Google DeepMind recently unveiled SIMA 2, a next-gen gaming AI agent capable of playing various games by interpreting screen data and controlling virtual interfaces. Leveraging the Gemini AI model, SIMA 2 offers advanced reasoning, actionable natural language insights, and self-optimization features, inching closer to artificial general intelligence (AGI).
This evolution from the original SIMA showcases a shift from basic command execution to complex reasoning and strategic navigation. SIMA 2’s hybrid training combines human gameplay demonstrations with AI-generated data, enhancing its ability to understand nuanced player instructions. A standout feature is its transfer of learned concepts across different gaming contexts, demonstrating genuine understanding rather than mere memorization.
With the integration of Genie 3, SIMA 2 can navigate procedurally generated environments, showcasing real-time adaptability. The AI also learns through independent gameplay, reflecting significant self-improvement capabilities. This research has implications beyond gaming, potentially advancing robotics and autonomous systems. SIMA 2 is currently available for limited research use, with broader deployment plans underway.
Source link
