Never had an AI amazed me: Genie 3 of Google Deepmind transforms a prompt into an interactive world

Genie 2 had already marked a CAP by allowing generate short interactive sequences from a simple textual description. But the new version, Genie 3, completely changes scale. Capable of creating in a few seconds a playable, persistent and realistic 3D environment, this “World Model”as Google names it, could well become the dream tool of game creators, trainers and researchers in AI. Rendered in high definition, credible physics, dynamic events … We go from a technical demonstration to a living experience.

Deepmind Genie 3: The AI that manufactures playable worlds in real time

Genie 3 is not content to produce a nice video: it gives life to a space in which you can move freely for several minutes. The sets are no longer vanished as soon as we look away: objects and landscapes persist, ready to be found as they have been left. The definition climbs to 720p, the fluidity reaches 24 frames per second, and light as water obeyed credible physical laws.

The feat does not stop there. Deepmind introduced the concept “Pilotable global events”: Just type “the night falls” or “it starts to rain” so that the universe adapts live. This flexibility opens the door to uses well beyond entertainment: robots training, simulations for research, video games, or vocational training.

But Genie 3 is far from being limitless. Interaction lasts only a few minutes, freedom left to IA agents remains supervised, and it remains unable to faithfully reproduce a real place. Above all, the tool is currently only accessible only to a limited circle of selected researchers and creators.

These restrictions resonate with The recent warning of Demis Hassabis, CEO of Deepmindwhich evokes general AI as a shock “Six times larger than the industrial revolution, and ten times faster”. Genie 3 illustrates in a tangible way what he means: a generation of AI capable not only to reason, but also to build his own learning grounds.

What nourish a vertiginous perspective: that of a future where to create a virtual world would be as simple as typing a sentence, and where AI could train there, play … or learn, exactly like us.

Want to save even more? Discover Our promo codes Selected for you.