Hacker News Reader

Comment by divbzero | Hacker News Reader

2019 was before ChatGPT. Generative AI could change the game for digital voice assistants.

I think it's weird that they (Alexa, Google etc) haven't tried to do something new with generative AI assistants.

I think they have all been trying [1][2][3] which is why assessments about voice assistants from 2019 might not apply today.

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

[2]: https://support.apple.com/guide/iphone/use-apple-intelligenc...

[3]: https://blog.google/products/assistant/google-assistant-bard...

"They" or in this case, Apple, has. This is largely why the latest Siri AI enhancements went over so poorly. Hallucinations when trying to get important personal data or accomplish important tasks are not viewed terribly fondly.

You're giving Apple too much credit in a way that their press releases are carefully crafted to encourage you to. The on-the-ground reality is that Siri will sometimes prompt me for permission to ship her mangled speech-to-text out to a very stale ChatGPT model and read what comes back. That's not failure by hallucination, that's failure by duct taping on an extra limb and wondering why it isn't useful.

They have. You can switch your Android phone from the old Google Assistant to Gemini. I don't like it, so I use Gemini separately by going to the Gemini URL, leaving the assistant as is. It may be dumb, but it's useful. Just like you wouldn't replace your Bash with a GUI, at least not for some values of 'you'.

Google Assistant is going away soon so this won't work anymore.

https://arstechnica.com/google/2025/03/end-of-life-gemini-wi...

I wonder how that will affect older devices, like Android TV streaming boxes, where you use the assistant through the remote.

I see that the article mentions this, but I'm skeptical. I think that the assistant will just be switched off, and that will be it; no more using the older Android TV box to ask for today's weather or search for videos.

I think older and low power devices would actually be easier to port: they capture the audio request, send it to the server, receive an audio response from the server and play it. No local "intelligence" at all.

So you could probably replace the whole backend while keeping things compatible as long as you maintain that simple API.

Weird that this got downvoted, considering that Amazon is currently rolling out Alexa+.

If the complaint was that the product wasn't profitable, wouldn't adding expensive LLM inference costs potentially make the problem worse at least at the outset?

The last thing I need is another bullshit generator.

I'll believe it when I see it.

Reply on news.ycombinator.com