Hacker News Reader

Comment by blenderob | Hacker News Reader

> there were huge misses on areas like AI

Honest question. What could they have possibly gained from AI? What did they miss out on by not getting into AI?

The only thing I can think of is if they had converted Alexa into a truly useful assistant and gotten people to sign up for a premium to it while everyone was still in awe of and having fun talking to LLMs.

Even if they executed this perfectly though, I think all it would do is increase the burn rate related to Alexa. Plus there's no moat, everyone has a phone with more capable hardware that would be a better place for this assistant to live long term.

I don't think they missed too much on the AI front. Bedrock was crap during formative LLM-solution building times, maybe that was a legitimate miss of theirs? SageMaker still is crap, another miss?

Bedrock is just hosting a bunch of different LLMs. It is just as capable as the third party LLMs it hosts. It hosts everything except for OpenAI and Gemini models as far as the popular ones. Even Nova, Amazon’s models, are good enough for the most part, cheaper and usually faster.

Bedrock imposes limitations on context window sizes and output sizes over and above those of the underlying model. There's also a bunch of undifferentiated heavy lifting that Amazon requires their customers to perform when they could have added value there.

I'm curious what are the shortcomings you find with SageMaker?

Have you noticed how bad their devices dept was gutted? I can easily imagine a proper AI system backing alexa, rolled out to all existing customers, which would potentially boost the viability of that whole product line.

right now I have no idea what Alexa is for anymore. It pings me when a package arrives but so does my phone. I can't order from it, because it makes me open the app. I can manage a shopping list I guess? I ask it the temp outside from time to time b/c I don't want to go look at the thermometer outside my window?

A proper device would discuss with me product options and Q&A, or something. Even piping Grok/GPT5 through to answer questions would be light years ahead of the current crap answers I get when I ask questions. And why doesn't it do a good job of identifying songs that are playing? That's trivial software, and it could recommend some songs like this (which is something I use GPT for).

I don't know I'm just a trad software engineer, but it sure seems like it's low risk / high improvement to at least have a modern LLM feature in that product.

There is absolutely nothing in the world I would like less than an AI device listening to all of my conversations at home.

I have privacy concerns, but an AI that worked as a servant would be a great help. I can't afford servants (or slaves), but I want something that listens in and takes action when I want something. Interrupt me when I'm planning vacation to remind me that there is some other event at home when I'd be gone. When I run low on baking powder ensure I get more.

Again, the privacy concerns need to be addressed. However there is a lot of potential for an AI in my house if it works.

We called that family and community for the last few millennia. It's a shame some people want to automate that out of their lives.

It works by listening for the keyword (it's name), and only then opening a connection to servers. You can verify this by removing internet connectivity and it'll still "turn on" when you say it's name, but will be unable to respond.

It can buffer your words as text until you call it. How many kB do one speak per day? 50?

Hard agree. Same for Siri.

I don't get why they just don't do it. There are jailbreaks, making it talk trash and so on, but is that the majority of users? I doubt it.

I never use Siri except when driving. But when driving, it's really wonderful to have, since my car does not have car play.

As a daily user, it's clear they have been actively messing with Siri still, particularly in the last year. But the changes seem to mostly make it worse. It now misses "hey siri" half the time for no apparent reason. I feel like they just don't know what they want to do with it.

The Alexa product as it is, is already lighting money on fire. Adding LLM integration to it would just be pouring gasoline on it; iirc inference is still not profitable even in easier sectors so Alexa would be a particularly bad place for it. (Even though it would be amazing UX, but I doubt enough people would pay the price it costs to break even).

Am I missing something? Alexa+ with LLMs was already released this year?

https://www.aboutamazon.com/news/devices/new-alexa-generativ...

Too little too late

I use almost daily my google nest to set timers hand free while cooking, and managing radio playbacks: I’m an avid web radio listener.

AWS is probably in last place among computing clouds on AI and the AI managed services from AWS are pretty second-rate.

Amazon had Alexa (Echos) everywhere and that was teed up to knock it out of the park putting a real AI assistant in every home. Then the devices team totally blew it and did almost nothing to take advantage of that huge lead. Folks are now basically just throwing these things in the trash and the devices unit is a huge pit of cash on fire.

The list goes on…

Not sure why you’re being downvoted. This is spot on. Likely Amazon and AWS folks in denial of where they really are relative to the market’s bar at the moment.

Surely b2b "model API provider" would've been a great fit for Amazon, but others are absolutely dominating this space right now.

The silver lining for Amazon is that nobody is making a profit by actually providing that service. It's just massive valuations and VC money flowing freely instead of a sustainable business model, and Amazon plays a much longer game - they can recover even if being so far behind is a "bad look" for their current management.

It's not hard to switch between OpenAI/Anthropic/Azure/Google/whatever and Amazon can surely eventually compete well in this space.

They had the infrastructure and muscle to put a dent into Google Search and Microsoft Enterprise software... but instead they've ceeded to OpenAI and Anthropic.

They had a bridgehead device of the type that OpenAI are struggling to create with Jonny Ive in Alexa... but nothing seems to be happening with Alexa now.

If Microsoft couldn’t compete with Google in search, how could Amazon? Even if the search were better, how do you get people to switch to it if Google is the default everywhere? Besides Alexa was losing billions when it was backed by the relatively cheap to run Amazon Lex pattern matching service. It would have lost more being backed by an LLM and it actually got worse from all indications.

How much of the search market will Agents or Chatbots take?

Google clearly believe : a chunk.

People switched to GPT and Claude...

Amazon risk becoming logistics and datacenters, what's wrong with that you ask? Well, it's not where the money is. The money is in the website and the application layer.

And how much of this switch do you think will happen as long as Google is the default search engine for every major browser and every major smartphone outside of China?

Even if “agents” aren’t overhyped, why would someone switch to Amazon instead of just using Google’s offerings?

Google has reported record revenues even after the advent of LLMs and have integrated AI overviews and a separate chat tab.

Why am I calling Anthropic/Google/OpenAI APIs for my AI inference and not AWS ones? Even Azure has some decent AI inference endpoints I encounter with clients.

Never seen a single person use AWS for this.

What if I told you that Amazon has spent more money on AI this year than any other company? On track to spend nearly 150B by the end of the year?

I’d say leadership should be fired for having so little to show for it

Short term - higher stock evaluation. Long term - depends on what value AI bubble will be able to deliver. But it's not like Amazon is sitting out of this game, they do invest into their AI (Nova models), with little return to show for it.

Reply on news.ycombinator.com