Without genuine understanding voice is just a terrible method for control. I don't know what the options are or what exactly I'm supposed to say. Alexa is overly verbose too. The control feels very clumsy, though it's sometimes good enough.
I am imagining GPT-4 on the other end with a suite of controls. If Alexa could understand what I'm talking about and all of the options available to it I think it would be a much better system. I could also imagine a system that's always listening and proactively acting on my behalf without my even giving directions. If hardware advances to the point where I could run a model this size locally? That would solve privacy concerns and improve latency. How far are we from running GPT-4 on commodity hardware? 20 years?
As someone else said, Alexa seems to be very “voice command line”. I think Google is similar, no? It can be wordy but it’s easier to parse correctly (and faster).
Siri has always tried to be more conversational (though not by too much). But that’s also probably one of the reasons it hasn’t been as expandable.
Much easier to add skills (in Amazon parlance) when the command can be “tell Foo to bar the baz” instead of trying to intuit which app you want to use based on heuristics or something.
I am imagining GPT-4 on the other end with a suite of controls. If Alexa could understand what I'm talking about and all of the options available to it I think it would be a much better system. I could also imagine a system that's always listening and proactively acting on my behalf without my even giving directions. If hardware advances to the point where I could run a model this size locally? That would solve privacy concerns and improve latency. How far are we from running GPT-4 on commodity hardware? 20 years?