On Monday, Google’s AI venture DeepMind showed off a live demo. fairyis a generative AI model that can learn game mechanics from hundreds of thousands of gameplay videos and then create playable games from simple prompts.
Genie, developed in collaboration between Google and the University of British Columbia, Generative Interactive Environments – Using a single image, you can create side-scrolling 2D platform games based on user prompts like Super Mario Brothers and Contra.
“Over the past few years, generative AI has emerged with models that can generate new, creative content through language, images, and even video,” Google DeepMind said. “Today we introduce Genie, a new paradigm for generative AI and generative conversational environments.”
Genie can create interactive, playable experiences from a single image prompt, thanks to what Google researchers described. A latent motion model that infers motion between video frames, a video tokenizer that converts raw video frames into individual tokens, and a dynamic model that determines the next frame.
“Focus on scale instead of adding inductive bias,” Google DeepMind developer Tim Rocktäschel said on Twitter. “We use a video dataset of over 200,000 hours of 2D platform games and train an 11B world model… (then) i“In an unsupervised way, Genie learns a variety of latent moves to control the character in a consistent way.”
Rocktäschel went on to say that Genie can also be used to convert other media types into games. In the included Google DeepMind research paperThe Genie can be prompted to create a virtual world that allows for a variety of motion controls through a variety of inputs.
“Our model can transform any image into a playable 2D world,” Rocktäsche said. “Genie can bring to life human-designed creations, such as sketches, such as the beautiful works of art by Seneca and Caspian, the world’s youngest creators.”
Genie is good at creating 2D worlds, but text or imageRocktäschel showed that AI models can do more than build side-scrollers. This includes the possibility of teaching other AI models or “agents” about the 3D world.
“We also train Genie on robotics data (RT-1) without motion and show that it is possible to learn a motion-controllable simulator from there as well,” he said. “We believe this is a promising step toward a general world model of AGI.”
Also known as . singularityAGI (Artificial General Intelligence) refers to AI that can understand and apply learned knowledge across a wide range of tasks, just like humans.
Google DeepMind said Genie’s dataset was created by filtering publicly available Internet videos, specifically those containing titles such as “dpeedrun” or “playthrough”, excluding words such as “movie” or “unboxing”. Yes.
Google DeepMind said advances in AI technology, hardware, and datasets have resulted in the ability to produce consistent, conversational language and “sharp, aesthetically pleasing” images.
“When choosing keywords, we manually looked through the results to ensure that they produced 2D platform gameplay videos that were not outnumbered by other types of videos that typically shared similar keywords,” the researchers continued.
“Genie allows you to train future AI agents with an endless curriculum of new, generated worlds,” Google DeepMind said. “Our paper provides a proof of concept that the potential behaviors learned by Genie can be transferred to real human-designed environments, but this only scratches the surface of what could be possible in the future.”
Thanks to the launch of OpenAI’s GPT-4 last year, tech companies including Google, Microsoft, and Amazon have been investing heavily in generative AI. Earlier this month, Google announced the launch of a subscription-based version of its Gemini AI model after rebranding from Google Bard.
Representatives from Google or the DeepMind program did not immediately respond to requests for comment. decryption.
Edited by Ryan Ozawa.
Stay up to date with cryptocurrency news and receive daily updates in your inbox.
