Google DeepMind introduces SIMA 2, an agentic AI that plays 3D games and self-improves via reinforcement learning.
Google DeepMind has launched SIMA 2, an agentic AI capable of playing 3D games using Gemini for language, planning, and precise keyboard/mouse control. SIMA 2 self-improves through reinforcement learning and generalizes to unseen environments, making it 10x better on long-horizon tasks than its predecessor. Currently in a limited research preview, SIMA 2 represents a critical step toward real-world robotics, with potential integration into systems like Tesla’s Optimus. This development highlights the growing trend of AI agents that can plan, act, and adapt over extended periods, moving beyond simple chatbot interactions. The technology could revolutionize industries ranging from gaming to manufacturing, where autonomous agents can perform complex tasks with minimal human intervention.
Source: NinjaAI