DeepMind reveals Genie 3 “world model” that creates real-time interactive simulations

Although nobody understood how to earn money from generative artificial intelligence, this did not prevent Google Deepmind from posting the limits of what is possible with a large heap of inference. The capacities (and costs) of these models were on an impressive ascending trajectory, a trend illustrated by the revelation of Genie 3. Only seven months after showing the “fundamental world model” of Genie 2, which was itself a significant improvement compared to its predecessor, Google Now A Genie 3.

With Genie 3, just an invite or an image to create an interactive world. Since the environment is generated continuously, it can be changed on the fly. You can add or modify objects, modify weather conditions or insert new characters – DeepMind calls these “prompts”. The ability to create modifiable 3D environments could make games more dynamic for players and offer developers new ways to prove concepts and level conceptions. However, many in the game industry have expressed their doubt that such tools would help.

It is tempting to think of Genie 3 simply as a way to create games, but Deepmind also considers this as a research tool. Games play an important role in the development of artificial intelligence because they provide difficult interactive environments with measurable progress. This is why Deepmind has already turned to games like Go And Starcraft To extend the limits of AI.

Global models bring this to the next level, generating an interactive global framework by image. This offers the possibility of refining how IA models – including so -called “embodied agents” – present themselves when they meet real situations. One of the main limits as a business that works towards the objective of general artificial intelligence (AG) is the scarcity of reliable training data. After essentially having every web and video page on the planet in AI models, researchers turn to synthetic data for many applications. Deepmind thinks that global models could be a key element of this effort, as they can be used to train AI agents with essentially unlimited interactive worlds.

Deepmind says that Genie 3 is significant progress because it offers much higher visual loyalty than Genie 2, and it is really in real time. Using the keyboard entry, it is possible to navigate in the simulated world in 720p resolution at 24 frames per second. Perhaps even even more important, Genie 3 can remember the world he creates.