@detective well, that's neat in multiple ways, one being that recall would I guess (edit: and I mean guess. I do not have access to any privileged information, this is me wildly speculating) in reality just run on basically anything, with hardware acceleration (assuming model doesn't have layers that are Annoying, but it runs on an NPU so it probably does not), and the other being that the NPUs on the copilot plus PCs are I guess going to be really easy to use for inferencing with whatever the hell models you want to play with