I worked in the Amazon Alexa division. The level of incompetence coupled with arrogance was astounding. Many of the people running Alexa had been there since 2012. Its was tyranny by those who started in that organization first. Backstabbing, politics, bad engineering, nepotistic promotions. The scientists were so far behind, almost all models were off the shelf implementations pulled from Github. Huge amounts of capital flushed down the drain.<p>The flip side might be that Alexa was ahead of its time, and that the ML capabilities werent there. But I bet Alexa spent more than OpenAI by a huge margin. Amazon's fundamental flaw is trying to solve innovative business problems with incremental improvement. This only works in operations heavy businesses, like retail and AWS. AWS is really just extremely competent operations on top of server management.<p>The whole company is a meat grinder, poor technical implementations with an army of SDEs keeping it running. It definitely works, they ship product, but the company is underperforming the S&P for five+ years now. Walmart has higher shareholder returns. They need to flush out the legacy employees and make room for hungry young workers.
I went all in on Google home, with the little pucks and also screens littered around my house. My wife and I have both noticed over the last couple of years that their usefulness seems to have diminished a lot. They used to seem much "smarter." Now, basically, they can tell me the time, set a timer, play music with like 65% accuracy in playing what I want, and tell me the weather outside. It's possible that they were always this bad and the novelty wore off, but it seems like the service just degraded.<p>I assumed that both Amazon and Google were underwhelmed by how much actual revenue these kinds of devices produced, so they were starving the backend services.<p>Now it looks like both companies are hoping that Generative AI is going to make them more valuable [0]<p>[0] <a href="https://www.cnbc.com/2023/08/01/google-reshuffles-assistant-unit-again-to-supercharge-it-with-ai.html" rel="nofollow noreferrer">https://www.cnbc.com/2023/08/01/google-reshuffles-assistant-...</a>
If Apple is smart, they will just integrate LLMs into Siri and across the iOS, macOS, speakers and TV and own everybody in "personal assistant" space. We already have very powerful devices in our hand (iPhone), on our desks (macs), around the house (speakers/tv), just integrate the latest AI tech into them in a smart and thoughtful way and it can be epic.
I’ve got about six Echo devices that I use basically as speakers, timers, grocery lists, weather radios, and some home control (lights and thermostats). I haven’t explored new features or asked it general questions in a while, but what I do use them for, I would miss.<p>It’s become second nature to say “Alexa, seven minute timer” while I’m cooking. Or “Alexa, add soy sauce to the shopping list” when I open the fridge and pull out an almost empty bottle.
They must have tried hooking Alexa up to LLMs with Whisper-like speech recognition. I can only imagine the cost projections of rolling that out were astronomical. Plus, everyone that uses Alexa daily counts on it acting in the same dumb ways it always does.<p>Wouldn't be surprised to see them roll out a new <i>optional</i> Alexa OS upgrade that you can opt into which has a subscription fee attached and provides AI coolness.
Alexa may not be the most capable voice assistant, but it is the most reliable. I use it and a few others daily for different things, and have never once got a random "I'm sorry, I can't do that right now" or "something went wrong try again in a few seconds". Also, it's the only one that I know of that can text reminders to me. I hope they don't ruin it.
I own a few Alexa devices, and I like them more or less, but doesn't seem like they(amazon) have invested much in making them truly useful devices anymore - the first few times you could talk to them were cool - but after that novelty wore off, the products seem to have stagnated quite a bit - seems the market agrees.
I tried to replace my Amazon Echo devices with Apple Homepod devices - but my wife wasn't having it. We just wanted whoever was in the room - even if it didn't recognize the voice - to add things to the shopping list. But nope. You have to have an apple device that's been trained on your voice to add things to the shopping list.<p>I'm still looking for another solution so I can get out of the hell of maintaining Alexa's random new notification setting of the day. I really hate all the new "suggestions" and new crap they keep shoving down your throat with no option to turn it off. I would pay a subscription to turn that it off the annoyance, but there is no option for it.
Former Alexa Skill developer here.<p>I gave it a shot for a year as a potential source of side income and released about 10 skills publicly (+20 more are half-done or abandoned).<p>I've learned quite a bit about this platform over the past few years. I'd say the only positives I got out of it are a monthly $50 credit on AWS and sporadic surveys that netted me anywhere from $5 - $50 each.<p>My notes:<p>- VUI is a terrible interface. There are too many variables (foreign accents, volume, word recognition, etc.) that just make it clunky and unusable. Testing and debugging in the Alexa Console does not work 100% like on a real device.<p>- The ISPs (In-Skill-Products) are confusing to developers and users. The documentation is often wrong.<p>- The Alexa forum posts are rarely addressed by anyone at Amazon. Instead, you start to see fellow developers venting their frustrations.<p>- The intent model and lack of state machine makes it incredibly difficult to make a skill worth someone paying for it. It wasn't uncommon for testing to work flawlessly and then the published skill would be so spotty as to be unusable. Imagine being in the middle of an Alexa game and then unceremoniously receiving a the default help response (required for Alexa skills) that restarts your progress.<p>- Alexa doesn't work like people expect it to. Again, the intent/slot paradigm is pretty terrible. There's a lot of grunt work to have it even recognize slight variations of slots, so you end up with a huge list of sample intents users could potentially use.<p>- Internationalization is a HUGE pain. There are different availabilities for ISPs, Alexa services, etc.<p>- The Alexa Console is very clunky. Today (just like every day), I loaded the Analytics for one of my skills and got a 500 error. Reloading two or three times usually solves this issue, but c'mon!<p>- Good luck trying to market your skill. You could game the system by publishing updates or implementing new Alexa features, but getting your skill onto someone's account is basically a guessing game of what will work or not.<p>- Amazon Skills are not very profitable for individual developers. This puts the platform in jeopardy, as only the big players care enough to have their brand out there and sole developers don't make enough to make useful Alexa skills.<p>- Other skill developers could just copy your skill and publish it! The certification process is weird and oftentimes seems to be arbitrary.<p>There's more I could gripe about, but I've abandoned the platform. I suspect that Amazon's real product was the unfettered access to real-world speech data. By having a surveillance device in a user's home, they were able to collect data and train their own software.
I’m assuming that this represents a massive pivot into LLM assistants and these jobs were in support of the legacy system which will be deprecated soon.
I stayed in the Circa hotel recently. They have an Amazon Show on the nightstand. It does nothing in terms of automation. It’s just there to replace the in room phone, show weather and nothing else. Can’t recall as it was very limited when I tried interacting with the voice assistant.<p>I haven’t seen Alexa evolve that much but maybe someone who worked at Amazon can give more on what’s evolved.
To be honest I really liked Alexa with Ring. I know there's controversy with Ring, but it's just amazing how I can setup some cameras outside my home and then buy cheap alexa monitors to place around the house which show a camera preview whenever there's movement detected. In the past this would have required much more hardware that would have cost thousands.
with things like chatgpt all of these assistant techs were immediately outdated.. I wouldn't have expected anything less that dumping whole old departments and starting over.