> It's amazing how on the one side of things we've got OpenAI, Llama, Gemini, Mistral, etc making mind-blowing LLMs that feel truly like the future.

> And at the same time we've got Siri, Alexa/Amazon Echo stuff, and Google Home stuff that already feels dated and is actively getting worse with each passing year.

Initially, when they were new, also Siri, Alexa and Google Home felt mindblowing and "truly like the future". My bet: in a few years these LLMs will feel similarly dated.

Concerning your "is actively getting worse with each passing year" point: do you remember these HN posts from the last months where people were complaining that ChatGPT and GPT-4 have become worse than they were initially? In other words: it's already happening. :-)

Siri/etc aren't just general disillusionment though, people often talk about features that have been removed and replaced with a web search.

LLM quality drops aren't disillusionment either - it's scale/cost optimization. And those are subscription models - Siri/Google assistant is bundled with device/services.

Coincidentally, those features were degraded so that those companies would have more compute power available to do inference for their LLMs

> Siri/etc aren't just general disillusionment though, people often talk about features that have been removed and replaced with a web search.

Recall that recently Google removed the image generation feature from Gemini after the diversity controversies. :-)

>Initially, when they were new, also Siri, Alexa and Google Home felt mindblowing and "truly like the future".

Strong disagree on that one. It was difficult to get them to do simple things like give the weather if a different location, play the right song. Even asking what time it is wasn't guaranteed to give you an answer if you didn't ask the right way. The marketting material made it sound great but the first hand experience was very underwhelming. The biggest use case at launch was making it say funny things.

At first yes, but about a year or two in was when you could do some really cool things, especially with Google. It could be astoundingly context aware, especially regarding a chain of on going queries. I have not seen that happen reliably in at least 5 years.

We are innovating faster both upwards and down!

Why has nobody glued ChatGPT or equivalent to a home smart speaker yet?

Because you need a reliable way to convert human speech into a structured representation of actions that the smart device can execute so the LLM can interface with the underlying system. For most smart devices this is probably locked away, and combined with issues like hallucination, lack of training data for the action representation, etc., it's hard to say whether such technology would be reliable enough for stable use.

We've seen a similar thing with big companies putting in LLM chatbots for customer service -- turns out LLMs can go off the rails really easily with the right prompts.

Because it’s expensive?

For some reason, I thought that was a plugin you could stuff into Mycroft.

Huh. Apparently not. I wonder why.