DeepMind Introduces Genie 3 for Advanced AI Training

DeepMind has announced Genie 3, a new world model that aims to advance the training of AI agents in interactive environments. This model is seen as a significant step towards achieving artificial general intelligence (AGI), as detailed in a company blog post. Genie 3 can generate dynamic 3D environments in real-time, maintaining consistency over several minutes at 720p resolution.

Genie 3 builds on its predecessors, Genie 1 and Genie 2, by offering real-time interaction and improved realism. It can simulate a wide range of environments using simple text prompts, allowing AI agents to learn and adapt through trial and error. This capability is crucial for training agents in general-purpose tasks, which is essential for reaching AGI.

The model's ability to remember past simulations and reason over long time horizons enables it to maintain physical consistency in its generated worlds. This feature allows AI agents to learn from their experiences in a manner similar to human learning. Despite its advancements, Genie 3 is still in research preview and not yet publicly available.