[BoxFour]

This strikes me as a pretty weak rationalization for "safe" always-on assistants. Even if the model runs locally, there’s still a serious privacy issue: Unwitting victims of something recording everything they said.

Friends at your house who value their privacy probably won’t feel great knowing you’ve potentially got a transcript of things they said just because they were in the room. Sure, it's still better than also sending everything up to OpenAI, but that doesn’t make it harmless or less creepy.

Unless you’ve got super-reliable speaker diarization and can truly ensure only opted-in voices are processed, it’s hard to see how any always-listening setup ever sits well with people who value their privacy.

[luxuryballs]

I wonder if the answer is that it is stored and processes in a way that a human can’t access or read, like somehow it’s encrypted and unreadable but tokenized and can be processed, I don’t know how but it feels possible.

[krupan]

It wouldn't matter of you did all that because you could still ask the AI, "what would my friend Bob think about this?" And the AI, who heard Bob talking in his phone when he thought he was alone in the other room, could tell you.

[luxuryballs]

Right but that’s where the controls could be, it would just pretend to not know about Bob due to consent controls etc, but of course this would limit the usefulness.

[ajuhasz]

We give an overview of our the current memory architecture at https://juno-labs.com/blogs/building-memory-for-an-always-on...

This is something we call out under the "What we got wrong" section. We're currently collecting an audio dataset that should help create a speech-to-text (STT) model that incorporates speaker identification and that tag will be weaved into the core of the memory architecture.

> The shared household memory pool creates privacy situations we’re still working through. The current design has everyone in the family shares the same memory corpus. Should a child be able to see a memory their parents created? Our current answer is to deliberately tune the memory extraction to be household-wide with no per-person scoping because a kitchen device hears everyone equally. But “deliberately chose” doesn’t mean “solved.” We’re hoping our in-house STT will allow us to do per-person memory tagging and then we can experiment with scoping memories to certain people or groups of people in the household.

[com2kid]

Heyas! Glad to see someone making this

I wrote a blog post about this exact product space a year ago. https://meanderingthoughts.hashnode.dev/lets-do-some-actual-...

I hope y'all succeed! The potential use cases for locally hosted AI dwarf what can be done with SaSS.

I hope the memory crisis isn't hurting you too badly.

[ajuhasz]

Yes! We see a lot of the same things that really should have been solved by the first wave of assistants. Your _Around The House_ reads similar to a lot of our goals though we would love the system to be much more pro-active than current assistants.

Feel free to reach out. Would love to swap notes and send you a prototype.

> I hope the memory crisis isn't hurting you too badly.

Oh man, we've had to really track our bill of materials (BOM) and average selling price (ASP) estimates to make sure everything stays feasible. Thankfully these models quantize well and the size-to-intelligence frontier is moving out all the time.

[com2kid]

I fired off an email!

Best of luck with the product launch, I've worked in consumer HW before, it is brutal!