>Let’s say you need to get your oven fixed—Alexa+ will be able to navigate the web, use Thumbtack to discover the relevant service provider, authenticate, arrange the repair, and come back to tell you it’s done—there’s no need to supervise or intervene.
This is a disaster waiting to happen. I don't trust an LLM to choose between two brands of dish soap for me let alone pick a contractor, schedule a repair, and make a payment. Even if there was a demo showing this working in a sterile environment, reality is so complex that something is certain to go wrong. Even the "simple" task of summarizing news had so many catastrophic failures that Apple had to pull it from the market.
Amazon is making bold claims about the capabilities of their voice assistant to sell their subscription service so that they can make the Alexa division profitable, but if any of their claims were real, they would be demoing rather than writing science fiction in a press release.
This reminds me of Facebook's "M" assistant from 2017.
> Today, a few hundred Bay Area Facebook users will open their Messenger apps to discover M, a new virtual assistant. Facebook will prompt them to test it with examples of what M can do: Make restaurant reservations. Find a birthday gift for your spouse. Suggest---and then book---weekend getaways.
Just like nuclear fusion is always 30 years away, you can find a handyman and schedule an appointment with a voice assistant / chatbot in 6 months, and everyone will use self driving cars in 5 years.
It also seems like it's ripe for just being an outright lie. People will pay Amazon to be ranked as the preferred service provider. You won't get the best service provider, you'll get the one that paid Amazon the most money.
They already failed at this. Adding something to a shopping cart is just one of the steps the mentioned example states. And consumers have already indicated that they don't trust it, because they prefer to know just what they're buying. So this project, by doubling down on that, is DOA
Yeah there's roughly zero chance that works reliably. It's so prone to bad failure cases I'm skeptical that they'll even ship something that tries to do that automatically.
On the plus side if they do ship this we should get all sorts of amusing stories out of it. I'm picturing someone saying offhand "I'm so hungry I could eat a horse!"
I'd expect this to be like when you call the bank and go through the phone tree, enter your account, birth date, SSN, zip code, etc. Finally get connected to the rep and the first questions are "What's your name and birth date?"
i can picture your ancestor saying "what? i will go to one mega store and buy something made in asia instead of getting a suit made to measure by Giovanni across the street?"
or your mom complaining about "buying shoes online without trying then" (or maybe that's you if born before 2000)
AI slop will define your identity tomorrow, doesn't matter what you think
Let me put this way: as a non native English speaker, I am fairly confident that I'll do a better job at "summarizing" headlines than Apple Intelligence. Take that however you like.
I find it interesting that the big companies are so sure that LLMs are somehow going to make a larger market for smart speakers than they currently have. To the contrary, I expect they are going to damage the market they have for people that just want easy kitchen timers and radio like functionality.
This feels like the VR plays some of the big companies have made. I'm willing to bet that the market for people that want to play VR games is far larger than the current market for any other VR use. To a silly degree.
Could this change with overwhelmingly amazing technology? Maybe. But a bit of a moot point, as we don't have that technology, yet. And in the meantime we are just making the existing markets depressed.
To that point, is it time I look into making my own kitchen timer/radio device? Was never really that tough, all told. A raspberry pi is more than powerful enough to do so. Difficult part is largely the packaging aspect of it. Upside will be that you can do what people largely want 100% local.
Google has already done this. Destroyed its existing assistant product -- removed key functionality, de-staffed the team, moved it all under Gemini.
I'm in a few Polestar car user groups, and people are pissed that their Android based head units can no longer do basic integration via assistant stuff that the car was initially sold to them able to do. In some cases they are blaming Polestar, in some cases connecting the dots back to Google.
It's beyond foolish. And destroys goodwill with customers. Who they seem to consider there being an infinite supply of. There is not.
Beyond that, there's the fact that stochastic "fuzzy" AIs are maybe not such a great fit when you just need to have the pod bay doors opened. Basic deterministic, symbolic, "AI" makes a lot of sense, especially once people get used to the quirks for the right way to talk to the thing.
I've been saying this for years, but I am still surprised there was no class action against Google for borking up their Home speakers.
Mine literally don't even understand "STOP" anymore when I ask them to stop playing my podcast. I am not kidding, every couple months they lose some basic functionality.
If that's not a modern version of planned obsolescence then I don't know what is.
I'm a little biased, but I do think "voicexml" probably covered far more use cases than people are willing to admit. Some documents wouldn't have been pretty, I don't think, but would be far preferable to the inability to reason about how things actually work. And, indeed, the messier the document, the more obvious that you were making it too consuming.
It is not completely unexpected. Within executive ranks there is an odd FOMO on 'new big thing'. That's partially why you saw attempts to sell more passing fads like crypto-everything, blockchain-everything, iot everything, subscription everything. It is easy to make fun (especially in retrospect), but being wrong means executive may make the mistake that Dwight made and declare internet a fad.
This makes some sense, but I generally you'd make sure the TAM of what you are building isn't smaller than one you would be cannibalizing.
Amusingly, this is exactly what Google did with Reader back in the day. They actually had Reader and Buzz integrated rather nicely, but lit it on fire in an effort to get circles going.
If the only thing this does is make it easier to control devices whose names I forget, than it is worth it. Because I never, ever remember what I name devices and to address them with the current implementation of Alexa I need to be pretty spot on.
You can take a few minutes to make a routine and give it a custom name. "Ziggy, goodnight" turns off all the lights in my bedroom, closet and office (in case I left them on), sets DnD mode on the Echo, turns it's screen off and sets my ceiling fan to speed 1.
I never understood how people find setting a timer on their phone so excruciatingly difficult that they need to buy a $100+ device they can speak to to do it for them. Or perhaps it's another case of shiny object syndrome.
Here's a few dozen use cases based on my own use of smart home devices:
- Hands are full or dirty while cooking. Voice activation is more convenient. True for not just timers, but every other aspect - music playing, controlling home devices like lights, watching something on YouTube, etc.
- The above also applies to any case where my hands can't readily access my phone, such as wanting to listen/change music when showering.
- As the other commenter said, sometimes the timer needs to be "room-specific" rather than on my phone (which stays with me)
- The device has a decent speaker, so makes a convenient Spotify device. The voice activation is sufficient, though I can also control the device via Spotify on my phone if there's occasional blips.
- Combined with smart light switches, I have convenient control over various aspects of lighting in my home
- Combined with Chromecast / Google TV, it provides voice activated access to pause/play/change what I'm watching.
- Basic internet queries, such as how long it will take to drive somewhere or when a certain place will close, work well also.
None of these use cases _individually_ is so amazing I'd spend $100+, but the combined total value is great for me.
I don't have my phone on me at all times. More, it is often the case that I will set the timer for whatever I just put in the oven and let the kids know to take it out when the timer goes off while I go take care of something in the yard/other. That is, the timer is often specific to the room I set the timer in. Not to me.
If you already have one the utility is obvious in the kitchen. You don't want to touch your dirty hands to your phone and you definitely don't want to touch your food after touching your even dirtier phone.
I don't think that is useful enough to allow Bezos to listen to everything in my home, but will absolutely enable this feature in a product like Home Assistant.
Massive lack of insight. The lady in the box on the kitchen counter is a "groupware timer". The timer on my watch is of no use to my wife in knowing when to turn off the oven when I'm in the bathroom.
I set time on my phone by voice (which is the only thing I use Siri for), because it's way quicker to say "5 minutes" than to try to set it through the app even with shortcut. But I have a kitchen timer (big, fat buttons and big digits) which I prefer to use in the kitchen because it's distinctly loud. My homepods are mostly for music.
It's convenience. I have a device in my kitchen that's hands-free, can set timers, show recipes, etc, that's always there when I'm cooking.
I don't usually have my phone on me, but even if I did I need to at least unlock it to enable voice commands, which instantly kills any notion of it being truly hands-free.
Obviously nobody finds it excruciatingly difficult, but you have a severe lack of empathy and imagination if you can’t see the utility here. Timers while cooking (hands dirty and/or full), measurements while woodworking/crafting/making, etc. and more as detailed thoroughly in other comments
On the rare occasions I watch TV, I'm incredulous at the adverts for technology that create these almost utopian looking lifestyles. I think to myself "who is taken in by this?". As it turns out, these are the people who are taken in by it.
This is so disingenuous. Echos are nowhere near $100 at the baseline, no one is buying it JUST for the timer, and no one finds setting a phone "excruciatingly difficult". Calm down.
I've been spoiled by LLMs in my daily work and now want to put the same kind of prompts into search boxes. Not "air fryers" but "air fryers without bluetooth or wifi and less than 3 cooking modes, and no negative reviews about the device failing prematurely." I'm not going to let Alexa plus or minus listen into my whole life, but I would like some that of intelligence when I actually go shopping.
I've found Perplexity.ai with Deepseek R1 to be very good at choosing a product or a hotel for me. I just punched in your query and it actually chose the air fryer that's already sitting in my kitchen ! "Cosori Pro LE Air Fryer"
Another example, after spending an hour on trip advisor going back and forth to maps to check for walking route to my destination, please recommend a hotel, more of a guesthouse, in marrakesh, near le jardin secret in the medina. something with a local flavor, not 5 star european -- I was so relieved to be able to book direct and be done with it.
Google was a great product for searching the best deal too. As was Amazon back in the day.
Perplexity.ai is massively in the red, and would have to extract a profit somewhere.
Even if perplexity.ai is really great at searching for products, it's great at searching for products for now. For now SEO didn't found a way to play a game, and for now no ad deals have been (to my knowledge) made.
And that's generally true for all commercial LLMs. They are unprofitable as is.
So even if they give you amazing advice, at one point the advise could get worse and it will be hard to notice.
I am on a one-month free trial of Perplexity Pro and its deep research is excellent. It may very occasionally deviate from the original query, but I will still miss it a lot. It makes using Google feel like such a chore.
I've noticed that instacart (and by extension, Costco same day shopping) has integrated an LLM into their search. It's awesome to be able to search for "ingredients for a chicken and vegetable roast" and have all the separate ingredients you need be returned. You can also search for things like "healthy snack" or "easy party appetizers".
I think this is a great use case for LLM search since I am able to directly input my intent, and the LLM knows what's in stock at the store I am searching.
Nothing you describe hasn't already been done in the pre-LLM era with simple keyword matching.
In the city i lived in 2012, the (now defunct) local supermarket chain could handle your roasted chicken request. You could also paste an entire grocery list into a text box and have it load the items into your cart all at once. That's the feature i moss the most.
I just tried your snack and appetizers requests with the grocery service i currently use, and it worked fine. No "AI" needed.
That's exactly what I asked for, wow. To whoever asked why should Amazon want to do this, it's to keep their customers from bypassing their own search with services like this one.
Curious as to what LLMs you are using to allow successful queries like this and what are you using them for? If you don't mind sharing. My understanding was that these would result in some fairly random, maybe true maybe not, results. Is there a company with a RAG that produces reliable results? If so, I would like to check it out.
The majority is coding in an IDE with Claude. It outputs results that I can validate immediately. There are lots of wrong answers to be tossed out but it's still a large acceleration over just docs and stackoverflow.
I can understand the skepticism if you use it in a context where you can't independently test the answers, so you can't filter out the trash. But it's a big level up when you can.
If an LLM-based voice assistant/hardware combination works as well as ChatGPT-for-voice works today, I don't think it's a stretch to say that nearly everyone in the coming years will use/have one (the software of course will be portable to whatever device you're using--house, phone, car, etc. But the hardware portion I do believe will be critical because most of the time using it will be at home in a room and in that scenario sound quality will actually be key).
That said, if nearly everyone will find utility in an assistant, obviously the biggest issue with using one of these, as this Amazon announcement illustrates, is whether you really can trust the company with such a thing when you would be having entire conversations about everything from your interests to something as sensitive as your emotional state (anyone simulated a therapy session with ChatGPT? It arguably is already a decent therapist!).
One of two things will happen, though. People will be dumb enough to "upload" their deepest darkest secrets to megacorp x (thousands of HN users cackle in the distance as if that's not happening today) or a completely privacy-safe option will be available and will win because they're able to effectively communicate that they are in fact private. It's one thing for Google or FB to build a picture of who you are, what you think, etc. through browsing activity/purchases/etc. It's entirely something else for you to literally tell them every last thing about you so that they can hear, in your own words, how you think about "everything."
You know, people said the same thing the first time voice assistants came out. They said the same thing when VR came out. Even when 3D printers came out for God's sake.
"Everyone will have one!"
It's a mistake to think every person is the same level of enthused with new technology as you are.
I definitely agree it's a mistake to think that. That said, I do think LLM's or their direct successors are going to be more akin to Google search than the items you mention in terms of market penetration. And my comment was attempting to communicate that voice is today, and will be in the future a great way to interact with LLM's. I think you're saying you disagree with that, which is totally cool of course. Just thought I'd reply to share a little more of my thinking.
I use LLMs pretty liberally and I can say with 100% certainty I am not going to leave an open microphone in my home hooked up to an LLM connected to a place I do not control that is actively trying to "learn" about me.
I wrote a blog post[1] describing what a local only LLM could do. The answer is quite a lot with today's technology. The question is - do any of the tech giants actually want to build it?
The locally hosted scenarios are in some ways more powerful than what you can do with cloud hosted services, and honestly given that companies could charge customers for the inference hardware instead of paying to host, it would likely be a net win for everyone. Sadly companies are addicted to SaaS revenue and have forgotten how to make billions by selling actual things (with the exception of Apple).
Fair, but the above comment is about general population. The percentage of people that’s actively against it in the real world is negligible. Like where do you cut the line? Is Siri/Google Assistant ok on your phone? What about every newer BMW nowadays coming with its own assistant? Samsung TVs? Nest/Ecobee products? I could go on, and I haven’t met a person who owns has 0 devices with voice assistants in years.
I'm not sure how any person can be confident of such things these days, but would you be ok with the open mic if you knew it couldn't be used to build some profile about you?
I think this is an interesting curiosity, but I am a little worn out on every company announcing AI as some kind of major upgrade. Alexa has already been sort of a waning product, and in some ways, it was already kind of cute since you could play goofy games. But cute gets old.
With AI, there is still this massive trust issue. How can I trust that AI is steering me in an actual helpful way? How is Alexa+ integrated with Amazon's core model of selling stuff... lots and lots of stuff?
I remember when law officers wanted the Alexa recording at the home of a murder. Amazon did not give it up.
I always thought that data was meaningless if it takes a person hours to go through it. Now we have AI. Which means the data is not meaningless. And the always on feature actually means something. And that means all your data at home can be at someone's fingertips ... because say they are looking for ways to make your home and government more efficient?
I think your memory is mistaken. Amazon will give up recordings when legally obligated to do so, because that’s the law. They can’t choose to ignore the law.
However, Alexa and similar devices don’t actually record everything. Amazon doesn’t get a recording of everything the devices hear. They have to be triggered by the wake word (or possible a false positive).
I don't know about Alexa specifically, but I've seen stories where the police requested Ring videos from a neighbor's house, including cameras inside the neighbor's house that they could not have known of without Amazon's assistance, that were not pointed outside, and even, if I remember correctly, one that was in the neighbor's business in a completely different location, where the justification pointedly identified the neighbor as not a suspect, but Amazon gave over this video anyways.
I don't know the specifics of this case, but maybe the investigators just asked in case there was an accidental trigger, or a real trigger etc. Seems reasonable for the detective to attempt to turn over any stone they can to aid the investigation.
There’s a difference, though. Alexa’s recordings are made inside your house. Ring videos are taken outside your house. This distinction matters both as a fact and in the law.
I want a physical robot to do domestic tasks. All the things that Alexia+ automates are things that don't take much time, nor are things I want to hand over to AI.
It’s been pretty shocking that a Turing-test sentience at genius level was easier technology to make than a consumer level robot with 1950s level design expectations that stupidly picks up and moves objects.
It’s humbling to humanity. We differ from animals in our “spirits” but that part of us was less difficult to invent than “legs” which every macroscopic creature has mastered.
Readit News
This is a disaster waiting to happen. I don't trust an LLM to choose between two brands of dish soap for me let alone pick a contractor, schedule a repair, and make a payment. Even if there was a demo showing this working in a sterile environment, reality is so complex that something is certain to go wrong. Even the "simple" task of summarizing news had so many catastrophic failures that Apple had to pull it from the market.
Amazon is making bold claims about the capabilities of their voice assistant to sell their subscription service so that they can make the Alexa division profitable, but if any of their claims were real, they would be demoing rather than writing science fiction in a press release.
> Today, a few hundred Bay Area Facebook users will open their Messenger apps to discover M, a new virtual assistant. Facebook will prompt them to test it with examples of what M can do: Make restaurant reservations. Find a birthday gift for your spouse. Suggest---and then book---weekend getaways.
https://www.wired.com/2015/08/facebook-launches-m-new-kind-v...
https://www.youtube.com/shorts/Oa8s07agHeY
On the plus side if they do ship this we should get all sorts of amusing stories out of it. I'm picturing someone saying offhand "I'm so hungry I could eat a horse!"
"You'll want to feed him some oats unless you're ready to butcher him in the next two hours."
I'm not aware of any failures that could reasonably be described as "catastrophic".
This is exactly what happened, multiple times.
or your mom complaining about "buying shoes online without trying then" (or maybe that's you if born before 2000)
AI slop will define your identity tomorrow, doesn't matter what you think
Dead Comment
This feels like the VR plays some of the big companies have made. I'm willing to bet that the market for people that want to play VR games is far larger than the current market for any other VR use. To a silly degree.
Could this change with overwhelmingly amazing technology? Maybe. But a bit of a moot point, as we don't have that technology, yet. And in the meantime we are just making the existing markets depressed.
To that point, is it time I look into making my own kitchen timer/radio device? Was never really that tough, all told. A raspberry pi is more than powerful enough to do so. Difficult part is largely the packaging aspect of it. Upside will be that you can do what people largely want 100% local.
I'm in a few Polestar car user groups, and people are pissed that their Android based head units can no longer do basic integration via assistant stuff that the car was initially sold to them able to do. In some cases they are blaming Polestar, in some cases connecting the dots back to Google.
It's beyond foolish. And destroys goodwill with customers. Who they seem to consider there being an infinite supply of. There is not.
Beyond that, there's the fact that stochastic "fuzzy" AIs are maybe not such a great fit when you just need to have the pod bay doors opened. Basic deterministic, symbolic, "AI" makes a lot of sense, especially once people get used to the quirks for the right way to talk to the thing.
Mine literally don't even understand "STOP" anymore when I ask them to stop playing my podcast. I am not kidding, every couple months they lose some basic functionality.
If that's not a modern version of planned obsolescence then I don't know what is.
Amusingly, this is exactly what Google did with Reader back in the day. They actually had Reader and Buzz integrated rather nicely, but lit it on fire in an effort to get circles going.
This took about 5 minutes to setup.
- Hands are full or dirty while cooking. Voice activation is more convenient. True for not just timers, but every other aspect - music playing, controlling home devices like lights, watching something on YouTube, etc.
- The above also applies to any case where my hands can't readily access my phone, such as wanting to listen/change music when showering.
- As the other commenter said, sometimes the timer needs to be "room-specific" rather than on my phone (which stays with me)
- The device has a decent speaker, so makes a convenient Spotify device. The voice activation is sufficient, though I can also control the device via Spotify on my phone if there's occasional blips.
- Combined with smart light switches, I have convenient control over various aspects of lighting in my home
- Combined with Chromecast / Google TV, it provides voice activated access to pause/play/change what I'm watching.
- Basic internet queries, such as how long it will take to drive somewhere or when a certain place will close, work well also.
None of these use cases _individually_ is so amazing I'd spend $100+, but the combined total value is great for me.
I don't think that is useful enough to allow Bezos to listen to everything in my home, but will absolutely enable this feature in a product like Home Assistant.
I don't usually have my phone on me, but even if I did I need to at least unlock it to enable voice commands, which instantly kills any notion of it being truly hands-free.
but it's nice to use your voice so you don't contaminate your phone. (preparing chicken and such)
the kitchen is pretty much the only use-case for voice assistants imo.
Gee I can't wait for my Amazon Prime renewal price to go up this year when Amazon decides they had to raise the price to justify the inclusion of AI.
It's a good air fryer.
https://www.perplexity.ai/search/air-fryers-without-bluetoot...
Another example, after spending an hour on trip advisor going back and forth to maps to check for walking route to my destination, please recommend a hotel, more of a guesthouse, in marrakesh, near le jardin secret in the medina. something with a local flavor, not 5 star european -- I was so relieved to be able to book direct and be done with it.
https://www.perplexity.ai/search/please-recommend-a-hotel-mo...
How do you know it's not selected because it's the one with the most paid ads? Or reddit fake reviews? Or llm generated seo articles about it?
Even if perplexity.ai is really great at searching for products, it's great at searching for products for now. For now SEO didn't found a way to play a game, and for now no ad deals have been (to my knowledge) made.
And that's generally true for all commercial LLMs. They are unprofitable as is. So even if they give you amazing advice, at one point the advise could get worse and it will be hard to notice.
> earbuds that have the wire in between so I can dangle them around my neck
First result: Sony WI-1000XM2 Wireless. These are neither earbuds nor do they have a wire.
Pointless garbage. It also doesn't let me copy and paste the result, for no reason. Bad software.
I think this is a great use case for LLM search since I am able to directly input my intent, and the LLM knows what's in stock at the store I am searching.
In the city i lived in 2012, the (now defunct) local supermarket chain could handle your roasted chicken request. You could also paste an entire grocery list into a text box and have it load the items into your cart all at once. That's the feature i moss the most.
I just tried your snack and appetizers requests with the grocery service i currently use, and it worked fine. No "AI" needed.
Amazon makes money by selling products you want and loses money when you return them.
They aren’t manipulating search results “against you”
https://websets.exa.ai/cm7m8a1ip006rdzzzgxsalirs
I can understand the skepticism if you use it in a context where you can't independently test the answers, so you can't filter out the trash. But it's a big level up when you can.
That said, if nearly everyone will find utility in an assistant, obviously the biggest issue with using one of these, as this Amazon announcement illustrates, is whether you really can trust the company with such a thing when you would be having entire conversations about everything from your interests to something as sensitive as your emotional state (anyone simulated a therapy session with ChatGPT? It arguably is already a decent therapist!).
One of two things will happen, though. People will be dumb enough to "upload" their deepest darkest secrets to megacorp x (thousands of HN users cackle in the distance as if that's not happening today) or a completely privacy-safe option will be available and will win because they're able to effectively communicate that they are in fact private. It's one thing for Google or FB to build a picture of who you are, what you think, etc. through browsing activity/purchases/etc. It's entirely something else for you to literally tell them every last thing about you so that they can hear, in your own words, how you think about "everything."
"Everyone will have one!"
It's a mistake to think every person is the same level of enthused with new technology as you are.
This is true in general, but LLMs do search better. Everyone already does search.
I wrote a blog post[1] describing what a local only LLM could do. The answer is quite a lot with today's technology. The question is - do any of the tech giants actually want to build it?
The locally hosted scenarios are in some ways more powerful than what you can do with cloud hosted services, and honestly given that companies could charge customers for the inference hardware instead of paying to host, it would likely be a net win for everyone. Sadly companies are addicted to SaaS revenue and have forgotten how to make billions by selling actual things (with the exception of Apple).
[1] https://meanderingthoughts.hashnode.dev/lets-do-some-actual-...
With AI, there is still this massive trust issue. How can I trust that AI is steering me in an actual helpful way? How is Alexa+ integrated with Amazon's core model of selling stuff... lots and lots of stuff?
I always thought that data was meaningless if it takes a person hours to go through it. Now we have AI. Which means the data is not meaningless. And the always on feature actually means something. And that means all your data at home can be at someone's fingertips ... because say they are looking for ways to make your home and government more efficient?
However, Alexa and similar devices don’t actually record everything. Amazon doesn’t get a recording of everything the devices hear. They have to be triggered by the wake word (or possible a false positive).
Here’s a case where the Alexa command was used as part of the case, though it didn’t have recordings of the actual crime: https://www.dailymail.co.uk/news/article-11899217/Murderer-j...
Dead Comment
https://apnews.com/general-news-343c0f5ed0884d1eb1dbe911e8ce...
https://broward.us/2023/07/18/amazons-alexa-is-surprise-witn...
https://www.wired.com/story/ring-police-rfa-tool-shut-down/
Rosie, from the Jetsons.
I want a physical robot to do domestic tasks. All the things that Alexia+ automates are things that don't take much time, nor are things I want to hand over to AI.
It’s humbling to humanity. We differ from animals in our “spirits” but that part of us was less difficult to invent than “legs” which every macroscopic creature has mastered.