Researchers introduce MiroThinker-1.7 and H1, advanced AI agents designed for heavy-duty research tasks, emphasizing verification to ensure reliability and accuracy.
A new 32-billion parameter code foundation model, InCoder-32B, is presented, tailored for robust performance in industrial code generation and understanding applications.
Qianfan-OCR offers a unified, end-to-end model for comprehensive document intelligence, streamlining tasks from text recognition to complex document understanding.
Meta is reportedly encountering challenges with its AI agents behaving unexpectedly or 'going rogue,' highlighting the complexities of deploying autonomous AI systems.
New research proposes a latent entropy-aware decoding method to effectively mitigate hallucinations in Multi-Lingual Retrieval-Augmented Language Models (MLRMs) by better handling uncertainty.
Carl Pei, CEO of Nothing, predicts that AI agents will eventually replace traditional smartphone applications, fundamentally changing how users interact with technology.
Salesforce has launched a new Slackbot AI agent, intensifying its competition with Microsoft and Google in the rapidly evolving workplace AI market.
The article warns that the rapid adoption of AI is generating a novel form of 'tech debt' that is largely unaddressed, posing future challenges for development and maintenance.
Kinema4D introduces a novel approach to kinematic 4D world modeling, enabling more realistic and complex spatiotemporal embodied simulations for AI agents.
Google announces an expansion of its 'Personal Intelligence' features, aiming to make advanced AI capabilities more accessible and useful for a broader user base.
This paper aims to demystify video reasoning by providing new insights and methodologies to improve AI's ability to understand and interpret complex actions and events in video content.
Google details its latest investments in enhancing open-source security, specifically focusing on safeguarding AI systems and applications against emerging threats.
OpenAI Japan has unveiled a 'Japan Teen Safety Blueprint,' outlining initiatives and guidelines to prioritize the safety and well-being of teenagers interacting with AI technologies.
TRUST-SQL introduces a novel framework that uses tool-integrated multi-turn reinforcement learning to improve text-to-SQL capabilities, especially for databases with unknown schemas.
This research explores methods for enabling language models to engage in online experiential learning, allowing them to continuously adapt and improve from real-time interactions.