AI / ML

Amazon OpenSearch Service now supports OpenSearch 3.7, introducing performance gains in vector search and relevance. Key updates include 1-bit scalar quantization for Faiss and Lucene engines to reduce storage and memory usage without accuracy loss. The release also brings faster vector retrieval via doc values and enhanced Search Relevance Workbench features for hybrid search optimization.

• Run OpenSearch 3.7 on Amazon OpenSearch Service immediately.
• Use 1-bit scalar quantization to compress vectors and cut memory overhead.
• Retrieve vectors faster using doc values without reindexing.
• Leverage new Search Relevance Workbench metrics and hybrid search tools.
• Access automated query recommendations and finished-queries cache for insights.

DeepMind has released Gemini Robotics 2, a system that applies whole-body intelligence to robotic control using large language models. This update moves beyond isolated limb manipulation to coordinate complex, multi-joint movements for more natural and adaptive robot behavior. The release highlights a shift toward unified AI architectures capable of handling diverse physical tasks through high-level reasoning.

• Whole-body intelligence allows coordinated multi-joint movement rather than isolated limb control.
• Large language models are now directly applied to physical robotic actuation strategies.
• DeepMind's release signals a trend toward unified AI architectures for diverse physical tasks.
• This approach aims to improve adaptability in unstructured environments for service robots.

Hugging Face has released LFM2.5-Encoders designed to accelerate long-context inference specifically on CPU hardware. This release targets practitioners who need to process extended text sequences without relying on GPU acceleration. The model architecture focuses on efficiency for CPU-bound workloads.

• Optimized for CPU execution, reducing dependency on GPU clusters for long-context tasks.
• Enables faster inference speeds when processing large context windows on standard hardware.
• Available via Hugging Face for immediate integration into existing CPU-based pipelines.