Google DeepMind Unveils SIMA 2 AI Agent with Advanced Capabilities
- SIMA 2, powered by Google’s Gemini model, can reason about goals and act in unfamiliar games.
- The AI agent achieved a task completion rate of 65%, significantly higher than the previous version’s rate of 31%.
- SIMA 2 adapts to new environments created by Genie 3, demonstrating improved generalization capabilities.
- DeepMind plans a limited research preview for developers and academics to explore SIMA’s potential applications.
Google DeepMind has introduced SIMA 2, an advanced AI agent designed to operate within virtual worlds using the Gemini model. This new iteration shows significant improvements in reasoning and adapting to new environments compared to its predecessor.
SIMA 2’s enhanced capabilities are evidenced by its ability to complete tasks at a much higher rate and adapt quickly to unfamiliar settings generated by Genie 3, marking a notable step towards more sophisticated AI systems. (Source)
