Google DeepMind has unveiled a new version of its Genie 3 artificial intelligence model, which can generate 3D environments that users and AI agents can interact with in real time. The developers say the model will remember where objects are, even when the user turns the camera away from them.
Google DeepMind has unveiled a new version of its Genie 3 artificial intelligence model, which can generate 3D environments that users and AI agents can interact with in real time. The developers say the model will remember where objects are, even when the user turns the camera away from them.
Genie 3 is one of a series of «world» models that can generate environments for entertainment or education. Based on a text prompt, they create a space like in a video game, where everything is generated on the fly using artificial intelligence. The Verge writes about this.
While such models have many drawbacks, in particular, the previous version of Genie 2 could only be played for a minute. In contrast, Genie 3 can generate worlds that support several minutes of continuous interaction, a significant improvement over the 10–20 seconds of interaction.
Google says the Genie 3 can hold spaces in visual memory for about a minute, so when a user looks away from something in the world and then comes back to it, things like paint on a wall or writing on a chalkboard will be in the same place. The worlds will also have a resolution of 720p and run at 24 frames per second.
Additionally, in Genie 3, users can create «world events.» This means that you can use a text prompt to change the weather conditions in the world or add new characters.
Currently, only a small group of scientists and content creators have access to Genie 3 so that AI model developers can better understand the risks and opportunities of its use.
