Google DeepMind has introduced a new version of its AI model, Genie 3. Now this system can easily generate interactive 3D worlds where users and even other AI can move around in real time.
Unlike previous models, it no longer has a time limit for interacting with something, so now the user can get distracted and come back and everything will be in the same place.
The system remembers well where you put a particular object, even if you are not looking at it.
All this ensures that the digital world interacts with you just as naturally as the real world.
Genie 3 is one of the “world models” of AI. It was created so that AI can learn to interact with objects and the environment in general.
This approach can be useful for education, training, or, alternatively, for games and testing various AI tools. Unfortunately, not many details have been disclosed about the system.
It is only known that the AI uses a simple text query instead of 3D assets, the construction of which was entrusted to the game developers.
That is, basically, the user types, for example, “go through the forest”, and here it is, everything is right in front of you. If earlier, in Genie 2, the time limit for interacting with worlds was only a few seconds, now with Genie 3 it has been extended to a maximum of a few minutes.
Elements will stay in place for about a minute, so if you look away, then come back, everything will remain in the same place, as in the picture, in drawings on the wall.
In addition, the past generation worked at 480p resolution and 15 frames per second, but now everything is already at 720p and 24 frames per second.
Genie 3 also has other useful advantages, for example, now you can independently make elements in the scene invisible with the help of text commands.
You can also use this to adjust the weather or add new characters. The older AI models that performed this task with the user had no such feature.
Genie 3 is currently being tested with a small group of researchers and creators. According to DeepMind, this will allow them to do a better risk assessment and enhance safety prior to further rollout.
The features of the playground are being still developed like, Enhanced Interactivity options and Text in generated environments are more readable.
The potential applications of Genie 3 abound, according to DeepMind. It may come in handy in classrooms to liven history or science classes. It can enable AI in robotics to practice realistic real-world tasks in a safe simulation environment.
Heder gives the example of game devs needing an easy way to test an idea quickly or develop a new level without having to design the level first.
The one key difference between Genie 3 and an AI tool is that Genie 3 is interactive since you can have it respond in real time to inputs while keeping the state of the world same.
It remembers what happened before and maintains details of the scene That’s a huge step up from the previous video-generation tools, which would just play back little animations without any interactivity.
To DeepMind Genie 3 goes beyond a mere technical upgrade. This is a stride toward creating smarter AI capable of learning and adapting closer to acquiring the latter half of the definition of artificial general intelligence.
Genie 3 might still be an early attempt, but it provides a glimpse of how generative AI has evolved thus far and some of the ways it might influence education, other gaming aspects/changes, and much more in the future.
