3 articles
Google DeepMind's new AI, D4RT, can reconstruct dynamic 3D scenes in four dimensions (including time) with unprecedented speed and accuracy. This breakthrough uses a single transformer model to handle motion, depth, and camera pose simultaneously, even tracking objects through occlusion.
OpenAI has unveiled GPT-5.4, a groundbreaking AI model featuring native computer use capabilities and outperforming human experts on complex tasks. The release also includes specialized financial tools and raises questions about AI's growing impact on the labor market.
NVIDIA's new Omnimatte Zero technique allows for real-time, seamless removal of objects, shadows, and reflections from videos. It leverages existing AI models and requires no additional training, marking a major leap in video editing capabilities.
