Ok here’s from the google dev blog
https://blog.google/products/ads-commerce/put-google-ai-to-work-with-search-ads/
Here’s how a model can store a localized profile from adsense and learn on it
Here’s how google tensor chips are literally built for that workflow to the detriment of other performance.
Google has an adsemse profile on you the user, that profile is added to by metadata from apps and sensors on the phone; then offloaded to google cloud servers when the phone is charging. No input from the user required.
Here’s a GrapheneOS security feature to prevent persistence and breaks the above workflow.
https://grapheneos.org/features#anti-persistence
I have an honest question. How much android kernel development or tensor chip work have you done?
Reread my response to you. It’s not a stream, it’s locally held as tokens and then streamed over wifi while charging. I don’t see any sources posted by you. I think we’re wasting each other’s time.