Google DeepMind’s CEO, Demis Hassabis, showcased the Genie 2 AI model, a significant advancement in world-building technology that has the potential to train robots. During a segment on 60 Minutes, correspondent Scott Pelley tested Astra, DeepMind’s AI assistant, which can recognize objects and generate creative narratives based on visual inputs. The demonstration highlighted Astra’s ability to interpret artwork and create stories, illustrating the AI’s understanding of emotional contexts.
DeepMind’s progress in generative AI was further exemplified by the Veo 2 model, which can produce photorealistic videos from text prompts, showcasing remarkable advancements in visual realism. Genie 2 can transform static images into interactive 3D environments, allowing users or AI agents to explore and interact with these generated worlds.
Hassabis explained the practical applications of this technology, emphasizing its potential in entertainment, gaming, and the training of AI and robots. Simulated environments can provide extensive data for training AI, reducing the need for costly real-world data collection. Additionally, Hassabis mentioned the exploration of using Google’s geographic data to enhance AI systems’ understanding of the real world, potentially making static images interactive and immersive.
