Google has opened a new Project Genie experience to qualifying Google AI Ultra subscribers, granting first access to the Genie 4 world model. Genie 4 lets users edit an existing world mid-session via natural-language prompts (switching weather, adding characters) and debuts native real-time audio that adapts to the scene; Genie 3 also gains up to 3-minute sessions, 360-degree panorama grounding and world chaining. All outputs carry SynthID watermarks, and the models remain frontier research.
World Model
世界模型
A modeling paradigm that learns environment dynamics and predicts future states, widely used in embodied AI and robot planning.
Curated Intelligence
Google is rolling out Genie 4 to eligible Google AI Ultra subscribers, enabling real-time natural-language environment editing, native audio, and sharing creations in a new community gallery. The updated Genie 3 model adds longer sessions, panorama grounding, and linked world sequences, while the new Project Genie agent builds custom worlds through conversational interaction. Google AI Pro members can only join the waitlist for now.
Google is rolling out Genie 4 to eligible Google AI Ultra subscribers, enabling real-time natural-language editing of interactive 3D environments, native audio, and a community gallery for sharing creations. The accompanying Project Genie agent builds custom worlds through conversational prompts, while Genie 3 gains longer sessions, panorama grounding, and linked world sequences.
AMD completed its acquisition of World Labs, the spatial intelligence startup co-founded by Fei-Fei Li, on November 13, issuing about 14.98 million shares in an all-stock deal valued at roughly $8.2 billion and closing in just 46 days, making it AMD's second-largest acquisition ever. Fei-Fei Li becomes Executive Vice President and Chief Scientist reporting directly to Lisa Su, with the World Labs team folded into AMD to bolster its 3D world model and physical AI capabilities.
Tongxin Intelligence, incubated by the Beijing Institute for General Artificial Intelligence and initiated by Zhu Songchun, has closed seed and angel rounds totaling hundreds of millions of RMB within six months of founding, with proceeds going to pre-tapeout architecture design. The company is building a novel inference chip for AGI algorithms that stays compatible with mainstream models while adding dedicated hardware support for logical, causal and value reasoning, aiming to define next-generation compute before the algorithms settle.
At CoRL 2026, Li Fei-Fei and AMD CEO Lisa Su appeared together publicly for the first time since AMD agreed to acquire World Labs for about $8.2 billion in an all-stock deal. The conversation covered AMD's robotics strategy, the limits of end-to-end foundation models and academia's compute shortage; Li will become AMD EVP and chief scientist reporting to Su, with the deal expected to close by year-end.
The author ran 600 experiments to systematically probe planning in JEPA world models, showing that changing only the planning procedure lets a model solve long-horizon physics tasks with zero additional training. The key finding is that transforming a memory graph to generate subgoals on the fly enables generalization to out-of-distribution goals, though the model still cannot choose its own subgoals without demonstrations.
Huawei Cloud has scheduled an online launch on November 16 for CloudRobo-WM1.2, the first Ascend-native world model under its Industry AI Dream Factory platform. The model targets embodied intelligence scenarios, consistent with Huawei Cloud's strategy of building platforms and models rather than robot hardware. Only the name and date have been disclosed so far, with no technical details yet.
Google researchers introduce SensorWM, a machine learning framework that converts passive smartwatch biometrics into a modifiable coachable health age and uses an action-conditioned world model to simulate personalized 12-week what-if habit trajectories. The framework extends to routine blood biomarkers for early cardiovascular-kidney-metabolic (CKM) risk tracking, grounding personalized coaching in validated clinical prognostics.
Daxiao Robotics unveiled a dual-brain embodied architecture in which a world model handles physical environment modeling and Astra executes long-horizon tasks, combined with embodied RSI for continuous iteration. The system completes an end-to-end hotel laundry loop covering pickup, transport, machine operation and delivery, targeting cold-start and deformable-object manipulation challenges in a high-frequency, quantifiable B2B scenario.