I’ve always theorized that it should be possible to have multiple wake words with different functions, some invisible to the user.
It has to be “always listening” for the wake word to function at all, so it clearly is doing that, what’s to stop them from having another wake word like “bomb” which it then starts recording and sends to the NSA for instance, or even “clip the last 30 seconds” like an xbox could be feasible. Or even have corporations pay to get on the “list” of secret trigger words, like Toyota pays and it hears “Toyota” or “new car” and starts serving ads for 2026 Celicas (I wish lol). It doesn’t even have to send much data back for that, just “ohp, said word, check box to join “toyota” ad group.”
I’m not saying they do that, but like, it sounds totally easily possible and I can’t be the first person that had this idea, why wouldn’t they?
Flipper@feddit.org 2 days ago
All the assistants listen all the time for their codeword. The new pixel phone show you a list of songs played around them and more. It is already happening all the time in the background.
serenissi@lemmy.world 2 days ago
That’s done locally. You can try training wake word models for any open assistant and see how much computing power it needs for even simple phrase.