I was looking for cables yesterday and went deep into my bins and found 4 Alexa devices that I haven’t been using, I kinda want to give them to the local “reuse center” but I am not so sure how to clear my AMZN id from them so it is safe.
nice thought; but does anyone else have a use for them either?
I never understood what the "voice assistant" hype was about, the advertised benefits looked ... non beneficial; the examples I saw in peoples homes didn't appear to work when they tried to use them.
Was it worth that much money and effort to have microphones in people's houses?
> I never understood what the "voice assistant" hype was about, the advertised benefits looked ... non beneficial
I'm an Apple user, and by extension, Homekit and Siri.
In a household full of kids and an all-around crazy schedule, having HomePods scattered amidst my house has proven useful; beyond just being great speakers for music.
I use them for everything from adding/completing items on my Reminders list. Controlling the home (lights, timers, etc.). Asking questions (this one is very hit/miss with Siri, but she's getting a lot better). And reading/sending messages.
Is it absolutely necessary? No. But when you're lugging around little ones throughout the house, fixing lunches, dinner, etc. -- they can make life a little easier. Sometimes.
I love our Alexa, but she's progressively getting worse. She's really only good at setting timers and reminders and even then it's hit and miss if I ask her about my next reminder. Asking her a question is no bueno... I have to pull out my phone and google it 9 times out of 10.
We have at least 8 Echos around the house, with clear reasons for a couple more, but Amazon simply isn't interested in the actual problems that the product solves, or the use cases customers care about.
They're completely backwards, here. If that means the whole thing collapses and we have to find another solution, great. Products and companies who work this way need more failures.
Same. I think the article's title lays it out well. "Amazon's Big Dreams". Amazon wanted to do a specific set of things w/ Alexa and they've found that the market doesn't want any of that. I'll never buy something through my Alexa, for example.
What it could/should have been is the end cap for the Amazon brand in homes. A loss leader itself, but a constant reminder of the value of the company and the desire to keep spending money there due to the convenience it provides.
What the article makes it sound like they want is something more tangible. Something that quantitatively generates direct ROI. They'll never get that, or rather, that number will never add up to one that justifies Alexa's existence.
An exercise in perverse incentives. It seemed clear to me from the start that Alexa devices were a Trojan Horse to insert Amazon dependence and ads into real-world situations like writing a grocery list or cooking a meal. Similarly, it's clear that Google's search is a Trojan Horse (at this point) to sell more ads.
Folks need to be discerning with tech purchases. If it isn't in a company's interest to make your experience better (and it is in the company's interest to make your experience worse, say, with ads) when push comes to shove they will prioritize profits over your customer experience. Instead of investing in rotten platforms like Alexa, I wish more people invested in open source and non-exploitative solutions where the incentives align to improve user experience, not destroy it.
Imagine a world where you could self-host or pay a small fee to cloud host your own smart assistant, tweaked to your preferences, not in any way linked into ad sales for Amazon or Google. That's where I'd like to see us go.
Amazon hasn’t been “customer obsessed” in years. The leadership there is “promotion obsessed.” Bloated management with too many empire builders. Make big flashy features with no real need or something that optimizes for short term target numbers. Working backwards from the customer is never a thing anymore. This is why terrible decisions like 3rd party sellers (diminishing Amazon’s trust while increasing profits) will continue to happen.
AWS and their massive logistics platform will support this terrible management for awhile but it will continue to destroy the brand.
Because the $30+ Billion they spent on voice recognition research won't make itself back if you only use Alexa as a voice kitchen timer or glorified radio / Spotify station.
My elderly parents struggle with voice assistants because they can't get their thoughts out fast enough. You really do have to queue up the whole sentence before starting to talk, then blurt it out pretty fast without pausing, or they don't work. Which sucks because the things could be a really great UI for them, otherwise.
I have basically stopped using Alexa in my house. It started as a vision for an intelligent personal assistant, and what we ended up with is a glorified kitchen timer that reads out Amazon ads all day.
“By the way, sign up for Amazon music.”
“Tip your Amazon driver.”
No thanks.
I wish they had actually done something with the tech rather than promptly handing it over to the marketing team once it got some users. I’m trying to think of a single major improvement to the experience since it first launched a decade ago, but am coming up empty.
Yeah you can google it, it's a real program that Alexa will notify you of sometimes (if you're in a participating region), just like the other ads, sorry, "reminders" it serves.
Mine just asks me to "thank" or compliment my driver. Initially they gave drivers money based on their thankfulness metrics but I think that program ended.
TBF it wasn´t like that at the beginning, when growing the ecosystem was their priority.
As pointed in the article, the move to monetization happened once the writing was on the wall that it wouldn´t become significantly different than what we have now.
Also they´ve always been priced so aggressively, I´m kinda glad I got the original puck at such a low price, to see first hand where the technology was and what to expect from it.
Glorified kitchen timer reminds me a little of my use case(s).
When uncorrected I have severe vision impairment (-10 + astigmatism). Without exaggeration I have trouble finding my glasses unless the room is well lit, and even then I have some panic if they aren't exactly where I think they are. I have backup glasses in a known place because if my glasses slide off the nightstand or something else I'm essentially completely blind.
At night, in a dark (or even dim) room, it's much worse. So, for me, the Echo does it's job just because I can tell it to turn on the lights. I imagine for other people with more significant accessibility issues it can be very useful - even though it drives all of us crazy with "Oh, by the way, did you know I can...".
It's also nice for a few other bedtime/dark room related things like lying down to go to sleep and get that panic "OH WAIT! Is that 7am meeting tomorrow? Better check the calendar and set my alarm accordingly." Or when I wake up at some point in the dark and need to decide if I should start my day I can ask Echo what time it is without struggling to find my phone and be blinded (even more so) by the backlight as I hold it (literally) 1-2" from my eye.
It can be really annoying but for me these (admittedly minor) use cases keeps Echo in my house (bedroom).
Motion and sound activated lights have been a thing forever. The only new innovation is that your request has to hit 3 AWS data centers and a server in China and Google Analytics before your bedside lamp can turn on. And if Wi-Fi goes out for some reason then you’re sitting in the dark.
You don't like Amazon/Echo and that's fine. I'm not coming to your house and forcing you to install one. No one is.
Server in China? I use Home Assistant with total local control of all of my smart home devices. On a separate VLAN and SSID. With no internet access. Echo is on a different VLAN and uses the (excellent) Home Assistant integration. AWS data center of course, but see my other comments[0] about the near-total uselessness of any local devices and implementations for wake word detection, ASR, and NLP. Speaking of Home Assistant, see my other comment about the graphing and notification system I built for Echo/Alexa[1].
Yes, I remember commercials for "The Clapper". I've been in bathrooms with motion activated lights that turn off while you're on the toilet (you know, the move where everyone starts frantically waving their arms).
Not only are these solutions also imperfect, you conveniently ignored my other use cases about checking the time in the dark, checking my calendar, setting alarms, etc.
I like my Echo. I like my "smart home" configuration and I've minimized the security issues and blast radius. I understand your concern and it's not without merit but you have the choice of whether or not to use internet connected devices in your environment. So do I.
It makes sense they’d want to support the service with ads, but it feels especially jarring to have a simple interface for simple tasks unpredictably and inconsistently interrupted with ads. I think especially because it tends to be a device we use for multitasking while cooking, cleaning, cuddling, etc. but also just because of the inconsistency. It’s like if your toaster or C compiler occasionally asked if you knew it could cook bagels or give warnings for unused variables, when you’re just trying to make breakfast or get your code to compile before a meeting.
Ooh ooh! And the fact that EVERY ALEXA IN THE HOUSE will rotate a mysterious yellow light if you get a special Amazon notification. Took me a good while to figure out wtf it is.
This is horrible, in NL we have the "international version" of Alexa and we don't get any of this crap, just a basic voice assistant that can play music and control smart home devices.
Theres another way to view alexa. It is in 100m+ homes. Its limited right now due to the historical limits of NLP technology. But with the advent of LLMs, amazon has a chance to dramatically upgrade the capabilities of alexa, which is already in so many homes. A product that was too early, but may end up being able to capture the uptrend in AI.
> with the advent of LLMs, amazon has a chance to dramatically upgrade the capabilities of alexa
This is a terrible idea. NLP, with or without LLMs, isn't remotely close to being able to interpret commands in someone's house. And who would want a non-deterministic voice assistant anyway?
Imagine you say, "Alexa, turn up the air conditioning," and Alexa interprets this to mean you want to increase the temperature. There are uncountable and unforeseeable failure modes.
Note that this happens even when talking to humans with near-perfect NLP/brains.
The major flaw with voice assistants in general is that they only make sense if they have a pretty narrow and pre-defined universe of accepted commands, which is incredibly confusing to someone who mistakes them for actual NLP.
It's just a really, really bad UI. I use mine all the time, but only to set timers and turn lights on or off.
You sound like you have no clue what youre talking about. The limitations of alexa and complaints about alexa are really complaints about the limits of NLP, which have just been expanded.
This comment still has no content in it other than "this is how it is" and "trust me bro". Do you have anything substantial to add? Do you have any reason to refute anything I said?
It doesn't matter if you're Amazon's ML expert or the world's best surgeon or a barista if all you do is make an unsubstantiated statement and give your credentials.
No because its proprietary, and secondly all of your arguments are made out of thin air. Do you want me to produce a design doc about why getting alexa to work correctly is hard?
I quasi-jokingly have suggested that we will end up with a conlang like lojban (which is designed to be unambiguous) as a kind of "command language" to machines.
It's definitely not a joke at all. It's exactly how I use the voice assistants. But unlike a traditional programming language, there's no documentation, it's non-deterministic, and the same command will suddenly stop working. It's insane.
So here's my process with Google Asisstant:
- guess what it will understand through trial and error
- use that command pattern, like "set [something] to [level]%", for a few months
- try it again one day, and suddenly it doesn't work
It makes you scared to trust it in almost any situation.
Siri has gone through phases at least twice where it forgot what the word "half" meant in relation to lights, then suddenly understood it again several months later.
I've extensively used Google Assistant and Siri. I've used Alexa a little bit, too. Google is like 4/10, Alexa is 5/10, and Siri is 0/10. Siri is stunningly bad. It understands the command ~50% of the time, and then the action it takes is useful (i.e. prevents me from needing to touch my phone) ~30% of the time.
> Imagine you say, "Alexa, turn up the air conditioning," and Alexa interprets this to mean you want to increase the temperature. There are uncountable and unforeseeable failure modes.
"up" and "off" sound similar too, and I've had assistants confuse them often.
Hypothetically, if the voice detection and parsing were superhuman - what capabilities does that enable which the device currently lacks?
Voice control is a miserable interface, and there are few applications which I would endeavor to use it. Time, weather, math, and maybe a trivial fact queried from Wikipedia.
Voice control using a central mic that’s always on seems to deliver a poor user experience.
I found myself shouting at these assistants while they try to interpret what I say mid sentence.
The novelty wore off very quickly. But voice seems like a very powerful interface if you can solve privacy and usability issues. Seems like we haven’t tried very hard at this yet.
What made Alexa miserable to me was the Amazon ecosystem. It is just way too limited to be more useful than a timer, news broadcast and perhaps home automation assistant. In that respect, a better voice interface wouldn´t help.
In comparison Google and Apple are competitive in so much more (in particular calendaring, contacts, music/videos).
I think Amazon missed the boat in not coming with an open standard and get cosier with independant makers like Sonos or Ikea etc. Basically they could have pulled an android for the voice assistant ecosystem, but preferred to try to build another walled garden.
I think the point is that there’s plenty of ways in which Alexa could have been useful besides ones requiring LLM and if Amazon couldn’t deliver on these simpler use cases then it’s optimistic to think they’ll be able to do any different with LLMs.
We threw all these listening devices out the door. They're annoying and not very useful. This includes the Google ones too. We would use it to dim the lights or turn on the air conditioning, which is cool, but using your phone to do it is just as fast, and less error prone.
"Alexa, I want to throw you in the garbage." is probably the phrase I say the most to that device. I use it as a kitchen assistant, and that is pretty much the only thing I want to use it for. All the extra crap that amazon keeps pushing on to the device just makes it worse. My main uses cases are: grocery lists, listening to music, setting timers, viewing recipes, making meal related announcements to the house. I really don't want to see any advertisements or other content it thinks I might be interested in. Just stop doing it please.
The only reason that I haven't moved to homepods is because apple can't figure out how to do shopping lists that any one in a family can access.
"Alexa - Add bread" takes one second. "Hey Siri - add bread" takes upwards of twenty seconds for the homepod to figure out who you are, and if you have permission, and add it to the reminder list. I really would just throw away the alexa's if apple could get that one thing figured out.
I hope they never take it away though! They don’t make money from me with it but it is an amazing cheap and no subscription kitchen timer, tell me the weather, play my sleep sounds device. For once, we seem to be benefitting from a company’s bad decisions. She’s also pretty great at answering basic knowledge type questions I may have throughout the day. Certainly an Einstein compared to Siri.
As soon as it becomes terminally unprofitable, they will kill her and brick everyone’s devices. It has happened before and it will happen again and again and again to every IoT device that relies on a cloud service. The only way to win this game is to refuse to p(l)ay.
The best use for Alexa is as a voice calculator for woodworking. She lives in my shop, so she can't listen to my regular conversations in the house. When I need to measure things, I can get a calculator out and do the fraction math (which I am not good at).
OR, I can say things like, "Alexa, what is 23 and 1/16th minus 2.1625" And she'll spit it out as a decimal and clarify as a fraction if I want.
I find it interesting that the culmination of so many amazing technologies result in a tool for americans to handle their crazy ancient measurement system more easily.
For miles, pounds, and ounces, I have no defense, but I will say that having a base 12 length measuring system (in/ft/yd) is much nicer for carpentry than base 10 (m/cm), since it can be divided evenly by 2 and 3. It's the same reason why having 12 notes in a scale works for music.
I genuinely don't understand what you mean here, and I say that as an American who has done a fair amount of carpentry. I can't think of a single time I've thought to myself that using fractions was superior to measuring straight units in mm/cm/m.
> I will say that having a base 12 length measuring system (in/ft/yd) is much nicer for carpentry than base 10 (m/cm), since it can be divided evenly by 2 and 3. It's the same reason why having 12 notes in a scale works for music.
When would you want to divide an octave into two or three intervals like that?
The normal way to divide an octave in "half" is at the fifth, which is a ratio of 3:2. If you instead go six half-steps up from the tonic, you'll be a half-step short of a fifth, at fi instead of so, with an irrational ratio of √2:1. That's the opposite of what you want. [From wikipedia: The tritone is a restless interval, classed as a dissonance in Western music from the early Middle Ages through to the end of the common practice period. This interval was frequently avoided in medieval ecclesiastical singing because of its dissonant quality. The first explicit prohibition of it seems to occur with the development of Guido of Arezzo's hexachordal system]
If you divide the octave into "quarters" by going up three half-steps, you get a minor third.
(Going for "thirds" by going up four half-steps will take you to mi and form a third [the name of the chord is not related to the idea of dividing an octave into three equal parts; it's called a "third" because "do re mi" is three notes], which is fine. But it's pretty clearly just a coincidence that that works out; the methodology of "there are twelve half-steps in the octave, so I'll divide that into equal parts and see what I get" just isn't related to the way you compose music.)
"An investigation attributed the failure to a measurement mismatch between two software systems: metric units by NASA and US customary units by spacecraft builder Lockheed Martin"
So, no fault of NASAs, or of the metric system.
For Lockheed to use anything but metric system in the modern world is bizarre.
I kind of like iOS Siri's limitations. Unlike Alexa, which gives you the illusion that it can do everything, Siri fails spectacularly on certain tasks, and, if you use it enough, you can understand the demarcation between Siri-capable tasks and things that you'd be better off doing "manually."
I don’t like this limitations. I’d love to get rid of my Echo and get a HomePod but Siri is just too clunky for the thing I use frequently - AnyList for groceries.
I can say “Alexa, add eggs” and eggs are added to my grocery list in AnyList. On Siri I have to say something like “Hey Siri, using AnyList add eggs to my grocery list”. The Siri prompt feels like I’m speaking a command line.
I have many issues with Siri but adding things to the reminder list across the family works fairly seamlessly for me, now that my wife has given permission in the 'home' app to use personal requests.
That said, my 11yo daughter said to me yesterday "why don't they just add AI to Siri" after failing to get Siri to tell her how long a timer was originally set for (Siri would only tell you how much time is left on the timer).
Someone did build a crazy ChatGPT-to-Siri bridge through Automations which I haven't tried yet but plan to at some point...
As often is the case with Apple the lack of interoperability to other apps is what kills me. I do not want to use the apple lists app but third-party support is mostly a miss.
Has anyone interacted with chatgpt in a siri/alexa voice interaction for an extended period of time? I too have wondered when AI arrives in these platforms. It could be massively disruptive.
I'd be perfectly fine with voice assistants understanding context better so I could ask thing A and after the response I could ask a clarifying questions.
> My main uses cases are: grocery lists, listening to music, setting timers, viewing recipes, making meal related announcements to the house.
The interesting thing is that with the new Whisper models these things can be done with minimal internet or no internet at all. I would pay good money for an offline only home assistant. Maybe it could receive updates if I specifically asked for them.
yeah, but they were never cool with you being cool with just that. It was created out of fomo in a grandiose attempt to like unseat the iPhone or something ("the platform"), now that it's just a music clock, they really have no idea how to recoup all that money and what to do next.
A friend of mine was an SDM for something like Alexa grocery ordering, a few years ago.
I remember him saying that the additional revenue generated by what they did couldn't pay for even a single developer on their team. And it wasn't going anywhere. But hey, management want it to exist so they're all getting paid to do it.
Meanwhile, my part of the company at the time (Fulfillment) had projects with 7, 8, maybe even 9 figure savings estimated and we just couldn't find the developers for it because all the new hires were sucked into the Alexa org.
I'm glad they're cutting their losses, but Amazon ought to have done it five years ago when it was clear this was never going to earn them additional revenue.
Alexa has worked well for remote support of blind/elderly/disabled:
landline call/answer via Echo Connect
voice/timer power control of lights/fans/aircon
voice-controlled microwave
motion detection to orient blind person
FM radio streaming via TuneIn
cable TV control /w Logitech Harmony
hourly chime + time announcement
remote maintenance of phonebook
remote-initiated audio or video call
remote cough monitoring
remote notification of motion/activity
Alexa was very helpful with my inlaws. We had them scattered throughout the house in case of falls. Neither of them would wear the life alert necklace. What I never found was a good app for medication reminders.
I use them as for when my housemate and wife, who both have disabilities and are fall risks, need help. All they need to do is call for help. I set myself as the emergency contact, so my cell phone and other Echo devices will tell me that they need help.
Yeah, stuff like this was the most interesting use case for me but a company like Amazon doesn't want to market its new, supposedly cool device as a speaking clock for the elderly, or a weather app for blind folks, even though it's super useful for both. I have never wanted one but have been cheered by the division's (apparent) success because I knew lots of people who are left behind by tech a lot found these things very useful.
1. Very few things directly monetized Alexa. Doing so was basically impossible. I was in a brain storming session in 2017 on this topic, I walked in very optimistic and walked out very pessimistic.
2. Voice interface is often slower than preexisting solutions. It's faster to pull a phone out of a pocket, unlock it, and load a weather app then it is to ask Alexa what the weather is. Loading the weather app will show you more information.
3. Alexa is kind of clumsy. It will trigger accidentally. It will not understand you.
4. ChatGPT is clearly better at answering questions. It's probably better at running commands. It doesn't directly compete with Alexa, but it highlights how outdated Alexa really is.
5. Alexa was awesome for some things. It was awesome for vision impairment. It was awesome for hands free situations, when you cannot spare the mechanical efforts of using another device or the visual attention.
> 2. Voice interface is often slower than preexisting solutions.
Yes. This is also why assistants with a screen are so much more useful. Google´s hubs aren´t blazing fast either (and get slower over time...), but getting the full visual information at the same time as the voice answer helps a lot.
> 4. ChatGPT is clearly better at answering questions.
It is stunning that Amazon wasn't the forerunner in using ChatGPT/LLM-based tech for Alexa. A mistake in the tens, if not hundred billion dollars when looking at the price MS paid for OpenAI.
Because LLMs are so unreliable. Alexa misinterpreting and giving a nonsensical/unrelated answer is annoying, but fine. If it was an LLM, when it was wrong it would be lying confidently and unless you're double checking everything it outputs (in which case just search it that way in the first place) then that means it'll be misinforming you without you having a clue, as opposed to the current system where it's obvious when it gets it wrong. That's clearly just worse.
And that's ignoring the fact that all the publicly accessible LLMs have been "jailbroken" and you can get them to say all manner of wild things, and Amazon is definitely aware that all it takes is one or two viral clips of Alexa saying something unhinged for their sales to start dropping, and the tech isn't currently sufficiently understood for them to avoid that.
I think you are missing the point that using LLMs to improve Alexa would certainly be hard, but it would have been massively worthwhile if it worked. Microsoft and Google are well aware that LLMs have the potential to upend the search dominance of Google.
I am also much less concerned about jailbroken LLMs, because Alexa is already so broken it can't get much worse. Normal users don't seem to care much about potential misuses (compare with search autocomplete biases of Google) as long as a service works for them. But unless you widely deploy a tool, you won't uncover how it is abused (see Sidney's launch).
> But unless you widely deploy a tool, you won't uncover how it is abused (see Sidney's launch).
ChatGPT was, while less widely deployed, being abused (very publicly) in very similar manners to the way Sydney was; Microsoft could have learned about those abuses by paying attention to the public information about ChatGPT.
I don't think you can learn about it much unless you have the tech deployed yourself. Microsoft obviously believed that their own prompt design and the learnings at OpenAI would prevent this, but was shown quickly otherwise. I believe they will use the beta to iterate this quickly to a point where we come to accept the residual level of abuse potential.
>I think you are missing the point that using LLMs to improve Alexa would certainly be hard, but it would have been massively worthwhile if it worked.
I really struggle to think how it would have worthwhile. My problem with smart speakers isn't that they aren't smart enough - it's that for anything slightly more complex, my phone has 100x better UX and information density. An Alexa LLM sounds cool in a sci-fi sort of way, but still feels like it would be useless.
I saw a demo the other day of a guy who had hooked Chat GPT up to some kind of voice assistant. The main benefit was that it supported chaining commands so you could say "turn on the light, then set the volume to 50% and then tell me the weather" and the device could parse it all and do the things you wanted in sequence. I can't find the video now but it was super impressive.
Google Assistant does that already. I know because I often tell it to "set volume to 1 and turn off the bedroom light"
Of course I only need to do this because I might listen to loud music during the day and if I forget to turn it down it'll blast "OK TURNING OFF THE MASTER BEDROOM LIGHT" at max volume right as the family is going to bed.
There isn't much of a need to hook up a LLM to Alexa's output. "Turn on the lights" doesn't need a bunch of text in response.
It's the input side of these things that needs a ton of improvements and the use of an LLM could be basically invisible to the user, just "summarize what the user said into a command"; my most common issues with voice assistants are if I say things out of the expected order or change my mind. "Alexa turn on the bedroom lights actually turn on the office lights" or somesuch, or something like "Remind me to make a reservation for Friday tomorrow" not distinguishing that "make a reservation for Friday" is the text of the reminder and "tomorrow" is the desired time.
> It is stunning that Amazon wasn't the forerunner in using ChatGPT/LLM-based tech for Alexa
I'm a current scientist at Amazon. Amazon has little ability to invest in longer term research. While you have scope to pursue projects without immediate product application to some extent (though far less than you would at Alphabet/Meta/Microsoft), that is less valued than at the other Big Tech companies. There's also a far more bureaucratic hurdles to jump through to publish at Amazon. The lack of care for non-product research shows at every instance. Amazon also has extremely low -- or I should say non-existent standards for hiring scientists. If you're putting a person with a decent and non-existent pub record on the same level, that is not going to yield great results.
Luckily for Amazon, their bread and butter -- Ads via retail and AWS -- are not immediately threatened by LLMs.
Honestly if they had taken the billions they burnt on Alexa and put it in research they would be far far far better off today. But try explaining that to the bean counter mentality of executives that Bezos cultivated...
This is a good point. For an outsider it always seems that Amazon (or at least AWS) excels at deploying microservice solutions which plug into an ever growing ecosystem of interconnected offerings. I would have imagined that Alexa org is setup similarly and allows teams to deploy a solution such as an LLM and then given success be rewarded with growth.
ChatGPT and Alexa are not comparable at all. ChatGPT build a language model on a giant but static corpus. Alexa is trying to give mostly timely answers do work in real time and in response to an active environment.
At first, it seemed like a pretty useful device, but as time has gone on, Alexa has become increasingly problematic. It seems to have trouble understanding even basic commands. How many times have you asked Alexa to do something, only to have it respond with something completely irrelevant? Like it is regressing, which is curious.
The fact that Alexa is always listening, even when you're not talking to it, is straight up creepy. Also just plain annoying. Do we really need a robot voice constantly interrupting our conversations and playing cheesy music on command?
But perhaps the most frustrating thing about Alexa is its constant attempts to upsell us on Amazon products. Every time you ask it for something, it tries to sell you on a product that you didn't ask for. It's like having a pushy salesperson in your own home.
It blows my mind how anybody who bought this couldn't see that it was a device to sell amazon products.. still, yeah, perhaps it could have had a better balance to both be useful and both sell stuff..
It actually came free with another product that I wanted to buy. These companies throw away their spy devices at dirt cheap prices in exchange for an ongoing stream of your personal information.
I think the regression is the natural evolution of these kind of tools.
At first the range of acceptable command is pretty small and once we find some that work well we start to rely on them. But I´d suspect we´re not using the optimal incatations or pronounciations for those, and they work because there is no competing commands for the recognized sounds/words.
But as the repertory grows, fuzzy orders will be harder to reliably map to the same commands every time, and the more the voice assistant needs precise and spot on instructions.
At least that´s my mental image of why ours results are degrading over time. Interestingly on the Google side, ours got less sensitive over time, with fewer random activations but it also became harder to trigger it.
Note that saying this to Alexa only works for the next 24 hours, but you can create a custom Alexa routine to run this exact command every day (I run it on the Alexa in my garage since it responds audibly).
In case you didn't know, you can go to the app and look at recent activity to see what it thought you said. That can sometimes at least make why it did the wrong thing make some sense.
For example I have lights on the left and right side of the room named, rather unimaginatively, "left" and "right". It sometimes hears "turn on right" as "turn on light".
Sometimes though it definitely does just go nuts. It will occasionally start answering "what time is sunset" with "Sunset today in Washington DC will be <time>". I am in Washington but the one on the other side of the country, which is about 2300 miles west of DC and (more important for sunset time) about 10° north of DC.
It's not like it has decided I'm in DC. Asking it "what time is sunset here" will give me "Sunset today in <my_city> will be <time>". This usually lasts maybe a day.
> The fact that Alexa is always listening, even when you're not talking to it, is straight up creepy. Also just plain annoying.
In principle a home assistant should maintain a ring-buffer that listens for its "wake word" and then immediately discards any audio that doesn't match the trigger. Only audio that includes and follows that word should ever exit the device to the network. That reduces the privacy risk to accidental triggers, which happen sometimes (but relatively rarely.)
I've been told by security experts that this is (or rather was, as of 2018) the behavior of Alexa. I'd like to believe that we live in a world where Amazon/Apple/Google respect privacy and won't change that behavior via a software update (and if they do, they will be smacked hard by regulators.) I'm not 100% convinced we do live in that world, unfortunately.
Whenever I think about this, I'm reminded of that time someone asked Amazon for their personal data via gdpr, and Amazon sent them recordings from alexia of entire conversations and even clips from in the shower -- from an entirely family by mistake!
> In principle a home assistant should maintain a ring-buffer that listens for its "wake word" and then immediately discards any audio that doesn't match the trigger.
Should this be easy to verify with network monitoring? audio data is quite heavy. I'm assuming if audio information gets uploaded, it will be quite obvious on the data. We haven't heard about it, and I'm for that reason I don't think it actually does not do it (i.e. the ring buffer hypothesis is likely correct).
I doubt the guts of something like an Echo would be capable of containing any kind of speech to text model. It'll probably be a lot more of a concern in a few years when speech models like that are getting baked into cheap chips for consumer devices, though.
Dragon Naturally Speaking required extensive training on the voice of the user, and the result wasn't as accurate as modern machine-learning speech-to-text models. You had to watch it and actively correct it's results.
IMO, the accuracy of modern speech-to-text models is still nowhere near accurate enough. Maybe they should bring back the per-user training
I'm not as doubtful: iPhones have had on-device speech recognition since at least 2019, with iOS 13[1]. Amazon might have legal or policy reasons for not doing offline speech recognition, but the technology has been there for a while.
$29 is the sale price, not the BOM price. The entire point of "smart" home devices is to reduce consumer friction around revenue-driving actions; it would be surprising if Amazon were to prioritize making a profit on a $29 device rather than the sales and subscriptions it facilitates.
I don't know if it's every smart home vendor's goal to drive revenue. Apple sells their smart speakers for $100 or $300, which doesn't seem like a "loss leader" price, and it doesn't ask you to buy anything else. Their marketing page mentions what it can integrate with (smart locks, smart lights, etc.) but those things are notably not subscriptions. If you buy a smart light switch and a smart speaker, then you can just say "Hey Siri turn off the lights in the kitchen" and nobody ever gets any more money, and you didn't have to go downstairs and turn off the lights in the kitchen.
This seems like a very justified smart home. I am skeptical of all things proprietary technology, but Apple's stuff bothers me the least.
True, but note that they're using the tiny model. In my experimentation, you need at least the small model to get transcription I'd call "good", which is still a bit slower than you'd like on a moderately fast laptop from 2019.
That said, whisper is incredible and the era of very good local speech-to-text on moderate hardware is basically here, or will be in the next year.
I’m a big fan of Whisper and whisper.cpp but doing reliable wake word detection with any kind of reasonable latency on a Raspberry Pi is likely to be a poor fit and very bad experience.
The Whisper model operates on 30 sec speech chunks. Input audio has to be padded to that length. So you’re constantly going to be recording audio, padding, looking for wake word, and then activating full recording upon detection. All on padded 30 sec chunks looking back…
Then there is model size and availability. Whisper base or maybe even tiny could potentially give decent results for wake word detection but I’m skeptical. Wake words can be surprisingly tricky.
That’s just for wake word assuming you’re going to stream audio after, as reliably doing ASR and NLP to figure out speaker intent is far too challenging and time consuming to be done on Raspberry Pi class hardware in anything approaching response times that would be considered acceptable. Whisper does pretty well with relatively high noise/low quality speech and far-field microphones are amazing but I doubt that's enough to provide anything approaching Echo/Alexa quality in the real world.
This is a carefully scripted demo[0] showing it takes a whopping 15 seconds to wake, ASR the speech, and return the result. The average person could easily take their phone out of their pocket, unlock it, look for the weather app, and read the weather in less time.
This demo[1] claims "real time" but looking at the example videos it clearly isn't and the accuracy leaves a lot to be desired. This is with three threads on a Raspberry Pi 4.
I just tried asking Echo what the weather is like and it was so fast I had trouble timing it - somewhere around one second.
Recently I made some progress on efficiently detecting short voice commands (wake words) on RPi4 [0]. Checkout the "command" example in whisper.cpp and it's "Guided mode" operation. There are additional improvements on the way too.
As I said I really appreciate and respect all of the work you’re doing on whisper.cpp but when it comes to things like wake words and commands I have to think Whisper is just fundamentally the wrong tool for the job. Tiny is a 39m parameter model with fairly poor accuracy and high latency (without GPU) that just about maxes out a Raspberry Pi - all for a few a few very carefully pronounced words under ideal conditions (in this case).
That said, there isn’t much in the open source space (that I’ve found) that’s even remotely competitive with Alexa/Echo so I’m all for any efforts and attention in this area. Perfect is the enemy of good but this thread started off with people wondering if Whisper or anything based on it was close to Alexa/Echo for wake word activated assistant tasks. I think it’s very safe to say it isn’t.
Again, I really appreciate your work on making Whisper more accessible to the masses for local ASR - please don’t take this as criticism for your efforts. If anything I’ve been involved in open source projects and it’s frustrating when people try to jam a square peg in a round hole, only to come back and complain your labor of love didn’t work for them.
This is a many years old dev board with Wi-Fi, Bluetooth, a multi-mic ring array, RBG LEDs, speaker amplifier, SD slot, etc. Even this old version supports wake word with ring buffer, etc. That said this is also a (fun!) dev board - not a finished product. Newer versions based on the S3 are much more powerful.
Teardowns of Echo devices show much more capable hardware but at Amazon’s scale with their engineering, supply chain, and manufacturing resources they’re probably not losing all that much on Echo devices - although the Echo unit at Amazon has been posting serious losses but I’ve never seen breakouts of where those losses are coming from.
I whipped up a full-blown Rube Goldberg contraption that involved a lot of pcap filters on my router and kludgy hacks to Home Assistant for graphing, etc. I even integrated it to a separate LED on an esp32 sitting next to the Echo where the esp32 LED would activate when traffic to/from the Echo was detected.
It's a tiny and anecdotal sample but even when looking at historical graphs I never once "caught" the Echo device transferring any even remotely suspicious amount of data unless it was wake-word activated.
It doesn't seem like it would be impossibly hard to program "cocaine" as a quiet (ie, no visual or audio activity when activated with that particular word), save the audio on either side of that wake weird word, then wait for a "normal" activation that sends audio to the cloud, and then uploads the "cocaine" audio at the same time. Not remotely claiming any of the assistants do this, but if I were trying to program an evil voice assistant, it's what I would do to avoid detection. If the devices have the hardware for it, even imperfectly, run the transcription locally and it's even easier to stenographically hide just the text in the later upload. It doesn't need to perfect transcription to English either. Recording the phonemes for the reciever to reconstruct the language from would be enough. It's more important to get the names in the conversation anyway.
Bonus points for also recording+transcribing phone numbers/DTMF tones it hears and then uploading all of them to the NSA.
Can it be turned on so Amazon / government employees could listen for 'quality assurance' purposes. The fact that we are guessing tells us it's probably an easy area to exploit now or in the future.
That article is literally talking about amazon workers listening to queries asked to Alexa. Are you mad that they have a feedback loop that lets them improve? They’re not spying on you talking about your mundane life. No profit in that.
I’m fairly certain some Amazon engineer working on Alexa posted a message here saying they didn’t send anything unless the wake word was spoken. Since this is HN, I sort of took them at their word.
The trick is in what it interprets as its wakeword.
I remember looking at the recordings google stores back in 2015-2016. I had never used hey google (at all, ever), but it sure recorded a ton of irrelevant crap at my work and other conversations. Ill probably never use a personal assistant. Too invasive.
> I'm not 100% convinced we do live in that world, unfortunately.
Yes. I think also in principle we should not trust any advertising or other business deeply entrenched in surveillance capitalism to respect our privacy in any way, shape or form. Valuing privacy and having an Alexa doesn't combine. Having such device is not just creepy, it is foolish.
I feel differently about this. I think that privacy-conscious people should be comfortable availing themselves of common technologies that are in widespread use. And we should pressure firms to make those technologies safe.
What is really annoying is trying to get a specific song or playlist going in Spotify thats not in your library. For example I tried asking it for the Elder Scrolls Oblivion soundtrack for me to sleep to but it didn't understand my command until I added a playlist with the specific tracks I wanted to my library. This has happened a few times with other playlists. Very very annoying.
Yup. The one in my home got unplugged years ago due to general un-utility and constant listening. So I never got to the point of pushy salesperson impersonation — that just seems like a perfect example of enshittification... (the tech version of Jumping The Shark).
Yeah, absolutely hate notifications which are just marketing spam. Getting very close to chucking it but kids love it. They ask it endless questions, so that’s good for their curiosity.
Yeah, some things in life seem "stupid" or wrong to be the default when viewed from one side, and seem smart and the obvious thing to do when viewed from the opposite side.
I had Alexa devices almost since day 1. The notification spam started only in last couple of years. Which annoys me as that means they enabled a disabled setting.
But it is business and if it really annoyed it me, I would have gotten rid of it. Now I just ignore notifications, until I am stepping out of the room. Then I would Alexa for notifications and walk away.
We used to use Alexa on our Sonos devices, and immediately moved to the Sonos voice command tool when it was released.
I still find it wild that, in my experience at least, even the very first public release was just so much better than Amazon’s solution despite a staggering resource gap between the companies.
Beyond all the other issues, voice just isn't that convenient an interface.
This is especially true for Amazon, whose massive catalogue of products and services all but requires the speed available from reading and typing to navigate effectively.
Plus if you live with other people, voice interface use is disruptive.
We use todoist for our home shopping (grocery) list. Connecting this to Alexa has been one of the better quality of life improvements for us. After a bit of trial and error we figured out that you don't have to say "Alexa, add milk to my todoist shopping list", just "Alexa, add milk". So, for this specific purpose I don't think there is a better interface than voice.
It does occur to me that taking my shopping list on the phone to Tesco and putting things in a basket manually is probably not the future that Bezos had in mind for Alexa. But ultimately I don't trust Alexa / Amazon to actually do the buying.
> Beyond all the other issues, voice just isn't that convenient an interface.
This is an excellent point, and not only related to voice.
We often try to invent "smarter" ways to communicate, that are actually harder to use. Like having an app to control your toaster, which seems to be the next step, but simply having a physical device on it that takes 1 second to press is a lot faster, easier and feels more rewarding from the tactile feedback (I love pressing clicky buttons).
Voice is the same, turning up the volume on your TV is easier by simply pressing a button on the remote (assuming you have it in reach) than saying a voice command. This is also true in human interactions, sometimes it's easier and faster to just do something by yourself than having to ask someone to do that thing for you. Also, voice commands have to be perfect in the same way physical buttons are, whenever you press them the result is always the same. It's hard to keep using an unreliable input method, that has only 60% chances of doing what you expect.
The more commands they add they worse it gets at all commands. This is the problem with an extensible ecosystem -- all voice parsing is affected by every change to the voice parsing, so as the complexity goes up, it gets shittier. I'm holding out for some really cool chat GPT-based assistant, preferably open source, that I could run on a raspberry pi or something. Even better would be something based on one of the open alternatives to chatgpt
I am like most people here. I have a bunch and use them for radio, kitchen timers, music (although less and less because the experience of getting a pitch for Amazon music every time is annoying, if it even plays what I want at all).
I also have it hooked up to a bunch of lights, the garage door opener. I have an Alexa in the garage as well for when I am wrenching on things.
I do not have one in the office, or the bedroom.
If they break, who cares. I may eventually try and create a custom web app to control all these things instead of using voice.
I hate the back talking and advertising. I already bought the device, I’m a prime member, why can’t you just tell me the temperature and shut the fuck back up?
I hate the stupid thing but my wife has a little more patience for her than me.
I have a question about Siri that I don't want to risk testing...
If you tell her "OH FUCK SIRI THERES SOMEONE IN THE HOUSE CALL 911 OH SHIIIIT!!!!"
Will siri use her chastizing tone not to speak to her that way and ignore you?
I feel like it would be hillarious if she did. She's also dumb as a bag of rocks in my experience so I suspect she would. Bitch can't even reliably play podcasts.
She's very pleasant if you tell her "thank you you classy bitch." But it has to be the entire proper "thank you." If you say "thanks" she gets offended.
Things like shopping on Alexa are downstream of Amazon’s quality/counterfeit problems. I can never say something like “Alexa, buy triple A batteries” because I can’t be confident of receiving authentic product at a fair price.
I never saw any use for voice assistants for most of the population. Yes, there are a few tasks in the kitchen that can be automated like setting a timer but that didn’t need the huge server side infrastructure they built out. Just the NLP to translate text and extract the timer info would have sufficed.
However, I have a feeling they will do better in the future with the conversational AI like GPT coming up. They could integrate it in the backend and provide more answers. I still don’t think if it will be monetizable though.
I had two Echo Dots 2017/2018. I needed them to do two things: dimm/switch my lights in kitchen and living room and play music.
The light thing worked 80% of the time, I'd say.
The music thing never worked. I mainly listen to tango/vals/milonga from the 1930's to 1950's.
Spotify has all of that stuff but Alexa is unable to find it.
E.g. saying: "Play tango from D'Arienzo on Spotify" would yield Alexa playing some latin music, usually from the last 20 years. On Spotify. I mean it got that part right ...
I never once managed to get it to play tango. After the light thing didn't improve notably over the course of those two years I sold the Dots end of 2018.
My parents have two Alexa devices in their home; one in the living room and one upstairs in a back bedroom. I once asked Alexa to play Christmas music from in the living room, and rap music started playing upstairs.
IMO the ultimate feature of Alexa is setting timers… but I feel like you could do this with a $10 device without internet connection… hell, it might have better voice recognition if it only needed to understand a few commands for setting the timers.
I was one of the earliest Alexa users, I was invited to buy the first Echo before it was generally available. It was handy mostly as a kitchen timer, occasionally for random questions when my hands were too busy/gooey to whip out my phone (e.g., "What is the safe cooking temperature for chicken thighs?").
But then one day the LED ring on the top of the Echo was flashing yellow, and when I asked Alexa why, it responded with an ad. I haven't used it since.
I was a software developer in that division in 2012 so I got to witness the early days of optimism.
After tasting success with Kindle Amazon went full throttle with their hardware investment. They were optimistic about replicating Kindle success so they threw the entire kitchen sink at it. The division hired 1000+ in product and development in a matter of weeks. It was truly heady days but equally chaotic as no one had a clue as to what were they supposed to be working on. Managers and execs promptly built their empires took promotions etc.
However it was not clear to anyone how would they monetize. Whenever some lowly developer asked the answer was standard, Amazon’s model is to grow big first and then figure out making money.
But for the interest rate hikes perhaps they could have continued down the path, who knows.
All that said, if I owned any of Alexa devices I’d be planning on sunsetting them. They will put the entire infrastructure on a strictly lights on mode. So expect things to break on the edges, at first.
I think the Alexa hardware has more or less reached maturity. And even the software side is probably going into regroup and harden mode. It's no longer going to be your digital butler but a smart speaker that can play music and set alarms and tell you the weather and maybe turn the lights on. Which is altogether a much smaller vision than 2012 but something extremely useful and with a still growing base of users.
> Amazon’s model is to grow big first and then figure out making money
I think this is the case with most start-ups nowadays. Attention is considered to be the most valuable thing you can get from a person, the assumption being is that you have the attention, you should easily be able to monetize it somehow. Also, the number of users is what investors mostly care about in the early stages and the "traction" is one of the highest valued KPIs.
I do think this is wrong though, from many, many points of view. The worst way in which is wrong is that we end up not creating good products or products that provide value, but only addicting products or those that provide instant-gratification and are purely focused on the short-term benefits.
> Attention is considered to be the most valuable thing you can get from a person, the assumption being is that you have the attention, you should easily be able to monetize it somehow.
This sounds like planned extortion.
"Look at this nice thing you are enjoying. Give us money or it goes away"
It feels a bit like it. Because the point is to first get you hooked/trapped, lure you in, and then find ways to take advantage of you.
It was the same with Google Search. It was useful and free. Now that they have everyone trapped, they can exploit the s*t out of our attention. They can put 5 ads in search and video ads every minute on YouTube and enough people will still use the service, as the users are already trapped, there are no alternatives due to them creating a monopoly.
"Predatory pricing is a pricing strategy, using the method of undercutting on a larger scale, where a dominant firm in an industry will deliberately reduce the prices of a product or service to loss-making levels in the short-term.[1] The aim is that existing or potential competitors within the industry will be forced to leave the market, as they are unable to effectively compete with the dominant firm without making a loss.[2] Once competition has been eliminated, the dominant firm now having a majority share of the market can raise its prices to monopoly levels in the long-term to recoup its losses."
217 comments
[ 5.4 ms ] story [ 245 ms ] threadnice thought; but does anyone else have a use for them either?
I never understood what the "voice assistant" hype was about, the advertised benefits looked ... non beneficial; the examples I saw in peoples homes didn't appear to work when they tried to use them.
Was it worth that much money and effort to have microphones in people's houses?
I think I bought one when they first came out, then I got three for free when I bought other things.
I'm an Apple user, and by extension, Homekit and Siri.
In a household full of kids and an all-around crazy schedule, having HomePods scattered amidst my house has proven useful; beyond just being great speakers for music.
I use them for everything from adding/completing items on my Reminders list. Controlling the home (lights, timers, etc.). Asking questions (this one is very hit/miss with Siri, but she's getting a lot better). And reading/sending messages.
Is it absolutely necessary? No. But when you're lugging around little ones throughout the house, fixing lunches, dinner, etc. -- they can make life a little easier. Sometimes.
[1]: https://www.amazon.com/l/9187220011
We have at least 8 Echos around the house, with clear reasons for a couple more, but Amazon simply isn't interested in the actual problems that the product solves, or the use cases customers care about.
They're completely backwards, here. If that means the whole thing collapses and we have to find another solution, great. Products and companies who work this way need more failures.
What it could/should have been is the end cap for the Amazon brand in homes. A loss leader itself, but a constant reminder of the value of the company and the desire to keep spending money there due to the convenience it provides.
What the article makes it sound like they want is something more tangible. Something that quantitatively generates direct ROI. They'll never get that, or rather, that number will never add up to one that justifies Alexa's existence.
Folks need to be discerning with tech purchases. If it isn't in a company's interest to make your experience better (and it is in the company's interest to make your experience worse, say, with ads) when push comes to shove they will prioritize profits over your customer experience. Instead of investing in rotten platforms like Alexa, I wish more people invested in open source and non-exploitative solutions where the incentives align to improve user experience, not destroy it.
Imagine a world where you could self-host or pay a small fee to cloud host your own smart assistant, tweaked to your preferences, not in any way linked into ad sales for Amazon or Google. That's where I'd like to see us go.
AWS and their massive logistics platform will support this terrible management for awhile but it will continue to destroy the brand.
“By the way, sign up for Amazon music.”
“Tip your Amazon driver.”
No thanks.
I wish they had actually done something with the tech rather than promptly handing it over to the marketing team once it got some users. I’m trying to think of a single major improvement to the experience since it first launched a decade ago, but am coming up empty.
Ha ha, what? Is that a real thing?
As pointed in the article, the move to monetization happened once the writing was on the wall that it wouldn´t become significantly different than what we have now.
Also they´ve always been priced so aggressively, I´m kinda glad I got the original puck at such a low price, to see first hand where the technology was and what to expect from it.
When uncorrected I have severe vision impairment (-10 + astigmatism). Without exaggeration I have trouble finding my glasses unless the room is well lit, and even then I have some panic if they aren't exactly where I think they are. I have backup glasses in a known place because if my glasses slide off the nightstand or something else I'm essentially completely blind.
At night, in a dark (or even dim) room, it's much worse. So, for me, the Echo does it's job just because I can tell it to turn on the lights. I imagine for other people with more significant accessibility issues it can be very useful - even though it drives all of us crazy with "Oh, by the way, did you know I can...".
It's also nice for a few other bedtime/dark room related things like lying down to go to sleep and get that panic "OH WAIT! Is that 7am meeting tomorrow? Better check the calendar and set my alarm accordingly." Or when I wake up at some point in the dark and need to decide if I should start my day I can ask Echo what time it is without struggling to find my phone and be blinded (even more so) by the backlight as I hold it (literally) 1-2" from my eye.
It can be really annoying but for me these (admittedly minor) use cases keeps Echo in my house (bedroom).
Server in China? I use Home Assistant with total local control of all of my smart home devices. On a separate VLAN and SSID. With no internet access. Echo is on a different VLAN and uses the (excellent) Home Assistant integration. AWS data center of course, but see my other comments[0] about the near-total uselessness of any local devices and implementations for wake word detection, ASR, and NLP. Speaking of Home Assistant, see my other comment about the graphing and notification system I built for Echo/Alexa[1].
Yes, I remember commercials for "The Clapper". I've been in bathrooms with motion activated lights that turn off while you're on the toilet (you know, the move where everyone starts frantically waving their arms).
Not only are these solutions also imperfect, you conveniently ignored my other use cases about checking the time in the dark, checking my calendar, setting alarms, etc.
I like my Echo. I like my "smart home" configuration and I've minimized the security issues and blast radius. I understand your concern and it's not without merit but you have the choice of whether or not to use internet connected devices in your environment. So do I.
[0] - https://news.ycombinator.com/item?id=35041435#35051644
[1] - https://news.ycombinator.com/item?id=35041435#35051766
This is a terrible idea. NLP, with or without LLMs, isn't remotely close to being able to interpret commands in someone's house. And who would want a non-deterministic voice assistant anyway?
Imagine you say, "Alexa, turn up the air conditioning," and Alexa interprets this to mean you want to increase the temperature. There are uncountable and unforeseeable failure modes.
Note that this happens even when talking to humans with near-perfect NLP/brains.
The major flaw with voice assistants in general is that they only make sense if they have a pretty narrow and pre-defined universe of accepted commands, which is incredibly confusing to someone who mistakes them for actual NLP.
It's just a really, really bad UI. I use mine all the time, but only to set timers and turn lights on or off.
- why ChatGPT's style of "convincing bullshit" is a bad fit for a user interface like Alexa
- failure modes of NLP in UIs
- why NLP still won't save Alexa if it gets to a human level of accuracy because humans aren't accurate enough
If you have any actual knowledge or argument other than, "You're ignorant, [technology that doesn't exist] is going to fix Alexa," then let's hear it.
My knowledge is that I know how these systems work, and why its so hard to get them to work properly.
It doesn't matter if you're Amazon's ML expert or the world's best surgeon or a barista if all you do is make an unsubstantiated statement and give your credentials.
So here's my process with Google Asisstant:
- guess what it will understand through trial and error
- use that command pattern, like "set [something] to [level]%", for a few months
- try it again one day, and suddenly it doesn't work
It makes you scared to trust it in almost any situation.
"up" and "off" sound similar too, and I've had assistants confuse them often.
Voice control is a miserable interface, and there are few applications which I would endeavor to use it. Time, weather, math, and maybe a trivial fact queried from Wikipedia.
I found myself shouting at these assistants while they try to interpret what I say mid sentence.
The novelty wore off very quickly. But voice seems like a very powerful interface if you can solve privacy and usability issues. Seems like we haven’t tried very hard at this yet.
In comparison Google and Apple are competitive in so much more (in particular calendaring, contacts, music/videos).
I think Amazon missed the boat in not coming with an open standard and get cosier with independant makers like Sonos or Ikea etc. Basically they could have pulled an android for the voice assistant ecosystem, but preferred to try to build another walled garden.
The only reason that I haven't moved to homepods is because apple can't figure out how to do shopping lists that any one in a family can access.
"Alexa - Add bread" takes one second. "Hey Siri - add bread" takes upwards of twenty seconds for the homepod to figure out who you are, and if you have permission, and add it to the reminder list. I really would just throw away the alexa's if apple could get that one thing figured out.
OR, I can say things like, "Alexa, what is 23 and 1/16th minus 2.1625" And she'll spit it out as a decimal and clarify as a fraction if I want.
It's amazing.
It doesn't really follow though, I think what's more important is the length of wood you start with.
It's easier to divide 3m into 2 x 1.5m or 3 x 1m than it is to divide 9'10" into 2 x 4'11" or 3 x 3'3.333333333"
If your blade was e.g. 2mm thick instead of 1/16", I think it would be easier to manage.
What's the name of the 12 ft unit? Why isn't the mile equal to 12×12×12×12 ft? Why isn't an lb equal to 12 oz?
It is; that's called Troy weight.
When would you want to divide an octave into two or three intervals like that?
The normal way to divide an octave in "half" is at the fifth, which is a ratio of 3:2. If you instead go six half-steps up from the tonic, you'll be a half-step short of a fifth, at fi instead of so, with an irrational ratio of √2:1. That's the opposite of what you want. [From wikipedia: The tritone is a restless interval, classed as a dissonance in Western music from the early Middle Ages through to the end of the common practice period. This interval was frequently avoided in medieval ecclesiastical singing because of its dissonant quality. The first explicit prohibition of it seems to occur with the development of Guido of Arezzo's hexachordal system]
If you divide the octave into "quarters" by going up three half-steps, you get a minor third.
(Going for "thirds" by going up four half-steps will take you to mi and form a third [the name of the chord is not related to the idea of dividing an octave into three equal parts; it's called a "third" because "do re mi" is three notes], which is fine. But it's pretty clearly just a coincidence that that works out; the methodology of "there are twelve half-steps in the octave, so I'll divide that into equal parts and see what I get" just isn't related to the way you compose music.)
So, no fault of NASAs, or of the metric system.
For Lockheed to use anything but metric system in the modern world is bizarre.
I can say “Alexa, add eggs” and eggs are added to my grocery list in AnyList. On Siri I have to say something like “Hey Siri, using AnyList add eggs to my grocery list”. The Siri prompt feels like I’m speaking a command line.
That said, my 11yo daughter said to me yesterday "why don't they just add AI to Siri" after failing to get Siri to tell her how long a timer was originally set for (Siri would only tell you how much time is left on the timer).
Someone did build a crazy ChatGPT-to-Siri bridge through Automations which I haven't tried yet but plan to at some point...
Has anyone interacted with chatgpt in a siri/alexa voice interaction for an extended period of time? I too have wondered when AI arrives in these platforms. It could be massively disruptive.
The interesting thing is that with the new Whisper models these things can be done with minimal internet or no internet at all. I would pay good money for an offline only home assistant. Maybe it could receive updates if I specifically asked for them.
I remember him saying that the additional revenue generated by what they did couldn't pay for even a single developer on their team. And it wasn't going anywhere. But hey, management want it to exist so they're all getting paid to do it.
Meanwhile, my part of the company at the time (Fulfillment) had projects with 7, 8, maybe even 9 figure savings estimated and we just couldn't find the developers for it because all the new hires were sucked into the Alexa org.
I'm glad they're cutting their losses, but Amazon ought to have done it five years ago when it was clear this was never going to earn them additional revenue.
Have you tried the WebMD app/skill? https://www.webmd.com/webmd-medreminders-syncing
For medication reminders and dosing, we use Hero Health devices. https://herohealth.com/
2. Voice interface is often slower than preexisting solutions. It's faster to pull a phone out of a pocket, unlock it, and load a weather app then it is to ask Alexa what the weather is. Loading the weather app will show you more information.
3. Alexa is kind of clumsy. It will trigger accidentally. It will not understand you.
4. ChatGPT is clearly better at answering questions. It's probably better at running commands. It doesn't directly compete with Alexa, but it highlights how outdated Alexa really is.
5. Alexa was awesome for some things. It was awesome for vision impairment. It was awesome for hands free situations, when you cannot spare the mechanical efforts of using another device or the visual attention.
Yes. This is also why assistants with a screen are so much more useful. Google´s hubs aren´t blazing fast either (and get slower over time...), but getting the full visual information at the same time as the voice answer helps a lot.
It is stunning that Amazon wasn't the forerunner in using ChatGPT/LLM-based tech for Alexa. A mistake in the tens, if not hundred billion dollars when looking at the price MS paid for OpenAI.
Why aren't they hooking up a LLM right now?
And that's ignoring the fact that all the publicly accessible LLMs have been "jailbroken" and you can get them to say all manner of wild things, and Amazon is definitely aware that all it takes is one or two viral clips of Alexa saying something unhinged for their sales to start dropping, and the tech isn't currently sufficiently understood for them to avoid that.
I am also much less concerned about jailbroken LLMs, because Alexa is already so broken it can't get much worse. Normal users don't seem to care much about potential misuses (compare with search autocomplete biases of Google) as long as a service works for them. But unless you widely deploy a tool, you won't uncover how it is abused (see Sidney's launch).
ChatGPT was, while less widely deployed, being abused (very publicly) in very similar manners to the way Sydney was; Microsoft could have learned about those abuses by paying attention to the public information about ChatGPT.
I really struggle to think how it would have worthwhile. My problem with smart speakers isn't that they aren't smart enough - it's that for anything slightly more complex, my phone has 100x better UX and information density. An Alexa LLM sounds cool in a sci-fi sort of way, but still feels like it would be useless.
“Alexa, be my friend and talk with me for a while, ok?”
Of course I only need to do this because I might listen to loud music during the day and if I forget to turn it down it'll blast "OK TURNING OFF THE MASTER BEDROOM LIGHT" at max volume right as the family is going to bed.
It's the input side of these things that needs a ton of improvements and the use of an LLM could be basically invisible to the user, just "summarize what the user said into a command"; my most common issues with voice assistants are if I say things out of the expected order or change my mind. "Alexa turn on the bedroom lights actually turn on the office lights" or somesuch, or something like "Remind me to make a reservation for Friday tomorrow" not distinguishing that "make a reservation for Friday" is the text of the reminder and "tomorrow" is the desired time.
I'm a current scientist at Amazon. Amazon has little ability to invest in longer term research. While you have scope to pursue projects without immediate product application to some extent (though far less than you would at Alphabet/Meta/Microsoft), that is less valued than at the other Big Tech companies. There's also a far more bureaucratic hurdles to jump through to publish at Amazon. The lack of care for non-product research shows at every instance. Amazon also has extremely low -- or I should say non-existent standards for hiring scientists. If you're putting a person with a decent and non-existent pub record on the same level, that is not going to yield great results.
Luckily for Amazon, their bread and butter -- Ads via retail and AWS -- are not immediately threatened by LLMs.
Honestly if they had taken the billions they burnt on Alexa and put it in research they would be far far far better off today. But try explaining that to the bean counter mentality of executives that Bezos cultivated...
The fact that Alexa is always listening, even when you're not talking to it, is straight up creepy. Also just plain annoying. Do we really need a robot voice constantly interrupting our conversations and playing cheesy music on command?
But perhaps the most frustrating thing about Alexa is its constant attempts to upsell us on Amazon products. Every time you ask it for something, it tries to sell you on a product that you didn't ask for. It's like having a pushy salesperson in your own home.
At first the range of acceptable command is pretty small and once we find some that work well we start to rely on them. But I´d suspect we´re not using the optimal incatations or pronounciations for those, and they work because there is no competing commands for the recognized sounds/words.
But as the repertory grows, fuzzy orders will be harder to reliably map to the same commands every time, and the more the voice assistant needs precise and spot on instructions.
At least that´s my mental image of why ours results are degrading over time. Interestingly on the Google side, ours got less sensitive over time, with fewer random activations but it also became harder to trigger it.
> Ok, I will snooze my suggestions for now.
For example I have lights on the left and right side of the room named, rather unimaginatively, "left" and "right". It sometimes hears "turn on right" as "turn on light".
Sometimes though it definitely does just go nuts. It will occasionally start answering "what time is sunset" with "Sunset today in Washington DC will be <time>". I am in Washington but the one on the other side of the country, which is about 2300 miles west of DC and (more important for sunset time) about 10° north of DC.
It's not like it has decided I'm in DC. Asking it "what time is sunset here" will give me "Sunset today in <my_city> will be <time>". This usually lasts maybe a day.
In principle a home assistant should maintain a ring-buffer that listens for its "wake word" and then immediately discards any audio that doesn't match the trigger. Only audio that includes and follows that word should ever exit the device to the network. That reduces the privacy risk to accidental triggers, which happen sometimes (but relatively rarely.)
I've been told by security experts that this is (or rather was, as of 2018) the behavior of Alexa. I'd like to believe that we live in a world where Amazon/Apple/Google respect privacy and won't change that behavior via a software update (and if they do, they will be smacked hard by regulators.) I'm not 100% convinced we do live in that world, unfortunately.
https://www.businessinsider.com/amazon-sends-alexa-shower-re...
[0] - https://www.amazon.com/gp/help/customer/display.html?nodeId=....
Should this be easy to verify with network monitoring? audio data is quite heavy. I'm assuming if audio information gets uploaded, it will be quite obvious on the data. We haven't heard about it, and I'm for that reason I don't think it actually does not do it (i.e. the ring buffer hypothesis is likely correct).
IMO, the accuracy of modern speech-to-text models is still nowhere near accurate enough. Maybe they should bring back the per-user training
[1]: https://developer.apple.com/documentation/speech/sfspeechrec...
This seems like a very justified smart home. I am skeptical of all things proprietary technology, but Apple's stuff bothers me the least.
That said, whisper is incredible and the era of very good local speech-to-text on moderate hardware is basically here, or will be in the next year.
The Whisper model operates on 30 sec speech chunks. Input audio has to be padded to that length. So you’re constantly going to be recording audio, padding, looking for wake word, and then activating full recording upon detection. All on padded 30 sec chunks looking back…
Then there is model size and availability. Whisper base or maybe even tiny could potentially give decent results for wake word detection but I’m skeptical. Wake words can be surprisingly tricky.
That’s just for wake word assuming you’re going to stream audio after, as reliably doing ASR and NLP to figure out speaker intent is far too challenging and time consuming to be done on Raspberry Pi class hardware in anything approaching response times that would be considered acceptable. Whisper does pretty well with relatively high noise/low quality speech and far-field microphones are amazing but I doubt that's enough to provide anything approaching Echo/Alexa quality in the real world.
This is a carefully scripted demo[0] showing it takes a whopping 15 seconds to wake, ASR the speech, and return the result. The average person could easily take their phone out of their pocket, unlock it, look for the weather app, and read the weather in less time.
This demo[1] claims "real time" but looking at the example videos it clearly isn't and the accuracy leaves a lot to be desired. This is with three threads on a Raspberry Pi 4.
I just tried asking Echo what the weather is like and it was so fast I had trouble timing it - somewhere around one second.
[0] - https://youtube.com/watch?v=Aor6CFkcWzU&si=EnSIkaIECMiOmarE
[1] - https://github.com/ggerganov/whisper.cpp/discussions/166
[0] https://twitter.com/ggerganov/status/1602759833312456704
That said, there isn’t much in the open source space (that I’ve found) that’s even remotely competitive with Alexa/Echo so I’m all for any efforts and attention in this area. Perfect is the enemy of good but this thread started off with people wondering if Whisper or anything based on it was close to Alexa/Echo for wake word activated assistant tasks. I think it’s very safe to say it isn’t.
Again, I really appreciate your work on making Whisper more accessible to the masses for local ASR - please don’t take this as criticism for your efforts. If anything I’ve been involved in open source projects and it’s frustrating when people try to jam a square peg in a round hole, only to come back and complain your labor of love didn’t work for them.
That said, pretty much ($35 with shipping qty 1):
ESP32-Korvo Development Board https://a.co/d/fmaQfuL
This is a many years old dev board with Wi-Fi, Bluetooth, a multi-mic ring array, RBG LEDs, speaker amplifier, SD slot, etc. Even this old version supports wake word with ring buffer, etc. That said this is also a (fun!) dev board - not a finished product. Newer versions based on the S3 are much more powerful.
Teardowns of Echo devices show much more capable hardware but at Amazon’s scale with their engineering, supply chain, and manufacturing resources they’re probably not losing all that much on Echo devices - although the Echo unit at Amazon has been posting serious losses but I’ve never seen breakouts of where those losses are coming from.
I whipped up a full-blown Rube Goldberg contraption that involved a lot of pcap filters on my router and kludgy hacks to Home Assistant for graphing, etc. I even integrated it to a separate LED on an esp32 sitting next to the Echo where the esp32 LED would activate when traffic to/from the Echo was detected.
It's a tiny and anecdotal sample but even when looking at historical graphs I never once "caught" the Echo device transferring any even remotely suspicious amount of data unless it was wake-word activated.
Transmission of customer's private conversations is now delayed until the wake word is spoken. See Bug 39482: "Suspicion aroused"
Bonus points for also recording+transcribing phone numbers/DTMF tones it hears and then uploading all of them to the NSA.
Thousands of Amazon Workers Listen to Alexa Users' Conversations https://time.com/5568815/amazon-workers-listen-to-alexa/
I remember looking at the recordings google stores back in 2015-2016. I had never used hey google (at all, ever), but it sure recorded a ton of irrelevant crap at my work and other conversations. Ill probably never use a personal assistant. Too invasive.
Yes. I think also in principle we should not trust any advertising or other business deeply entrenched in surveillance capitalism to respect our privacy in any way, shape or form. Valuing privacy and having an Alexa doesn't combine. Having such device is not just creepy, it is foolish.
Ugh
Ridiclous that it's on by default
But it is business and if it really annoyed it me, I would have gotten rid of it. Now I just ignore notifications, until I am stepping out of the room. Then I would Alexa for notifications and walk away.
I still find it wild that, in my experience at least, even the very first public release was just so much better than Amazon’s solution despite a staggering resource gap between the companies.
This is especially true for Amazon, whose massive catalogue of products and services all but requires the speed available from reading and typing to navigate effectively.
Plus if you live with other people, voice interface use is disruptive.
It does occur to me that taking my shopping list on the phone to Tesco and putting things in a basket manually is probably not the future that Bezos had in mind for Alexa. But ultimately I don't trust Alexa / Amazon to actually do the buying.
This is an excellent point, and not only related to voice.
We often try to invent "smarter" ways to communicate, that are actually harder to use. Like having an app to control your toaster, which seems to be the next step, but simply having a physical device on it that takes 1 second to press is a lot faster, easier and feels more rewarding from the tactile feedback (I love pressing clicky buttons).
Voice is the same, turning up the volume on your TV is easier by simply pressing a button on the remote (assuming you have it in reach) than saying a voice command. This is also true in human interactions, sometimes it's easier and faster to just do something by yourself than having to ask someone to do that thing for you. Also, voice commands have to be perfect in the same way physical buttons are, whenever you press them the result is always the same. It's hard to keep using an unreliable input method, that has only 60% chances of doing what you expect.
Never once had it tried!
I also have it hooked up to a bunch of lights, the garage door opener. I have an Alexa in the garage as well for when I am wrenching on things.
I do not have one in the office, or the bedroom.
If they break, who cares. I may eventually try and create a custom web app to control all these things instead of using voice.
I hate the stupid thing but my wife has a little more patience for her than me.
Siri on the other hand, never seems to miss.
If you tell her "OH FUCK SIRI THERES SOMEONE IN THE HOUSE CALL 911 OH SHIIIIT!!!!"
Will siri use her chastizing tone not to speak to her that way and ignore you?
I feel like it would be hillarious if she did. She's also dumb as a bag of rocks in my experience so I suspect she would. Bitch can't even reliably play podcasts.
https://images-na.ssl-images-amazon.com/images/I/51ecGcPE54L...
However, I have a feeling they will do better in the future with the conversational AI like GPT coming up. They could integrate it in the backend and provide more answers. I still don’t think if it will be monetizable though.
The light thing worked 80% of the time, I'd say.
The music thing never worked. I mainly listen to tango/vals/milonga from the 1930's to 1950's.
Spotify has all of that stuff but Alexa is unable to find it.
E.g. saying: "Play tango from D'Arienzo on Spotify" would yield Alexa playing some latin music, usually from the last 20 years. On Spotify. I mean it got that part right ...
I never once managed to get it to play tango. After the light thing didn't improve notably over the course of those two years I sold the Dots end of 2018.
But then one day the LED ring on the top of the Echo was flashing yellow, and when I asked Alexa why, it responded with an ad. I haven't used it since.
After tasting success with Kindle Amazon went full throttle with their hardware investment. They were optimistic about replicating Kindle success so they threw the entire kitchen sink at it. The division hired 1000+ in product and development in a matter of weeks. It was truly heady days but equally chaotic as no one had a clue as to what were they supposed to be working on. Managers and execs promptly built their empires took promotions etc.
However it was not clear to anyone how would they monetize. Whenever some lowly developer asked the answer was standard, Amazon’s model is to grow big first and then figure out making money.
But for the interest rate hikes perhaps they could have continued down the path, who knows.
All that said, if I owned any of Alexa devices I’d be planning on sunsetting them. They will put the entire infrastructure on a strictly lights on mode. So expect things to break on the edges, at first.
I think this is the case with most start-ups nowadays. Attention is considered to be the most valuable thing you can get from a person, the assumption being is that you have the attention, you should easily be able to monetize it somehow. Also, the number of users is what investors mostly care about in the early stages and the "traction" is one of the highest valued KPIs.
I do think this is wrong though, from many, many points of view. The worst way in which is wrong is that we end up not creating good products or products that provide value, but only addicting products or those that provide instant-gratification and are purely focused on the short-term benefits.
This sounds like planned extortion.
"Look at this nice thing you are enjoying. Give us money or it goes away"
It feels a bit like it. Because the point is to first get you hooked/trapped, lure you in, and then find ways to take advantage of you.
It was the same with Google Search. It was useful and free. Now that they have everyone trapped, they can exploit the s*t out of our attention. They can put 5 ads in search and video ads every minute on YouTube and enough people will still use the service, as the users are already trapped, there are no alternatives due to them creating a monopoly.
I think this is the definition of Vendor lock-in: https://en.wikipedia.org/wiki/Vendor_lock-in
Or maybe Antitrust: https://en.wikipedia.org/wiki/Competition_law
Leading to Predatory pricing: https://en.wikipedia.org/wiki/Predatory_pricing
"Predatory pricing is a pricing strategy, using the method of undercutting on a larger scale, where a dominant firm in an industry will deliberately reduce the prices of a product or service to loss-making levels in the short-term.[1] The aim is that existing or potential competitors within the industry will be forced to leave the market, as they are unable to effectively compete with the dominant firm without making a loss.[2] Once competition has been eliminated, the dominant firm now having a majority share of the market can raise its prices to monopoly levels in the long-term to recoup its losses."
I also however changed the trigger from Alexa to Computer.
Which makes watching episodes of Star Trek... interesting.
What a sad goal. The bond produced by parents reading a bedtime story to their children is difficult to put into words. But it is real and enduring.