Amazon Says It Has Over 10k Employees Working on Alexa, Echo | Hacker News Reader

Amazon Says It Has Over 10k Employees Working on Alexa, Echo

wsj.com

129 pointsdeegles194 comments

And those employees are doing a rather poor job on the AI side of things. Hardware itself is OK, it hears me better than its Google counterpart thanks to the phased microphone array. But both speech recognition and natural language understanding are piss poor and other than ordering stuff from Amazon and setting the time it's not useful for much, nor does it understand me in a sufficiently high percentage of interactions, nor is it in any way "conversational", the way Google Home is. Alexa app is also lackluster and feels like it's held together by duct tape, like all other Amazon apps.

For me the quality of AI in this case outweighs the creep factor of Google. FWIW, I replaced my Echo with a Google Home Mini.

Full disclosure: I'm an ex-Googler who nevertheless thinks that ads are cancer, and pervasive tracking shouldn't be legal.

I have a few Echos that we bought a year ago over Christmas. We use it to play music and control home automation. It works relatively well. My husband has a thick russian accent, and with him it struggles sometimes.

I don't use the app and so have no use for that. What are other things the Google home does better? Like help me understand the use case. You say Google Home is more "conversational", but how?

You can mostly speak to it like you'd speak to a human, and in the vast majority of cases it will understand. No need for rigid commands. You can also ask clarifying questions and it will keep context. Its speech synthesis and response language is more "human" and "cheerful". I just wish it also had an echo-like microphone array. When it's playing music it sometimes can be difficult to get it to hear commands.

Previous Echo models had the same issue with music. The latest Echo model will lower the music volume when it hears the wake word.

Alexa can make phone calls that dial a landline number, which is not an option with Google.

> The latest Echo model will lower the music volume when it hears the wake word.

I got my Echo a while ago before the current generation was even announced. It would always lower the volume as soon as it hears the wake word.

Then it was only Echo v1 which didn't have this feature.

Hrm? Google Home can make calls (uses the google voice # of the account paired to it).

Echo can use a landline for the outbound call.

How is this better than the Google Home/Google Voice integration? Seems significantly more complex, and reliant on the end user having a working home phone line

Echo can dial (VOIP) without a landline, same as Google Voice/Home.

If you have a landline and want your outbound Echo calls to have the same caller ID as the landline (so that people can return/screen calls), then Echo is the only option for voice dialing.

I’ve only used an Echo and I’m definitely interested in hearing how the user experience of the Google Home is better. How is it more conversational?

You can get most of the experience with the Assistant app which is available for both ios and android.

Search-related portions are available through that "microphone" icon (which nobody clicks) on Google.com front page. The main trick that makes it conversational is you can usually ask it a question about somebody or something, and then ask follow up questions using pronouns and without referencing the entity in question directly, much like you would with a human. Try e.g. "what's the population of Seattle", then "when was it founded", then "in what state is it", etc. Eventually this breaks down, but it mostly works on the first and second follow-up question. I.e. "is Costco open right now", then "how long would it take me to get there".

From an AI practitioner: this is god-mode level work.

> Search-related portions are available through that "microphone" icon (which nobody clicks)

I've noticed no one my age clicks this. My son and his friends as well as a friends nieces and nephews who live a couple thousand miles away from us use Google in a completely different way. They don't go to google.com and type in a search query, they click the microphone and ask google their question directly. They are also more likely to use the assistant on their devices than any of the adults I know. I think this is one of those generational shifts as they don't need to be able to type in their queries in the best way to get the results they are looking for, they can just use natural language for the most part.

Yes, a significant fraction of _mobile_ searches on Google are voice. According to Google it was 20% in 2016, the number has likely gone up significantly since then. Even old farts like myself are getting on the bandwagon, this is no longer early-adopter stuff.

https://www.google.com/search?q=what+fraction+of+google+sear...

Just tried it:

"What is one plus one?" "2"

"Plus one?" <Random movie reference>

"Add one to that?" "I don't understand"

"Add that to one?" "3"

Feels like an old text adventure.

This is my experience with Google Home. Haven't tried others.

It is almost always somewhere between faster and much faster to pull out my phone, switch app, type, than to waken up the device, ask it, wait for an excuse, rephrase my question and wait for it to read it.

That said it shows promise and if I still trusted Google I'd be somewhat enthusiastic :-| (Now I'm mostly scared.)

Since I do not trust them I have it in my office and turn the mic off when I'm not experimenting with it.

Next up: figure out if that mic is actually disabled in hardware if I "turn it off".

@Googlers: a good step forward wrt trust would be to allow the semi paranoid ones like me to turn on the mic on my phone on demand. I can live with the annoyance of tapping some button three times or whatever.

After China and a bunch of other stuff I do not trust you to value your relationship with your customers over your relationship with your bank.

I concur. For me it's similar wrt. annoyance/efficiency factor, but I have the opposite needs in terms of the microphone. I don't use my phone's Google Assistant because there are only two workflows where I would find it useful, and I can get neither of them to work.

Workflow one is leaving it on a desk and having it listen for the hotword all the time. No "screen must be unlocked" or "phone must be charging" nonsense; if I have to come to the desk and fiddle with the phone, I'll do what I need faster with my fingers. Workflow two is always listening when I'm on a Bluetooth headset. I can accept pressing a button on the headset to turn it on, but again, if I have to fiddle with my phone to get it to listen, it defeats the entire purpose of the tool.

(I'm also an old curmudgeon that is convinced speech recognition can be done entirely off-line, and there's no direct reason to send voice data to the mothership. But I guess it's fighting with windmills at this point.)

I've used both pretty heavily. We just gave away all our google home's except one and all our mini's.

Natural language voice recognition is better on google home. No question there - it's flat out better. I found google smarter.

But some deal breaker negatives. I basically want a totally predictable / reliable system. I don't need smarts, I need something useful (honestly, most commands to these things are probably weather, alarm/calendar, music etc).

The google app in a multi-person household does not work - sharing / inviting others to manage the house is totally broken. There is some kind of eventually consistency issue somewhere and the internal state gets terribly messed up. Maybe your partner is not into the automation as much as you are are, BUT if you give them the app with total control so they can see / do more, they will be more into it.

Some core features for most folks are things like what is on my calendar, set a timer etc. I'm a heavy google user, and I cannot get it to handle my google apps calendar. It's a known limitation. I can get alexa to handle my google calendar! Read here for more: https://support.google.com/assistant/thread/222810?hl=en

Let that sink in. ALEXA CAN handle my google calendar. Google CANNOT! Talk about a surprise when you are setting things up, I was convinced I had done something wrong.

This runs through things. The alexa tech is more limited, but they are more customer oriented, listen more. With google, even if it's the #1 request for literally years, and seems obvious (ie, google home should be able to read the events from the google calendar which is under the exact login used for the app) it is flat out ignored. They are simply not customer facing.

Another example I don't care about. Folks kept asking for apple music support on alexa. Amazon ultimately did what apple wanted to get music on alexa, even if it impacted amazon.com etc (ie, seem to have cracked down on fake apple products etc). Even though this was an apple limitation, amazon basically spent time to solve this for users.

Alexa routines are a lot of fun, worth checking out if you haven't.

And a tip -> give amazon some feedback, I've seen them listen.

You should have said "G Suite calendar". Technically I suppose it's "Google" calendar, so you're not wrong, but a number of things aren't really available on G Suite, which Google positions as their "business" offering. And a lot of other things aren't available on consumer GMail. That's probably why you're having issues adding others to a family group.

Should it be available? I suppose it should. But it's not really a problem for 99.99% of their users. The rest can use Alexa. :-)

I switched our company to google - I’m someone heavily in the google ecosystem and not supported by google.

Alexa does support me even though I don’t pay them for email. This is just one point, but illustrates the point.

Google has better tech, but weird product delivery approach. And I moved to AWS from google compute for a similar set of reasons.

Google works very simply on the inside: you get rewarded for launching _new_ impressive stuff. You rarely if ever get rewarded for fixing old and busted _existing_ stuff.

Whether the newly launched stuff is suitable for the user is a tertiary concern. So yeah, no disagreement there, Google is not very user-focused, and has never been. But they're so far ahead technologically that in a lot of cases I'm willing to forgive them their ADHD.

That would totally explain this! I'm often like, this tech is amazing. But wait, this doesn't even work with this other thing it should work with. It's definitely the fix / polish / integrate corner cases.

Google+ was a nightmare as well with youtube and other services - again, I was like how are you listening to users with these approaches.

I have both a Google home and Amazon echo literally next to each other. I have found that the Amazon device is really poor at hearing you.

The Google device I can talk to from any room in my apartment and it hears me correctly 95% of the time, it even triggers and understands me if I am brushing my teeth with an electronic toothbrush at the same time. It is impressive.

The Amazon device on the other hand has a 50/50 chance of triggering, and I often have to walk right up to it to clearly enunciate "computer" before it wakes up. Not sure if this because I speak en-gb instead of en-us.

(I have both only because the Google one does not reliably work with Samsung smartthings any more (it used to work fine!) but the Amazon one does. Amazon one is also good for music but Google tries to sell you a YouTube subscription. Google one is hugely better at random day to day questions/queries/translation/usefulness etc - I guess that is b cause they are a search engine and not an e-commerce site)

Even Amazon's music search sucks. It can never find the song unless you provide an accurate name. Its fuzzy matching sucks too, doesn't even account for regional trend in songs.

Apple's music search sucks too, especially if your music tastes are even slightly outside the mainstream. It simply doesn't personalize per user, so when I search, I have to wade through a list of unfamiliar band and musician names until I type almost the whole thing. For a company whose motto once upon a time was "design is how it works", this is inexcusable.

I found that small positioning changes can make a big difference with my Echo -- it used to be on the a kitchen counter in a small corner formed by a beam, I had to be close and very clearly enunciate "Alexa" to get it to wake up. One day after cleaning the counter, I put it out in front of the beam instead of the corner, and it was a night and day difference, it could suddenly hear me across the room.

> ads are cancer

> pervasive tracking shouldn't be legal

> I replaced my Echo with a Google Home Mini

Huh?

I'm ok with Google knowing what I do on Google. I'm not OK with Google knowing what I do everywhere else. Home does not have ads. If it ever gets them, it'll go straight into the bin.

> I'm an ex-Googler who nevertheless thinks that ads are cancer

Just curious: did this opinion about ads formed before, during or after you worked at Google?

I'd be interested to know what team in Google you were in. Most Googlers who design and use systems that interact with private user data come to trust them...

I hope at least some of them are working on Alexa's seeming inability to do any kind of learning on the go. For instance if, as you repeatedly inform me, there's no station called Summer FM Groove Salad, then maybe, just maybe, I mean Soma FM Groove Salad, like the one in ten times you hear me correctly.

Oh wow, I had completely forgotten about that station! Thanks for mentioning it!

Space Station Soma for the win.

Secret Agent and Illinois Street Lounge for me.

I’d settle for the seemingly vastly simpler simple task of producing an app that doesn’t show me content that I can’t consume.

I’m an Amazon Prime video subscriber currently located outside the US and the Prime Video app on Android, Roku, and PC, insists on listing all content, rather than what I am actually permitted to watch where I’m located.

90% of the titles shown subsequently show a “This title isn’t available in your region” once you actually click on it and bring up the summary page.

It’s truly absurd that one needs to click through every single video one at a time to determine what is actually watchable. Netflix has no such issue and filters based on location.

This. I never tell Alexa to set a timer for 15 or 50 minutes (only 16 or 49) due to tendency to confuse.

These are sometimes an issue for native speakers.

16 doesn't get confused with 60?

I guess it takes a lot of people to listen to the accumulated recordings from all of those people's living rooms!

debby downer

Wouldn't be the first time, won't be the last. For example...

https://www.businessinsider.com/expensify-is-proud-to-use-hu...

Amazon is a competently run technology company with smart people from my limited experience. I wonder why Alexa is so much worse compared to Google Assistant then.

Perhaps natural language processing / understanding applications aren’t as well researched or developed at Amazon? Maybe it’s the difference in culture between the two companies with Google being much more academic in nature and research oriented compared to the more process oriented operations / logistics focus of Amazon that has caused such a widening gap to build up?

Seems pretty simple to me, Google has more data and users that keep adding even more data to optimize their offerings. I doubt Alexa has even 1/10th the users as Google's assistant.

This is the unfortunate barrier to entry when it involves big data. Even behemoths Amazon and Microsoft are stuck with smaller pools of voice data than Apple or Google. It makes it very difficult to compete...imagine being an even smaller player, you have no shot. If Amazon and Microsoft weren't cloud competitors I bet they'd do a joint venture. They are already playing the most friendly with each other.

Amazon has still sold millions of these devices and thus has inordinate amounts of data. When you're talking about training sets that are, say, 1 trillion large vs 10 trillion, does it really make that much of a difference? It's diminishing returns. 10X the data size comes nowhere close to a 10X difference in quality at that scale.

I neither work on Alexa nor on Google Assistant. But based on my experience, going from 99% accuracy to 99.9% would require much more data than 0.9%. Same with adding each subsequent "9".

And the perceived quality difference between 99% accuracy and 99.99% accuracy is much much bigger.

In my experience, like with high availability, each additional 9 requires between double to 10x the resources to achieve.

A friend of mine who worked at Amazon, not in an Alexa related division so take this with a grain of salt, said that Alexa is basically a huge series of if statements.

I imagine they're working on improving it to more powerful natural langue processing, but it wouldn't surprise me if decisions they made to get it to market years ago are preventing it from being as good as Google right now.

Take what I say with another grain of salt, but I work at Amazon and have a large number of friends that moved over to the Alexa org.

It's not a huge series of if statements. Maybe it was at one time, but 10k developers can do better than that.

I would be surprised if 10k developers could do much of anything. The communication overhead would be enormous.

There's a reason most really nice things are designed by one person.

Amazon is not a perfect company, but they have done a very good job at figuring out how to organize large scale software organizations. Emphasis on SOA whenever possible, teams that aren't too large, and principle engineers who focus on large scale organizational problems.

Amazon very much has a "get it done" culture. Get a bunch of good engineers in a room to solve a very specific problem. They've never been really focused on iterating fuzzy problems.

  > competently run technology company

Not all parts. Not even remotely.

Source: worked there.

You’re at Google now right? You people think anybody not at Google is an idiot.

The reality is Amazon is extremely competently run given its size. The pace of innovation is astonishing. We might not be ubermensch but we get the job done, and we are trusted by more people in this country than either Google or Facebook are.

Your at Amazon, right? You should be aware that many other Seattleites in tech despise those who are part of FAANG, vast data mining (FANG) & extreme internal secrecy (Apple, Amazon) have created an environment where ethics are ignored, and only bootlickers are promoted.

This has come back to haunt some of these companies already, but fixing the structural rot is unlikely to happen anytime soon.

I think the answer is simply Google the search engine. Google has a huge corpus of semantic data that it gathered from both structured and unstructured data from around the web. Think about how many times you search Google and get a One Box result.

However, I have to disagree to a certain extent. Having both Google (on my phone) and Alexa in my house, overall I have been extremely happy with Alexa. I think it being "so much worse" is very subjective.

This is definitely part of it. Alexa only answers my most basic questions. Google answered me a fact that it was able to parse out from youtube and then pre-labeling the answer with "According to youtube..." That was mind blowing.

Alexa is new type of interface, just doing better AI/voice recognition is not enough. Amazon never were good at User Interfaces. Google is getting better at it I think, has more designers and UX.

Google had all of the pieces before Alexa even existed. The Google Assistant was basically a bunch of existing Search tech duct taped together at the start.

Some of this investment is negated by the amazing progress in neural nets since Google was founded, and diffusion of Search knowledge, but not all of it.

Amazon is really not that good from a tech standpoint. They are good at selling and shipping at scale. Tech (including AWS) has always been a secondary or tertiary concern for them.

I think this is where a company gets an advantage by doing real research.

That is, Microsoft and Google famously hire a lot of ex-accadmics and let them do basic research not immediately related to a product. That lets them build up 'secret' technology within the company that other people just don't have. However this is an expensive thing to do.

I think this is because google (and microsoft) are companies with a big cash cow and an unclear future path. They need to find and dominate the next big thing in order to one day replace their current thing.

Amazon on the other hand has a number of avenues for future growth. They just need to eat all of retail and internet infrastructure. Therefore they can grow just by solving their customers problems in a focussed way. Low cost, well engineered, current but not next generation stuff.

Has anyone tried to build an open-source voice assistant using the $100 Respeaker far-field mike array dev board, https://respeaker.io & http://wiki.seeedstudio.com/ReSpeaker_Core_v2.0/?

> ReSpeaker Core v2.0 is designed for voice interface applications. It is based on the Rockchip RK3229, a quad-core ARM Cortex A7, running up to 1.5GHz, with 1GB RAM. The board features a six microphone array with speech algorithms including DoA (Direction of Arrival), BF (Beam-Forming), AEC (Acoustic Echo Cancellation), etc. ReSpeaker Core v2.0 runs a GNU/Linux operating system ... Our BF and AEC algorithms are provided by Alango, a professional audio DSP company used in the automotive industry.

Voice assistants are magical when they do-what-you-mean. If there was an active OSS community for voice assistants, it would enable experimentation with privacy-preserving architecture and customization for niche use cases and language.

- https://github.com/NaomiProject/Naomi

- http://mycroft.ai

- http://snips.ai

That reminds me that Google has DIY kits at Target, but they are designed as Google service extensions.

https://aiyprojects.withgoogle.com

Hardware seems to be by far the easiest component of the equation. Google home minis are practically free.

Mozilla was doing some work with speech recognition. Don't know if that project is still active.

But yeah, these things needs lots of data to work well. That would be a massive project.

Mozilla DeepSpeech is pretty active considering the holidays: https://github.com/mozilla/DeepSpeech/pulse

Looks like they'll be having their v0.4 release sometime this month?

My guess is ~9950 of those employees just do data labeling.

That still implies 50 of them are trying to make it seem like anything but a toy.

I was given a Google Home Mini as a Christmas gift (the resident tech god for most people's lives doesn't have one? what, how can this be!? someone rectified that for me), and I've been toying with it... and its the Google assistant in my phone, but in a stand alone device.

I mean, don't get me wrong, that's a lot more useful than people realize because.... IT ACTUALLY WORKS. Ridiculously loud kitchen fan hood roaring along at levels that are probably considered unsafe by OSHA standards? "Hey Google, do the thing." "Okay, I'm doing the thing." Upstairs, far away from it? Not even yelling at it, it understands me reasonably and does things alarmingly often.

The best impossible magic? "Okay Google, turn my television off." My Chromecast is not named "television" nor "my television", is only referenced in my Home config as the default TV for that room (of which the TV and Mini share), and somehow turns the TV off even though the Chromecast does not expose this functionality anywhere in the Home app (and was assumed to only know how to issue exactly one HDMI CEC command (turn TV on and switch input to here) due to some undocumented and strange hardware issue... hundreds of threads across the Internet all about how can they turn the TV off even though they can turn the TV on, none of them ever figuring out how to issue CEC off or even if the Chromecast can... it can, and the assistant in my tiny speakers knows how to issue this probably undocumented command to the Chromecast).

My experience with Alexa? Ideal listening conditions, no background noise, only me talking. "Alexa, do the skill that you know how to do and is advertised in the commercials on TV, any off them." "I'm not quite sure how to help you with that".

Alexa is like a little kid, you just have to repeat yourself often, and sometimes she pipes up in the middle of a conversation because she wants to do something else.

That said, she’s quite capable of turning my TV and lights on and off.

Seems Siri is much better than Alexa as well - I often hear my roommate struggling with getting Alexa to do what she wants, and various other friends struggling with Alexa as well, usually because of accents.

Smart assistants in general have a ways to go still, but perhaps in the next 5-10 years they will be very good.

I have been using Alexa for years. I've been controlling my TV and I have WiFi smart plugs. It is super useful and it has never had issues understanding what I was saying.

I have noticed people who have issues they place the device next to a sound source such as a TV or fan or something and obviously this will mess it up.

I also don't understand why so much hate for Alexa. Everyone in this thread just seems like a Debbie downer, shi*ing on everything all the time.

I find Alexa "voice recognition" quite, if not amazingly, good. Native English speaker but definitely not news anchor voice. The problem is in conversational understanding that drops off quickly beyond "What's the weather?"

Because it's a spying device people are installing in their homes. These are the same people that are going to complain that it was a huge invasion of privacy in 5 years. It's going to be Facebook all over again. People willingly hand all their data over to daddy Bezos and then wonder why Amazon National Healthcare is charging them insane premiums.

Edit: The same people developing facial recognition for the DoD is developing eavesdropping devices for Americans to install in their homes. Why, after the Snowden revelations, would any rational person think the big-brother doesn't have a direct line into your home? Listening to your every word, to your childrens words, to their opinions in formation.. Seems willingly selfish to not only hand over your privacy, but everyone you invite into your life.

Edit 2: Considering the investment being put into a smart-microphone, I can only imagine the market for the device is a skunkworks operation. Selling a perfect surveillance device to the CPC perhaps? Beat Google and their project dragonfly to the punch? Hmm

> I also don't understand why so much hate for Alexa. Everyone in this thread just seems like a Debbie downer, shi*ing on everything all the time.

My issue is that it hears me correctly, it just doesn't understand what I'm saying. You can't use "AND" to combine stuff ("Turn off the living room AND kitchen light"), trying to play a podcast by reciting the name doesn't work 90% of the time. I can barely get it to play my Spotify and do shuffle play, it usually takes like 3 failed commands, and I think "wow, I could of just pulled my phone out of my pocket and done it faster - without getting annoyed".

I've had it for a year now, and it hasn't improved at all as far as I can tell. Kind of pathetic when you are the #1 selling device and I assume have multiple times the data of their competitors for training.

I'll give you a concrete example. The primary light fixture in my small apartment is a standing lamp in the corner of the living room. I bought a smart plug for the lamp and during set-up I chose the name "lamp", since that is what we call it (in our household you don't have to say /which/ lamp, since it's obvious). Now, Alexa doesn't know what to do if I ask it to "turn off the light", which is disappointing enough, but it also fails to understand "turn off the lamp" -- instead I have to remember to drop the 'the' and say "turn off lamp", which is so unnatural. It feels like I have to bend and deform natural language into something Alexa can parse, which is the exact opposite of what assistants promise to do.

I’ve set up a few single device “groups” in the Alexa app to serve as aliases. This has helped quite a bit.

Well, naming devices for voice control is more tedious than you'd think. You have one device that you named poorly. Try having a house with over 150 devices! Naming matters. Putting a little thought into it makes a big difference in usability.

I've got one that gets used everyday. It's named "Breakfast Table". 'Alexa, turn on Breakfast Table' works quite reliably. Now, am I actually turning on the table? No, of course not, but that rolls off the tongue a lot better than "Alexa, turn off pendant lamp over the breakfast table". Or recessed ceiling cans as opposed to just "living room ceiling". We really don't have many devices with 'lamp' or 'light' as part of their name. Because we're not really asking for control over a light, we're asking for light for an activity at a location.

But there still ends up being a few that are clunky. "Family room endtable" for a reading lamp near that end of the sectional, or 'Family room sofa" for the two on a console table behind the other part. Haven't hit upon better names for those. Activity naming is an option but we really don't call for lighting in a scene oriented kind of way. Some folks seem to like that, go figure.

Bearing in mind with an open floorplan just about everything on a level is within line of sight and earshot. The placing of multiple Alexa devices took a little fine tuning to overcome reflection from lots of wood and drywall surfaces. That allows for unexpected pickup. Oh, the units handle avoiding overlap with each other, but sometimes facing one way in a room leads to the sound being picked up by the one in that direction. As you'd expect sound waves would travel. But without a LOT more intrusive sensors (cameras, motion, position) it's handling things remarkably well just with voice.

The great tragedy is the leap to conclusions people make. "Promise to do".... where? By whom?

Oh, you want a TV cartoon equivalent of Rosie the Robot... we're not there yet. But given the hilariously low price point for these devices, we're getting quite a lot of bang for the buck in the meanwhile.

Can you rename it to "the lamp"?

> I've been controlling my TV and I have WiFi smart plugs. It is super useful and it has never had issues understanding what I was saying.

> I also don't understand why so much hate for Alexa.

Those are nice features. You absolutely shouldn't have to send you voice over the internet and give up all of your privacy just to make these work. Does it work when your internet goes out?

I don't know anything about how good Alexa works since requiring an internet connection is a non starter. This isn't why I hate Alexa.

Agreed, technically it's amazing.

My wife and I ave different native languages yet we can speak to it, combining both languages in a single inquiry/conversation and it replies to me in my language and to her in hers.

Ingenious stuff, whatever people think of the privacy issues.

W/o Paywall: https://outline.com/vqmqwg

This org could be it's own company (like many other Amazon orgs)! For context, number of employees at:

  Motorola (2014): 40,000
  Nvidia: 11,528
  AMD: 8,900

I bet there are more Alexa customers than nvidia ones. And way more than direct AMD ones.

Amazon has sold about 35 million Echo units, which is a little under half of the total number of PlayStation 3 units sold to date, each of which has an Nvidia GPU in it.

And about 10 million Nintendo Switches, which are also Nvidia powered.

100 million, they released numbers the other day https://www.reddit.com/r/amazonecho/comments/ad2drt/100_mill...

1 employee per 10,000 echos doesn't seem like much.

In an interview with the Verge this week, Amazon themselves finally revealed a relatively solid sales figure (by Amazon standards) - over 100 million units with Alexa pre-installed. Sure, this will presumably include some non-Amazon hardware, such as Sonos etc.

It's unquestionably less than the number of devices containing some kind of Nvidia IP. However, that's still a pretty impressive number of devices, especially when being an Echo/Alexa interface is likely the primary function of many of these devices - something you can't say about Siri or Google Assistant to anything like the same extent given it comes as part of the OS on many phones.

> https://www.theverge.com/2019/1/4/18168565/amazon-alexa-devi...

Sure but they have plenty of R&D cash to spend and I feel like their goal isn't to sell a single device but to become the technology leader in that field. They didn't put Alexa into a microwave for no reason, they want it everywhere.

Like someone else said too, most of them are probably just doing data labeling.

PS3 is made of plastic that doesn't make a purchaser oil company's customer. Same with the chips in it. Sony was nvidia's customer.

No way Alexa has anywhere near as many customers as Nvidia. Not only does Nvidia ride the Cryptocurrency wave, they are also the leading manufacturer of PC GPUs, and their processors are found in literally billions of devices, like game consoles and smartphones.

There are tens of millions of Echos. They're successful, but not ubiquitous like Nvidia devices.

Nvidia has around 100 direct customers. ~20 graphic card OEMs + laptop/car manufacturers etc.

They make founders edition cards, so consumers are direct customers.

And yet, Google has way better speech recognition than Alexa.

and google cloud customers might get a google home for free.

To be fair, Google has been working on speech recognition for longer and likely possess more data than Amazon.

Is this why Google gave everyone free phone calls and voice mail for years?

Yes, along with their 411 service at the start.

Almost certainly.

They also pushed their voice assistant on Android almost as hard as Apple pushed Siri. Integrating it into default home screens, keyboards, etc.

https://en.m.wikipedia.org/wiki/GOOG-411

> Google had stated that the company originally implemented GOOG-411 to build a large phoneme database from users' voice queries. This phoneme database, in turn, allowed Google engineers to refine and improve the speech recognition engine that Google uses to index audio content for searching

Google made a 411 service a long time ago for this exact reason. https://en.wikipedia.org/wiki/GOOG-411

I'm surprised at the number of people saying there's a noticeable difference. I've been using Dots for a few months, now, and it pretty much understands me every time, even with the vent hood of the range on, etc.

I did have one odd error, where when I named one of my lamps 'halogen', Alexa would always ask back "did you mean halogen?" with the exact same pronunciation I had used. Then proceed to turn it off or on like I'd asked in the first place. This struck me as more of a bug than anything else, and I worked around it by just using a different name. But still, if that's the only listening error I've dealt with after several months' use, I'm impressed.

> when I named one of my lamps 'halogen', Alexa would always ask back "did you mean halogen?" with the exact same pronunciation I had used.

Maybe due to https://en.wikipedia.org/wiki/H-dropping ?

An interesting thought. In my case, though, I live and work in the bay area of California, so no H-dropping, and the accent I use is probably the one that is the most common in terms of use for a training set.

(I was saying 'hale-oh-gen', as was Alexa, FWIW. Now I'm curious if others can repeat the same error!)

Does anyone have an approx breakdown by role?

e.g. how many software engineers vs how many data labelers

or how many are full-time vs contractors, etc

10,000 people, any my Echo still doesn't understand the difference between any of my smarthome devices that contain the word "lamp" in their names.

I have many alexa devices in my home and I love it. They solve lot of problem for us. From automate my home to a speaker(via bluetooth).

But one thing concern me a lot is Alexa is constantly stream/read data to internet. I'm not sure what it's but it consume to 114Mbps per device.

With that huge data it is sending back to Alexa, I imagine they may have a huge team of scientist to work on it.

One thing I noticed is Alexa understand my broken English way better than my wife(native). Probably I have speak to it to much and it learned my voice.

It's massively insane that people put microphones into their house and willingly stream all conversations to a private 3rd party in exchange for minor convenience.

What's the liability here when I go into someone's house... ? Do I give implied consent? What about states with two party consent?

"Please take off your shoes"

"Please turn off your Alexa"

"Alexa, please set an alarm for 12 hours from now to remind me that you are always listening"

Won't this be the same with any IoT service? You walk past peoples houses who have Ring, it's recording you.

Youre walking by in public; no expectation of privacy.

http://www.americancriminallawreview.com/files/5114/9515/418...

The third party doctrine may be extended to guests visiting a home, but existing law should adapt to require the visitor’s consent. One party consent laws will no longer be feasible in this context without an exception, as they would completely eradicate any expectation of privacy when visiting another’s home.

A more pragmatic approach would be to extend the two party consent laws to owners of recording devices, whereby the duty to disclose that conversations are being recorded would most efficiently be placed on the owner of the recording device.

Consent laws must be modified in order to match the reasonable expectation of privacy of guests visiting another’s home. Owners of an Echo device, similar to banking institutions, should have a duty to disclose to their home guests that conversations may be recorded. The duty of disclosure would best be reflected by a uniform application of the two party consent laws.

That’s a good question— maybe people just need to have a disclaimer on their front door that conversations may be monitored!

>all conversations

I believe Alexa only streams what you say after the trigger word and not all conversations. Assuming you believe Amazon.

if its technically possible to do it without you knowing then all that is left is how much you trust Amazon with that. Note that Amazon may have to surrender its devices access to three letters agencies for random reasons as well. You have to live with that assumption.

Sure. But most people trust corporations with almost all their private correspondences, photos, internet browsing history, shopping behavior, etc. So at that point how much is there to lose by adding everything you say into the mix as well?

Some companies, notably Google and Amazon, have maintenance of customer trust as a core principle. This seems to be working. Generally, people trust a lot of their private data to these companies and the companies do not abuse it.

Microsoft believes in this too, but had to learn the hard way back around 2000 that customer trust matters a lot.

Facebook seems to have had a harder time learning this lesson. Perhaps they've learned it now, but it'll take a while before consumers really believe it.

It's widely understood how the wake-word system works. You can monitor how much data is being uploaded and when. So no, it's not possible to just stream everything you say undetected.

I believe 99 percent of Alexa users wouldn't notice the increased data usage. If "three letter agencies" carefully choose who they spy on, it might go undetected for a very long time.

If a three letter agency is interested in you then you would be lucky if your Alexa is the best bugging device they have.

I notice Amazon's wording about this is pretty careful. The device could still be processing conversations locally to some extent and still be adhering to the wording.

It records every conversation locally in a two second loop I believe, and then streams it after a trigger word.

Correction, it keeps a rolling loop going so that when the CPU hears the wake word it has enough audio buffered for voice recognition to have a clean starting point.

If it only started recording when the wake word was processed as being heard there would be an added delay before you could start speaking a command. Might as well push a button and wait for a beep for that added delay.

And the wake word detection is done offline, as can be easily proven by disconnecting your internet and saying 'Alexa report this conversation to the NSA' - she will wake up, but then blink red due to lack of network connectivity.

Minor correction but it could still do the wake word detection locally, then fail with the red light because the "report this conversation to the NSA" part could not be sent.

Exactly. Everything after the first trigger word you said when you unwrapped and plugged it in is streamed.

Yes, but now I don't have to get up to switch my light off.

I don't get this. Yes, people are lazy, but now everybody has to go to the gym, because otherwise they won't even have a little exercise.

Look, I know how to make my own candles from tallow (including how to bring down the beast to get the fat). Does that mean using electric lights is lazy?

Of course not.

I have my hands full cooking. Is asking "Alexa, set a potato timer for 12 minutes" any lazier than stopping my stirring the sauce, wiping my hands, picking up the timer and punching buttons on it to set a timer? I can keep going about my tasks and have the Alexa device help. Likewise, "Alexa, turn on dinner table" or "Alexa, announce 'Everybody get down here, dinner is ready'" These are little conveniences that keep a busy process going smoothly.

Could I live without them, sure. But I could live without electric lights and plumbing too. But why would I when they exist and are available at a reasonable expense?

Luddites, the lot of you.

It's not about being lazy, it's a huge help if you are multi-tasking and are time poor ,like most parents are. A pretty normal scenario for me is trying to get the kids out the door, whilst doing their shoelaces you tell Alexa to turn off the tv, turn off lights, tell you the weather forecast and when the next bus is due to arrive out the front.

Cell phones and tablets could already be doing this, it's not unique to the ladies in cans.

“Ladies in cans” have a power source that allows them to continuously listen and send back audio.

And so does you phone.. how often do you have your phone on you and it's NOT on?

No, it really doesn’t. Phones run on batteries and continuously listening to and streaming audio would cause them to quickly die.

Maybe they actually do? For example, the 2017 version of the Nokia 3310, according to Nokia, can have calls of ~22h. And that's for a small 1300mAh battery. So it's technically possible to stream audio for a very long time. Maybe smartphones are doing the same in a very low-powered stand by mode?

Yeah but they're not. It would be obvious from the packet traffic if your phone was constantly streaming everything you said to Google or Apple's servers. It would also be a violation of privacy policy and potentially state law. I think there would be huge backlash if either Google or Apple did this.

It doesn't need to constantly stream in order to record everything and send it back in periodic batches. Would it also be possible to determine that from the packet traffic? (not intending to be confrontational here)

The Alexa devices don't have the hardware within to do that. People have torn them down and looked. People have also constructed their own devices (rpi with a mic) to mimic the same functionality and it's consuming the same sort of bandwidth.

Now, of course, nothing will ever placate the conspiracy nitwits.

Thanks for the answer.

I don't see how there's a conspiracy here. Questioning the intentions and actions of an actor that could very likely be hostile to the user is reasonable and has precident.

My parents bought us one for Christmas and I politely refused. There's no way I'm letting those into my house without a huge benefit...

Is that a typo? While I don't doubt that Alexa sends some amount of data to the internet, I find it hard to believe that it uses 114 mbps per device, maybe you meant 14 mbps?

no, it isn't a typo. I really have no idea what it's. They differ from 80Mbps -> 114Mbps.

This is the data to back if up:

https://ibb.co/2Zm11RL https://ibb.co/zmYVP6F

That is the connection speed from the device to the router. Not what it's actually transmitting.

Oh thanks you so much. Now it makes more sense.

Do you have any idea why it takes so much data between the echo and the router then? Other devices in home didn't take that much :(.

You still are not understanding, the tx and rx are simply what speed they are connected at, not any different than your ethernet connection reporting 100 mbps or 1000 mbps, it does not mean it is actually transmitting that much data.

Sorry for my ignorance. You seems to know this stuff very well.

So to clarify, it's mean the speed when the device is connected at, but I saw they changing up and down? How does that get calculated if they aren't the actually transmiting data?

Automatic negotiation with your wireless router. The max tx/rx speed will change regularly, especially if you’re in noisy WiFi area.

If an echo would require 114Mbps, it would barely work for any household, since internet with that speed is far from common. The real required bandwidth should be in the region that is required for transmitting a bit of audio (< 10mbps), and only if audio is active.

You mean Kbps? 114 Mbps is the same as a 4k Blu-Ray per device which means you'd have to have >gigabit internet.

There's no way. Alexa could be streaming raw analog sensor data and not use that much bandwith.

This is definitely not universally true. I monitor usage on all my devices pretty carefully - I have several Echos in our house and they never use more than a few hundred kbps and that only rarely. Even with several devices streaming video at once I rarely peak above 50mbps. (1gbps FTTH connection).

This is 100% false.

Network traffic seems a bit frequent with a set of packets every 5 minutes or so, but they are very small.

It is def not using 114Mbps all the time. So we are clear, that is 1.2 terabytes PER DAY.

"With that huge data it is sending back to Alexa, I imagine they may have a huge team of scientist to work on it."

OK, there are 100 million or so Alexa devices supposedly sending 1.2 terabytes per day. Where does this stuff come from (seriously).

Folks, most home routers you can actually monitor packet usage by device / port. If interested please do, the results DO NOT match the posters comments.

You mention your evidence for this assertion further down ( https://ibb.co/2Zm11RL https://ibb.co/zmYVP6F ) but apearson points out "That is the connection speed from the device to the router. Not what it's actually transmitting."

Thank you. That makes more sense now. I have been tracking down a network slow recently and this assumption make me think it's the echo. Now I understand and this is helpful for me to eliminate echo during that troubleshooting.

I cannot edit this anymore. But just want to let everyone know that I mis-understand the router report page. The report is what between the router and echo device. Though I'm not sure why the communicated between router and echo device are that much...

Seems like an 80/20 rule might apply here: 8K of them are in marketing, with the remaining 2K also being split 80/20 to 1,600 tech/infrastructure support and 400 actual ML/AI, device & application developers.

Alexa seems to not understand simple questions like "current temperature in Celsius" which is really a deal breaker for me.

(I don't own any Alexa device but when I visit Seattle they have one in the room).

I don't care how many people they have working on Alexa or Echo - those things will never be allowed inside my house. Massive security risk.

Do you have a smart phone? It's not any different.

Edit: But it also constantly tracks your location and has a built in camera.

Smart phones don't transmit voice data back to third parties, nor do they transmit photos taken in the native camera apps, so it is different

Has there ever been a documented case of a smart phone randomly burst out in creepy laughter or sending private conversations to random contacts?

Here is Android sending private conversations to random contacts: https://daringfireball.net/linked/2011/01/06/android-sms

Even if it was exactly the same (which it isn't), why mke the problem even worse?

More people makes it even worse in my opinion.

If those 10k employees earn 90k$ on average, that would be 900 million dollars per year just for salaries.

The echo dot costs 30$.

I'm sure Amazon just wants to make a great user-centric product and will not agressively violate your privacy to make more money! </sarcasm>

How many of them are working on tagging manually the audio tracks for machine learning?

Amazon must have a very decentralized recruiting methodology because I've been contacted by recruiters for Alexa at least 4 times.

It seems like Alexa, Go, and AWS have massive staffing targets based on my interactions w/ recruiters.

Yes, every group in Amazon has its own recruiters who largely find candidates independently. However, Amazon does have a good central database of everybody who is contacted, who last talked with them, what happened with the contact, etc. And they're usually pretty good about having only one group active with a candidate at a time.

Wow. Is the software that complex that it takes 10k employees to work with it?!! :)

Me: That's a pretty complex thing to build. I'm guessing it'll take about 12 months.

Management: Great! We'll hire 10000 people and have it done before lunch!

Me: ...

There's a force that makes big companies grow bigger to the extent that their success allows them to, not to the extent that it's needed.

Software and organizations are always much more complicated than you would realize as an outsider. Most of those 10k employees are probably not software engineers. Businesses have many other functions. From a software standpoint, any system that support 100M clients with rich functionality is going to be shockingly complex.

Do people like having these devices in their house? Don't you feel like they're listening to everything you say? My wife says these will never be allowed in my house.

I'll never own one. The privacy issues are mostly speculative but I just hate taking to computers. On my phone, I've disabled everything I can related to speech.

Your wife is wise. I won't have one in my house either. It's crazy to me that anyone would be ok with them post Snowden revelations.

Is there any public info on the numbers for Google and Apple?

Apple likely has fewer people in their entire software engineering organization; they generally hire fewer people than you’d expect for most projects.

Really hudge, how they are going to make it very lucrative? Monthly payment? Or sponsored hidden ads?

And it's still no match for Google.

My 6yo son thinks Alexa is an actual person... he might be on to something.

I wonder if any of the 10,000 people are actually named Alexa, and if so do they get any preferential treatment?

Most likely they would rather get truckloads of horrible jokes.

I always feel sorry for all the Alexas. And the Harry Potters who were living their lives just fine before the books.

My friend is named Tressa Green. Her dad assured me he had no idea that most people pronounced Tres-sa as Tree-sa.

[edit] this is a fun read: https://www.radiotimes.com/news/film/2018-03-22/meet-the-rea...

If you look at the lists of ranked baby names by year, you'll see a massive drop off of Alexas about when Amazon shipped.

One of my coworkers used to go by Alexa, and has since just switched to Alex. She is not happy about it, but none of the available wake words are ideal. I say computer and amazon too often to use those, and it baffles me that we can't just choose our own, which is something that the very first Droid phone supported.

Maybe 9,900 of those employees are hired to step in anytime Alexa doesn't know how to interpret a voice command.

If it worked real time, that would make for quick accumulation of training data. But I don't think a human can manually solve the queries fast enough.

Relevent podcast miniseries about exactly this idea: https://www.gimletmedia.com/sandra

the link below points to the source code for alexa based personal assistants in a can. https://www.amazon.com/gp/help/customer/display.html?nodeId=.... you can also access alexa by browser, but you need to sign in

https://www.amazon.com/ap/signin?showRmrMe=1&openid.return_t....

there is also a way of saddling an SDcard into the echodot debug port and booting whatever you have put on the SD thus you may run custom boot up on the dot and hax around.

https://medium.com/@micaksica/exploring-the-amazon-echo-dot-....

One of the most interesting things to get into is the MP3files that make up the dots responses to you, writing your own skills is cool too.

and here are a bunch of fidjitty tricks to do to your dot.

https://hackaday.com/?s=echo+dot

Once Amazon/Google/etc manage to get their audio surveillance embedded into every device on earth, will breathing oxygen and speaking aloud be considered "agreeing to Amazon/Google/etc privacy policies"?

Reply on news.ycombinator.com