Cutting AI inference costs without new hardware

The New Stack describes Chip Huyen’s advice on cutting AI inference costs without new hardware. It also quotes Huyen on a distinction between generated and visible output: “The first generated token might not be the same as the first visible token.”