[Taikonerd]

I think the real problem was that the product just wasn't what people imagined it would be.

When Alexa was first announced in 2014, I thought, "OK, right now it can only do simple things like play a song on Spotify or tell you the weather. But they're going to iterate rapidly, and in 10 years we'll be talking to it like a computer from Star Trek."

The rapid progress of GPT-3 and friends seemed to confirm this.

But here we are in 2022 and... we're nowhere near Star Trek, nor even GPT-3-ish levels of conversation. There seemed to be a failure of imagination: you wanted some cool sci-fi stuff, maybe shared storytelling like AI Dungeon[0]. But you got triggers for home automation, which 90% of people don't care about.

[0] https://gpt3demo.com/apps/aidungeon-io

[ozim]

I have different point of view - 90's Start Trek stuff is here - I can tell Alexa turn on the lights and it will happen. I can tell specific name of a room I want to off/on/dim/bright that I typed in while setting up my smart lights even.

In 90's Star Trek they also don't talk with computer much more than saying very specific commands and getting some response just like one would ask Alexa for a weather. I can also setup some sensors so it would give me status report if I want to.

You take voice recognition for granted - it is not easy - it was really not possible not that long ago on level we have now like even in 2000's.

From 1990's to today - I feel that I live in the future.

If you take Doctor the hologram or holosuite characters - well yeah AI of today is nowhere near and I believe it never will be - but as I say my mind is blowing up when I see level of voice recognition, image recognition, game engines of today compared to what was possible in 1990's or even 2000's.

[clnq]

Amazon is in a great market position to implement GPT-3 (or GPT-3-like AI) in Echos. I would pay a considerable premium on these assistants if Amazon could adapt their current features to the new generation of language models in useful ways.

The wealth of knowledge that language models like these have is fantastic. Although as Meta's Galactica has demonstrated, you can't just slap an input box on a pre-trained model and ship it. It needs to be much more integrated with useful services, and robust with fact-checking. But a company like Amazon should not have problems doing both (and more) if they would commit to it.

It's so sad to read about the mismanagement of this technology in the article. They have a tremendous stake in the world of AI assistants and I wish Amazon were more optimistic about the technology. Perhaps a conversational AI assistant could drive revenues itself (without the need of complex monetization).

[ghaff]

The situation with Alexa and its ilk (as with GPT-3, Stable Diffusion, autonomous driving, etc.) seems to be yet another case where at one level the results are impressive but it's not nearly good enough to enable the really interesting use cases. So it devolves to basically short formulaic commands.