Ep 210: Google just demoed an AI that builds a whole interactive 3D world with a flying mouse at a…

From Models & Agents for Beginners — Video Edition · 11:49

Show by Patrick

Google just demoed an AI that builds a whole interactive 3D world with a flying mouse at a Victorian tea party—in real time. Today we're diving into Genie 4, Google's new world simulator that lets you steer scenes with plain words while it adds sound and movement on the fly. We'll also unpack why language models sometimes invent details when the facts run out, see how anyone can use ChatGPT or Claude to brainstorm thoughtful gifts, and check a few quick updates that show where AI is heading next. Everything is explained so you can follow along even if you've never used these tools before. The Genie 4 demo shows a mouse in Victorian garb at high tea, a porcupine walking by, the mouse growing wings, and then flying to the teacup, all updating instantly as you walk around the scene. This kind of real-time control used to require teams of game developers and artists working for months. The Big Story Google DeepMind just showed off Genie 4, a new kind of AI that creates living, interactive 3D worlds you can walk around inside. You type a simple prompt like “a mouse at a Victorian high tea” and the scene appears right away, complete with moving characters, changing weather, and even sound. When you add new instructions such as “the mouse grows wings and flies to the teacup,” the world updates instantly while you keep exploring. The system uses another AI called Gemini as its “brain” to plan what should happen next, keep the story consistent, and generate natural dialogue and background audio on the spot. Think of it like a video game level that builds itself while you play—no designers or programmers needed in the moment. Behind the scenes, Gemini translates your requests into actions, plans complex events, watches the scene unfold, and adjusts as you change the story. Genie 4 also introduces ambient audio and real-time interactive human dialogue that adapts on the fly. Early testers are exploring it for filmmaking, game design, education, and robotic simulation, with examples including VFX pre-visualization and bringing historical moments to life using verified archives. Testers noted improved photorealism, dynamic text events, and real-time audio as the biggest leaps forward. This matters because it moves AI from answering questions on a screen to letting you step inside made-up places that feel alive. For anyone who loves games, storytelling, or just messing around with creative ideas, it opens the door to experiences that used to require whole teams of artists and coders. Students could one day build history lessons you can walk through, or friends could co-create short animated stories together in minutes. The model includes strict safety guardrails, and all generated audio and video is watermarked with SynthID so viewers know the content was created by AI. Right now the model is still in early testing and will roll out first to Google AI Ultra subscribers and some partners through Project Genie and the Gemini API, so you can’t try the full version yet. But the short clips already online show how far real-time world building has come in just a couple of years, and that progress usually reaches free or low-cost tools surprisingly fast. The demo also shows a porcupine walking by and the mouse flying to the teacup, all happening in real time with free camera movement. This combination of natural language control, consistent physics, and native audio marks a clear step beyond earlier world models that lacked sound or struggled with longer sessions. Source: x.com Explain Like I'm 14 You know how when you’re texting a friend and the conversation has a gap—maybe you forgot a detail from last week—your brain sometimes fills it in with the most likely guess so the chat keeps flowing? Now picture an AI that does the same thing, except it has no memory of which parts it actually read and which parts it just made up to sound complete. Classical scholars have spent centuries working with ancient writings that survive only as scattered quotes or single sentences. Of everything Thales supposedly wrote, not one authenticated word survives. Anaximander wrote the first prose treatise in the Greek tradition, but today it exists as roughly one sentence preserved because Aristotle quoted it and Simplicius copied that quotation almost a thousand years later. Heraclitus comes down in scattered aphorisms, most embedded in sermons of church fathers who quoted him to refute him. They developed a simple rule: when they guess at a missing word, they put the guess in brackets or a different font so everyone can see exactly where the real evidence ends and the scholar’s best guess begins. The standard edition of the Presocratics, called Diels-Kranz, physically separates the fragments—what the philosopher actually wrote—from the testimonia, which are later authors’ comments about him. Aristotle made the distinction theoretical by separating necessary evidence that compels a conclusion from a probable sign that only supports a claim but…

More episodes of Models & Agents for Beginners — Video Edition

More episodes

PodSearch is a free, independent podcast search engine. Read about PodSearch, the privacy policy or contact us.