Google Unveils Genie 3: Sora's Rival, Redefining AI World Models
Genie 3: Google's breakthrough AI world model creates interactive, realistic environments in real-time. A leap toward AGI.
"AI Disruption" Publication 7400 Subscriptions 20% Discount Offer Link.
Genie 3 is one of the most advanced world models ever created.
Through text alone, it can generate fully interactive, highly consistent worlds in real-time.
It represents not only the culmination of DeepMind's accumulated expertise but also a crucial step toward AGI and embodied intelligent agents.
But how was Genie 3 built? What will future world models look like?
Just recently, Google DeepMind research scientist Jack Parker-Holder and research director Shlomi Fruchter shared their insights in an interview with a16z.
This conversation provided first-hand insights into Genie 3.
Host Justine Moore tweeted: "Genie 3 has caused a sensation online."
He summarized the key points from their in-depth discussion:
Genie 3 is the result of collaboration between two DeepMind projects (Veo 2 and Genie 2).
Real-time, interactive world models have many potential applications.
But applications are not the main driver of research—they emerge naturally from users working with the models.
Genie 3 can retain spatial memory for up to one minute.
Physical laws are a "natural byproduct" of the model and continue to improve with the scale and depth of training data.
There is currently no "ultimate model" that possesses all the capabilities of both Veo 3 and Genie 3.