GOOGLE'S EXTREME VECTOR COMPRESSION

Summary from the TLDR AI newsletter

TurboQuant is a quantization method that reduces vector memory overhead while preserving performance. This improves key-value cache efficiency and accelerates vector search.