Google DeepMind has unveiled SIMA 2, an advanced AI agent that enhances interaction within virtual 3D environments. Leveraging Google’s Gemini model, SIMA 2 can reason about goals, explain its plans, and adapt quickly to newly generated worlds using Genie 3 technology. This marks a significant evolution from the first SIMA version, which focused on basic commands. SIMA 2 is designed to think independently, interpret high-level objectives, and collaborate effectively within games. Its performance has improved, achieving a 65% task completion rate compared to 31% by its predecessor, and it demonstrates better generalization across varied tasks, including logic prompts and visual interpretations. Though SIMA 2 faces challenges in handling complex, multi-step tasks, it represents a significant leap towards Artificial General Intelligence (AGI) and potential applications in robotics and navigation. DeepMind aims to provide a limited preview for developers and researchers to explore this groundbreaking technology.
Source link
Share
Read more