dav43

I come with frustrations. My new iPhone 16 max is a dog. I use to use dictation a lot, has it changed from on device because I can’t get a straight answer from Apples documentation.

(1) Support website says yes on device : Dictation requests are processed on your device in many languages—no internet connection is required. When dictating in a search box, dictated text may be sent to the search provider in order to process the search. (https://support.apple.com/en-au/guide/iphone/iph2c0651d2/ios)

(2) Device documentation says it will show you if it’s on device or on cloud (it never does) : https://i.postimg.cc/d3pxT9cH/IMG-1661.jpg

I have heard this mentioned, online and in casual conversation

** During business hours these endpoints may get so highly utilised that accuracy and quality decline (ignore latency) (effective available token capacity is reduced etc?**

Has anyone run studies/experiments to show (true or false) - across a large dataset - that the performance of these endpoints during peak usage hours vs, low usage hours changes or does not change (I assume some statistical significance test required).

I don't have the knowledge to answer this question or understand if it's a valid hypothesis. Anyone got resources on this? I could only find tangential mentions in these papers.

[0] https://arxiv.org/html/2507.18007v1 "These challenges contribute to bottlenecks during peak workloads, ultimately affecting inference service quality, scalability, and responsiveness, which requires accurate resource profiling for LLM inference task."

[1] Asking Gemini - https://share.google/aimode/bWb5w9dpf2ZeggqSj