Yep, that was before it became this bluntly obvious how much overhead there seems to be for on-device AI on iDevices and Macs. You'd think Apple would find a way of sorting it out for 8GB, especially if they can manage to do it in 1GB on a watch if the math co-processor ("neural engine") is good enough. But that might require them to admit that even 13 series iPhones or so are "good enough", and with how well they still seem to handle everything EXCEPT Siri, that might be dangerous for the bottom line.