and decisions.
This paper introduces a new type of language model called NCP-ArchPreview that can generate text by predicting both individual tokens and larger concepts, and how this approach can improve performance. Practitioners might care about this because it could lead to better language models that can handle more complex tasks.
This paper introduces ZGCM-1, a highly efficient foundation model for math and agentic search that combines internal thinking with external tool use, and shows it can perform well on various benchmarks despite its compact size. Practitioners may care about the efficiency improvements and scalable architecture of ZGCM-1.
This paper explores a way for language models to remember information from previous updates, even after many changes, and finds that combining different techniques can lead to much better retention than using any one technique alone.
This paper introduces Qwen-UI-Agent, a type of artificial intelligence system that can perform tasks on various devices, such as smartphones and computers, and improve its abilities on its own. Practitioners might care about this research because it aims to create more practical and autonomous AI systems that can be used in real-world scenarios.
This paper creates a benchmark for evaluating AI systems in real-world customer delivery scenarios, where agents must clarify ambiguous requirements, work in live systems, and uncover unstated constraints. Practitioners might care because it highlights the challenges of delivering effective AI systems in complex, real-world environments.
This paper develops a method to turn agent trajectories into reusable environments, allowing for more efficient testing and interaction with the agent. Practitioners might care because this approach can help scale agent training and improve performance in various applications.
This paper introduces a new type of AI model called memory foundation models, which allows the model to learn and retain information internally, rather than relying on external memory modules. This could be useful for practitioners who want to build more efficient and flexible AI agents.
This paper introduces OpenART, a platform for testing AI agents in open-ended environments, to evaluate their safety in complex and evolving scenarios. Practitioners can use OpenART to identify potential risks and improve the robustness of their AI systems.
This paper discovers that language models can learn from unrelated text after being trained on a specific task, and introduces a method called Active Taskless Distillation (ATD) that allows this capability transfer to happen without using target-task examples or teacher parameters. Practitioners might care because it could lead to more efficient and effective language model training.
This paper introduces GroundingPI, a new AI model that improves its ability to understand and interact with the physical world by learning to precisely identify objects and their locations, even in cluttered environments. Practitioners may care about GroundingPI's potential to enhance applications like robotics, autonomous driving, and object recognition.
This paper creates a benchmark (HarnessDev) to test whether large language models (LLMs) can design and improve their own execution infrastructure, called the agent harness, which affects their performance. Practitioners might care because it explores how models can adapt to changing environments and potentially improve efficiency.
This paper develops proactive video agents that can understand and respond to user intent in real-time, improving interaction quality and response timing. Practitioners can benefit from this research to create more intuitive and responsive video interfaces.
This paper evaluates GPT-6-Astra's robotic manipulation and navigation capabilities, demonstrating its ability to perform tasks that require physical interaction and adaptation. Practitioners in AI and robotics might care about this research as it challenges traditional approaches to building specialized models for individual tasks and could lead to more efficient and effective embodied intelligence.
This paper develops a method to efficiently fine-tune large pretrained models for brain encoding, allowing for the adaptation of models with up to 104B parameters. Practitioners might care because this approach enables better performance and interpretability for brain encoding tasks.
This paper surveys the development of video world models that capture the underlying dynamics of the world depicted in videos, and how they relate to video foundation models. Practitioners in AI and machine learning might care about this research because it aims to create more realistic and interactive video models that can be used in applications such as virtual reality, robotics, and autonomous systems.
This paper introduces ASPIRE, a benchmark for self-evolving models that can learn from vague goals, and explores how agents can interpret and operationalize these goals to improve their performance. Practitioners may care about how models can learn from unclear objectives in real-world applications.
This paper investigates whether video generation models, which are a type of world model, have a basic understanding of object permanence and can be improved through training. Practitioners interested in building human-like physical intelligence may care about this research as it could lead to more realistic and interactive AI systems.
This paper investigates how Large Language Models (LLMs) behave when "told not to think", i.e., when instructed to produce answers without using reasoning. Practitioners might care about understanding how to control this behavior, as it can impact the accuracy of LLMs in certain tasks.
This paper creates a benchmark to test how well coding agents can build agents that meet their users' needs, and it finds that most agents fail to deliver what they promise. Practitioners might care because it highlights the need for better evaluation methods to ensure agents are built to meet user requirements.
This paper proposes a framework for self-evolving agents that can adapt to changing environments by modifying their software scaffolding without changing their model weights. This approach has implications for reliability, accountability, and knowledge sharing in various domains, including scientific research and organizational workflows.
This paper teaches language agents to reflect on their actions and adapt to changing environments, allowing them to learn from feedback and correct mistakes more effectively. Practitioners can benefit from this approach in real-world applications where agents need to handle noisy feedback and user corrections.
This paper shows how to make large language models improve themselves without needing external guidance or supervision, by using their own internal consistency to correct mistakes. Practitioners might care about this because it could lead to more robust and self-sufficient AI models.
This paper introduces UniEvo-VL, a self-improvement framework for multimodal models that uses their own feedback to learn and improve, without relying on an external teacher. Practitioners might care because it can lead to better image generation capabilities and more efficient models.
This paper introduces VGI-bench, a new benchmark for evaluating video generation models' visual reasoning capabilities, and finds that current models can solve some visually grounded tasks but still struggle with reliability. Practitioners may care about developing more reliable video generation models that can perform better on this benchmark.
This paper compares the effectiveness of on-policy and off-policy learning in a distillation setting, finding that rollout policy is not as crucial as previously thought, and that learning rate and token-level KL direction play more significant roles in task performance and forgetting.
This paper identifies a problem in self-evolving search agents where the proposer and solver agree on errors, leading to improved internal reward without better external performance. Practitioners might care about this issue to develop more reliable search agents that can learn from their mistakes.
This paper develops a unified model that combines perception, memory, and response for streaming video interaction, allowing for efficient and accurate processing of video content in real-time. Practitioners in the field of natural language processing and computer vision may care about this research as it aims to improve the performance of streaming video understanding models.
This paper evaluates the ability of Deep Search agents to conduct end-to-end research in real-world open Web environments, where they must autonomously search for, access, and integrate relevant evidence. Practitioners may care about this research to develop agents that can effectively search and retrieve information in complex, real-world settings.
This paper introduces Macaron-V1, an open agent model family that enables continual learning and self-improvement in real-world environments, and explores its potential for collective intelligence. Practitioners might care about this work if they're interested in developing AI systems that can learn and adapt over time.
This paper develops a new approach to understanding videos by predicting when specific events or evidence occur within the video. Practitioners working on video analysis and AI models might care about this research because it could lead to more accurate and robust video understanding systems.
This paper investigates how to improve the effectiveness of On-Policy Distillation, a technique that trains a student model on its own generated data with teacher feedback. By prioritizing supervision reliability over alignment coverage, the researchers found that using a smaller subset of shared vocabulary at aligned positions can lead to comparable accuracy to using the full shared vocabulary.
This paper investigates how skills work in LLM agents, when they help, and why they fail, to improve their performance. Practitioners can benefit from understanding these factors to build more reliable self-evolving agents.
This paper investigates the net value of agent memory in long-horizon agents, finding that dedicated memory systems often underperform without memory in many cases. Practitioners may care because understanding the net value of memory can help design more reliable and effective memory systems.
This paper proposes a method to determine which past update evidence in a large language model is still relevant and useful after subsequent training, to prevent wasting compute and potentially degrading the model's performance. Practitioners in the field of autonomous systems and language models might care about this problem because it can lead to better model performance and efficiency in adapting to changing domains and requirements.
This paper introduces RRSI, a method to improve the performance of large language model (LLM) agents by iteratively editing their harness, which includes prompts and control flow, to achieve better performance on various tasks. Practitioners may care about this paper because it presents a way to regularize the self-improvement process of LLM agents to avoid overfitting.
This paper improves the efficiency of extracting the target speaker from audio-visual data in real-world acoustic scenes, which is crucial for applications like voice assistants. Practitioners may care about this research if they work on latency-sensitive voice recognition systems.
This paper proposes a new training method called Progress-Aligned Recurrence (PAR) to improve the performance of sequential state-tracking models on longer problems. By training the model to match the state after a certain number of operations and rescaling recurrent states, PAR enables models to generalize better to longer problems.
This paper introduces MemT-1.0, a new architecture that makes memory a part of foundation models, allowing them to accumulate and utilize information. Practitioners may care because this innovation can improve the efficiency and effectiveness of AI models in various tasks.
This paper helps vision-language models (VLMs) think and explore in 3D space by interacting with the environment through actions and observations, rather than receiving pre-rendered images. Practitioners might care because it can improve the accuracy of VLMs on 3D benchmarks.
This paper introduces SolarWM, an open framework for training video world models from diverse datasets and using different video backbones, allowing for more consistent and reproducible results. Practitioners can use SolarWM to build interactive video world models that can be applied to various real-world applications.
This paper introduces a method to generate high-quality long-horizon trajectories for terminal agents, allowing them to complete complex tasks in a realistic and autonomous way. Practitioners in AI/ML research and development can benefit from this approach to improve the performance and reliability of autonomous systems.
This paper introduces RAGU, an open-source GraphRAG engine that improves large language models with structured knowledge by separating extraction and consolidation, and trains a compact extractor that outperforms larger models on knowledge-graph construction and GraphRAG tasks. Practitioners might care because RAGU can efficiently generate more accurate and complete context for tasks like factoid-level evidence recall and multi-hop question answering.
This paper introduces FrontierChallenge, a benchmark to evaluate the completion of end-to-end scientific workflows, and investigates the performance of various models in completing these workflows. Practitioners might care about this research to improve the accuracy of scientific workflow completion evaluations.
This paper addresses a challenge in large language models that use latent visual reasoning to make decisions, where it's hard to know what visual evidence they're using to make those decisions. Practitioners caring about model interpretability and performance might find this research interesting.
This paper explores how agentic systems can improve on their own through co-evolution, where multiple agents and their environment adapt to each other, and discusses the challenges and benefits of building such systems that can learn beyond human design.
This paper creates a platform to help large language models generate code for data pipelines, which can then be edited and used to automate data processing workflows. Practitioners might care about this because it can help reduce the time and cost of developing and maintaining these pipelines.
This paper proposes a framework, LongHorizon-Harness, to help large language model agents tackle long-horizon tasks by explicitly tracking task states and verifying facts from the environment. Practitioners might care because it can improve the performance of these agents on real-world tasks.
This paper introduces TurboVLA, a new vision-language-action model that reduces computation and memory overhead by directly exchanging information between visual observations and language instructions, allowing for faster and more efficient robotic manipulation. Practitioners might care about this approach for building more efficient and effective VLA models.
This paper develops a method to generate high-quality terminal tasks for training agents, ensuring that the tasks accurately reflect the original instruction and environment, and providing a way to validate and improve the generated tasks. Practitioners in AI and robotics may care about this work because it addresses a common challenge in training terminal agents.
This paper evaluates whether a multimodal generative model can reason about the physical world by testing its ability to integrate information from different modalities, such as text, images, video, and audio. Practitioners might care about this research because it can help develop more advanced models that can better understand and generate complex scenarios.
This paper develops a system for generating speech and audio for various applications, including animation and video production, without reference recordings. Practitioners can use this system to create customized voices and control speaker styles, and to generate high-quality audio in complex scenarios.
This paper develops a benchmark for evaluating the memory capabilities of large audio language models (LALMs) in conversational systems. Practitioners can benefit from understanding how well these models retain information from past interactions and how they perform across different memory operations and acoustic evidence types.
This paper develops a new AI model called EVOKE that can generate interactive worlds without running out of memory or taking too long. Practitioners might care about EVOKE because it can support long-term generation and adaptation in complex environments.
This paper explores how AI can be applied across different stages of game development, from playing games to designing and testing them, and how to reuse capabilities across these stages. Practitioners might care about how to apply AI to improve game development efficiency and effectiveness.
This paper introduces EarlyEval, a method to reduce the cost of evaluating large language model (LLM) agents by predicting their outcomes early, allowing for earlier termination of agent runs. Practitioners may care because it can significantly reduce the computational cost of agent development.
This paper develops a new benchmark, LiveMACEBench, to evaluate the capabilities of large language model (LLM) agents in evolving financial markets. It shows that simply looking at the outcomes achieved by LLM agents can be misleading and introduces a new way to measure their capabilities.
This paper introduces LatentPress, a method to compress conversational histories and documents into a continuous memory token format that allows language models to directly read and process the context without text reconstruction. Practitioners might care about this because it could lead to faster and more efficient language model inference.
This paper develops a model that can automatically generate and adapt agent harnesses, which are crucial for the performance of AI models, to improve their ability to perform tasks. Practitioners might care about this because it could lead to more efficient and effective AI systems.
This paper introduces a new way for robots to learn from each other's experiences, without needing human corrections, to improve their performance in real-world tasks. Practitioners might care because it could make robots more reliable and efficient in complex tasks.
This paper proposes a new method called cache engineering that allows large language model agents to control their active cache without modifying the trajectory or re-prefilling retained context, which can improve task accuracy and reduce costs. Practitioners might care about this because it can help them optimize their LLM agents for complex tasks.
This paper explores how large language models can turn local observations of a city into reliable actions, and whether these models can sustain goal-directed behavior in complex urban environments. Practitioners in AI/ML and urban planning might care about the limitations and potential of current models in navigating real-world cities.
This paper presents a model called Gander that can understand and interact with users in multiple ways, like talking, typing, and watching videos, in real-time. A practitioner might care about this research because it could lead to more natural and expressive human-AI interactions.
This paper introduces StepAudio 3 Realtime, a model that enables real-time spoken interaction with deep reasoning, prompt responses, and fluid turn-taking. Practitioners might care because this technology can improve human-computer interaction, enabling more natural and efficient conversations.
This paper investigates whether the intelligence of vision-language models (VLMs) can be transferred to robotic control, enabling robots to learn and adapt in both digital and physical worlds. Practitioners might care because it could lead to more human-like robots that can seamlessly adapt to new environments and tasks.
This paper proposes a new method to process images only when necessary in diffusion vision-language models, reducing unnecessary computations and maintaining high accuracy. Practitioners caring about efficient model performance might be interested in this approach.
This paper introduces LoopArena, a benchmark for evaluating how well a model can guide a coding agent through a long-running task, and finds room for improvement in long-horizon loop control. Practitioners working on AI development might care because it could lead to more efficient and reliable development processes.
This paper presents GigaBrain-0.7, a new embodied foundation model that achieves strong generalization across diverse robot embodiments and tasks, by improving the architecture and scaling it to large amounts of data. Practitioners might care about this research if they're working on developing generalist robots that can adapt to new tasks and environments.
This paper introduces IQuest-Q1, a large and flexible AI model designed to handle complex decision-making tasks, which could be useful for developers building agentic systems that require flexible and adaptive behavior.
This paper investigates why some AI models can generate excessively long responses and proposes a solution to mitigate this issue by aligning the models' termination tokens. Practitioners might care about this problem because it can lead to wasted generation budgets and inefficient AI applications.
This paper develops a method to efficiently scale agent research loops, allowing for more effective self-improvement and reusable improvements across diverse environments. Practitioners might care about this research because it could lead to significant cost savings and improved performance in automated code completion and generation tasks.
This paper proposes a method to optimize prompts for multi-agent systems by identifying which agent's modification resolves a failure, and then using that agent's output as supervision to extract a fine-grained gradient. Practitioners might care because it could improve the performance of large language model-based multi-agent systems.
This paper proposes a new method for fine-tuning large language models (LLMs) in reinforcement learning (RL) tasks with long horizons, using evolution strategies (ES) instead of traditional backpropagation-based training. Practitioners might care because it allows for more efficient and flexible fine-tuning of LLMs, enabling them to tackle complex tasks with larger models and longer interactions.
This paper improves autoregressive video distillation methods by aligning the initialization and distribution matching stages, focusing on matching the target distribution's mode coverage rather than just visual quality. Practitioners can benefit from this approach to generate higher-quality videos with better diversity and coverage.
This paper investigates whether on-policy distillation (OPD) truly improves student policies by analyzing the effects of noisy teacher supervision. It finds that OPD works by suppressing low-probability tokens, which can be achieved without a teacher, and introduces a new method called On-Policy Self-Adaptation (OPSA) that outperforms OPD and traditional reinforcement learning methods.
This paper proposes a new framework for training autonomous agents to perform multi-turn tool-calling tasks, addressing the challenge of dealing with vast solution spaces by using a diamond topology-aware approach. Practitioners may care about this research because it can help improve the performance of autonomous agents in complex tasks.
This paper proposes a technique called Mid-Harness to improve the reliability of actions taken by terminal agents, which are models that act in environments by generating and verifying actions. By allocating test-time compute at the boundary between the model and harness, Mid-Harness can increase the success rate of actions and reduce the cost of testing.
This paper proposes a model-system framework called Long-WAM to improve real-time robot control by scaling the context of causal world-action models. A practitioner might care because it can help robots learn from their past actions and make better decisions in complex tasks.
This paper proposes a new method for agent orchestration that improves performance by allowing agents to learn from self-play and skill withdrawal, and shows it outperforms existing methods on various tasks.
This paper introduces a framework called DeepSearch-Evolve, which helps train self-improving web agents by iteratively refining their performance using their own experience. Practitioners might care because this approach can lead to more efficient and effective agents that can learn from their own mistakes.
This paper evaluates how well large language model agents can infer and apply habits from incomplete or weak evidence, and how this relates to their performance in real-world conversations. Practitioners caring about conversational AI might care about this research because it can help improve agents' ability to understand user behavior and adapt to changing habits.
This paper introduces a system called MotorMind that allows a general-purpose vision-language model to control a robot without relying on external models or tools, enabling zero-shot manipulation in new tasks and environments. Practitioners might care because it could simplify and improve robotic manipulation tasks.
This paper develops a system called SoftwareGym that creates diverse and reliable software environments for training computer-use agents. By using these environments, agents can be trained to perform well on a wide range of tasks, which can improve their performance in real-world scenarios.