There's a recently published paper on something called Starling. As far as I can tell, it's a more efficient DiskANN. It claims enormous performance benefits.

There's a recently published paper on something called Starling. As far as I can tell, it's a more efficient DiskANN. It claims enormous performance benefits. The paper is tedious and near-unreadable, so I didn't finish reading it. Maybe there's something here, maybe not. https://arxiv.org/pdf/2401.02116.pdf https://arxiv.org/pdf/2401.02116.pdf