So for as far as I understand it:
Yes, this is a big issue on the developers part as this should not have happened without consent.
No, this is not a direct security breach for every and all home assistant users. You do not need to immediately change all your passwords (due to it being partial hashes)
Did I understand correctly? I'm not that deep into how all this works as I am mostly a enthusiast user without a lot of computer science knowledge
It can have it's use cases. Family members don't have to be as precise with correct sentences for example. If you want to use it with music assistant a simple LLM works better in my experience then trying to do this with intent. Can you do all this with the local intent recognition? Absolutely, but you will experience edge cases and it takes a while to get dialled in correctly.
An other big plus, LLMs often automatically support multi language. This is big for a lot of European users in my family we already speak 4 different languages and various dialects and now no matter who visits or stays over, they can use the voice control in native langue.
Personally, I use both. If the intent recognition doesn't catch it, the LLM will. So for me it is just an extra layer that protects me from angry family members :)
Setup:
It works reasonably well. Not any better or worse then Google Assistant did. Setup:
I use it to control the TV, play music, control vacuum and control lights. I have a script for the TV functions and a script that controls different scenes in my house which are both exposed. This works really well as I have commands setup to trigger automations or I can just ask it to do something via the LLM. This last one takes a bit longer (up to 45 seconds or so) but it is workable as things rarely have to happen immediately. I'm upgrading the laptop to a desktop with gtx1660S soon though which will be a huge improvement.
I also have automations setup that automatically lowers music when the wake word triggers and goes back to volume when it is done listening which helps tremendously as the Voice assistant preview does have trouble understanding in a noisy room. The LLM also partially solved this though
Just walk over it?
We have also not found a way of merging quantum physics with regular physics models so there is still a possibility al be it slim
China is not a Russian ally though. China is an opportunistic state. They make agreements and trade with anyone who they think will further their goals. Thus they trade with Russia because by doing so they gain influence in a diminishing state that controls a lot of resources (Russia's far east is increasingly becoming Chinese by population and companies for a reason)
China doesn't make friends, they make deals with anyone and everyone as long as they think it is positive for China.
If so, maybe asml should team up with China to produce these machines. Completely cut out the hypocrites in the US out of the picture.
This is sometimes still a thing. I know of at least one religious news organisation that takes the website offline on Sundays.
Or maybe another form of renewable energy?
No I can't because for spoken language I don't need a camera, just a mic and I also wouldn't want to be constantly bombarded with a translation of every foreign language. If I want to translate something I can't read I can get my phone to either type it or use the camera function on that thing. These glasses will not replace my phone and that already has a camera with better privacy option than these glasses will ever have. So no. I do not see a need for a camera that videos everything I see all the time except for some very niece scenarios. If you think you need something like this, then great, but don't expect people taking it kindly that you are filming their every move.
But "perfect white teeth" aka veneers are often a cosmetic procedure and are thus generally not covered by universal healthcare (there are exceptions ofcourse).
I'd argue a rootcanal treatment is very necessary so I don't think they are comparable situations. That is ofcourse just the opinion of a random internet person.
So isn't the solution then to just feed the Claude output through a second tiny LLM with the prompt to slightly rewrite the input? I mean you can probably use a 1b model to do that and that can run on practically anything nowadays