Unlike the previous version, SIMA 2 uses the AI model Gemini for the first time. Thanks to this, it not only follows instructions but also understands the user's overall goal, analyzes the situation, adapts to new gaming conditions, and achieves results in unfamiliar environments.
DeepMind emphasizes that the main goal is to prepare the agent for work in the real world. The gaming environment serves merely as a simulation for developing skills that can later be applied in robotic systems and practical tasks.
Recall that Google introduced the new Gemini model, which solves tasks more efficiently than Elon Musk's xAI. In 2024, the new AI from Google DeepMind was trained to play new 3D games alongside gamers.