Running AI is so expensive that Amazon will probably charge you to use Alexa in future, says outgoing exec

Guess we'll find out when they finally pull the trigger

But he acknowledged that Alexa will need to drastically improve before that happens.

I get tired of the outrage-headline game.

I do wonder how much of those voice assistants could run on-device. Most of what I use Bixby for (I know. I KNOW.) is setting timers. I think simple things like that can run entirely on the phone. It's got a shocking amount of processing in it.

If IBM actually manages to convert COBOL into Java like they're advertising, they'll end up killing their own cash cow

So much still runs on COBOL

It's not even A.I. either

Setting all my Alexa's to UK English got rid of all marketing "by the ways." I still regret going with the Alexa ecosystem but at least for now there is a workaround for the most rage inducing part of it.

By the way, did you know that you can find out more about telemarketing with an audio book from audible on the subject. Would you like to hear a preview of that now?

We have a Google mini. They listen to my six year old request the song Poopy Bum Bum all day. The ads get interesting.

That's often the case. They can have their cake and eat it too. Shareholders would expect nothing less.

I think the data is probably less valuable than people think, especially if the users expect an AI response whenever a data point can be collected from them.

I wake my echo and then grind coffee beans.

Let me in your house and I'll observe it for you for less money

Google did that with Nest Aware years ago as well. It's super annoying.

And much of it can be listened to by staff that are hired to label it to train the model.

Yeah, they're all pretty disappointing. I'd love to have something that feels like how movies portray digital assistants. Movie assistants never misunderstand you or say "I'm sorry, I couldn't recognize your voice". I've mostly used the Google one and it's so bad at doing what I feel like is feasible even with inaccuracy.

Eg, I've tried to tell my assistant to like a song that was currently playing on YTM but could not find a voice command that worked (and some commands backfired by making it skip to the next song). I've had very poor success with getting assistant to cast something to my Chromecast with my voice. It sometimes works, but it fails or gets it wrong so often that it's not worth the time.

Sometimes I use it for rewinding (e.g., "ok google, rewind 30 seconds") because many apps don't have granular rewind buttons and tracking on the track bar is way too inaccurate. But lol, it's so slow! It takes a few seconds to figure out what I said (so I have to ask it to rewind more than I wish) and it seems every app is unoptimized for rewinding, as it usually takes several seconds of loading.

It can't really do any kind of research either. You basically can just ask it to google things and it sometimes is able to extract the meaningful part from simple questions. It's a far way from how Hollywood thinks a digital assistant will work.

From the article:

Amazon has bet big on AI, with the company unveiling a new, AI-powered version of Alexa alongside updated versions of its Echo Frames and Carrera smart glasses last week.

I still use mine for voice commands with home assistant. Works great.

All mine to is turn lights on and off.. very occasionally they might be used to find a phone, or set a reminder, but I wouldn't miss it if that went.

I wondered if I was unusual in not using the voice features much, but according to this thread it seems I'm not.

Yes?

I'll happily join in bashing Amazon for plenty of reasons, but training an AI is a step that adds significant value to the material. Just like any other product that is an effort that people are willing to pay for.

"Alexa, release the Roombas"

30 roombas make their way out from under my bed

Yeah, can you imagine trusting that thing to buy stuff for you? Bit scary, actually.

They had those "re-order product" physical buttons for a while which you were supposed to glue to your washing machine so you could reorder when your detergent ran out.

Besides legal issues (at least over here all they could do is put things in your shopping cart) apparently the primary customers of those buttons were hardware hackers, turning them into all kinds of stuff.

Exactly. I never did find another use for that thing. Had one 2014-2022. It didn’t survive my last move. Was voted off the island.

You just typed that question on one. See: GPT4All You can download many models and run them locally. They were about 5-16GB in size the last time I downloaded one. Pretty slow if you don't have a hefty GPU, but it works!

That is sort of the issue when mixing good conscience with capitalism. Either the goods are valued at what we're willing to pay, or either they're valued at what we think the profit margin of the business should be, but mixing the two ultimately leads us to fall for PR crap. Business are quick to gather sympathy when the margins are low, and we fall for this PR crap, but then as soon they own a part of the market it turns into raising the price as much as they possibly can.

That being said, Amazon became what it is because Bezos was hell bent on not rug pulling customers, at least in the early years, so it is possible they would decrease prices eventually to gain market advantage, that's their whole strategy.

but the hardware will continue to improve and get cheaper.

Eh. I mean sure the likes of A100s will invariably get cheaper because they're overpriced AF, but there isn't really that much engineering going into those things hardware-wise: Accelerating massive chains of fmas is quite a smaller challenge than designing a CPU or GPU. Meanwhile moore's law is -- well maybe not dead but a zombie. In the past advances in manufacturing meant lower price per transistor, that hasn't been true for a while now and the physics of everything aren't exactly getting easier, they're now battling quantum uncertainty in the lithography process itself.

Where there might still be significant headways to be made is by switching to analogue, but, eeeh. Reliability. Neural networks are rather robust against small perturbations but it's not like digital systems can't make use of that by reducing precision, and controlling precision is way harder in analoge. Everything is harder there, it's an arcane art.

tl;dr: Don't expect large leaps, especially not multiple. This isn't a naughts "buy a PC twice as fast at half the price two years later" kind of situation, AI accelerators are silicon like any other they already make use of the progress we made back then.

Mine also finds my phone, it's primary use.

in my eyes they put themselves in an awkward position by garnering a reputation of always collecting more user data than justified, and at this point i assume they do the same with paid products as it's an industry norm. however I'm not ok with it and will never pay when the product doesn't respect privacy. the saying used to be "if you don't pay, you're the product", but it is increasingly shifting to: you're the product and also you have to pay so that our shareholders can experience more infinite growth

Yo ho, yo ho.

Bubble about to pop

You can record and edit videos on your own devices, but that doesn't mean it's suddenly free for Netflix or YouTube to stream their videos to you.

Surely a local version of Alexa could be developed, but that development would come with its own costs.
Some things simply can't be done locally, such as a web search. Often your route calculations for a map application are also done in the cloud.

Dave Limp 💀

Here is an alternative Piped link(s):

https://youtu.be/nwPtcqcqz00?si=QNG8ElAz7yZDGQUx

Piped is a privacy-respecting open-source alternative frontend to YouTube.

I'm open-source; check me out at GitHub.

I love when satire.