Latest news with #GoogleGoogleDeepMind

Time of India

Google rolls out new version of AI model that can create video game worlds with real-time interaction and improved memory

Image credit: Google Google DeepMind has started releasing a new version of its AI "world" model, Genie 3. This AI model can generate 3D environments that both users and other AI agents can interact with in real time. The company has promised that users will be able to engage with these worlds for a significantly longer duration than with previous models. Genie 3 is also designed to "remember" the location of objects even when they are not in the user's line of sight. This update marks a step forward in creating more interactive and persistent virtual environments. What are AI world models and what's new in Genie 3 World models are a type of AI system that simulates digital environments for apps needed for education, entertainment or training AI agents. These systems generate spaces based on prompts, allowing users to move around in them like in a video game. Unlike traditional games that rely on handcrafted 3D assets, these environments are built using AI. Google has been exploring this area, with its Genie 2 model introduced last year. This AI model is capable of creating interactive worlds from an image. The company has also formed a dedicated world models team, led by Tim Brooks, a former co-lead of OpenAI's Sora video tool. However, the current capabilities of such models remain limited. For example, Genie 2 only allowed for about a minute of interaction. Another interactive video tool, developed by a startup backed by Pixar's cofounder, offered an experience that resembled a blurred version of Google Street View, with visuals that shifted unexpectedly. Genie 3 offers a few technical improvements. According to a blog post, users will be able to interact with generated environments for a few minutes, up from Genie 2's 10–20 seconds. The model can remember visual elements for about a minute, so if a user looks away and then returns to a scene, features like wall markings or writing on a chalkboard are likely to remain in place. These environments will render at 720p resolution and 24 frames per second. DeepMind is also incorporating " promptable world events " into Genie 3, which will allow users to adjust elements such as weather or add characters through written prompts. Access to Genie 3 will be restricted at launch. Google is releasing it as a limited research preview for a small group of academics and creators to study risks and refine safety measures. Current limitations include narrow interaction options and difficulty generating legible text unless it is included in the initial prompt. Google has indicated it may expand access in the coming days. Apple Confirms: Majority of iPhones Sold in US Are Now Made in India AI Masterclass for Students. Upskill Young Ones Today!– Join Now