Edge Ai

Google and NVIDIA have optimized the Gemma 4 family of AI models for efficient local execution across devices, including NVIDIA RTX GPUs and DGX Spark supercomputers. These models support various tasks such as reasoning, coding, and multimodal interactions, enabling applications like OpenClaw to create capable local agents that automate workflows using personal context and files.

Ubotica's research focuses on enabling satellites to process data in real-time using AI. By employing a teacher-student model approach, they reduce model size to ~3.73M parameters and eliminate the need for costly human labels. Their tests on flood detection and crop classification show promising results, achieving 95–98% of baseline performance while generalizing effectively to new regions.