AI

Decart unveils MirageLSD, the world's first real-time, unlimited diffusion video model capable of transforming any video stream into immersive worlds within 40ms, supported by Karpathy.

The GThinker model from CASIA introduces a novel cue-guided rethinking approach, enabling multimodal models to revisit and improve visual understanding, surpassing previous models like O4-mini.

Apple releases its 2025 foundational AI model report, highlighting multi-language, multi-modal models, and new architectures, with leadership transition from Ruoming Pang to Zhifeng Chen.

PresentAgent, developed by a collaborative team, can transform long documents into synchronized presentation videos with voice explanations, showcasing advanced multimodal AI capabilities.

Chemia, an open-source framework from Shanghai AI Lab, simplifies chemical AI model training with a 'configuration as code' approach, accelerating drug discovery and material design.

LeadDisFlow, developed by a Chinese team, offers a rapid, precise, and interpretable workflow for drug discovery, successfully advancing candidate drugs into Phase II trials.

Discover how Oasis enables multimodal instruction data synthesis using just images, eliminating manual prompts and boosting data quality for AI training at ICCV2025.

Zebra Smart Cabin scores top in IDC's 2025 assessment, leading industry with L3 capabilities, innovative AI models, and real-world applications, setting new standards in intelligent vehicle cabins.

Exploring the deep connection between AI and mathematics, this forum highlights breakthroughs, challenges, and future paths in AI×Math, promising limitless innovation.

Le Chat has launched major new features, aiming to fully rival ChatGPT with advanced research modes, voice interaction, multi-language reasoning, and integrated image editing, showcasing Europe's AI ambitions.