halcyoncmdr@halcyoncmdr@piefed.social

Replying to an earlier post

So I finally let my Android Auto update from Google Assistant to Gemini, which it’s been wanting for months.

First thing I did was prompt it to play a specific artist, just like I do at least once a week since it forces me to use the voice activation instead of the on screen keyboard, even when there are passengers that can type. It responded saying it is not capable of playing music at all. Wtf.

So I then rephrase and tell it to play the artist on YouTube Music… And it does so, but doesn’t get the artist even close to correct. I had to prompt it 5 times with varying emphasis to get it to finally get it right.

So literally the first thing it did wa like to me… and then it worked correctly less than 20% of the time… So much better than the Assistant it replaced that never gave me a problem. Totally better.

hietsu@hietsu@sopuli.xyz

Replying to @⁨1984@lemmy.today⁩

At least all the things NSA and CIA are interested at the time…

All new things always have an intelligence aspect to it:

  • Find my network: Find any device globally by its ”anonymized” identifier. Make a device you hack remotely send these BLE pings and voilà, some Apple or Google device is surely nearby to hear it and tell you where it is.
  • Voice assistants: Hear what people talk about privately by injecting new wake-up words like ”Trump”. Cannot be used en masse yet perhaps because it’s usually only the device detecting the trigger word, and sending ”uncertain interpretations” too often would get noticed by someone.
  • LLM chats: Well the same as Google searches etc. historically but now way more nuanced and with more context of what the target in intending/thinking. Building a psychological profile of them through features like ”memory”. And as LLM is such an undeterministic black box, you can easily get away with injecting all sorts of ideas through its responces.