Not sure how enthusiastic I am about those pcs you linked.
While I didn’t care to look too hard into it a while back, was a major complaint of Google Pixels not to do with something similar? They spent so much time on their inferior Tensor chips saying that they were built for dedicated AI tasks, but then 90% of the AI tasks run on a stock pixel are done in the cloud anyways.
I don’t really see why Microsoft would be making PCs to run Copilot locally when they can do so on the cloud and potentially use that for invasive privacy practices.
I just want to point ouf that for the Qualcomm copilot pcs+, those are compatibe with certain Linux distros like Ubuntu. This mean you could both leverage the chip NPU performances, and preserve your privacy, I have no idea however how well LInux leverages NPUs.
Face grouping and facial/object recognition, and possibly some of the other auto-magic features of Google Photos is one additional example.
Google translate is another example of a Google app that is feature limited and less capable if used without an active connection to Google servers (and the features that are limited appear to be those that most likely rely on ML/AI under the hood).
interesting, because my experience has been the complete opposite
I can think of a view possibilities. Either (1) things may have changed since I last actively used Google Photos (> a year or two ago), (2) the hardware or software may have improved enough that Google now does something on device that was previously reliant on the cloud (3) or one of us is simply mistaken.
edit: I just briefly tested on an old phone (stock Android).
While offline, I cannot search for objects, people, or things, in Google Photos, and I get an error about being ‘unable to connect’.
When online, the functionality works as expected.
I’d like to test face recognition and grouping, but I’m not exactly sure how I’d go about testing that quickly.