Google DeepMind has introduced the latest version of its AI world model, which according to the company \u201cunlocks unprecedented realism\u201d and specializes in simulating human beings.<\/p>\n
Launched today as a limited research preview among a small group of early testers, the new model, called Genie 4, can generate photorealistic videos which users can modify in real-time via simple natural language prompts. Unlike its predecessor Genie 3<\/a>\u2014which Google first unveiled in August of last year\u2014the new model can also generate audio. And it can do so automatically, to fit whatever visual scene a user has crafted. \u201cEvery world carries the ambient sound of its own setting, from wind in the trees to waves crashing at the beach,\u201d Google wrote in a blog post<\/a> published Tuesday. Google also highlights the new model\u2019s ability to simulate real-world physics, creating scenes that respond dynamically to shifting camera angles, changes in lighting, and so on.<\/p>\n But the model\u2019s real differentiator compared with earlier versions, according to Google, is its ability to simulate humans. Bodily motion and facial expressions are now more fluid and realistic, whereas Genie 3\u2019s depictions tended to be more mechanical. \u201cThe most dramatic leap is how Genie 4 simulates people, creating characters that move naturally, respond organically to their environments, and can be interactively steered in real-time,\u201d Google wrote in its blog post. That leap extends to the simulation of human voices, too: Users can prompt AI-generated characters to converse with one another and respond to events unfolding in a video.<\/p>\n \u201cWorld models\u201d have become a major area of investment for frontier AI labs as they try to build systems that can respond to virtually any kind of input from limited training data. They should be able to learn by observing and simulate the actual mechanics of the physical world, rather than just improving their ability to coherently string words together. Some experts, most notably the French computer scientist and Turing Award-winner Yann LeCun, have suggested that world models\u2014rather than LLMs\u2014are the most likely route to the development of true artificial general intelligence. Google appears to share that view: \u201cUnderstanding and simulating the world is a crucial aspect of building AGI,\u201d the company wrote in its Tuesday blog post.<\/p>\n Google says Genie 4 is being rolled out with \u201cstrict safety guardrails\u201d akin to those that the company has put in place around its other AI tools, and that it was subject to \u201crigorous safety evaluations to prevent the generation of harmful or malicious content.\u201d The model is still in its infancy, and Google says it still occasionally struggles to maintain visual consistency and audio accuracy, particularly when simulations get longer and more complex. All videos and audio generated by Genie 4 will be tagged with SynthID, Google\u2019s invisible watermark, according to the blog post. By limiting the model\u2019s initial availability exclusively to a cohort of testing partners, Google says it\u2019s aiming to collect feedback and make any necessary safety upgrades before a more general release, starting with Google AI Ultra subscribers. (The blog post didn\u2019t specify when that broader release would happen, and Google didn\u2019t immediately respond to Gizmodo\u2019s request for comment.)<\/p>\n Among that group of early testers are video game developers, who according to Google are actively experimenting with Genie 4 to quickly build working prototypes via natural language prompts\u2014call it \u201cvibe gaming.\u201d It\u2019s not the first time one of Google\u2019s AI tools has threatened to upend the old guard of gaming: the January launch<\/a> of its Project Genie web app reportedly<\/a> caused video game stocks to plummet.<\/p>\n","protected":false},"excerpt":{"rendered":"Google DeepMind has introduced the latest version of its AI world model, which according to the company \u201cunlocks…\n","protected":false},"author":2,"featured_media":816105,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[6],"tags":[554,23804,59,1946,86,56,54,55,107597,127791],"class_list":["post-816104","post","type-post","status-publish","format-standard","has-post-thumbnail","category-technology","tag-ai","tag-artificial-general-intelligence","tag-gb","tag-google-deepmind","tag-technology","tag-uk","tag-united-kingdom","tag-unitedkingdom","tag-world-models","tag-yann-lecun"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts\/816104","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/comments?post=816104"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts\/816104\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/media\/816105"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/media?parent=816104"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/categories?post=816104"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/tags?post=816104"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}