AI model releases & ML systems

Google Research has released TurboQuant, a new algorithm suite that compresses AI memory usage by 6x and boosts performance by 8x, reducing costs by over 50%. TurboQuant addresses the "Key-Value cache bottleneck" in large language models without retraining, maintaining accuracy while cutting memory needs. The software-only solution is publicly available and aims to enhance AI efficiency on existing hardware, potentially impacting enterprise AI deployments and memory hardware markets.

ConvertWithMoss, a free open-source multi-sample converter tool, has released a significant update focused on vintage Akai equipment. The update enhances compatibility and functionality for users working with Akai samples, improving the tool's utility for music production and sound design involving classic Akai hardware.

Oracle announced new agentic AI capabilities for its AI Database, featuring a Unified Memory Core that processes vector, JSON, graph, relational, spatial, and columnar data in a single ACID-transactional engine without sync layers. The release includes Vectors on Ice for vector indexing on Apache Iceberg tables, a standalone Autonomous AI Vector Database service, and an Autonomous AI Database MCP Server for agent access. Oracle aims to address data fragmentation and consistency challenges in enterprise AI deployments.