US EditionGoogle Deepmind Is Using Gemini to Train Agents Inside Goat Simulator 3
SIMA 2, powered by Gemini, doubles predecessor performance by mastering complex open-ended tasks and self-improving within virtual and photorealistic worlds, DeepMind says.
- On Thursday, Google DeepMind shared a research preview of SIMA 2 that integrates Gemini, and DeepMind says SIMA 2 doubles the performance of SIMA 1.
- Trained on human gameplay from eight commercial video games including No Man's Sky and Goat Simulator 3 plus three company-created virtual worlds, SIMA 2 builds on complex multi-step tasks driving agent research.
- Users control SIMA 2 by text, voice or drawing on the screen, while Gemini enables it to reason internally and solve tasks like navigating to the red house in No Man's Sky.
- DeepMind says SIMA 2 is a step toward general-purpose robots but declined to provide a timeline, noting robotics foundation models were trained separately recently.
- Tested in environments it hadn't seen before, SIMA 2 completed complex tasks in previously unseen environments and used self-generated experiences to learn from its mistakes.
16 Articles
16 Articles
Google's SIMA 2 agent uses Gemini to reason and act in virtual world
“SIMA 2 is a more general agent. It can complete complex tasks in previously unseen environments. And it's a self improving agent... which is a step towards more general purpose robots and AGI systems more generally.”
Google DeepMind has introduced SIMA 2, the further development of their AI agent for virtual 3D environments. While the first version of SIMA (which stands for Scalable Instructable Multiworld Agent) could only follow simple instructions, the agent is now developing into an interactive game partner that thinks about its goals and continuously improves itself – at least the big promise. SIMA 2 integrates a current Gemini model (which Google does …
Massive World Model Release & AI Agent Action! Marble & Google’s SIMA 2!
Two Game Changers in Gen AI! Today I’ve got a hands-on deep dive into World Labs’ Marble (Gaussian-splat virtual sets you can edit, expand, and animate—now widely available with new features and surprising pricing) and a breakdown of Google Deepmind’s SimA-2 research—an agent learning inside generated worlds via Genie 3 and guided by Gemini. I show how I used Marble to build a virtual set for my short “Alarm,” demo multi-image world bashing, key…
Google DeepMind Unveils SIMA 2: AI Agent Capable Of Playing, Reasoning, And Learning In 3D Virtual Worlds
AI arm of the technology company Google, Google DeepMind introduced SIMA 2, the latest version of its Scalable Instructable Multiworld Agent, marking a move toward more capable and general-purpose AI agents. Built on the advanced reasoning capabilities of Gemini models, the system expands beyond basic instruction-following in virtual environments and now functions as an interactive companion that can interpret goals, converse with users, and re…
Google DeepMind Introduces SIMA 2, a Gemini-Powered AI Agent That Can Play Video Games
Google DeepMind has unveiled SIMA 2, an upgraded Gemini powered AI agent that can reason about its actions, plan tasks and interact with users through text while playing 3D open world games. The system shows better generalisation to new titles, accepts multimodal prompts and improves through self generated training data. DeepMind sees SIMA 2 as a step toward natural language controlled real world robotics.
Coverage Details
Bias Distribution
- 100% of the sources are Center
Factuality
To view factuality data please Upgrade to Premium
