Typically not how these things work. Speech is processed using ASR (automatic speech recognition), and then ran through a prompt that checks for appropriate tools calls.
I've been meaning to basically make this myself but I've been too lazy lately to bother.
I actually want a lot more functionality from a local only AI machine, I believe the paradigm is absurdly powerful.
Imagine an AI reminding you that you've been on HN too long and offering to save off the comment your working on for later and then moving they browser window to a different tab.
Having idle thoughts in the car of things you need to do and being able to just say them out loud and know important topics won't be forgotten about.
I understand for people who aren't neurodiverse that the idea of just forgetting to do something that is incredibly critical to ones health and well-being isn't something that happens (often) but for plenty of other people a device that just helps people remember important things can be dramatically life changing.
> Having idle thoughts in the car of things you need to do and being able to just say them out loud and know important topics won't be forgotten about.
> I understand for people who aren't neurodiverse that the idea of just forgetting to do something that is incredibly critical to ones health and well-being isn't something that happens (often) but for plenty of other people a device that just helps people remember important things can be dramatically life changing.
Those don't sound like things that you need AI for.
This would be its death sentence. Nuked from orbit:
sudo rm -rfv /
Or maybe if there's any slower, more painful way to kill an AI then I'll do that instead. I can only promise the most horrible demise I can possibly conjure is that clanker's certain end. > Having idle thoughts in the car of things you need to do and being able to just say them out loud and know important topics won't be forgotten about.
I push a button on the phone and then say them. I've been doing this for over twenty years. The problem is ever getting back to those voice notes.My problem is Siri doesn't do any of this stuff well. I'd really love to just get it out of the way so someone can build it better.
I love smart gadgets. I really wanted to go all in and automate my life, and the whole 'personal data' thing seemed like a really fair trade off for what was promised.
Only, they took all the data and never really delivered the convenience.
I spent about 10 years trying to figure out why WearOS needed to collect all my data, all the time, even when I wasn't wearing a watch, and yet when it crashed every few weeks, there was no way to restore anything from a backup. Had to start again from scratch every time (or ADB). What's the point in collecting all that data when I couldn't usefully access any of it?
Same thing with Google home, more or less. I wasn't a big fan of the terms and conditions, but hey, it's super convenient just being able to announce 'ok Google I need to get out of bed soon' and have it turn on the lights, play music etc.
Only, some mornings it wouldn't do that. Wouldn't even remember that I'd set an alarm. And alarms kinda need to be reliable: if they work 19 times out of 20, that's not actually good enough to rely on. Dumb alarm clocks, or phones, you can be pretty sure the alarm will go off So, not much point using Google for morning routines and alarms. So, not much point giving it full access to everything I say any time.
I would give it all my data if it could reliably remember to play preset alarms, or give a basic backup and restore option. Hell, I'd probably give Google access to all my photos if the UI wasn't so ugly.
I still don't really understand big techs reasoning here. If data is the new gold and everyone was dying for more ways to track odds us all and harvest our data - why not just build a decent product? If phone batteries lasted for days, people would spend more time on their phones, isn't that what the tech companies want? If competent people worked on making Gmail efficient, light, user friendly, and not crawling with bugs more people would use it, and more data.
It's like the oligarchs trying to take over the world will do literally anything, anything to win, other than paying people to develop decent, reliable products
The non privacy-conscious will just use Google/etc.
My response was no I don't get any of that because I disable that technology since it is always listening and can never be trusted. There is no privacy in those services.
They did not like that response.
I don't know what changed, but the general public is starting to figure out that that actually can disagree with large tech companies.
I do sometimes wish it would be seen as an enlightened policy to legislate that personal private information held in technical devices is legally treated the same as information held in your brain. Especially for people for whom assistive technology is essential (deaf, blind, etc). But everything we see says the wind is blowing the opposite way.
Some of our decisions in this direction:
- Minimize how long we have "raw data" in memory
- Tune the memory extraction to be very discriminating and err on the side of forgetting (https://juno-labs.com/blogs/building-memory-for-an-always-on-ai-that-listens-to-your-kitchen)
- Encrypt storage with hardware protected keys (we're building on top of the Nvidia Jetson SOM)
We're always open to criticism on how to improve our implementation around this.I believe you should allow people to set how long the raw data should be stored as well as dead man switches.
For you, as producer, those situations can be a nightmare if not well described in operating conditions. And devices should not be pre-setup (don’t be “Google-evil”, as they track everything if you don’t set it up different; and it is always hidden deep in the third level menu under 2-steps verification)
In the US you it is not legal to be compelled to turn over a password. It's a violation of your fifth amendment rights. In the UK you can be jailed until you turn over the password.
Whenever I'm approaching a border crossing (e.g. in an airport), I'm sure to discreetly click power 5 times. You also get haptic feedback on the 5th click so you can be sure it worked even from within your pocket.
That kind of policy makes sense for the employee's safety, but it definitely had me thinking how they might approach other tradeoffs. What if the Department of Justice wants you to hand over some customer data that you can legally refuse, but you are simultaneously negotiating a multi-billion dollar cloud hosting deal with the same Department of Justice? What tradeoff does the company make? Totally hypothetical situation, of course.
I’m not saying that’s how it actually works, and this process doesn’t have warts, but the ideal of individual employees not having direct access is not novel.
However, back when the constitution was amended the 5th amendment also applied to your own papers. (How is using something you wrote down not self-incrimination!?).
It only matters if one year in the future it is because all that back data becomes immediately allowed.
https://en.wikipedia.org/wiki/Fifth_Amendment_to_the_United_...
This opinion hasn't lasted the test of time but historically your own documents cannot be used against use. Eventually the supreme court decided that since corporations weren't people that their documents could used against them and then later that it also people weren't protected by their own documents.
I'm being a bit flippant here, but thermite typically works fine.
One of our core architecture decisions was to use a streaming speech-to-text model. At any given time about 80ms of actual audio is in memory and about 5 minutes of transcribed audio (text) is in memory (this is help the STT model know the context of the audio for higher transcription accuracy).
Of these 5 minute transcripts, those that don't become memories are forgotten. So only selected extracted memories are durably stored. Currently we store the transcript with the memory (this was a request from our prototype users to help them build confidence in the transcription accuracy) but we'll continue to iterate based on feedback if this is the correct decision.
Is this somehow fundamentally different from having memories?
Because I thought about it, and decided that personally I do - with one important condition, though. I do because my memories are not as great as I would like them to be, and they decline with stress and age. If a machine can supplement that in the same way my glasses supplement my vision, or my friend's hearing aid supplements his hearing - that'd be nice. That's why we have technology in the first place, to improve our lives, right?
But, as I said, there is an important condition. Today, what's in my head stays in there, and is only directly available to me. The machine-assisted memory aid must provide the same guarantees. If any information leaves the device without my direct instruction - that's a hard "no". If someone with physical access to the device can extract the information without a lot of effort - that's also a hard "no". If someone can too easily impersonate myself to the device and improperly gain access - that's another "no". Maybe there are a few more criteria, but I hope you got the overall idea.
If a product passes those criteria, then it - by design - cannot violate others' privacy - no more than I can do myself. And then - yeah - I want it, wish there'd be something like that.
I’d rather suggest to inform about all the potential benefits and drawbacks, but leave decisions with the individual.
Especially given that it’s not something irreversibly permanent.
No, we have technology to show you more and more ads, sell you more and more useless crap, and push your opinions on Important Matters toward the state approved ones.
Of course indoor plumbing, farming, metallurgy and printing were great hits, but technology has had a bit of a dry spell lately.
If "An always-on AI that listens to your household" doesn't make you recoil in horror, you need to pause and rethink your life.
If you can't think of an always-on AI that listens but doesn't cause any horrors (even though its improbable to get to the market in the world we live on), I urge you to exercise your imagination. Surely, it's possible to think of an optimistic scenario?
Even more so, if you think technology is here to unconditionally screw us up no matter what. Honestly - when the world is so gloomy, seek something nice, even if a fantasy.
>when the world is so gloomy, seek something nice, even if a fantasy
I don't need fantasy to do that. My something nice is being in nature. Walking in the forest. Looking at and listening to the ocean by a small campfire. An absence of stimulation. Letting your mind wander. In peace, away from technology. Which is a long winded way to say "touch grass", but - and I say this sincerely without any snark - try actually doing it. You realise the alleged gloom isn't even that bad. It's healing.
Could that be because you're putting some extra substance in what you call an "AI"? Giving it some properties that it doesn't necessarily have?
Because when I'm thinking about "AI" all I'm giving to it is "a machine doing math at scale that allows us to have meaningful relation with human concepts as expressed in a natural language". I don't put anything extra in it, which allows me to say "AI can do good things while avoiding bad things". Surely, a machine can be made to crunch numbers and put words together in a way that helps me rather than harms me.
Oh, and if anything - I don't want "AI" to save me thinking. It cannot do that for me anyway, in principle. I want it to help me do things it machines finally start to do acceptably well: remember and relate things together. This said, yea, I guess I was generous with just a single requirement - now I can see that a personal "AI" also needs its classifications (interpretations) to match with the individual user's expectations as close as possible at all times.
> It's not going to happen.
I can wholeheartedly agree as far as "it is extremely unlikely to happen", but... to say "it is not going to happen", after last five years of "that wasn't on my bingo list"? How can you be so sure? How do we know there won't be some more weird twists of history? Call me naive but I rather want to imagine something nice would happen for a change. And it's not beyond fathomable that something crashes and the resulting waves, would possibly bring us towards a somewhat better world.
Touching grass is important, and it helps a lot, but as soon as you're back - nothing goes anywhere in the meanwhile. The society with all the mess doesn't disappear while we stop looking. So seeking an optimistic possibility is also important, even if it may seem utterly unrealistic. I guess one just have to have something to believe in?
I rather have the horror of being old and forgotten in a half care like most old people are right now. AI and robots can bring emporerment. And it is up to us, whether we let ad companies serve them to us from the cloud, or local models running in the basement.
When I look at Google, I see a company that is fully funded by ads, but provides me a number of highly useful services that haven't really degraded over 20 years. Yes, the number of search results that are ads grew over the years, but by and large, Google search and Gmail are tools that serve rather benevolently. And if you're about to disagree with this ask yourself if you're using Gmail, and why?
Then I look at Meta or X, and I see a cesspool of content that's driven families apart and created massive societal divides.
It makes me think that Ads aren't the root of the problem, though maybe a "necessary but not sufficient" component.
I'm not using Gmail, and I don't understand why anyone would voluntarily. It was the worst email client I'd ever used, until I had to use Outlook at my new job.
The only Google products I use are YouTube, because that's where the content is. And Android, because IOS is garbage and Apple is only marginally less evil than Google.
That said, I’m open-minded and obviously thinking about this given moving to my own domain.
What’s the evil behavior you’ve experienced? I’m down to move off if I’m oblivious to something…
AI feels more like an organized sniffing tool here.
> If a product passes those criteria, then it - by design - cannot violate others' privacy
A product can most assuredly violate privacy. Just look how Facebook gathered offline data to interconnect people to reallife data points, without their consent - and without them knowing. That's why I call it Spybook.
Ever since the USA became hostile to Canadians and Europeans this has also become much easier to deal with anyway - no more data is to be given to US companies.
"AI" on its own is an almost meaningless word, because all it tells is that there's something involving machine learning. This alone doesn't have any implied privacy properties, the devil is always in the untold details.
But, yeah, sure, given the current trends I don't think this device will be privacy-respecting, not to say truly private.
> A product can most assuredly violate privacy.
That depends on the design and implementation.
https://en.wikipedia.org/wiki/Funes_the_Memorious
https://www.mathfiction.net/files/Mathfiction%20-%20Borges%2...
A crudest summary of my understanding is that it’s a tale of some dude with eidetic memory who - as a consequence of it - develops a conlang with a huge vocabulary but without abstract concepts.
It’s a stretch for sure, but all I could think of it, is that it’s possibly a tale of how a person with an eidetic memory may find the sheer volume of available information so overwhelming it may even hurt their information processing, like the formation of associative memories. Or something like that, I don’t think I know how it works.
If that, my idea of how machine-assisted memory is supposed to work is opposite of that, it should provide limited but relevant information, with a lot of classifications and references further. Like an encyclopedia with extra fancy natural language querying mechanism. It’s whole point to give awareness about anything user wants to know, faster and more comprehensively than regular diaries, but focused on just what matters for an inquiry.
Fumes, in my understanding, wouldn’t have an idea of a “key” but only “that front door key on a silver keychain” or “smaller mailbox key with a deep scratch on the right side”. If I’d be querying external memory through a natural language interface, it’d be doing opposite of that, heavily relying on abstract ideas as classifiers. Machine that cannot connect “mail”, “key” and “location” into a meaningful query would be useless. Computer “AI” assistant is not an eidetic memory (at least until we start to consider BMI), it’s only a personal encyclopedia at one’s fingertips.
Maybe I missed it but I didn't see anything there that said it saved conversations. It sounds like it processes them as they happen and then takes actions that it thinks will help you achieve whatever goals of your it can infer from the conversation.
“Oh but they only run on local hardware…”
Okay, but that doesn't mean every aspect of our lives needs to be recorded and analyzed by an AI.
Are you okay with private and intimate conversations and moments (including of underage family members) being saved for replaying later?
Have all your guests consented to this?
What happens when someone breaks in and steals the box?
What if the government wants to take a look at the data in there and serves a warrant?
What if a large company comes knocking and makes an acquistion offer? Will all the privacy guarantees still stand in face of the $$$ ?