Firehose

This paper introduces MintAct, a unified AI model that can navigate and interact with digital environments, such as mobile apps and websites, and perform tasks like using visual tools, in a way that rivals specialized models for each environment. Practitioners might care because MintAct's approach could enable more efficient and scalable AI development for real-world applications.

This paper creates a system called CodeMidas that turns existing open-source code into environments for training coding agents using reinforcement learning. Practitioners might care because it could lead to more diverse and effective coding agents that can learn to fix issues, construct code, and verify their own work.

This paper develops a system for training and evaluating AI models that can have natural-sounding conversations with users, using both audio and video input. Practitioners might care because this research could lead to more human-like chatbots that can understand and respond to users in a more intuitive way.

This paper proposes a method to improve a pre-trained robot's performance on long-horizon tasks by focusing on specific subtasks that the robot struggles with, allowing for minimal human intervention. Practitioners might care about this approach as it can significantly increase the success rate of tasks that require precise manipulation.

This paper optimizes the performance of large language models (LLMs) on fine-grained visual perception tasks by learning to selectively focus on relevant regions of the image, rather than relying on high-resolution visual encoding. By doing so, it can improve accuracy with fewer visual tokens, making it more efficient and effective for real-world applications.

This paper explores whether supervising both agent actions and environment observations during reinforcement learning improves agent exploration and performance. Practitioners might care because it could lead to better initialization for reinforcement learning tasks.

This paper introduces ProgramDistill, a benchmark that evaluates coding agents on their ability to infer behavior from working software and implement it in an incomplete application. Practitioners in AI/ML and web development might care about this work because it provides a scalable and controlled benchmark for evaluating and training coding agents.

This paper creates a system called ScienceIDE that converts scientific code into environments that can be used to train artificial agents to perform scientific tasks. Practitioners might care because this could lead to more efficient and effective ways to develop scientific intelligence.

This paper investigates a common problem in reinforcement learning for language models called Value Flattening, where critics fail to accurately estimate state values, and proposes a new method, SP^3O, to mitigate this issue by supervising only a few well-separated states per response.

This paper proposes a new framework for Mixture-of-Agents that allows query routing and agent fine-tuning to evolve together, improving the ability of agents to adapt to changing capabilities. Practitioners might care about this approach because it can lead to more efficient and effective data-driven specialization in complex tasks.

This paper proposes a framework for training-free skill evolution in GUI agents, allowing them to adapt to dynamic graphical user interfaces without additional training. Practitioners may care about this research if they develop GUI agents that need to handle changing environments.

This paper introduces EvolveTrade, a self-evolving framework that allows large language model trading agents to refine their policies over time, enabling them to adapt to changing market regimes and improve their performance. Practitioners in finance and AI may care about this research as it provides a way to build more robust and adaptive trading agents.